From 903cd19ce31c27edb7de49d5c77c09c397813de7 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 21 Apr 2023 17:37:00 -0700
Subject: [PATCH 001/194] Put back the split-k gemm code. (#684)

* simplify karg in device/grid split-k op

* fix mk_kn_mn instances

* add more instances

* use name from tensor layout

---------

Co-authored-by: carlushuang <carlus.huang@amd.com>
---
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp | 464 ++---------------
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  | 486 +++++++++++++-----
 ...l_splitk_f16_f16_f16_mk_kn_mn_instance.cpp |  27 +-
 3 files changed, 438 insertions(+), 539 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 0d2aeaeb7e..1f08cec67e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -73,157 +73,18 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     static constexpr auto I2 = Number<2>{};
     static constexpr auto I3 = Number<3>{};
 
-    static constexpr auto K1Number = Number<K1>{};
-
-    static auto
-    MakeAGridDescriptor_KBatch_K0_M_K1(index_t M, index_t K, index_t StrideA, int KBatch, int KPad)
-    {
-        assert(KPad % (K1 * KBatch) == 0);
-
-        const index_t K0 = KPad / (K1 * KBatch);
-
-        const auto a_grid_desc_m_k = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
-            a_grid_desc_m_k,
-            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-            return transform_tensor_descriptor(
-                a_grid_desc_m_kpad,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_right_pad_transform(M, PadM)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                a_grid_desc_m_kpad,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_pass_through_transform(M)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-    }
-
-    static auto
-    MakeBGridDescriptor_KBatch_K0_N_K1(index_t K, index_t N, index_t StrideB, int KBatch, int KPad)
-    {
-        assert(KPad % (K1 * KBatch) == 0);
-
-        const index_t K0 = KPad / (K1 * KBatch);
-
-        const auto b_grid_desc_k_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
-            }
-        }();
-
-        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
-            b_grid_desc_k_n,
-            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-            return transform_tensor_descriptor(
-                b_grid_desc_kpad_n,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                b_grid_desc_kpad_n,
-                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1Number)),
-                           make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
-    {
-        const auto c_grid_desc_m_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-    }
-
-    static auto GetKPad(index_t K, index_t KBatch)
-    {
-        const index_t K0   = math::integer_divide_ceil(K, K1 * K0PerBlock * KBatch) * K0PerBlock;
-        const index_t KPad = KBatch * K0 * K1;
-        return KPad;
-    }
-
-    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_KBatch_K0_M_K1(1, 1, 1, 1, 1));
-    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_KBatch_K0_N_K1(1, 1, 1, 1, 1));
-    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
-    // GridwiseGemm
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
         ADataType, // TODO: distinguish A/B datatype
         AccDataType,
         CDataType,
-        InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
+        ALayout,
+        BLayout,
+        CLayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -253,236 +114,64 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         CBlockTransferScalarPerVector_NWaveNPerXDL,
         CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
 
-    // GridwiseGemm
-    using GridwiseGemmAtomicAdd = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
-        BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
-        AccDataType,
-        CDataType,
-        InMemoryDataOperationEnum::AtomicAdd,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
-        AElementwiseOperation,
-        BElementwiseOperation,
-        CElementwiseOperation,
-        MPerBlock,
-        NPerBlock,
-        K0PerBlock,
-        MPerXDL,
-        NPerXDL,
-        K1,
-        MXdlPerWave,
-        NXdlPerWave,
-        ABlockTransferThreadClusterLengths_K0_M_K1,
-        ABlockTransferThreadClusterArrangeOrder,
-        ABlockTransferSrcAccessOrder,
-        ABlockTransferSrcVectorDim,
-        ABlockTransferSrcScalarPerVector,
-        ABlockTransferDstScalarPerVector_K1,
-        false, // AThreadTransferSrcResetCoordinateAfterRun,
-        ABlockLdsAddExtraM,
-        BBlockTransferThreadClusterLengths_K0_N_K1,
-        BBlockTransferThreadClusterArrangeOrder,
-        BBlockTransferSrcAccessOrder,
-        BBlockTransferSrcVectorDim,
-        BBlockTransferSrcScalarPerVector,
-        BBlockTransferDstScalarPerVector_K1,
-        false, // BThreadTransferSrcResetCoordinateAfterRun,
-        BBlockLdsAddExtraN,
-        CShuffleMRepeatPerShuffle,
-        CShuffleNRepeatPerShuffle,
-        CBlockTransferScalarPerVector_NWaveNPerXDL,
-        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
-
-    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
-        decltype(GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}));
-
-    using Block2CTileMap = typename GridwiseGemm::CBlockClusterAdaptor;
-
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t M,
-                 index_t N,
-                 index_t K,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 index_t M01,
-                 index_t N01,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op,
-                 index_t k_batch)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              a_grid_desc_kbatch_k0_m_k1_{},
-              b_grid_desc_kbatch_k0_n_k1_{},
-              c_grid_desc_m_n_{},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              k_batch_{k_batch}
-        {
-            int KPad = DeviceGemmXdlSplitKCShuffle::GetKPad(K, k_batch_);
-
-            a_grid_desc_kbatch_k0_m_k1_ =
-                DeviceGemmXdlSplitKCShuffle::MakeAGridDescriptor_KBatch_K0_M_K1(
-                    M, K, StrideA, k_batch_, KPad);
-            b_grid_desc_kbatch_k0_n_k1_ =
-                DeviceGemmXdlSplitKCShuffle::MakeBGridDescriptor_KBatch_K0_N_K1(
-                    K, N, StrideB, k_batch_, KPad);
-            c_grid_desc_m_n_ = DeviceGemmXdlSplitKCShuffle::MakeCGridDescriptor_M_N(M, N, StrideC);
-
-            block_2_ctile_map_ =
-                GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_kbatch_k0_m_k1_,
-                                           b_grid_desc_kbatch_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n_);
-            }
-        }
-
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        AGridDesc_K0_M_K1 a_grid_desc_kbatch_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_kbatch_k0_n_k1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock c_grid_desc_mblock_mperblock_nblock_nperblock_;
-        Block2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t k_batch_;
-    };
+    using Argument = typename GridwiseGemm::Argument;
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceGemmXdlSplitKCShuffle::Argument;
 
-        void Print(const Argument& arg)
-        {
-            std::cout << "arg.a_grid_desc_kbatch_k0_m_k1_{"
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0) << ", "
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1) << ", "
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I2) << ", "
-                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I3) << "}" << std::endl;
+        void Print(const Argument& karg) { karg.Print(); }
 
-            std::cout << "arg.b_grid_desc_kbatch_k0_n_k1_{"
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I0) << ", "
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I1) << ", "
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I2) << ", "
-                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I3) << "}" << std::endl;
-
-            std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                      << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
-        }
-
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
         {
             if(stream_config.log_level_ > 0)
             {
-                Print(arg);
+                Print(karg);
             }
 
-            const auto kbatch = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0);
+            const auto kbatch = karg.k_batch;
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
-                                            arg.b_grid_desc_kbatch_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(karg))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2 has invalid setting");
+                    "wrong! GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2 has invalid "
+                    "setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
-
-            const auto K0 = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1);
+            index_t gdx, gdy, gdz;
+            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(karg);
+            const auto K0           = karg.K0;
 
             const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
 
             float ave_time = 0;
 
             const auto Run = [&](const auto& kernel) {
-                hipGetErrorString(hipMemset(
-                    arg.p_c_grid_,
-                    0,
-                    arg.c_grid_desc_mblock_mperblock_nblock_nperblock_.GetElementSpaceSize() *
-                        sizeof(CDataType)));
+                if(kbatch > 1)
+                    hipGetErrorString(
+                        hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(CDataType)));
 
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_,
-                                           arg.p_b_grid_,
-                                           arg.p_c_grid_,
-                                           arg.a_grid_desc_kbatch_k0_m_k1_,
-                                           arg.b_grid_desc_kbatch_k0_n_k1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.block_2_ctile_map_);
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             };
 
             if(has_main_k0_block_loop)
             {
                 if(kbatch == 1)
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        true>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             true,
+                                                             InMemoryDataOperationEnum::Set>;
 
                     Run(kernel);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemmAtomicAdd,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        true>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             true,
+                                                             InMemoryDataOperationEnum::AtomicAdd>;
 
                     Run(kernel);
                 }
@@ -491,37 +180,19 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
             {
                 if(kbatch == 1)
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        false>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             false,
+                                                             InMemoryDataOperationEnum::Set>;
 
                     Run(kernel);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r4r2<
-                        GridwiseGemmAtomicAdd,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::BGridDesc_K0_N_K1>,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::
-                                               CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
-                        AElementwiseOperation,
-                        BElementwiseOperation,
-                        CElementwiseOperation,
-                        remove_reference_t<DeviceGemmXdlSplitKCShuffle::Block2CTileMap>,
-                        false>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
+                                                             false,
+                                                             InMemoryDataOperationEnum::AtomicAdd>;
 
                     Run(kernel);
                 }
@@ -544,12 +215,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         return true;
     }
 
-    static bool IsSupportedArgument(const Argument& arg)
+    static bool IsSupportedArgument(const Argument& karg)
     {
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
-                                           arg.b_grid_desc_kbatch_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(karg);
     }
 
     // polymorphic
@@ -567,9 +235,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op,
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation,
                              index_t KBatch)
     {
         return Argument{p_a,
@@ -581,11 +249,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         StrideA,
                         StrideB,
                         StrideC,
-                        1,
-                        1,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op,
+                        GridwiseGemm::CalculateMPadded(M),
+                        GridwiseGemm::CalculateNPadded(N),
+                        GridwiseGemm::CalculateKPadded(K),
+                        GridwiseGemm::CalculateK0(K, KBatch),
                         KBatch};
     }
 
@@ -601,9 +268,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op,
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation,
                                                       ck::index_t KBatch = 1) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
@@ -615,11 +282,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                           StrideA,
                                           StrideB,
                                           StrideC,
-                                          1,
-                                          1,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op,
+                                          GridwiseGemm::CalculateMPadded(M),
+                                          GridwiseGemm::CalculateNPadded(N),
+                                          GridwiseGemm::CalculateKPadded(K),
+                                          GridwiseGemm::CalculateK0(K, KBatch),
                                           KBatch);
     }
 
@@ -630,31 +296,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     }
 
     // polymorphic
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "DeviceGemmXdlSplitKCShuffle"
-            << "<"
-            << BlockSize << ", "
-            << MPerBlock << ", "
-            << NPerBlock << ", "
-            << K0PerBlock << ", "
-            << K1 << ", "
-            << MPerXDL << ", "
-            << NPerXDL << ", "
-            << MXdlPerWave << ", "
-            << NXdlPerWave << ", "
-            << ABlockTransferSrcScalarPerVector << ", "
-            << ABlockTransferDstScalarPerVector_K1 << ", "
-            << BBlockTransferSrcScalarPerVector << ", "
-            << BBlockTransferDstScalarPerVector_K1
-            << ">";
-        // clang-format on
-
-        return str.str();
-    }
+    std::string GetTypeString() const override { return GridwiseGemm::GetTypeString(); }
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 190194f1eb..727f180e96 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -18,60 +18,23 @@
 namespace ck {
 
 template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_B_K0_M_K1,
-          typename BGridDesc_B_K0_N_K1,
-          typename CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename CBlockClusterAdaptor,
-          bool HasMainKBlockLoop>
+          bool HasMainKBlockLoop,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r4r2(const FloatAB* __restrict__ p_a_grid,
-                                  const FloatAB* __restrict__ p_b_grid,
-                                  FloatC* __restrict__ p_c_grid,
-                                  const AGridDesc_B_K0_M_K1 a_b_k0_m_k1_grid_desc,
-                                  const BGridDesc_B_K0_N_K1 b_b_k0_n_k1_grid_desc,
-                                  const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
-                                      c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                  const AElementwiseOperation a_element_op,
-                                  const BElementwiseOperation b_element_op,
-                                  const CElementwiseOperation c_element_op,
-                                  const CBlockClusterAdaptor c_block_cluster_adaptor)
+        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
+    constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
 
-    __shared__ FloatAB p_shared_block[shared_block_size];
+    __shared__ uint8_t p_shared[shared_size];
 
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
-                                                  p_b_grid,
-                                                  p_c_grid,
-                                                  static_cast<void*>(p_shared_block),
-                                                  a_b_k0_m_k1_grid_desc,
-                                                  b_b_k0_n_k1_grid_desc,
-                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  c_block_cluster_adaptor);
+    GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
+        karg, static_cast<void*>(p_shared));
 #else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = a_b_k0_m_k1_grid_desc;
-    ignore = b_b_k0_n_k1_grid_desc;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = c_block_cluster_adaptor;
+    ignore = karg;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -79,13 +42,13 @@ template <index_t BlockSize,
           typename FloatAB,
           typename FloatAcc,
           typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_B_K0_M_K1,
-          typename BGridDesc_B_K0_N_K1,
-          typename CMNGridDesc,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
           index_t MPerBlock,
           index_t NPerBlock,
           index_t K0PerBlock,
@@ -126,10 +89,238 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     static constexpr auto I7 = Number<7>{};
 
     // K1 should be Number<...>
-    static constexpr auto K1 = Number<K1Value>{};
+    static constexpr auto K1  = Number<K1Value>{};
+    static constexpr auto M01 = 1;
+    static constexpr auto N01 = 1;
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    struct Argument : public ck::tensor_operation::device::BaseArgument
+    {
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t KPadded;
+        index_t K0;
+        index_t k_batch;
+
+        Argument(const FloatAB* p_a_grid_,
+                 const FloatAB* p_b_grid_,
+                 FloatC* p_c_grid_,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_,
+                 index_t MPadded_,
+                 index_t NPadded_,
+                 index_t KPadded_,
+                 index_t K0_,
+                 index_t k_batch_)
+            : p_a_grid(p_a_grid_),
+              p_b_grid(p_b_grid_),
+              p_c_grid(p_c_grid_),
+              M(M_),
+              N(N_),
+              K(K_),
+              StrideA(StrideA_),
+              StrideB(StrideB_),
+              StrideC(StrideC_),
+              MPadded(MPadded_),
+              NPadded(NPadded_),
+              KPadded(KPadded_),
+              K0(K0_),
+              k_batch(k_batch_)
+        {
+        }
+
+        void Print() const
+        {
+            std::cout << "arg {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "KP:" << KPadded << ", "
+                      << "K0:" << K0 << ", "
+                      << "KB:" << k_batch << "}" << std::endl;
+        }
+    };
+
+    __host__ __device__ static auto CalculateGridSize(const Argument& karg)
+    {
+        return std::make_tuple(math::integer_divide_ceil(karg.N, NPerBlock),
+                               math::integer_divide_ceil(karg.M, MPerBlock),
+                               karg.k_batch);
+    }
+
+    // prefer this to be called on host
+    __host__ __device__ static auto CalculateMPadded(index_t M)
+    {
+        return (M + MPerBlock - 1) / MPerBlock * MPerBlock;
+    }
+
+    __host__ __device__ static auto CalculateNPadded(index_t N)
+    {
+        return (N + NPerBlock - 1) / NPerBlock * NPerBlock;
+    }
+
+    __host__ __device__ static auto CalculateK0(index_t K, index_t K_Batch = 1)
+    {
+        // k_batch * k0 * k0_per_block * k1
+        auto K_t = K_Batch * K0PerBlock * K1;
+        return (K + K_t - 1) / K_t * K0PerBlock;
+    }
+
+    __host__ __device__ static auto CalculateKPadded(index_t K, index_t K_Batch = 1)
+    {
+        auto K0 = CalculateK0(K, K_Batch);
+        return K_Batch * K0 * K1;
+    }
+
+    __host__ __device__ static auto MakeAGridDescriptor_KBatch_K0_M_K1(index_t M,
+                                                                       index_t MPad,
+                                                                       index_t K,
+                                                                       index_t StrideA,
+                                                                       index_t KBatch,
+                                                                       index_t K0,
+                                                                       index_t KPad)
+    {
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    __host__ __device__ static auto MakeBGridDescriptor_KBatch_K0_N_K1(index_t K,
+                                                                       index_t NPad,
+                                                                       index_t N,
+                                                                       index_t StrideB,
+                                                                       index_t KBatch,
+                                                                       index_t K0,
+                                                                       index_t KPad)
+    {
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
+            b_grid_desc_k_n,
+            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, K0, K1)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    __host__ __device__ static auto
+    MakeCGridDescriptor_M_N(index_t M, index_t N, index_t MPad, index_t NPad, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(c_grid_desc_m_n,
+                                               make_tuple(make_right_pad_transform(M, MPad - M),
+                                                          make_right_pad_transform(N, NPad - N)),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
     __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
     {
         constexpr auto max_lds_align = K1;
@@ -178,45 +369,68 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                          c_block_size * sizeof(FloatC));
     }
 
-    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    template <typename Block2CTileMap>
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_B_K0_M_K1& a_b_k0_m_k1_grid_desc,
-                  const BGridDesc_B_K0_N_K1& b_b_k0_n_k1_grid_desc,
-                  const CMNGridDesc& c_m_n_grid_desc,
-                  const Block2CTileMap& block_2_ctile_map)
+    __host__ __device__ static constexpr bool CheckValidity(const Argument& karg)
     {
-        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
-                      "wrong! K1 need to be known at compile-time");
-
-        static_assert((MPerBlock % (MPerXDL * MRepeat) == 0) &&
-                          (NPerBlock % (NRepeat * NPerXDL)) == 0,
-                      "Invalid tuning param!");
-
-        const auto M      = a_b_k0_m_k1_grid_desc.GetLength(I2);
-        const auto N      = b_b_k0_n_k1_grid_desc.GetLength(I2);
-        const auto K0     = a_b_k0_m_k1_grid_desc.GetLength(I1);
-        const auto KBatch = a_b_k0_m_k1_grid_desc.GetLength(I0);
-
-        if(!(M == c_m_n_grid_desc.GetLength(I0) && N == c_m_n_grid_desc.GetLength(I1) &&
-             K0 == b_b_k0_n_k1_grid_desc.GetLength(I1) &&
-             K1 == a_b_k0_m_k1_grid_desc.GetLength(I3) &&
-             K1 == b_b_k0_n_k1_grid_desc.GetLength(I3) &&
-             KBatch == b_b_k0_n_k1_grid_desc.GetLength(I0)))
-            return false;
-
-        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K0 % K0PerBlock == 0))
-            return false;
-
-        if(!block_2_ctile_map.CheckValidity(c_m_n_grid_desc))
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
         {
-            return false;
+            if(!(karg.M % MPerBlock == 0))
+                return false;
+        }
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(karg.N % NPerBlock == 0))
+                return false;
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(karg.K % ABlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+        else
+        {
+            if(karg.M % ABlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(karg.N % BBlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+        else
+        {
+            if(karg.K % BBlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+        {
+            if(karg.N % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+                return false;
+        }
+        else
+        {
+            if(karg.M % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+                return false;
         }
 
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
         return true;
     }
 
+    __host__ __device__ static auto GetKPad(index_t K, index_t KBatch)
+    {
+        const index_t K0   = math::integer_divide_ceil(K, K1 * K0PerBlock * KBatch) * K0PerBlock;
+        const index_t KPad = KBatch * K0 * K1;
+        return KPad;
+    }
+
     __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
     {
         const bool has_main_k0_block_loop = K0 > K0PerBlock;
@@ -224,8 +438,9 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         return has_main_k0_block_loop;
     }
 
+    template <typename CGridDesc>
     __host__ __device__ static constexpr auto
-    MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(const CMNGridDesc& c_m_n_grid_desc)
+    MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc& c_m_n_grid_desc)
     {
         const auto M = c_m_n_grid_desc.GetLength(I0);
         const auto N = c_m_n_grid_desc.GetLength(I1);
@@ -242,10 +457,11 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     }
 
     // return block_id to C matrix tile idx (m0, n0) mapping
+    template <typename CGridDesc>
     __host__ __device__ static constexpr auto MakeCBlockClusterAdaptor(
-        const CMNGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
+        const CGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
     {
-        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CMNGridDesc>(
+        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc>(
             c_m_n_grid_desc, 8, KBatch);
     }
 
@@ -262,24 +478,25 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        Number<CShuffleNRepeatPerShuffle * NWave * NPerXDL>{}));
     }
 
-    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
-        decltype(MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(CMNGridDesc{}));
-    using CBlockClusterAdaptor = decltype(MakeCBlockClusterAdaptor(CMNGridDesc{}, 1, 1, 1));
-
-    template <bool HasMainKBlockLoop>
-    __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
-                               const FloatAB* __restrict__ p_b_grid,
-                               FloatC* __restrict__ p_c_grid,
-                               void* __restrict__ p_shared_block,
-                               const AGridDesc_B_K0_M_K1& a_b_k0_m_k1_grid_desc,
-                               const BGridDesc_B_K0_N_K1& b_b_k0_n_k1_grid_desc,
-                               const CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
-                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
-                               const AElementwiseOperation& a_element_op,
-                               const BElementwiseOperation& b_element_op,
-                               const CElementwiseOperation& c_element_op,
-                               const CBlockClusterAdaptor& c_block_cluster_adaptor)
+    template <bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation>
+    __device__ static void Run(const Argument& karg, void* __restrict__ p_shared_block)
     {
+        const FloatAB* p_a_grid          = karg.p_a_grid;
+        const FloatAB* p_b_grid          = karg.p_b_grid;
+        FloatC* p_c_grid                 = karg.p_c_grid;
+        const auto a_b_k0_m_k1_grid_desc = MakeAGridDescriptor_KBatch_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
+        const auto b_b_k0_n_k1_grid_desc = MakeBGridDescriptor_KBatch_K0_N_K1(
+            karg.K, karg.NPadded, karg.N, karg.StrideB, karg.k_batch, karg.K0, karg.KPadded);
+        const auto c_grid_desc_m_n =
+            MakeCGridDescriptor_M_N(karg.M, karg.N, karg.MPadded, karg.NPadded, karg.StrideC);
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
+        const AElementwiseOperation a_element_op = AElementwiseOperation{};
+        const BElementwiseOperation b_element_op = BElementwiseOperation{};
+        const CElementwiseOperation c_element_op = CElementwiseOperation{};
+
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_b_k0_m_k1_grid_desc.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
@@ -289,26 +506,16 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
         const auto K0 = a_b_k0_m_k1_grid_desc.GetLength(I1);
 
-        // divide block work by [M, N]
-        const auto block_work_idx =
-            c_block_cluster_adaptor.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
-
-        if(!c_block_cluster_adaptor.ValidCTileIndex(
-               make_tuple(block_work_idx[I1], block_work_idx[I2]),
-               make_tuple(c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
-                          c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
-        {
-            return;
-        }
-
-        const index_t k_batch_id = block_work_idx[I0];
+        const index_t block_m_id = __builtin_amdgcn_readfirstlane(blockIdx.y);
+        const index_t block_n_id = __builtin_amdgcn_readfirstlane(blockIdx.x);
+        const index_t k_batch_id = __builtin_amdgcn_readfirstlane(blockIdx.z);
 
         // HACK: this force m/n_block_data_idx_on_grid into SGPR
         const index_t m_block_data_idx_on_grid =
-            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * MPerBlock);
+            __builtin_amdgcn_readfirstlane(block_m_id * MPerBlock);
 
         const index_t n_block_data_idx_on_grid =
-            __builtin_amdgcn_readfirstlane(block_work_idx[I2] * NPerBlock);
+            __builtin_amdgcn_readfirstlane(block_n_id * NPerBlock);
 
         // lds max alignment
         constexpr auto max_lds_align = K1;
@@ -444,7 +651,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
                                                                 FloatAB,
@@ -647,7 +853,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                 {c_block_desc_mblock_mperblock_nblock_nperblock,
                  make_multi_index(0, 0, 0, 0),
                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                 make_multi_index(block_work_idx[I1], 0, block_work_idx[I2], 0),
+                 make_multi_index(block_m_id, 0, block_n_id, 0),
                  c_element_op};
 
             constexpr auto mxdlperwave_forward_step =
@@ -716,6 +922,48 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             });
         }
     }
+
+    template <typename Layout>
+    struct LStr
+    {
+        static std::string Get() { return ""; }
+    };
+
+    template <>
+    struct LStr<ck::tensor_layout::gemm::RowMajor>
+    {
+        static std::string Get() { return "R"; }
+    };
+
+    template <>
+    struct LStr<ck::tensor_layout::gemm::ColumnMajor>
+    {
+        static std::string Get() { return "C"; }
+    };
+
+    static std::string GetTypeString()
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "GemmXdlSplitKCShuffle_"
+            << getGemmSpecializationString(GemmSpec) << "_"
+            << std::string(ALayout::name)[0]
+            << std::string(BLayout::name)[0]
+            << std::string(CLayout::name)[0]
+            << "_"
+            << "B" << BlockSize << "_"
+            << "Vec" << ABlockTransferSrcScalarPerVector << "x"
+            << BBlockTransferSrcScalarPerVector << "x"
+            << CBlockTransferScalarPerVector_NWaveNPerXDL << "_"
+            << MPerBlock << "x"
+            << NPerBlock << "x"
+            << K0PerBlock << "x"
+            << K1 ;
+        // clang-format on
+
+        return str.str();
+    }
 };
 
 } // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index 9b5ff40484..c4680db831 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -26,7 +26,8 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
@@ -35,14 +36,22 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>
     // clang-format on
     >;
 

From 3eecbfb6ec231cd8012faceb8b6fbc87199db60d Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Mon, 24 Apr 2023 12:40:00 +0800
Subject: [PATCH 002/194] Revise layout of group convolution (#675)

* [What] Remove pure conv int8 instance
[Why] We will never use pure int8 conv in AI, use int8 quantization instead

* Change layout

* Share the kernel parameter

* Support more type of NHWGC for group conv

* Revise client example of conv 2d, use NHWGC layout

* Add instance to cmake

* Revise layout of group conv quantization instance

* Revise layout of external api of group conv quantization

* Revise layout of group conv quantization client example

* Fix clang format

* Add comment to describe meaning of each parameter
---
 .../grouped_conv2d_fwd.cpp                    |  74 +++-----
 ..._fwd_bias_relu_perchannel_quantization.cpp |  42 +++--
 ...2d_fwd_bias_relu_perlayer_quantization.cpp |  26 +--
 ..._fwd_bias_tanh_perchannel_quantization.cpp |  28 +--
 ...2d_fwd_bias_tanh_perlayer_quantization.cpp |  26 +--
 .../conv2d_fwd_perchannel_quantization.cpp    |  40 +++--
 .../conv2d_fwd_perlayer_quantization.cpp      |  24 +--
 ...d_bias_perchannel_quantization_example.inc |   6 +-
 ...fwd_bias_perlayer_quantization_example.inc |   6 +-
 ...2d_fwd_perchannel_quantization_example.inc |   6 +-
 ...nv2d_fwd_perlayer_quantization_example.inc |   6 +-
 .../gpu/grouped_convolution_forward.hpp       |  58 +++----
 ...n_bias_forward_perchannel_quantization.hpp |  34 ++--
 ...ion_bias_forward_perlayer_quantization.hpp |  34 ++--
 ...lution_forward_perchannel_quantization.hpp |  22 +--
 ...volution_forward_perlayer_quantization.hpp |  22 +--
 .../gpu/grouped_conv2d_fwd/CMakeLists.txt     |   6 +-
 .../device_grouped_conv2d_fwd_common.hpp      |  53 ++++++
 ..._fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp | 112 ++++--------
 ..._fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp | 116 ++++---------
 ...fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp | 104 -----------
 .../device_grouped_conv2d_fwd_dl_instance.hpp |  50 ++++++
 ...wd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp | 162 ++++--------------
 ...fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp | 160 ++++-------------
 ...fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp | 132 ++++----------
 ...wd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp | 125 --------------
 ...device_grouped_conv2d_fwd_xdl_instance.hpp | 105 ++++++++++++
 ...wd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |  66 +++++++
 ...fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp | 160 ++++-------------
 ...fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |  66 +++++++
 .../conv2d_fwd/conv2d_quantization_common.hpp |   4 +-
 ..._perchannel_quantization_int8_instance.cpp |  57 ++++--
 ...as_perlayer_quantization_int8_instance.cpp |  57 ++++--
 .../device_conv2d_dl_int8_instance.hpp        |   7 +-
 ..._perchannel_quantization_int8_instance.cpp |  38 ++--
 ...dl_perlayer_quantization_int8_instance.cpp |  38 ++--
 ..._perchannel_quantization_int8_instance.cpp |  57 ++++--
 ...as_perlayer_quantization_int8_instance.cpp |  57 ++++--
 .../device_conv2d_xdl_int8_instance.hpp       |  39 +++--
 ..._perchannel_quantization_int8_instance.cpp |  38 ++--
 ...dl_perlayer_quantization_int8_instance.cpp |  38 ++--
 41 files changed, 1079 insertions(+), 1222 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp

diff --git a/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
index ece6e30c56..0a798be270 100644
--- a/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
+++ b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
@@ -17,22 +17,22 @@ using InDataType  = ck::half_t;
 using WeiDataType = ck::half_t;
 using OutDataType = ck::half_t;
 
-using InLayout    = ck::tensor_layout::convolution::GNHWC;
+using InLayout    = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout   = ck::tensor_layout::convolution::GKYXC;
-using OutLayout   = ck::tensor_layout::convolution::GNHWK;
+using OutLayout   = ck::tensor_layout::convolution::NHWGK;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr ck::index_t NumDimSpatial = 2;
 static constexpr ck::index_t G             = 32;
-static constexpr ck::index_t N             = 256;
-static constexpr ck::index_t K             = 192;
-static constexpr ck::index_t C             = 192;
-static constexpr ck::index_t Y             = 3;
-static constexpr ck::index_t X             = 3;
-static constexpr ck::index_t Hi            = 28;
-static constexpr ck::index_t Wi            = 28;
-static constexpr ck::index_t Ho            = 28;
-static constexpr ck::index_t Wo            = 28;
+static constexpr ck::index_t N             = 256; // batch size
+static constexpr ck::index_t K             = 64;  // output channel
+static constexpr ck::index_t C             = 32;  // input channel (per group)
+static constexpr ck::index_t Y             = 3;   // filter H
+static constexpr ck::index_t X             = 3;   // filter W
+static constexpr ck::index_t Hi            = 28;  // input H
+static constexpr ck::index_t Wi            = 28;  // input W
+static constexpr ck::index_t Ho            = 28;  // output H
+static constexpr ck::index_t Wo            = 28;  // output W
 
 struct SimpleDeviceMem
 {
@@ -52,50 +52,24 @@ struct SimpleDeviceMem
 
 int main()
 {
-    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, Hi, Wi, C};
-    std::array<ck::index_t, NumDimSpatial + 3> in_strides{0, 0, 0, 0, 1};
-
-    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, Y, X, C};
-    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{0, 0, 0, 0, 1};
-
-    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, Ho, Wo, K};
-    std::array<ck::index_t, NumDimSpatial + 3> out_strides{0, 0, 0, 0, 1};
-
-    std::partial_sum(rbegin(in_lengths),
-                     std::prev(rend(in_lengths)),
-                     std::next(rbegin(in_strides)),
-                     std::multiplies<>{});
-    std::partial_sum(rbegin(wei_lengths),
-                     std::prev(rend(wei_lengths)),
-                     std::next(rbegin(wei_strides)),
-                     std::multiplies<>{});
-    std::partial_sum(rbegin(out_lengths),
-                     std::prev(rend(out_lengths)),
-                     std::next(rbegin(out_strides)),
-                     std::multiplies<>{});
-
-    // transpose GNHWC/GKYXC/GNHWK to GNCHW/GKCYX/GNCHW
-    std::rotate(
-        rbegin(in_lengths), std::next(rbegin(in_lengths)), std::next(rbegin(in_lengths), 3));
-    std::rotate(
-        rbegin(in_strides), std::next(rbegin(in_strides)), std::next(rbegin(in_strides), 3));
-    std::rotate(
-        rbegin(wei_lengths), std::next(rbegin(wei_lengths)), std::next(rbegin(wei_lengths), 3));
-    std::rotate(
-        rbegin(wei_strides), std::next(rbegin(wei_strides)), std::next(rbegin(wei_strides), 3));
-    std::rotate(
-        rbegin(out_lengths), std::next(rbegin(out_lengths)), std::next(rbegin(out_lengths), 3));
-    std::rotate(
-        rbegin(out_strides), std::next(rbegin(out_strides)), std::next(rbegin(out_strides), 3));
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
+    std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 5> wei_lengths{G, K, C, Y, X};
+    std::array<ck::index_t, 5> wei_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
+    std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
 
     std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
     std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
     std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
     std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * G * N * Hi * Wi * C);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
     SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Ho * Wo * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
                                                                                  InLayout,
@@ -155,9 +129,9 @@ int main()
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
             std::size_t flop      = std::size_t(2) * G * N * K * C * Ho * Wo * Y * X;
-            std::size_t num_bytes = sizeof(InDataType) * G * N * Hi * Wi * C +
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
                                     sizeof(WeiDataType) * G * K * Y * X * C +
-                                    sizeof(OutDataType) * G * N * Ho * Wo * K;
+                                    sizeof(OutDataType) * N * Ho * Wo * G * K;
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
diff --git a/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
index a10dd3e006..43a4779f5f 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
@@ -17,26 +17,26 @@ using BiasDataType         = int32_t;
 using RequantScaleDataType = float;
 using OutDataType          = int8_t;
 
-using InLayout           = ck::tensor_layout::convolution::GNHWC;
+using InLayout           = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
 using BiasLayout         = ck::tensor_layout::convolution::G_K;
 using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 using PassThrough        = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp       = ck::tensor_operation::element_wise::Relu;
 using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul2_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
-static constexpr ck::index_t N             = 4;   // batch size
-static constexpr ck::index_t K             = 64;  // output channel
-static constexpr ck::index_t C             = 192; // input channel
-static constexpr ck::index_t Y             = 3;   // filter H
-static constexpr ck::index_t X             = 3;   // filter W
-static constexpr ck::index_t Hi            = 71;  // input H
-static constexpr ck::index_t Wi            = 71;  // input W
-static constexpr ck::index_t Ho            = 36;  // output H
-static constexpr ck::index_t Wo            = 36;  // output W
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;  // batch size
+static constexpr ck::index_t K             = 32; // output channel
+static constexpr ck::index_t C             = 64; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 71; // input H
+static constexpr ck::index_t Wi            = 71; // input W
+static constexpr ck::index_t Ho            = 36; // output H
+static constexpr ck::index_t Wo            = 36; // output W
 struct SimpleDeviceMem
 {
     SimpleDeviceMem() = delete;
@@ -55,8 +55,11 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
@@ -64,17 +67,18 @@ int main(int argc, char* argv[])
     std::array<ck::index_t, 5> requant_scale_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> requant_scale_strides{K, 0, 1, 0, 0};
     std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * K, Ho * Wo * K, 1, Wo * K, K};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem bias(sizeof(BiasDataType) * K * Y * X * C);
-    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * K);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
         NumDimSpatial,
diff --git a/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
index b8e6a493ef..2ff91fe966 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
@@ -16,19 +16,19 @@ using WeiDataType  = int8_t;
 using BiasDataType = int32_t;
 using OutDataType  = int8_t;
 
-using InLayout     = ck::tensor_layout::convolution::GNHWC;
+using InLayout     = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout    = ck::tensor_layout::convolution::GKYXC;
 using BiasLayout   = ck::tensor_layout::convolution::G_K;
-using OutLayout    = ck::tensor_layout::convolution::GNHWK;
+using OutLayout    = ck::tensor_layout::convolution::NHWGK;
 using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp = ck::tensor_operation::element_wise::Relu;
 using OutElementOp = ck::tensor_operation::element_wise::Add_Activation_Mul_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t G             = 4;
 static constexpr ck::index_t N             = 4;    // batch size
-static constexpr ck::index_t K             = 64;   // output channel
-static constexpr ck::index_t C             = 192;  // input channel
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
 static constexpr ck::index_t Y             = 3;    // filter H
 static constexpr ck::index_t X             = 3;    // filter W
 static constexpr ck::index_t Hi            = 71;   // input H
@@ -55,23 +55,27 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> bias_strides{K, 0, 1, 0, 0};
     std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * K, Ho * Wo * K, 1, Wo * K, K};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem bias(sizeof(BiasDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp =
         ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
diff --git a/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
index a0e1865d32..6ea5dd2231 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
@@ -17,21 +17,21 @@ using BiasDataType         = int32_t;
 using RequantScaleDataType = float;
 using OutDataType          = int8_t;
 
-using InLayout           = ck::tensor_layout::convolution::GNHWC;
+using InLayout           = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
 using BiasLayout         = ck::tensor_layout::convolution::G_K;
 using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 using PassThrough        = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp       = ck::tensor_operation::element_wise::TanH;
 using OutElementOp =
     ck::tensor_operation::element_wise::Add_Mul2_Activation_Mul_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t G             = 4;
 static constexpr ck::index_t N             = 4;    // batch size
-static constexpr ck::index_t K             = 64;   // output channel
-static constexpr ck::index_t C             = 192;  // input channel
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
 static constexpr ck::index_t Y             = 3;    // filter H
 static constexpr ck::index_t X             = 3;    // filter W
 static constexpr ck::index_t Hi            = 71;   // input H
@@ -58,8 +58,11 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
@@ -67,17 +70,18 @@ int main(int argc, char* argv[])
     std::array<ck::index_t, 5> requant_scale_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> requant_scale_strides{K, 0, 1, 0, 0};
     std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * K, Ho * Wo * K, 1, Wo * K, K};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem bias(sizeof(BiasDataType) * K * Y * X * C);
-    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * K);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
         NumDimSpatial,
diff --git a/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
index 7637f5c785..33407c9a1c 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
@@ -16,19 +16,19 @@ using WeiDataType  = int8_t;
 using BiasDataType = int32_t;
 using OutDataType  = int8_t;
 
-using InLayout     = ck::tensor_layout::convolution::GNHWC;
+using InLayout     = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout    = ck::tensor_layout::convolution::GKYXC;
 using BiasLayout   = ck::tensor_layout::convolution::G_K;
-using OutLayout    = ck::tensor_layout::convolution::GNHWK;
+using OutLayout    = ck::tensor_layout::convolution::NHWGK;
 using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp = ck::tensor_operation::element_wise::TanH;
 using OutElementOp = ck::tensor_operation::element_wise::Add_Mul_Activation_Mul_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t G             = 4;
 static constexpr ck::index_t N             = 4;    // batch size
-static constexpr ck::index_t K             = 64;   // output channel
-static constexpr ck::index_t C             = 192;  // input channel
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
 static constexpr ck::index_t Y             = 3;    // filter H
 static constexpr ck::index_t X             = 3;    // filter W
 static constexpr ck::index_t Hi            = 71;   // input H
@@ -56,23 +56,27 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> bias_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> bias_strides{K, 0, 1, 0, 0};
     std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * K, Ho * Wo * K, 1, Wo * K, K};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem bias(sizeof(BiasDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem bias(sizeof(BiasDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp =
         ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
diff --git a/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
index 6439c22e7f..6f51124707 100644
--- a/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
@@ -16,25 +16,25 @@ using WeiDataType          = int8_t;
 using RequantScaleDataType = float;
 using OutDataType          = int8_t;
 
-using InLayout           = ck::tensor_layout::convolution::GNHWC;
+using InLayout           = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
 using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 using PassThrough        = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp       = PassThrough;
 using OutElementOp       = ck::tensor_operation::element_wise::Activation_Mul2_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
-static constexpr ck::index_t N             = 4;   // batch size
-static constexpr ck::index_t K             = 64;  // output channel
-static constexpr ck::index_t C             = 192; // input channel
-static constexpr ck::index_t Y             = 3;   // filter H
-static constexpr ck::index_t X             = 3;   // filter W
-static constexpr ck::index_t Hi            = 71;  // input H
-static constexpr ck::index_t Wi            = 71;  // input W
-static constexpr ck::index_t Ho            = 36;  // output H
-static constexpr ck::index_t Wo            = 36;  // output W
+static constexpr ck::index_t G             = 4;
+static constexpr ck::index_t N             = 4;  // batch size
+static constexpr ck::index_t K             = 32; // output channel
+static constexpr ck::index_t C             = 64; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 71; // input H
+static constexpr ck::index_t Wi            = 71; // input W
+static constexpr ck::index_t Ho            = 36; // output H
+static constexpr ck::index_t Wo            = 36; // output W
 
 struct SimpleDeviceMem
 {
@@ -54,23 +54,27 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> requant_scale_lengths{G, N, K, Ho, Wo};
     std::array<ck::index_t, 5> requant_scale_strides{K, 0, 1, 0, 0};
     std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * K, Ho * Wo * K, 1, Wo * K, K};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * K);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem requant_scale(sizeof(RequantScaleDataType) * G * K);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp =
         ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
diff --git a/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
index f7c46a95fe..6a11f9fc26 100644
--- a/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
@@ -15,18 +15,18 @@ using InDataType  = int8_t;
 using WeiDataType = int8_t;
 using OutDataType = int8_t;
 
-using InLayout     = ck::tensor_layout::convolution::GNHWC;
+using InLayout     = ck::tensor_layout::convolution::NHWGC;
 using WeiLayout    = ck::tensor_layout::convolution::GKYXC;
-using OutLayout    = ck::tensor_layout::convolution::GNHWK;
+using OutLayout    = ck::tensor_layout::convolution::NHWGK;
 using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
 using ActivationOp = PassThrough;
 using OutElementOp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<ActivationOp>;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t G             = 4;
 static constexpr ck::index_t N             = 4;    // batch size
-static constexpr ck::index_t K             = 64;   // output channel
-static constexpr ck::index_t C             = 192;  // input channel
+static constexpr ck::index_t K             = 32;   // output channel
+static constexpr ck::index_t C             = 64;   // input channel (per group)
 static constexpr ck::index_t Y             = 3;    // filter H
 static constexpr ck::index_t X             = 3;    // filter W
 static constexpr ck::index_t Hi            = 71;   // input H
@@ -53,20 +53,24 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
+    // We have NHWGC/GKYXC/NHWGK (x, weight, y) in memory space
+    // However, CK's API only accept length and stride with order of GNCHW/GKCYX/GNCHW
+    // Hence, we need to adjust the order of stride
     std::array<ck::index_t, 5> in_lengths{G, N, C, Hi, Wi};
-    std::array<ck::index_t, 5> in_strides{N * Hi * Wi * C, Hi * Wi * C, 1, Wi * C, C};
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
     std::array<ck::index_t, 5> weight_lengths{G, K, C, Y, X};
     std::array<ck::index_t, 5> weight_strides{K * Y * X * C, Y * X * C, 1, X * C, C};
     std::array<ck::index_t, 5> out_lengths{G, N, K, Ho, Wo};
-    std::array<ck::index_t, 5> out_strides{N * Ho * Wo * K, Ho * Wo * K, 1, Wo * K, K};
+    std::array<ck::index_t, 5> out_strides{C, Ho * Wo * G * C, 1, Wo * G * C, G * C};
+
     std::array<ck::index_t, 2> in_left_pad{1, 1};
     std::array<ck::index_t, 2> in_right_pad{1, 1};
     std::array<ck::index_t, 2> conv_strides{2, 2};
     std::array<ck::index_t, 2> conv_dilations{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
-    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * G * K);
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<NumDimSpatial,
                                                                                  InLayout,
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
index 1587c614da..5675db77fa 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
@@ -190,11 +190,11 @@ int run_conv2d_fwd_bias_perchannel_quantization_example(const OutElementOp& out_
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
 
-    using InLayout           = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
+    using InLayout           = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout          = ck::tensor_layout::convolution::KYXGC;
     using BiasLayout         = ck::tensor_layout::convolution::G_K;
     using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-    using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+    using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
index 455e0804d4..9fd19c1c4e 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
@@ -178,10 +178,10 @@ int run_conv2d_fwd_bias_perlayer_quantization_example(const OutElementOp& out_el
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
 
-    using InLayout   = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout  = ck::tensor_layout::convolution::GKYXC;
+    using InLayout   = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout  = ck::tensor_layout::convolution::KYXGC;
     using BiasLayout = ck::tensor_layout::convolution::G_K;
-    using OutLayout  = ck::tensor_layout::convolution::GNHWK;
+    using OutLayout  = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
index 8e75c27746..cacedfdadd 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
@@ -180,10 +180,10 @@ int run_conv2d_fwd_perchannel_quantization_example(const OutElementOp& out_eleme
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
 
-    using InLayout           = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout          = ck::tensor_layout::convolution::GKYXC;
+    using InLayout           = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout          = ck::tensor_layout::convolution::KYXGC;
     using RequantScaleLayout = ck::tensor_layout::convolution::G_K;
-    using OutLayout          = ck::tensor_layout::convolution::GNHWK;
+    using OutLayout          = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
index 926c033c58..77332cb6d9 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
@@ -162,9 +162,9 @@ int run_conv2d_fwd_perlayer_quantization_example(const OutElementOp& out_element
     const auto in_element_op  = InElementOp{};
     const auto wei_element_op = WeiElementOp{};
 
-    using InLayout  = ck::tensor_layout::convolution::GNHWC;
-    using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-    using OutLayout = ck::tensor_layout::convolution::GNHWK;
+    using InLayout  = ck::tensor_layout::convolution::NHWGC;
+    using WeiLayout = ck::tensor_layout::convolution::KYXGC;
+    using OutLayout = ck::tensor_layout::convolution::NHWGK;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index a8df7f0d5b..175932e632 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -117,20 +117,6 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -159,20 +145,21 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances(
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
+                                                              NHWGK,
+                                                              BF16,
+                                                              BF16,
                                                               Empty_Tuple,
-                                                              int8_t,
+                                                              BF16,
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-// grouped conv2d forward, NHWGC/GKYXC/NHWGK
+
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -187,6 +174,20 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
 // grouped conv3d forward, GNDHWC/GKZYXC/GNDHWK
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -385,12 +386,6 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
-            {
-                add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(op_ptrs);
-                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances(op_ptrs);
-            }
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
@@ -398,7 +393,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
-                // no instance
+                add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
@@ -409,12 +404,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                               is_same_v<WeiDataType, ck::bhalf_t> &&
                               is_same_v<OutDataType, ck::bhalf_t>)
             {
-                // no instance
-            }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
-            {
-                // no instance
+                add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
             }
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
index 793dc8d04a..daec48050c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
@@ -17,14 +17,14 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_F32_Tuple,
@@ -36,10 +36,10 @@ void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -52,10 +52,10 @@ void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
 void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_F32_Tuple,
@@ -68,10 +68,10 @@ void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
 void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_F32_Tuple,
@@ -83,10 +83,10 @@ void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -99,10 +99,10 @@ void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
 void add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_F32_Tuple,
@@ -154,9 +154,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<DsDataType, I32_F32_Tuple> && is_same_v<OutDataType, int8_t>)
@@ -220,9 +220,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<DsDataType, I32_F32_Tuple> && is_same_v<OutDataType, int8_t>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
index c570f76750..b7d81021e2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
@@ -17,14 +17,14 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_Tuple,
@@ -36,10 +36,10 @@ void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -51,10 +51,10 @@ void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -67,10 +67,10 @@ void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
 void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                      GNHWC,
+                                                      NHWGC,
                                                       GKYXC,
                                                       GK_Tuple,
-                                                      GNHWK,
+                                                      NHWGK,
                                                       int8_t,
                                                       int8_t,
                                                       I32_Tuple,
@@ -82,10 +82,10 @@ void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -97,10 +97,10 @@ void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -152,9 +152,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<DsDataType, I32_Tuple> && is_same_v<OutDataType, int8_t>)
@@ -218,9 +218,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<DsDataType, I32_Tuple> && is_same_v<OutDataType, int8_t>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
index 089343fe6f..2d54879ea6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
@@ -17,13 +17,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -35,10 +35,10 @@ void add_device_conv2d_dl_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -50,10 +50,10 @@ void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -65,10 +65,10 @@ void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -119,9 +119,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                      is_same_v<WeiLayout, GKYXC> && is_same_v<DsLayout, GK_Tuple> &&
-                     is_same_v<OutLayout, GNHWK>)
+                     is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
index e570027eb5..f278cfa224 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
@@ -17,13 +17,13 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// grouped conv2d forward, GNHWC/GKYXC/GNHWK
+// grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -35,10 +35,10 @@ void add_device_conv2d_dl_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -50,10 +50,10 @@ void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -65,10 +65,10 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -117,8 +117,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
         {
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
index 5ef1b6866e..a36e1b47ca 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
@@ -3,11 +3,11 @@ add_instance_library(device_grouped_conv2d_fwd_instance
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
    # NHWGC, GKYXC, NHWGK
+   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
    device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-   #dl 
+   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+   #dl
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
new file mode 100644
index 0000000000..b4de825fb6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
@@ -0,0 +1,53 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index fc18b3c735..f7e575df2b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,100 +1,54 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "device_grouped_conv2d_fwd_dl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using InDataType  = ck::half_t;
-using WeiDataType = ck::half_t;
-using AccDataType = float;
-using OutDataType = ck::half_t;
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-using InLayout  = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-using OutLayout = ck::tensor_layout::convolution::GNHWK;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto Filter1x1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto Filter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Pad0_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,  Filter1x1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Stride1Pad0_instances =
-    std::tuple<
-        // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|          Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|              Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation|       Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |                     |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp, Filter1x1Stride1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              InLayout,
-                                                              WeiLayout,
+                                                              GNHWC,
+                                                              GKYXC,
                                                               Empty_Tuple,
-                                                              OutLayout,
-                                                              InDataType,
-                                                              WeiDataType,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
                                                               Empty_Tuple,
-                                                              OutDataType,
-                                                              InElementOp,
-                                                              WeiElementOp,
-                                                              OutElementOp>>>& instances)
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances{});
+                                   device_grouped_conv2d_fwd_dl_f16_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
 
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
 
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_Filter1x1Stride1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 648b39637e..85300b4e44 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,104 +1,54 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "device_grouped_conv2d_fwd_dl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using InDataType  = float;
-using WeiDataType = float;
-using AccDataType = float;
-using OutDataType = float;
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-using InLayout  = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-using OutLayout = ck::tensor_layout::convolution::GNHWK;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto Filter1x1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto Filter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances = std::tuple<
-    // clang-format off
-           // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Pad0_instances = std::tuple<
-    // clang-format off
-           // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,  Filter1x1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Stride1Pad0_instances =
-    std::tuple<
-        // clang-format off
-           // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|          Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|              Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation|       Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |                     |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp, Filter1x1Stride1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              InLayout,
-                                                              WeiLayout,
+                                                              GNHWC,
+                                                              GKYXC,
                                                               Empty_Tuple,
-                                                              OutLayout,
-                                                              InDataType,
-                                                              WeiDataType,
+                                                              GNHWK,
+                                                              F32,
+                                                              F32,
                                                               Empty_Tuple,
-                                                              OutDataType,
-                                                              InElementOp,
-                                                              WeiElementOp,
-                                                              OutElementOp>>>& instances)
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances{});
+                                   device_grouped_conv2d_fwd_dl_f32_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
 
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
 
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_Filter1x1Stride1Pad0_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
deleted file mode 100644
index 1cb5d06999..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
+++ /dev/null
@@ -1,104 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using InDataType  = int8_t;
-using WeiDataType = int8_t;
-using AccDataType = int32_t;
-using OutDataType = int8_t;
-
-using Empty_Tuple = ck::Tuple<>;
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-using InLayout  = ck::tensor_layout::convolution::GNHWC;
-using WeiLayout = ck::tensor_layout::convolution::GKYXC;
-using OutLayout = ck::tensor_layout::convolution::GNHWK;
-
-static constexpr auto ConvSpec =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-static constexpr auto Filter1x1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-static constexpr auto Filter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmPadingSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,       ConvSpec,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Pad0_instances = std::tuple<
-    // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp,  Filter1x1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-    // clang-format on
-    >;
-
-using device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Stride1Pad0_instances =
-    std::tuple<
-        // clang-format off
-           // ########################################|        NDim|     InData|     WeiData|    MultpleD|     OutData|     AccData| InLayout| WeiLayout|   MultipleD| OutLayout|           In|           Wei|           Out|          Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-           // ########################################|     Spatial|       Type|        Type|        Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|              Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-           // ########################################|            |           |            |            |            |            |         |          |            |          |    Operation|     Operation|     Operation|       Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-           // ########################################|            |           |            |            |            |            |         |          |            |          |             |              |              |                     |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2, InDataType, WeiDataType, Empty_Tuple, OutDataType, AccDataType, InLayout, WeiLayout, Empty_Tuple, OutLayout,  InElementOp,  WeiElementOp,  OutElementOp, Filter1x1Stride1Pad0,    GemmPadingSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
-        // clang-format on
-        >;
-
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              InLayout,
-                                                              WeiLayout,
-                                                              Empty_Tuple,
-                                                              OutLayout,
-                                                              InDataType,
-                                                              WeiDataType,
-                                                              Empty_Tuple,
-                                                              OutDataType,
-                                                              InElementOp,
-                                                              WeiElementOp,
-                                                              OutElementOp>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_instances{});
-
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Pad0_instances{});
-
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_int8_Filter1x1Stride1Pad0_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
new file mode 100644
index 0000000000..bcda22006b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "device_grouped_conv2d_fwd_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
+    // clang-format off
+           // ########################################|        NDim| InData| WeiData|    MultpleD| OutData| AccData| InLayout| WeiLayout| MultipleD| OutLayout|          In|          Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+           // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+           // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+           // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
+    // clang-format on
+    >;
+
+template <typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_dl_f32_instances = std::tuple<
+    // clang-format off
+        // clang-format off
+           // ########################################|        NDim| InData| WeiData|    MultpleD| OutData| AccData| InLayout| WeiLayout| MultipleD| OutLayout|          In|          Wei|           Out|    Convolution|              GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+           // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+           // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+           // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index 29f3310314..40593a0efb 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -1,137 +1,14 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using BF16 = ck::bhalf_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // OddC
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,    64,    32,   8,   8,   32,   32,    4,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    1,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -146,8 +23,41 @@ void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 6a4a3d2a45..7088028bf5 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,137 +1,14 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // OddC
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,    64,    32,   8,   8,   32,   32,    4,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    1,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -147,7 +24,40 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances{});
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 1fec35fd9f..919274c503 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,109 +1,14 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -119,7 +24,40 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances{});
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
deleted file mode 100644
index 59b0121340..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instance.cpp
+++ /dev/null
@@ -1,125 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-using device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances = std::tuple<
-    // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-                                                                                                                                                                                                                
-        // Filter1x1Stride1Pad0                                                                                                                                                                                 
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  GNHWC,  GKYXC, Empty_Tuple,  GNHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
-void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_int8_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
new file mode 100644
index 0000000000..2858671ee9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
@@ -0,0 +1,105 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "device_grouped_conv2d_fwd_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_xdl_f16_instances =
+    std::tuple<
+        // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_xdl_bf16_instances =
+    std::tuple<
+        // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        // clang-format on
+        >;
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_xdl_f32_instances =
+    std::tuple<
+        // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+        // clang-format on
+        >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..25caf61df1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              BF16,
+                                                              BF16,
+                                                              Empty_Tuple,
+                                                              BF16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                NHWGK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
index 8aca730433..b997cfb672 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -1,137 +1,14 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NHWGC = ck::tensor_layout::convolution::NHWGC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using NHWGK = ck::tensor_layout::convolution::NHWGK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
-using device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // OddC
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  8, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  4, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4,  2, 8>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   256,   256,    64,    32,   8,   8,   32,   32,    4,    1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 32, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2,  NHWGC,  GKYXC, Empty_Tuple,  NHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    ConvFwdOddC, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    1,    2,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<2, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -147,7 +24,40 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances{});
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..3256a2a826
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               NHWGK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
index b231f8c956..672cdba65d 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
@@ -19,9 +19,9 @@ using Empty_Tuple = ck::Tuple<>;
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
-using GNHWC       = ck::tensor_layout::convolution::GNHWC;
+using NHWGC       = ck::tensor_layout::convolution::NHWGC;
 using GKYXC       = ck::tensor_layout::convolution::GKYXC;
-using GNHWK       = ck::tensor_layout::convolution::GNHWK;
+using NHWGK       = ck::tensor_layout::convolution::NHWGK;
 using GK          = ck::tensor_layout::convolution::G_K;
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 using Relu        = ck::tensor_operation::element_wise::Relu;
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
index ae5c1d7c32..d4b5484d8b 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -23,19 +23,28 @@ void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
 {
     // dl
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Mul2_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Mul2_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Mul2_Clamp,
                                                                            ConvFwd1x1S1P0,
@@ -44,10 +53,10 @@ void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -58,19 +67,28 @@ void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
 {
     // dl
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Relu_Mul2_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Relu_Mul2_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Relu_Mul2_Clamp,
                                                                            ConvFwd1x1S1P0,
@@ -79,10 +97,10 @@ void add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -93,19 +111,28 @@ void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
 {
     // dl
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Mul2_TanH_Mul_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Mul2_TanH_Mul_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_F32_Tuple,
                                                                            Add_Mul2_TanH_Mul_Clamp,
                                                                            ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
index 192d5c9a55..7db4b8d869 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
                                                               Add_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Mul_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Mul_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Mul_Clamp,
                                                                            ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -56,21 +65,30 @@ void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
                                                               Add_Relu_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Relu_Mul_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Relu_Mul_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Relu_Mul_Clamp,
                                                                            ConvFwd1x1S1P0,
@@ -79,10 +97,10 @@ void add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -92,21 +110,30 @@ void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
                                                               Add_Mul_TanH_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Mul_TanH_Mul_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Mul_TanH_Mul_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            I32_Tuple,
                                                                            Add_Mul_TanH_Mul_Clamp,
                                                                            ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
index 3c4987f159..7eefbe0381 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
@@ -12,7 +12,10 @@ namespace device {
 namespace instance {
 
 // clang-format off
-template <typename DsLayout,
+template <typename InLayout,
+          typename WeiLayout,
+          typename DsLayout,
+          typename OutLayout,
           typename DsDatatype,
           typename OutElementOp,
           ConvolutionForwardSpecialization ConvSpec,
@@ -23,7 +26,7 @@ using device_grouped_conv2d_dl_int8_instances =
         // ###########################################|     Spatial|   Type|    Type|       Type|        Type|        Type|         |          |      Layout|          |  Elementwise|   Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // ###########################################|            |       |        |           |            |            |         |          |            |          |    Operation|     Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
         // ###########################################|            |       |        |           |            |            |         |          |            |          |             |              |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< NDimSpatial, int8_t,  int8_t, DsDatatype,      int8_t,     int32_t,    GNHWC,     GKYXC,    DsLayout,     GNHWK,  PassThrough,   PassThrough,  OutElementOp,       ConvSpec,          GemmSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5, DstScalarPerVector>
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK< NDimSpatial, int8_t,  int8_t, DsDatatype,      int8_t,     int32_t, InLayout, WeiLayout,    DsLayout, OutLayout,  PassThrough,   PassThrough,  OutElementOp,       ConvSpec,          GemmSpec,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5, DstScalarPerVector>
     >;
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
index d45c1c1ee7..c8f5f7042c 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_dl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_dl_perchannel_quantization_int8_instances(
                                                               Mul2_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            F32_Tuple,
                                                                            Mul2_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            F32_Tuple,
                                                                            Mul2_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            F32_Tuple,
                                                                            Mul2_Clamp,
                                                                            ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_dl_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -56,19 +65,28 @@ void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
                                                               Relu_Mul2_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            F32_Tuple,
                                                                            Relu_Mul2_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            F32_Tuple,
                                                                            Relu_Mul2_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           GK_Tuple,
+                                                                           NHWGK,
                                                                            F32_Tuple,
                                                                            Relu_Mul2_Clamp,
                                                                            ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
index f4b9479502..d7f7384ff7 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_dl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_dl_perlayer_quantization_int8_instances(
                                                               Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
                                                                            Empty_Tuple,
                                                                            Mul_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
                                                                            Empty_Tuple,
                                                                            Mul_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
                                                                            Empty_Tuple,
                                                                            Mul_Clamp,
                                                                            ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_dl_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -56,19 +65,28 @@ void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
                                                               Relu_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
                                                                            Empty_Tuple,
                                                                            Relu_Mul_Clamp,
                                                                            ConvFwdDefault,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
                                                                            Empty_Tuple,
                                                                            Relu_Mul_Clamp,
                                                                            ConvFwd1x1P0,
                                                                            4>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_dl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_dl_int8_instances<NHWGC,
+                                                                           GKYXC,
+                                                                           Empty_Tuple,
+                                                                           NHWGK,
                                                                            Empty_Tuple,
                                                                            Relu_Mul_Clamp,
                                                                            ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
index b6e8ee1590..658aa8370a 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
                                                               Add_Mul2_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Mul2_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Mul2_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Mul2_Clamp,
                                                                             ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -56,19 +65,28 @@ void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
                                                               Add_Relu_Mul2_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Relu_Mul2_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Relu_Mul2_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Relu_Mul2_Clamp,
                                                                             ConvFwd1x1S1P0,
@@ -77,10 +95,10 @@ void add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_F32_Tuple,
@@ -90,19 +108,28 @@ void add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(
                                                               Add_Mul2_TanH_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Mul2_TanH_Mul_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Mul2_TanH_Mul_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_F32_Tuple,
                                                                             Add_Mul2_TanH_Mul_Clamp,
                                                                             ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
index 70f92cec3a..7102e9b259 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
                                                               Add_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Mul_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Mul_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Mul_Clamp,
                                                                             ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -56,21 +65,30 @@ void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
                                                               Add_Relu_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Relu_Mul_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Relu_Mul_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Relu_Mul_Clamp,
                                                                             ConvFwd1x1S1P0,
@@ -79,10 +97,10 @@ void add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               I32_Tuple,
@@ -92,21 +110,30 @@ void add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(
                                                               Add_Mul_TanH_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Mul_TanH_Mul_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Mul_TanH_Mul_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             I32_Tuple,
                                                                             Add_Mul_TanH_Mul_Clamp,
                                                                             ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
index 262ec06b7f..90f8791aa9 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
@@ -12,30 +12,33 @@ namespace device {
 namespace instance {
 
 // clang-format off
-template <typename DsLayout,
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
           typename DsDatatype,
           typename OutElementOp,
           ConvolutionForwardSpecialization ConvSpec,
           index_t DstScalarPerVector>
 using device_grouped_conv2d_xdl_int8_instances =
     std::tuple <
-        //########################################|     NumDim|      A|      B|       Ds|      E|  AData|  BData| AccData| CShuffle|         Ds|  EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|     CBlockTransfer|
-        //########################################|    Spatial| Layout| Layout|   Layout| Layout|   Type|   Type|    Type| DataType|   DataType|   Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|    ScalarPerVector|
-        //########################################|           |       |       |         |       |       |       |        |         |           |       |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|      _NWaveNPerXdl|
-        //########################################|           |       |       |         |       |       |       |        |         |           |       |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                   |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,  GNHWC,  GKYXC, DsLayout,  GNHWK, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>
+        //########################################|     NumDim|       A|       B|       Ds|       E|  AData|  BData| AccData| CShuffle|         Ds|  EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|     CBlockTransfer|
+        //########################################|    Spatial|  Layout|  Layout|   Layout|  Layout|   Type|   Type|    Type| DataType|   DataType|   Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|    ScalarPerVector|
+        //########################################|           |        |        |         |        |       |       |        |         |           |       |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|      _NWaveNPerXdl|
+        //########################################|           |        |        |         |        |       |       |        |         |           |       |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                   |
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial, ALayout, BLayout, DsLayout, ELayout, int8_t, int8_t, int32_t,  int32_t, DsDatatype, int8_t, PassThrough, PassThrough, OutElementOp,       ConvSpec,       GemmSpec,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 2>, DstScalarPerVector>
     >;
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
index b0f86b5c2e..9d69377085 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
                                                               Mul2_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             F32_Tuple,
                                                                             Mul2_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             F32_Tuple,
                                                                             Mul2_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             F32_Tuple,
                                                                             Mul2_Clamp,
                                                                             ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
 
 void add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               GK_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               F32_Tuple,
@@ -56,19 +65,28 @@ void add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(
                                                               Relu_Mul2_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             F32_Tuple,
                                                                             Relu_Mul2_Clamp,
                                                                             ConvFwdDefault,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             F32_Tuple,
                                                                             Relu_Mul2_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             8>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<GK_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            GK_Tuple,
+                                                                            NHWGK,
                                                                             F32_Tuple,
                                                                             Relu_Mul2_Clamp,
                                                                             ConvFwd1x1S1P0,
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
index 4812ceecfd..d6f87335b2 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
@@ -9,10 +9,10 @@ namespace device {
 namespace instance {
 void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -22,19 +22,28 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
                                                               Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
                                                                             Empty_Tuple,
                                                                             Mul_Clamp,
                                                                             ConvFwdDefault,
                                                                             16>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
                                                                             Empty_Tuple,
                                                                             Mul_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             16>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
                                                                             Empty_Tuple,
                                                                             Mul_Clamp,
                                                                             ConvFwd1x1S1P0,
@@ -43,10 +52,10 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
 
 void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
-                                                              GNHWC,
+                                                              NHWGC,
                                                               GKYXC,
                                                               Empty_Tuple,
-                                                              GNHWK,
+                                                              NHWGK,
                                                               int8_t,
                                                               int8_t,
                                                               Empty_Tuple,
@@ -56,19 +65,28 @@ void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
                                                               Relu_Mul_Clamp>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
                                                                             Empty_Tuple,
                                                                             Relu_Mul_Clamp,
                                                                             ConvFwdDefault,
                                                                             16>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
                                                                             Empty_Tuple,
                                                                             Relu_Mul_Clamp,
                                                                             ConvFwd1x1P0,
                                                                             16>{});
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_xdl_int8_instances<Empty_Tuple,
+                                   device_grouped_conv2d_xdl_int8_instances<NHWGC,
+                                                                            GKYXC,
+                                                                            Empty_Tuple,
+                                                                            NHWGK,
                                                                             Empty_Tuple,
                                                                             Relu_Mul_Clamp,
                                                                             ConvFwd1x1S1P0,

From 8b9cbba823024e9c2085ab54c5ac6f849e0c80e6 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Mon, 24 Apr 2023 08:07:39 -0500
Subject: [PATCH 003/194] reduce inital number for half_t splitk (#685)

---
 profiler/include/profiler/profile_gemm_splitk_impl.hpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index e5d5f8765e..233fb15c0a 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -72,8 +72,8 @@ bool profile_gemm_splitk_impl(int do_verification,
     {
     case 0: break;
     case 1:
-        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{0, 1});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-1, 1});
         break;
     default:
         a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});

From 8bb2bb4a051527bb11d0839f73ee0e82d5988708 Mon Sep 17 00:00:00 2001
From: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Date: Mon, 24 Apr 2023 22:43:36 +0200
Subject: [PATCH 004/194] Grouped Gemm + SplitK + simplified Kernel Args (#669)

* simplify karg in device/grid split-k op

* fix mk_kn_mn instances

* add more instances

* B2C with 3D grid for KSplit

* Remove unused code.

* Use default B2C (3D grid) in grid gemm v2r4r2.

* Device gemm splitk use B2C map.

* Device GroupedGemmXdlSplitKCShuffle

* Example for GroupedGemm Xdl SplitK

* Introduce Device GroupedGemmSplitK

* Fix updating kbatch size.

* Add instance mk-nk-mn

* Enable set kbatch in profiler.

* Add GGemmSplitK mk-kn-mn instances

* Add more instances & split into multiple files.

* minor fix

* tuning

* clean

* disabled failed instances

* use pipe v2

* Ignore arg on not supported arch.

* fix warning

---------

Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
Co-authored-by: root <root@ctr-ubbsmc15.amd.com>
---
 example/15_grouped_gemm/CMakeLists.txt        |   4 +-
 .../grouped_gemm_xdl_splitk_fp16.cpp          |  97 +++
 .../run_grouped_gemm_example.inc              |   1 +
 .../gpu/device/device_grouped_gemm.hpp        |   2 +-
 .../gpu/device/device_grouped_gemm_splitk.hpp |  39 ++
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |  20 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp | 612 ++++++++++++++++++
 .../gpu/grid/block_to_ctile_map.hpp           |  48 ++
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  87 ++-
 .../gpu/grouped_gemm.hpp                      |  58 ++
 .../gpu/grouped_gemm/CMakeLists.txt           |   4 +
 ...l_splitk_f16_f16_f16_mk_kn_mn_instance.cpp |  80 +++
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |  87 +++
 ...l_splitk_f16_f16_f16_mk_nk_mn_instance.cpp |  75 +++
 ...16_f16_f16_mk_nk_mn_irregular_instance.cpp |  81 +++
 .../profiler/profile_grouped_gemm_impl.hpp    |  31 +-
 profiler/src/profile_grouped_gemm.cpp         |  43 +-
 17 files changed, 1329 insertions(+), 40 deletions(-)
 create mode 100644 example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp

diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index e3080c4a9a..9df256c38d 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -5,6 +5,7 @@ add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
 add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
 add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
 add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
+add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
 
 
 add_dependencies(example_grouped_gemm_xdl
@@ -12,7 +13,8 @@ add_dependencies(example_grouped_gemm_xdl
                  example_grouped_gemm_xdl_fp16
                  example_grouped_gemm_xdl_bfp16
                  example_grouped_gemm_xdl_int8
-                 example_grouped_gemm_multiple_d_dl_fp16)
+                 example_grouped_gemm_multiple_d_dl_fp16
+                 example_grouped_gemm_xdl_splitk_fp16)
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
new file mode 100644
index 0000000000..a89937b2e2
--- /dev/null
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
@@ -0,0 +1,97 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F16;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemmXdlSplitKCShuffle
+    // clang-format off
+//######| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+//######|        |        |         |        |      Type|      Type|        Type|         DataType|       Type|      Type| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+//######|        |        |         |        |          |          |            |                 |           |          |   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+//######|        |        |         |        |          |          |            |                 |           |          |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
+// clang-format on
+
+#include "run_grouped_gemm_example.inc"
+
+int main(int argc, char* argv[])
+{
+    ProblemSize problem_size;
+    ExecutionConfig config;
+
+    problem_size.group_count = 16;
+
+    problem_size.Ms = {
+        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    for(int i = 0; i < problem_size.group_count; i++)
+    {
+        problem_size.Ns.push_back(768);
+        problem_size.Ks.push_back(4608);
+
+        problem_size.stride_As.push_back(problem_size.Ks[i]);
+        problem_size.stride_Bs.push_back(problem_size.Ks[i]);
+        problem_size.stride_Cs.push_back(problem_size.Ns[i]);
+    }
+
+    if(argc == 4)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=n0, 1=yes)\n");
+        exit(0);
+    }
+
+    return !run_grouped_gemm(problem_size, config);
+}
diff --git a/example/15_grouped_gemm/run_grouped_gemm_example.inc b/example/15_grouped_gemm/run_grouped_gemm_example.inc
index 324e177280..bceff29b63 100644
--- a/example/15_grouped_gemm/run_grouped_gemm_example.inc
+++ b/example/15_grouped_gemm/run_grouped_gemm_example.inc
@@ -147,6 +147,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
 #else
         a_tensors_device[i]->ToDevice(a_tensors[i].mData.data());
         b_tensors_device[i]->ToDevice(b_tensors[i].mData.data());
+	c_tensors_device[i]->SetZero();
 #endif
 
         p_a.push_back(a_tensors_device[i]->GetDeviceBuffer());
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
index 181ee4b428..4b1106c122 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
@@ -31,7 +31,7 @@ struct DeviceGroupedGemm : public BaseOperator
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
 
-    static_assert(DsLayout::Size() == DsDataType::Size(), "wrong! inconsisiten NumDTensor");
+    static_assert(DsLayout::Size() == DsDataType::Size(), "wrong! inconsistent NumDTensor");
 
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(std::vector<const void*>& p_a,
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
new file mode 100644
index 0000000000..06d180d30f
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
@@ -0,0 +1,39 @@
+#pragma once
+#include <iostream>
+#include <vector>
+
+#include "device_grouped_gemm.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
+struct DeviceGroupedGemmSplitK : public DeviceGroupedGemm<ALayout,
+                                                          BLayout,
+                                                          DsLayout,
+                                                          ELayout,
+                                                          ADataType,
+                                                          BDataType,
+                                                          DsDataType,
+                                                          EDataType,
+                                                          AElementwiseOperation,
+                                                          BElementwiseOperation,
+                                                          CElementwiseOperation>
+{
+    virtual void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 1f08cec67e..776f96e8e6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -114,7 +114,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         CBlockTransferScalarPerVector_NWaveNPerXDL,
         CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
 
-    using Argument = typename GridwiseGemm::Argument;
+    using Argument              = typename GridwiseGemm::Argument;
+    using DefaultBlock2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
 
     // Invoker
     struct Invoker : public BaseInvoker
@@ -138,8 +139,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                     "setting");
             }
 
+            const auto b2c_map = DefaultBlock2CTileMap{};
             index_t gdx, gdy, gdz;
-            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(karg);
+            std::tie(gdx, gdy, gdz) = b2c_map.CalculateGridSize(karg.M, karg.N, karg.k_batch);
             const auto K0           = karg.K0;
 
             const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
@@ -152,7 +154,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(CDataType)));
 
                 ave_time = launch_and_time_kernel(
-                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg, b2c_map);
             };
 
             if(has_main_k0_block_loop)
@@ -162,7 +164,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                     const auto kernel =
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              true,
-                                                             InMemoryDataOperationEnum::Set>;
+                                                             InMemoryDataOperationEnum::Set,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
@@ -171,7 +174,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                     const auto kernel =
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              true,
-                                                             InMemoryDataOperationEnum::AtomicAdd>;
+                                                             InMemoryDataOperationEnum::AtomicAdd,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
@@ -183,7 +187,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                     const auto kernel =
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              false,
-                                                             InMemoryDataOperationEnum::Set>;
+                                                             InMemoryDataOperationEnum::Set,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
@@ -192,7 +197,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                     const auto kernel =
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              false,
-                                                             InMemoryDataOperationEnum::AtomicAdd>;
+                                                             InMemoryDataOperationEnum::AtomicAdd,
+                                                             DefaultBlock2CTileMap>;
 
                     Run(kernel);
                 }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
new file mode 100644
index 0000000000..26a4319eaa
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -0,0 +1,612 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/ck.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename GridwiseGemm,
+          typename GemmDesc,
+          bool HasMainKBlockLoop,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
+                                       const index_t group_count)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+    constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
+    __shared__ uint8_t p_shared[shared_size];
+
+    const index_t block_id = get_block_1d_id();
+    const auto gemm_desc_ptr =
+        reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
+
+    index_t left     = 0;
+    index_t right    = group_count;
+    index_t group_id = index_t((left + right) / 2);
+    while((!(block_id >= gemm_desc_ptr[group_id].block_start_ &&
+             block_id < gemm_desc_ptr[group_id].block_end_)) &&
+          left <= right)
+    {
+        if(block_id < gemm_desc_ptr[group_id].block_start_)
+        {
+            right = group_id;
+        }
+        else
+        {
+            left = group_id;
+        }
+        group_id = index_t((left + right) / 2);
+    }
+
+    GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
+        gemm_desc_ptr[group_id].karg_,
+        static_cast<void*>(p_shared),
+        gemm_desc_ptr[group_id].block_2_ctile_map_);
+#else
+    ignore = gemm_descs_const;
+    ignore = group_count;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          ck::index_t NumPrefetch,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t KPerBlock,
+          ck::index_t AK1,
+          ck::index_t BK1,
+          ck::index_t MPerXDL,
+          ck::index_t NPerXDL,
+          ck::index_t MXdlPerWave,
+          ck::index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched = make_default_loop_scheduler(),
+          // Current implementation does not support multiple D fusions.
+          enable_if_t<AK1 == BK1 && is_same_v<DsLayout, ck::Tuple<>> &&
+                          is_same_v<DsDataType, ck::Tuple<>>,
+                      bool> = false>
+struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayout,
+                                                                           BLayout,
+                                                                           DsLayout,
+                                                                           ELayout,
+                                                                           ADataType,
+                                                                           BDataType,
+                                                                           DsDataType,
+                                                                           EDataType,
+                                                                           AElementwiseOperation,
+                                                                           BElementwiseOperation,
+                                                                           CDEElementwiseOperation>
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static_assert(KPerBlock % AK1 == 0);
+    static constexpr index_t K0PerBlock = KPerBlock / AK1;
+
+    using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
+        BlockSize,
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        EDataType,
+        ALayout,
+        BLayout,
+        ELayout,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        GemmSpec,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerXDL,
+        NPerXDL,
+        AK1,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CDEBlockTransferScalarPerVector_NPerBlock,
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+
+    using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
+    using Block2ETileMapKSplit =
+        BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
+    // Block2CTileMap configuration parameter.
+    static constexpr index_t B2E_M01 = 8;
+    using GroupedGemmBlock2ETileMap  = OffsettedBlockToCTileMap<Block2ETileMapKSplit>;
+    using KernelArgument             = typename GridwiseGemm::Argument;
+
+    struct GemmTransKernelArg
+    {
+        KernelArgument karg_;
+        GroupedGemmBlock2ETileMap block_2_ctile_map_;
+        index_t block_start_, block_end_;
+
+        GemmTransKernelArg() = default;
+        GemmTransKernelArg(KernelArgument&& karg,
+                           GroupedGemmBlock2ETileMap&& b2c_map,
+                           index_t block_start,
+                           index_t block_end)
+            : karg_{karg},
+              block_2_ctile_map_{b2c_map},
+              block_start_{block_start},
+              block_end_{block_end}
+        {
+        }
+    };
+
+    static constexpr index_t DefaultKBatch = 1;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+
+        Argument(std::vector<const void*>& p_As,
+                 std::vector<const void*>& p_Bs,
+                 std::vector<void*>& p_Es,
+                 std::vector<GemmDesc>& gemm_descs)
+            : Argument(p_As, p_Bs, p_Es, gemm_descs, DefaultKBatch)
+        {
+            // TODO: use occupancy api to calculate appropriate batch size.
+        }
+
+        Argument(std::vector<const void*>& p_As,
+                 std::vector<const void*>& p_Bs,
+                 std::vector<void*>& p_Es,
+                 std::vector<GemmDesc>& gemm_descs,
+                 index_t kbatch)
+            : K_BATCH{kbatch}
+        {
+            grid_size_   = 0;
+            group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
+
+            if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
+                 group_count_ == ck::type_convert<ck::index_t>(p_Bs.size()) &&
+                 group_count_ == ck::type_convert<ck::index_t>(p_Es.size())))
+            {
+                throw std::runtime_error("wrong! group_count_ != p_As/b/c.size");
+            }
+
+            gemm_kernel_args_.reserve(group_count_);
+
+            skipped_group_count_ = 0;
+
+            for(std::size_t i = 0; i < gemm_descs.size(); ++i)
+            {
+                const index_t M = gemm_descs[i].M_;
+                const index_t N = gemm_descs[i].N_;
+                const index_t K = gemm_descs[i].K_;
+
+                if(M == 0)
+                {
+                    skipped_group_count_++;
+                    continue;
+                }
+
+                const index_t stride_a = gemm_descs[i].stride_A_;
+                const index_t stride_b = gemm_descs[i].stride_B_;
+                const index_t stride_c = gemm_descs[i].stride_C_;
+
+                const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
+                const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
+                const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
+                const index_t k0       = GridwiseGemm::CalculateK0(K, K_BATCH);
+
+                const auto c_grid_desc_m_n =
+                    GridwiseGemm::MakeCGridDescriptor_M_N(M, N, m_padded, n_padded, stride_c);
+
+                const auto local_b2c_tile_map =
+                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+                const index_t block_start = grid_size_;
+                const index_t block_end   = grid_size_ + grid_size_grp;
+
+                grid_size_ += grid_size_grp;
+
+                // block-to-e-tile map
+                auto grouped_block_2_ctile_map =
+                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
+
+                auto karg = KernelArgument{type_convert<const ADataType*>(p_As[i]),
+                                           type_convert<const BDataType*>(p_Bs[i]),
+                                           type_convert<EDataType*>(p_Es[i]),
+                                           M,
+                                           N,
+                                           K,
+                                           stride_a,
+                                           stride_b,
+                                           stride_c,
+                                           m_padded,
+                                           n_padded,
+                                           k_padded,
+                                           k0,
+                                           K_BATCH};
+
+                gemm_kernel_args_.emplace_back(
+                    std::move(karg), std::move(grouped_block_2_ctile_map), block_start, block_end);
+            }
+        }
+
+        /**
+         * @brief      Recalculate group grid size for all gemms and update B2C maps.
+         *
+         * @param[in]  kbatch  The new splitK parameter value.
+         */
+        void UpdateKBatch(index_t kbatch)
+        {
+            K_BATCH    = kbatch;
+            grid_size_ = 0;
+
+            for(std::size_t i = 0; i < gemm_kernel_args_.size(); ++i)
+            {
+
+                auto& karg = gemm_kernel_args_[i].karg_;
+
+                const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
+                const index_t k0       = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
+
+                const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(
+                    karg.M, karg.N, karg.MPadded, karg.NPadded, karg.StrideC);
+
+                const auto local_b2c_tile_map =
+                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+                const index_t block_start = grid_size_;
+                const index_t block_end   = grid_size_ + grid_size_grp;
+
+                grid_size_ += grid_size_grp;
+
+                // block-to-e-tile map
+                auto grouped_block_2_ctile_map =
+                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
+
+                karg.KPadded                            = k_padded;
+                karg.K0                                 = k0;
+                karg.k_batch                            = K_BATCH;
+                gemm_kernel_args_[i].block_2_ctile_map_ = grouped_block_2_ctile_map;
+                gemm_kernel_args_[i].block_start_       = block_start;
+                gemm_kernel_args_[i].block_end_         = block_end;
+            }
+        }
+
+        //  private:
+        index_t K_BATCH;
+        index_t group_count_;
+        index_t skipped_group_count_;
+
+        std::vector<GemmTransKernelArg> gemm_kernel_args_;
+        index_t grid_size_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            index_t K0                       = arg.gemm_kernel_args_[0].karg_.K0;
+            bool all_have_kbatch_gt_one      = arg.gemm_kernel_args_[0].karg_.k_batch > 1;
+            bool all_have_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
+
+            for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
+            {
+                const auto& karg = arg.gemm_kernel_args_[i].karg_;
+                if(stream_config.log_level_ > 0)
+                {
+                    karg.Print();
+                }
+
+                auto kbatch = karg.k_batch;
+
+                if(!GridwiseGemm::CheckValidity(karg))
+                {
+                    std::ostringstream err;
+                    err << "Group id: " << i << " has invalid GridwiseGemm settings!" << __FILE__
+                        << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+
+                K0 = karg.K0;
+                bool not_all_have_main_k0_block_loop_same =
+                    all_have_main_k0_block_loop xor GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
+                bool not_all_have_kbatch_value_same = all_have_kbatch_gt_one xor (kbatch > 1);
+
+                if(not_all_have_main_k0_block_loop_same)
+                {
+                    std::ostringstream err;
+                    err << "Not all gemms have same value for main_k0_block_loop! in " << __FILE__
+                        << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+
+                if(not_all_have_kbatch_value_same)
+                {
+                    std::ostringstream err;
+                    err << "Not all gemms have same kbatch value (=1 or >1)! "
+                        << "group [" << i << "], kbatch: " << kbatch
+                        << ", group [0], kbatch: " << arg.gemm_kernel_args_[0].karg_.k_batch
+                        << " in " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
+                    throw std::runtime_error(err.str());
+                }
+            }
+
+            hip_check_error(hipMemcpy(arg.p_workspace_,
+                                      arg.gemm_kernel_args_.data(),
+                                      arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                                      hipMemcpyHostToDevice));
+
+            float ave_time = 0;
+
+            const auto Run = [&](const auto& kernel) {
+                if(all_have_kbatch_gt_one)
+                {
+                    for(const auto& trans_arg : arg.gemm_kernel_args_)
+                    {
+                        const auto& karg = trans_arg.karg_;
+                        hip_check_error(
+                            hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(EDataType)));
+                    }
+                }
+
+                ave_time =
+                    launch_and_time_kernel(stream_config,
+                                           kernel,
+                                           dim3(arg.grid_size_),
+                                           dim3(BlockSize),
+                                           0,
+                                           cast_pointer_to_constant_address_space(arg.p_workspace_),
+                                           arg.gemm_kernel_args_.size());
+            };
+
+            if(all_have_main_k0_block_loop)
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       true,
+                                                       InMemoryDataOperationEnum::AtomicAdd>;
+
+                    Run(kernel);
+                }
+                else
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       true,
+                                                       InMemoryDataOperationEnum::Set>;
+
+                    Run(kernel);
+                }
+            }
+            else
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       false,
+                                                       InMemoryDataOperationEnum::AtomicAdd>;
+
+                    Run(kernel);
+                }
+                else
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       false,
+                                                       InMemoryDataOperationEnum::Set>;
+
+                    Run(kernel);
+                }
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
+            arg.skipped_group_count_) != arg.group_count_)
+        {
+            return false;
+        }
+
+        bool supported = true;
+        for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
+        {
+            const auto& a        = arg.gemm_kernel_args_[i].karg_;
+            bool group_arg_valid = GridwiseGemm::CheckValidity(a);
+#if DEBUG_LOG
+            if(not group_arg_valid)
+            {
+                std::cout << "[" << __func__ << "] group id: " << i << " is not supported!\n";
+                a.Print();
+            }
+#endif // DEBUG_LOG
+            supported &= group_arg_valid;
+        }
+        return supported;
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::vector<const void*>& p_As,
+                             std::vector<const void*>& p_Bs,
+                             std::vector<std::array<const void*, NumDTensor>>&,
+                             std::vector<void*>& p_Es,
+                             std::vector<GemmDesc> gemm_descs,
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CDEElementwiseOperation)
+    {
+        return Argument{p_As, p_Bs, p_Es, gemm_descs};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::vector<const void*>& p_As,
+                        std::vector<const void*>& p_Bs,
+                        std::vector<std::array<const void*, NumDTensor>>&,
+                        std::vector<void*>& p_Es,
+                        std::vector<GemmDesc>& gemm_descs,
+                        AElementwiseOperation,
+                        BElementwiseOperation,
+                        CDEElementwiseOperation) override
+    {
+        return std::make_unique<Argument>(p_As, p_Bs, p_Es, gemm_descs);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedGemm_XdlSplitK"
+            << "<"
+            << std::string(ALayout::name)[0] << ","
+            << std::string(BLayout::name)[0] << ","
+            << std::string(ELayout::name)[0] << ","
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+
+    size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
+    {
+        return dynamic_cast<const Argument*>(p_arg)->gemm_kernel_args_.size() *
+               sizeof(GemmTransKernelArg);
+    }
+
+    static void SetKBatchSize(Argument& arg, index_t kbatch) { arg.UpdateKBatch(kbatch); }
+
+    // polymorphic
+    void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const override
+    {
+        return SetKBatchSize(*dynamic_cast<Argument*>(p_arg), kbatch);
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index fe4dce2b9d..9bd860f397 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -587,4 +587,52 @@ struct OffsettedBlockToCTileMap
     index_t block_start_;
 };
 
+/**
+ * @brief      Simple tile mapping which creates 3D grid of block of threads.
+ *
+ * @paragraph  Description
+ *             This Block-to-C-tile-map creates a 3D grid (n_blocks, m_blocks, z_blocks) of thread
+ *             blocks. The first 2D are regular 2D tiles created by division of output GEMM
+ *             dimenions by corresponding tile size. The third dimension (Z) is a k-split dimension,
+ *             which denotes the number of blocks we use to divide work on GEMM K dimension onto.
+ *
+ * @tparam     MPerBlock  Output block tile size in M dimension.
+ * @tparam     NPerBlock  Output block tile size in N dimension.
+ */
+template <index_t MPerBlock, index_t NPerBlock>
+struct BlockToCTileMap_3DGrid_KSplit
+{
+
+    __host__ __device__ BlockToCTileMap_3DGrid_KSplit() = default;
+
+    __host__ __device__ constexpr auto
+    CalculateGridSize(index_t M, index_t N, index_t k_split) const
+    {
+        // Create 3D grid
+        const auto M0 = math::integer_divide_ceil(M, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N, NPerBlock);
+
+        return std::make_tuple(N0, M0, k_split);
+    }
+
+    template <typename TopIdx>
+    __device__ constexpr auto CalculateBottomIndex(const TopIdx&) const
+    {
+        return make_tuple(blockIdx.z, blockIdx.y, blockIdx.x);
+    }
+
+    template <typename CTileIdx, typename CTileDim>
+    __host__ __device__ bool ValidCTileIndex(const CTileIdx& /* c_tile_idx */,
+                                             const CTileDim& /* c_tile_dim */) const
+    {
+        return true; // always valid provided that user gets grid size from CalculateGridSize()
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const
+    {
+        return true;
+    }
+};
+
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 727f180e96..4a2a77ce0f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -15,16 +15,20 @@
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+
 namespace ck {
 
 template <typename GridwiseGemm,
           bool HasMainKBlockLoop,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation>
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename Block2CTileMap>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg)
+        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg,
+                                             const Block2CTileMap& b2c_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
@@ -32,9 +36,10 @@ __global__ void
     __shared__ uint8_t p_shared[shared_size];
 
     GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
-        karg, static_cast<void*>(p_shared));
+        karg, static_cast<void*>(p_shared), b2c_map);
 #else
     ignore = karg;
+    ignore = b2c_map;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -478,8 +483,21 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        Number<CShuffleNRepeatPerShuffle * NWave * NPerXDL>{}));
     }
 
-    template <bool HasMainKBlockLoop, InMemoryDataOperationEnum CGlobalMemoryDataOperation>
-    __device__ static void Run(const Argument& karg, void* __restrict__ p_shared_block)
+    // return block_id to C matrix tile idx (m0, n0, k_split) mapping
+    __host__ __device__ static constexpr auto MakeDefaultBlock2CTileMap()
+    {
+        return BlockToCTileMap_3DGrid_KSplit<MPerBlock, NPerBlock>();
+    }
+
+    using CGridDesc_M_N         = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1, 1, 1))>;
+    using DefaultBlock2CTileMap = remove_cvref_t<decltype(MakeDefaultBlock2CTileMap())>;
+
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+              typename Block2CTileMap>
+    __device__ static void Run(const Argument& karg,
+                               void* __restrict__ p_shared_block,
+                               const Block2CTileMap& block_2_ctile_map)
     {
         const FloatAB* p_a_grid          = karg.p_a_grid;
         const FloatAB* p_b_grid          = karg.p_b_grid;
@@ -504,11 +522,21 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
 
-        const auto K0 = a_b_k0_m_k1_grid_desc.GetLength(I1);
+        // divide block work by [KBatch, M, N]
+        const auto block_work_idx =
+            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
 
-        const index_t block_m_id = __builtin_amdgcn_readfirstlane(blockIdx.y);
-        const index_t block_n_id = __builtin_amdgcn_readfirstlane(blockIdx.x);
-        const index_t k_batch_id = __builtin_amdgcn_readfirstlane(blockIdx.z);
+        if(!block_2_ctile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
+                          c_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
+        {
+            return;
+        }
+
+        const index_t block_m_id = __builtin_amdgcn_readfirstlane(block_work_idx[I1]);
+        const index_t block_n_id = __builtin_amdgcn_readfirstlane(block_work_idx[I2]);
+        const index_t k_batch_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
 
         // HACK: this force m/n_block_data_idx_on_grid into SGPR
         const index_t m_block_data_idx_on_grid =
@@ -651,6 +679,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
+#if 1
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
                                                                 FloatAB,
@@ -662,6 +691,20 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                                                                 MRepeat,
                                                                 NRepeat,
                                                                 K1>{};
+#else
+        auto blockwise_gemm = BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<
+            BlockSize,
+            FloatAB,
+            FloatAcc,
+            decltype(a_k0_m_k1_block_desc),
+            decltype(b_k0_n_k1_block_desc),
+            MPerXDL,
+            NPerXDL,
+            MRepeat,
+            NRepeat,
+            K1>{};
+
+#endif
 
         auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
 
@@ -680,6 +723,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
             p_b_block, b_k0_n_k1_block_desc.GetElementSpaceSize());
 
+#if 0
         // preload data into LDS
         {
             a_blockwise_copy.RunRead(a_b_k0_m_k1_grid_desc, a_grid_buf);
@@ -725,6 +769,31 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
             blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
         }
+#else
+        // gridwise GEMM pipeline
+        const auto gridwise_gemm_pipeline =
+            GridwiseGemmPipeline_Selector<PipelineVersion::v2, 1, LoopScheduler::Default>();
+
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
+            (a_b_k0_m_k1_grid_desc.GetLength(I1) * a_b_k0_m_k1_grid_desc.GetLength(I3)) /
+            (K0PerBlock * K1));
+
+        gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_b_k0_m_k1_grid_desc,
+                                                               a_b_k0_m_k1_block_desc,
+                                                               a_blockwise_copy,
+                                                               a_grid_buf,
+                                                               a_block_buf,
+                                                               a_block_slice_copy_step,
+                                                               b_b_k0_n_k1_grid_desc,
+                                                               b_b_k0_n_k1_block_desc,
+                                                               b_blockwise_copy,
+                                                               b_grid_buf,
+                                                               b_block_buf,
+                                                               b_block_slice_copy_step,
+                                                               blockwise_gemm,
+                                                               c_thread_buf,
+                                                               num_k_block_main_loop);
+#endif
 
         // output: register to global memory
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
index fa7462dbd9..e38dad1650 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
@@ -68,6 +68,58 @@ void add_device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances);
+
 template <typename ALayout,
           typename BLayout,
           typename ELayout,
@@ -109,11 +161,17 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<ELayout, Row>)
             {
                 add_device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
+                    op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<ELayout, Row>)
             {
                 add_device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
+                    op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<ELayout, Row>)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
index 82beb2ace2..b973b70aac 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
@@ -3,4 +3,8 @@ add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..764ec06198
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[m, k] * b[k, n] = e[m, n]
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // Currently AK1 must equal BK1 !
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..0385b0fc0c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,87 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough                    = ck::tensor_operation::element_wise::PassThrough;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // Currently AK1 must equal BK1 !
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Row,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..cffb0fce17
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough                 = ck::tensor_operation::element_wise::PassThrough;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// a[m, k] * b[n, k] = e[m, n]
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..5933ff61ec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,81 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough                    = ck::tensor_operation::element_wise::PassThrough;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemm<Row,
+                                                  Col,
+                                                  Empty_Tuple,
+                                                  Row,
+                                                  F16,
+                                                  F16,
+                                                  Empty_Tuple,
+                                                  F16,
+                                                  PassThrough,
+                                                  PassThrough,
+                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_grouped_gemm_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
index 2c28850e65..23dca244dc 100644
--- a/profiler/include/profiler/profile_grouped_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
@@ -8,6 +8,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp"
@@ -39,7 +40,8 @@ bool profile_grouped_gemm_impl(int do_verification,
                                const std::vector<int>& Ks,
                                const std::vector<int>& StrideAs,
                                const std::vector<int>& StrideBs,
-                               const std::vector<int>& StrideCs)
+                               const std::vector<int>& StrideCs,
+                               int kbatch = 1)
 {
 
     bool pass = true;
@@ -96,8 +98,6 @@ bool profile_grouped_gemm_impl(int do_verification,
             a_m_k[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0}, num_thread);
             b_k_n[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}, num_thread);
         }
-
-        c_m_n_device_results[i].GenerateTensorValue(GeneratorTensor_0<CDataType>{}, num_thread);
     }
 
     using AElementOp = ck::tensor_operation::element_wise::PassThrough;
@@ -132,13 +132,12 @@ bool profile_grouped_gemm_impl(int do_verification,
             std::make_unique<DeviceMem>(sizeof(ADataType) * a_m_k[i].mDesc.GetElementSpaceSize()));
         b_device_buf.emplace_back(
             std::make_unique<DeviceMem>(sizeof(BDataType) * b_k_n[i].mDesc.GetElementSpaceSize()));
-
         c_device_buf.emplace_back(std::make_unique<DeviceMem>(
             sizeof(CDataType) * c_m_n_device_results[i].mDesc.GetElementSpaceSize()));
 
         a_device_buf[i]->ToDevice(a_m_k[i].mData.data());
         b_device_buf[i]->ToDevice(b_k_n[i].mData.data());
-        c_device_buf[i]->ToDevice(c_m_n_device_results[i].mData.data());
+        c_device_buf[i]->SetZero();
 
         gemm_descs.push_back({Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
 
@@ -197,6 +196,28 @@ bool profile_grouped_gemm_impl(int do_verification,
         {
             std::string gemm_name = gemm_ptr->GetTypeString();
 
+            if(kbatch > 1)
+            {
+                using DeviceOpSplitK =
+                    ck::tensor_operation::device::DeviceGroupedGemmSplitK<ALayout,
+                                                                          BLayout,
+                                                                          ck::Tuple<>,
+                                                                          CLayout,
+                                                                          ADataType,
+                                                                          BDataType,
+                                                                          ck::Tuple<>,
+                                                                          CDataType,
+                                                                          AElementOp,
+                                                                          BElementOp,
+                                                                          CElementOp>;
+
+                if(dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get()) != nullptr)
+                {
+                    dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get())
+                        ->SetKBatchSize(argument_ptr.get(), kbatch);
+                }
+            }
+
             float ave_time =
                 invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
diff --git a/profiler/src/profile_grouped_gemm.cpp b/profiler/src/profile_grouped_gemm.cpp
index 871b2edfd5..34647adabf 100644
--- a/profiler/src/profile_grouped_gemm.cpp
+++ b/profiler/src/profile_grouped_gemm.cpp
@@ -52,20 +52,24 @@ std::vector<int> argToIntArray(char* input)
 
 int profile_grouped_gemm(int argc, char* argv[])
 {
-    if(!(argc == 14))
+    if(argc < 14)
     {
-        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
-        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
-        printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
-        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
-        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
-        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
-        printf("arg4: verification (0: no; 1: yes)\n");
-        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
-        printf("arg6: print tensor value (0: no; 1: yes)\n");
-        printf("arg7: time kernel (0=n0, 1=yes)\n");
-        printf("arg8 to 13: Ms, Ns, Ks, StrideAs, StrideBs, StrideCs (e.g., 256,256 128,128 64,64 "
-               "64,64 64,64 128,128)\n");
+        std::cout
+            << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+            << "arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n"
+            << "arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n"
+            << "                     1: A[m, k] * B[n, k] = C[m, n];\n"
+            << "                     2: A[k, m] * B[k, n] = C[m, n];\n"
+            << "                     3: A[k, m] * B[n, k] = C[m, n])\n"
+            << "arg4: verification (0: no; 1: yes)\n"
+            << "arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+            << "arg6: print tensor value (0: no; 1: yes)\n"
+            << "arg7: time kernel (0=n0, 1=yes)\n"
+            << "arg8 to 13: Ms, Ns, Ks, StrideAs, StrideBs, StrideCs (e.g., 256,256 128,128 64,64 "
+               "64,64 64,64 128,128)\n"
+            << "arg15: kbatch value (default 4)\n"
+            << std::endl;
+
         exit(1);
     }
 
@@ -83,6 +87,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     const auto StrideAs = argToIntArray(argv[11]);
     const auto StrideBs = argToIntArray(argv[12]);
     const auto StrideCs = argToIntArray(argv[13]);
+    const int kbatch    = argc == 15 ? std::stoi(argv[14]) : 1;
 
     if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
@@ -101,7 +106,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
@@ -120,7 +126,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
     {
@@ -139,7 +146,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
     {
@@ -158,7 +166,8 @@ int profile_grouped_gemm(int argc, char* argv[])
                                                                                    Ks,
                                                                                    StrideAs,
                                                                                    StrideBs,
-                                                                                   StrideCs);
+                                                                                   StrideCs,
+                                                                                   kbatch);
     }
     else
     {

From 7613c1d9b9dc612a5de79ab968c534ea58e7cbe4 Mon Sep 17 00:00:00 2001
From: Jun Liu <Liu.Jun@amd.com>
Date: Wed, 26 Apr 2023 13:41:03 -0700
Subject: [PATCH 005/194] [CK] suppress unsafe buffer warn (#687)

incomplete fix from https://github.com/ROCmSoftwarePlatform/composable_kernel/pull/670

So it does not only happen in gtest but also in CK code:

We need to fix them as a quality improvement, but for now suppressing this warning in immediate releases:
http://compiler-ci.amd.com/blue/rest/organizations/jenkins/pipelines/compiler-psdb-amd-stg-open/runs/2540/nodes/282/steps/3202/log/?start=0

e.g.
```
[2023-04-26T17:26:31.524Z] /jenkins/workspace/compiler-psdb-amd-stg-open/Libs/MIOpen/deps_hip/cget/build/tmp-a3db5da587a64213bde99fb856db1b43/composable_kernel-0f98035df1cc5ba3e90ab03187e672b426a25b00/include/ck/utility/generic_memory_space_atomic.hpp:52:19: error: unsafe pointer arithmetic [-Werror,-Wunsafe-buffer-usage]
[2023-04-26T17:26:31.524Z]         atomicAdd(c_style_pointer_cast<float*>(p_dst) + 1, vx.template AsType<float>()[I1]);
[2023-04-26T17:26:31.524Z]                   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
```
[2023-04-26T17:26:31.523Z] /jenkins/workspace/compiler-psdb-amd-stg-open/Libs/MIOpen/deps_hip/cget/build/tmp-a3db5da587a64213bde99fb856db1b43/composable_kernel-0f98035df1cc5ba3e90ab03187e672b426a25b00/include/ck/utility/amd_inline_asm.hpp:62:20: error: 'p_a_half2' is an unsafe pointer used for buffer access [-Werror,-Wunsafe-buffer-usage]
[2023-04-26T17:26:31.523Z]     const half2_t* p_a_half2  = c_style_pointer_cast<const half2_t*>(&a);
[2023-04-26T17:26:31.523Z]     ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
```
---
 cmake/EnableCompilerWarnings.cmake | 1 +
 1 file changed, 1 insertion(+)

diff --git a/cmake/EnableCompilerWarnings.cmake b/cmake/EnableCompilerWarnings.cmake
index 87bcb08e83..369cd0b54c 100644
--- a/cmake/EnableCompilerWarnings.cmake
+++ b/cmake/EnableCompilerWarnings.cmake
@@ -92,6 +92,7 @@ else()
                 -Wno-unused-command-line-argument
                 -Wno-weak-vtables
                 -Wno-covered-switch-default
+                -Wno-unsafe-buffer-usage
             )
         else()
             if (CMAKE_${COMPILER}_COMPILER_ID MATCHES "GNU" AND ${COMPILER} MATCHES "CXX")

From 54c90aae13622fa5cc9457af22bd55540a6f794b Mon Sep 17 00:00:00 2001
From: Haocong WANG <haocwang@amd.com>
Date: Thu, 27 Apr 2023 04:58:57 +0800
Subject: [PATCH 006/194] add vector load check (#680)

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../device_gemm_multiple_d_wmma_cshuffle.hpp  | 89 ++++++++++++++++++-
 .../gpu/device/impl/device_gemm_wmma.hpp      | 71 ++++++++++++++-
 2 files changed, 158 insertions(+), 2 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
index b59bb2b303..750df31a3f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
@@ -273,7 +273,10 @@ struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
               N01_{N01},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
+              cde_element_op_{cde_element_op},
+              MRaw_{M},
+              NRaw_{N},
+              KRaw_{K}
         {
             a_grid_desc_k0_m_k1_ = DeviceOp::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
             b_grid_desc_k0_n_k1_ = DeviceOp::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
@@ -335,6 +338,11 @@ struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
         AElementwiseOperation a_element_op_;
         BElementwiseOperation b_element_op_;
         CDEElementwiseOperation cde_element_op_;
+
+        // for checking vector load/store
+        index_t MRaw_;
+        index_t NRaw_;
+        index_t KRaw_;
     };
 
     // Invoker
@@ -488,6 +496,85 @@ struct DeviceGemmMultipleD_Wmma_CShuffle : public DeviceGemmMultipleD<ALayout,
         {
             return false;
         }
+        // check vector load/store
+        {
+            using Row = ck::tensor_layout::gemm::RowMajor;
+            using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+            // check vector load of A
+            if constexpr(is_same_v<ALayout, Row> && ABlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && ABlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.MRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector laod of B
+            if constexpr(is_same_v<BLayout, Col> && BBlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<BLayout, Row> && BBlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.NRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector load of Ds
+            // only support RowMajor for now
+            bool all_valid = true;
+
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                if constexpr(!is_same_v<DLayout, Row>)
+                {
+                    all_valid = false;
+                }
+            });
+
+            if(!all_valid)
+            {
+                return false;
+            }
+
+            // check vector store of E
+            // only support RowMajor for now
+            if constexpr(is_same_v<ELayout, Row>)
+            {
+                if(arg.NRaw_ % CDEShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+        }
 
         return GridwiseOp::CheckValidity(arg.a_grid_desc_k0_m_k1_,
                                          arg.b_grid_desc_k0_n_k1_,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
index 9b0ff3f46f..03ffcf8e59 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
@@ -239,7 +239,10 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
               N01_{N01},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
+              c_element_op_{c_element_op},
+              MRaw_{M},
+              NRaw_{N},
+              KRaw_{K}
         {
             a_grid_desc_k0_m_k1_ =
                 DeviceGemmWmma_CShuffle::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
@@ -276,6 +279,10 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
         AElementwiseOperation a_element_op_;
         BElementwiseOperation b_element_op_;
         CElementwiseOperation c_element_op_;
+        // for checking vector load/store
+        index_t MRaw_;
+        index_t NRaw_;
+        index_t KRaw_;
     };
 
     // Invoker
@@ -417,6 +424,68 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
             return false;
         }
 
+        // check vector load/store
+        {
+            using Row = ck::tensor_layout::gemm::RowMajor;
+            using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+            // check vector load of A
+            if constexpr(is_same_v<ALayout, Row> && ABlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<ALayout, Col> && ABlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.MRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector laod of B
+            if constexpr(is_same_v<BLayout, Col> && BBlockTransferSrcVectorDim == 2)
+            {
+                if(arg.KRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else if constexpr(is_same_v<BLayout, Row> && BBlockTransferSrcVectorDim == 1)
+            {
+                // FIXME: not rigorous
+                if(arg.NRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+
+            // check vector store of C
+            // only support RowMajor for now
+            if constexpr(is_same_v<CLayout, Row>)
+            {
+                if(arg.NRaw_ % CShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
                                            arg.b_grid_desc_k0_n_k1_,
                                            arg.c_grid_desc_m_n_,

From 4feebedd413212fcd455da4a2bfcb40e9e6cab5a Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 28 Apr 2023 16:22:59 -0700
Subject: [PATCH 007/194] Syncing up from internal repo to enable MI300. (#690)

* enable gfx940

* switch between intrinsic mfma routines on mi100/200 and mi300

* fix mfma_int8 on MI300

* disable 2 int8 examples on MI300

* Update cmake-ck-dev.sh

* restore gitignore file

* modify Jenkinsfile to the internal repo

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 example/31_batched_gemm_gemm/CMakeLists.txt   |  4 +-
 .../41_grouped_conv_conv_fwd/CMakeLists.txt   |  5 +-
 include/ck/ck.hpp                             | 18 +++---
 .../device_gemm_xdl_waveletmodel_cshuffle.hpp |  6 +-
 ...gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  6 +-
 ...tk_contraction_multiple_d_xdl_cshuffle.hpp |  6 +-
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |  6 +-
 .../device_batched_gemm_e_permute_xdl.hpp     |  3 +-
 .../device_batched_gemm_gemm_xdl_cshuffle.hpp |  6 +-
 .../impl/device_batched_gemm_multi_d_xdl.hpp  |  6 +-
 ...ultiple_d_gemm_multiple_d_xdl_cshuffle.hpp |  6 +-
 ...evice_batched_gemm_reduce_xdl_cshuffle.hpp |  3 +-
 ...gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  6 +-
 ...batched_gemm_softmax_gemm_xdl_cshuffle.hpp |  6 +-
 .../device/impl/device_batched_gemm_xdl.hpp   |  3 +-
 ...ce_contraction_multiple_d_xdl_cshuffle.hpp |  6 +-
 ...evice_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp |  3 +-
 .../impl/device_gemm_bias_e_permute_xdl.hpp   |  6 +-
 ...gemm_multiple_d_layernorm_xdl_cshuffle.hpp |  6 +-
 ...emm_multiple_d_multiple_r_xdl_cshuffle.hpp |  6 +-
 .../device_gemm_multiple_d_xdl_cshuffle.hpp   |  6 +-
 .../gpu/device/impl/device_gemm_xdl.hpp       |  2 +-
 .../device/impl/device_gemm_xdl_cshuffle.hpp  |  3 +-
 .../device_gemm_xdl_layernorm_cshuffle.hpp    |  3 +-
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |  6 +-
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  3 +-
 ...bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp |  3 +-
 ...fwd_multiple_d_multiple_r_xdl_cshuffle.hpp |  5 +-
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |  5 +-
 .../device/impl/device_grouped_gemm_xdl.hpp   |  3 +-
 ...e_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp |  3 +-
 .../gridwise_gemm_reduce_xdl_cshuffle_v1.hpp  |  3 +-
 .../grid/gridwise_gemm_xdl_cshuffle_v1.hpp    |  3 +-
 ...ridwise_gemm_xdl_layernorm_cshuffle_v1.hpp |  3 +-
 .../grid/gridwise_gemm_xdlops_bwd_weight.hpp  |  3 +-
 .../gridwise_gemm_xdlops_skip_b_lds_v1.hpp    |  3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |  3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4.hpp    |  3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r1.hpp    |  3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r2.hpp    |  3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r3.hpp    |  3 +-
 .../tensor_operation/gpu/warp/xdlops_gemm.hpp | 60 ++++++++++++++++++-
 include/ck/utility/amd_xdlops.hpp             | 40 ++++++++++++-
 include/ck/utility/data_type.hpp              |  2 +
 script/cmake-ck-dev.sh                        |  3 +-
 script/cmake-ck-release.sh                    |  3 +-
 47 files changed, 228 insertions(+), 72 deletions(-)

diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index d79248251c..ad40c96b41 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -1,7 +1,9 @@
 add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
 add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
 add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
-add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+if(NOT GPU_TARGETS MATCHES "gfx940")
+	add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+endif()
 
 if(USE_BITINT_EXTENSION_INT4)
 add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index 9cb30f6176..4eb79371a7 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -1,8 +1,9 @@
 add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
 add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
 add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
-
+if(NOT GPU_TARGETS MATCHES "gfx940")
+	add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+endif()
 if(USE_BITINT_EXTENSION_INT4)
 add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
 endif(USE_BITINT_EXTENSION_INT4)
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 036ca24a44..cb20ea2492 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -31,7 +31,7 @@
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_BUFFER_RESOURCE_3RD_DWORD -1
 #elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx90a__) // for GPU code
+    defined(__gfx90a__) || defined(__gfx940__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
 #elif defined(__gfx1030__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
@@ -43,8 +43,8 @@
 #ifndef __HIP_DEVICE_COMPILE__                   // for host code, define nothing
 #elif defined(__gfx803__) || defined(__gfx900__) // for GPU code
 #define CK_USE_AMD_V_MAC_F32
-#elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx1030__) // for GPU code
+#elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__) || \
+    defined(__gfx940__) // for GPU code
 #define CK_USE_AMD_V_FMAC_F32
 #define CK_USE_AMD_V_DOT2_F32_F16
 #define CK_USE_AMD_V_DOT4_I32_I8
@@ -53,14 +53,18 @@
 // MFMA instruction
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_MFMA
-#elif defined(__gfx908__) || defined(__gfx90a__) // for GPU code
+#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) // for GPU code
 #define CK_USE_AMD_MFMA
 #endif
 
-#if defined(__gfx90a__)
+#if(defined(__gfx90a__) || defined(__gfx940__))
 #define CK_USE_AMD_MFMA_BF16_1K_OP
 #endif
 
+#if defined(__gfx940__)
+#define CK_USE_AMD_MFMA_GFX940
+#endif
+
 // WMMA instruction
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_WMMA
@@ -80,13 +84,13 @@
 // buffer atomic add: floating point
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
-#elif defined(__gfx908__) || defined(__gfx90a__) // for GPU code
+#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
 #else // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 0
 #endif
 
-#if defined(__gfx90a__) // for GPU code
+#if(defined(__gfx90a__) || defined(__gfx940__)) // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 1
 #else
 #define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 0
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
index d985d0f92e..af38f14254 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -47,7 +47,8 @@ __global__ void
                 e_grid_desc_mblock_mperblock_nblock_nperblock,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -416,7 +417,8 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 70f3d0277d..0b1db28465 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -43,7 +43,8 @@ __global__ void
             const B1ElementwiseOperation b1_element_op,
             const CElementwiseOperation c_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -678,7 +679,8 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index 8eab1cdee5..70990e795c 100644
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -56,7 +56,8 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t num_blocks_per_batch =
@@ -938,7 +939,8 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index 2237ad944c..9bf8f5ccd9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -56,7 +56,8 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t num_blocks_per_batch =
@@ -839,7 +840,8 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
index 01f5e17d91..2018445869 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -74,7 +74,8 @@ __global__ void
                                           const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
                                           const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
index 3b87e56337..20e9920d93 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
@@ -60,7 +60,8 @@ __global__ void
             const index_t batch_count,
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -588,7 +589,8 @@ struct DeviceBatchedGemmGemm_Xdl_CShuffle : public DeviceBatchedGemmGemm<ALayout
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index e34c19bdf0..0df3460941 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -83,7 +83,8 @@ __global__ void
                                 const Block2ETileMap block_2_etile_map)
 {
 
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -579,7 +580,8 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index 4fc8e69d2c..196dc86da1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -68,7 +68,8 @@ __global__ void
             const index_t batch_count,
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -804,7 +805,8 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
index 080e26ea89..ef9b90ba7c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
@@ -59,7 +59,8 @@ __global__ void
             const ComputeBasePrtOfBatch compute_base_ptr_of_batch_,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 037867d5f4..0c6c0ef7ad 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -67,7 +67,8 @@ __global__ void
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch,
             const C0MatrixMask c0_matrix_mask)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -714,7 +715,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         arg.Print();
 #endif
 
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
index 1f21f2d712..84edde63ec 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
@@ -62,7 +62,8 @@ __global__ void
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch,
             const C0MatrixMask c0_matrix_mask)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -612,7 +613,8 @@ struct DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index 48a2244560..d35f194171 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -75,7 +75,8 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index 7a4c8bf267..1eaffe705a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -52,7 +52,8 @@ __global__ void
                 e_grid_desc_mblock_mperblock_nblock_nperblock,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -581,7 +582,8 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
index 31c761e09f..d52879cd90 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
@@ -55,7 +55,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / num_batches);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
index e95bf2f81d..9f9fe0f1c9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
@@ -51,7 +51,8 @@ __global__ void
                                        e_grid_desc_mblock_mperblock_nblock_nperblock,
                                    const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -456,7 +457,8 @@ struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOp
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
index b53927a9ed..a383b0bb7d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -63,7 +63,8 @@ __global__ void
             const Block2ETileMap block_2_etile_map,
             index_t NRaw)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemmWelford::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemmWelford::template Run<HasMainKBlockLoop>(
@@ -854,7 +855,8 @@ struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index f1185357a4..4c1c3ab7bf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -60,7 +60,8 @@ __global__ void
             const RsGridDescriptor_MBlock_MPerBlock rs_grid_desc_mblock_mperblock,
             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -554,7 +555,8 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index 36e8105122..2488101484 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -51,7 +51,8 @@ __global__ void
                                                 e_grid_desc_mblock_mperblock_nblock_nperblock,
                                             const Block2ETileMap block_2_etile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -490,7 +491,8 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
index 48fb646375..a5051455bf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
@@ -428,7 +428,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                 return false;
             }
         }
-        else if(ck::get_device_name() == "gfx90a")
+        else if(ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx940")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index 6621f25bde..7cd0ff72e8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -574,7 +574,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
index 2a3e74a4cb..8ee138f827 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
@@ -648,7 +648,8 @@ struct DeviceGemmLayerNorm_Xdl_CShuffle : public BaseOperator
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 03d9e26a46..76dd5a3666 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -37,7 +37,8 @@ __global__ void
             const BElementwiseOperation b_element_op,
             const CDEElementwiseOperation cde_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -703,7 +704,8 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a"))
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 4cef404447..92c20a308f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -130,7 +130,8 @@ __global__ void
             const Block2ETileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
index 34db9f2a52..c921c9f1b4 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
@@ -78,7 +78,8 @@ __global__ void
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
index b60f94d141..de40d71293 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -155,7 +155,8 @@ __global__ void
             const Block2ETileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -810,7 +811,7 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
                 return false;
             }
         }
-        else if(get_device_name() == "gfx90a")
+        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index 7bab2d0408..02458bf02a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -135,7 +135,8 @@ __global__ void
             const Block2ETileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -684,7 +685,7 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
                 return false;
             }
         }
-        else if(get_device_name() == "gfx90a")
+        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index a009bcb3bf..e3795060be 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -38,7 +38,8 @@ __global__ void
                                 const BElementwiseOperation b_element_op,
                                 const CDEElementwiseOperation c_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
index 16ba23280d..bebcdceb43 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
@@ -66,7 +66,8 @@ __global__ void
             const ReduceGridDescriptor_MBlock_MPerBlock reduce_grid_desc_mblock_mperblock,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
index 2fe5506844..a3f5324713 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
@@ -54,7 +54,8 @@ __global__ void
             const ReduceGridDescriptor_MBlock_MPerBlock reduce_grid_desc_mblock_mperblock,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index ecc528a7ed..1213cdc263 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -44,7 +44,8 @@ __global__ void
                                         c_grid_desc_mblock_mperblock_nblock_nperblock,
                                     const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
index 94e181cd45..2d4ebe7076 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
@@ -57,7 +57,8 @@ __global__ void
             const C0GridDescriptor_NBlock_NPerBlock c0_grid_desc_nblock_nperblock,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     // TODO ANT: separate into MMA + Epilogue
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index 2da92466b5..1979331d07 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -165,7 +165,8 @@ __global__ void
                                       const CElementwiseOperation c_element_op,
                                       const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
index 2aad7128f0..8d86f3c1d7 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
@@ -44,7 +44,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index 51c578385f..775b77118c 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -43,7 +43,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
index 949d564836..55f465a037 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
@@ -42,7 +42,8 @@ __global__ void
                                 const CElementwiseOperation c_element_op,
                                 const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 4a2a77ce0f..b393c4897e 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -30,7 +30,8 @@ __global__ void
         kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg,
                                              const Block2CTileMap& b2c_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
 
     __shared__ uint8_t p_shared[shared_size];
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
index ffb2926c87..8259927fec 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
@@ -46,7 +46,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
index 7e6dbb3b2e..5d5fdae170 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
@@ -49,7 +49,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
index fb1e34b985..dc83f8e984 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
@@ -53,7 +53,8 @@ __global__ void
             const CElementwiseOperation c_element_op,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index 4d53f0d816..319487bc05 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -27,6 +27,8 @@ enum struct MfmaInstr
     mfma_f32_16x16x8bf16,
     mfma_i32_32x32x8i8,
     mfma_i32_16x16x16i8,
+    mfma_i32_32x32x16i8,
+    mfma_i32_16x16x32i8,
     mfma_f64_16x16x4f64
 };
 
@@ -386,6 +388,50 @@ struct mfma_type<MfmaInstr::mfma_i32_16x16x16i8>
     }
 };
 
+template <>
+struct mfma_type<MfmaInstr::mfma_i32_32x32x16i8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_i32_32x32x16i8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_i32_16x16x32i8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_i32_16x16x32i8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
 template <>
 struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
 {
@@ -524,17 +570,29 @@ struct MfmaSelector
 #endif
     }
 
+#if defined(CK_USE_AMD_MFMA_GFX940)
+    template <>
+    static constexpr auto GetMfma<int8_t, 32, 32>()
+    {
+        return MfmaInstr::mfma_i32_32x32x16i8;
+    }
+    template <>
+    static constexpr auto GetMfma<int8_t, 16, 16>()
+    {
+        return MfmaInstr::mfma_i32_16x16x32i8;
+    }
+#else
     template <>
     static constexpr auto GetMfma<int8_t, 32, 32>()
     {
         return MfmaInstr::mfma_i32_32x32x8i8;
     }
-
     template <>
     static constexpr auto GetMfma<int8_t, 16, 16>()
     {
         return MfmaInstr::mfma_i32_16x16x16i8;
     }
+#endif
 
     static constexpr auto selected_mfma = mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops>()>{};
 
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index b4be0cbee7..a742496fc1 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -297,6 +297,44 @@ struct intrin_mfma_i32_16x16x16i8<16, 16>
     }
 };
 
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_i32_32x32x16i8;
+
+template <>
+struct intrin_mfma_i32_32x32x16i8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const int8x8_t& reg_a, const int8x8_t& reg_b, FloatC& reg_c)
+    {
+        reg_c.template AsType<int32x16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_i32_32x32x16_i8(bit_cast<int64_t>(reg_a),
+                                                  bit_cast<int64_t>(reg_b),
+                                                  reg_c.template AsType<int32x16_t>()[Number<0>{}],
+                                                  0,
+                                                  0,
+                                                  0);
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_i32_16x16x32i8;
+
+template <>
+struct intrin_mfma_i32_16x16x32i8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const int8x8_t& reg_a, const int8x8_t& reg_b, FloatC& reg_c)
+    {
+        reg_c.template AsType<int32x4_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_i32_16x16x32i8(bit_cast<int64_t>(reg_a),
+                                                 bit_cast<int64_t>(reg_b),
+                                                 reg_c.template AsType<int32x4_t>()[Number<0>{}],
+                                                 0,
+                                                 0,
+                                                 0);
+    }
+};
+
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f64_16x16x4f64;
 
@@ -306,7 +344,7 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
     template <class FloatC>
     __device__ static void Run(const double& reg_a, const double& reg_b, FloatC& reg_c)
     {
-#ifdef __gfx90a__
+#if defined(__gfx90a__) || defined(__gfx940__)
         reg_c.template AsType<double4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f64_16x16x4f64(
             reg_a, reg_b, reg_c.template AsType<double4_t>()[Number<0>{}], 0, 0, 0);
 #else
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index 079b0cb860..101061191e 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -898,6 +898,8 @@ struct vector_type<T, 256>
     }
 };
 
+using int64_t = long;
+
 // fp64
 using double2_t = typename vector_type<double, 2>::type;
 using double4_t = typename vector_type<double, 4>::type;
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index 8f462237f5..426f68d443 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -12,9 +12,8 @@ cmake
 -save-temps=$PWD"                                                                                 \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=ON                                                                                   \
--D GPU_TARGETS="gfx908;gfx90a"                                                                    \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
 
-#-D AMDGPU_TARGETS=gfx90a;gfx908
diff --git a/script/cmake-ck-release.sh b/script/cmake-ck-release.sh
index 268b1ebf9b..787eabbf96 100755
--- a/script/cmake-ck-release.sh
+++ b/script/cmake-ck-release.sh
@@ -11,9 +11,8 @@ cmake
 -D CMAKE_CXX_FLAGS="-O3"                                                                          \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=OFF                                                                                  \
--D GPU_TARGETS="gfx908;gfx90a"                                                                      \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
 
-#-D AMDGPU_TARGETS=gfx90a;gfx908

From f53ede26e5cb58cb90d9c6c96dafc9a61d1dccc8 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 2 May 2023 10:30:23 -0500
Subject: [PATCH 008/194] fixed init range (#691)

---
 profiler/include/profiler/profile_gemm_splitk_impl.hpp | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index 233fb15c0a..4cc62509d7 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -72,8 +72,8 @@ bool profile_gemm_splitk_impl(int do_verification,
     {
     case 0: break;
     case 1:
-        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{0, 1});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-1, 1});
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-1, 2});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-1, 2});
         break;
     default:
         a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
@@ -94,7 +94,7 @@ bool profile_gemm_splitk_impl(int do_verification,
 
     a_device_buf.ToDevice(a_m_k.mData.data());
     b_device_buf.ToDevice(b_k_n.mData.data());
-    c_device_buf.ToDevice(c_m_n_device_result.mData.data());
+    c_device_buf.SetZero();
 
     using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<ALayout,
                                                                     BLayout,

From 86e0190ec9abcf19b8864b6d4a35dc22f4be6350 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 3 May 2023 08:18:10 -0700
Subject: [PATCH 009/194] update daily build from rocm 5.4.3 to 5.5 (#693)

---
 Jenkinsfile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 6cb458031a..83559c223b 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -19,7 +19,7 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.ROCMVERSION != "5.5" && params.ROCMVERSION != "5.6"){
+    if (params.ROCMVERSION != "5.6"){
        if (params.COMPILER_VERSION == "") {
            img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
        }
@@ -597,7 +597,7 @@ def process_results(Map conf=[:]){
 
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
 CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true
-                                              0 21 * * * % ROCMVERSION=5.4.3;COMPILER_VERSION=release;COMPILER_COMMIT=
+                                              0 21 * * * % ROCMVERSION=5.5;COMPILER_VERSION=release;COMPILER_COMMIT=
                                               0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {

From 4a51d2da9de3524e96f2abed5f843ff9da535db3 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 3 May 2023 08:25:25 -0700
Subject: [PATCH 010/194] Fix grouped_gemm_splitk kernels on MI300. (#694)

* replace amd_buffer_atomic_add with hip_atomic_add

* fix grouped_gemm_splitk kernels on mi300

* fix syntax

* revert experimental atomic_add changes

---------

Co-authored-by: Jing Zhang <jizhan@amd.com>
---
 example/15_grouped_gemm/run_grouped_gemm_example.inc           | 2 +-
 .../device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp    | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/example/15_grouped_gemm/run_grouped_gemm_example.inc b/example/15_grouped_gemm/run_grouped_gemm_example.inc
index bceff29b63..320870e0de 100644
--- a/example/15_grouped_gemm/run_grouped_gemm_example.inc
+++ b/example/15_grouped_gemm/run_grouped_gemm_example.inc
@@ -147,7 +147,7 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
 #else
         a_tensors_device[i]->ToDevice(a_tensors[i].mData.data());
         b_tensors_device[i]->ToDevice(b_tensors[i].mData.data());
-	c_tensors_device[i]->SetZero();
+        c_tensors_device[i]->SetZero();
 #endif
 
         p_a.push_back(a_tensors_device[i]->GetDeviceBuffer());
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 26a4319eaa..467a8429ab 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -34,7 +34,8 @@ __global__ void
         kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
                                        const index_t group_count)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
     __shared__ uint8_t p_shared[shared_size];
 

From b8635a25b2ce87f70433b32e00858c6ae0f39fde Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 3 May 2023 16:27:04 -0700
Subject: [PATCH 011/194] Fix the group of quantization_int8 kernels on MI300.
 (#695)

* replace amd_buffer_atomic_add with hip_atomic_add

* fix grouped_gemm_splitk kernels on mi300

* fix syntax

* revert experimental atomic_add changes

* fix the group of kernels from ticket 723 on MI300

---------

Co-authored-by: Jing Zhang <jizhan@amd.com>
---
 .../device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp | 5 +++--
 .../gpu/device/impl/device_gemm_multiple_d_dl.hpp            | 5 +++--
 2 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 66b47abb62..5880f5f601 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -135,7 +135,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) || \
-    defined(__gfx90a__) || defined(__gfx908__))
+    defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -710,7 +710,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
 
         // check device
         if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
-             ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908"))
+             ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908" ||
+             ck::get_device_name() == "gfx940"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
index 6368469309..4397b6f99a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
@@ -51,7 +51,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx90a__) || defined(__gfx1030__))
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__))
 
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
@@ -552,7 +552,8 @@ struct DeviceGemmMultipleD_Dl : public DeviceGemmMultipleD<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
-           ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030")
+           ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030" ||
+           ck::get_device_name() == "gfx940")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_);

From b076a02ad25892192be9fefb08a53e89fa48160c Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Thu, 4 May 2023 10:25:47 -0500
Subject: [PATCH 012/194] Optimize bf16 conversion (#664)

* Add TypeConvert class and start refactoring

* Refactor TypeConvert as a struct

* Get back to template functions type_convert

* Add a type_convert_bf16_rtn, set rtz as default

* Clean up

* Add UnaryConvertPrecision struct for high-precision workloads

* Format

* Update type_convert to UnaryConvert on threadwise level

* Update UnaryConvertPrecision

* Format

* Fix chmod

* Add a flag to pick converion method

* Format

* Remove the added flag

* Merge elementwise op with type conversion

* Move type_convert to elemwise op, update the op

* Update type_convert_precision -> bf16_convert_rtn

* Clean up

* Update comments

* Update the CK_WORKAROUND_DENORM_FIX flag handling

* Update the unneeded op to work but warn user

* Remove the message

* Use a PassThrough instead of ConvertBF16RTN to calcaulate reference

* Format

* Add missing include
---
 include/ck/ck.hpp                             |  5 +-
 .../element/unary_element_wise_operation.hpp  | 23 +++++
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |  2 +-
 .../grid/gridwise_gemm_xdlops_bwd_weight.hpp  |  2 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |  2 +-
 .../threadwise_tensor_slice_transfer_v3r1.hpp | 27 ++----
 include/ck/utility/data_type.hpp              | 88 ++++++++++++-------
 .../cpu/reference_gemm.hpp                    | 23 ++++-
 8 files changed, 116 insertions(+), 56 deletions(-)

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index cb20ea2492..1626597ed2 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -175,7 +175,10 @@
 // denorm test fix, required to work around dissue
 #ifndef CK_WORKAROUND_DENORM_FIX
 #define CK_WORKAROUND_DENORM_FIX 0
-#endif
+#elif
+// enable only on MI200
+#define CK_WORKAROUND_DENORM_FIX = CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
+#endif // CK_WORKAROUND_DENORM_FIX
 
 namespace ck {
 
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 2987def02a..ef250b8bfd 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -56,6 +56,12 @@ struct PassThrough
         y = type_convert<bhalf_t>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<bhalf_t, half_t>(bhalf_t& y, const half_t& x) const
+    {
+        y = type_convert<bhalf_t>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<int8_t, int8_t>(int8_t& y, const int8_t& x) const
     {
@@ -86,6 +92,23 @@ struct UnaryConvert
     }
 };
 
+struct ConvertBF16RTN
+{
+    // convert to bf16 using round to nearest (rtn)
+    template <typename Y, typename X>
+    __host__ __device__ void operator()(Y& y, const X& x) const
+    {
+        // check Y datatype
+        static_assert(is_same<Y, bhalf_t>::value, "Data type is not supported by this operation!");
+
+        // check X datatype
+        static_assert(is_same<X, float>::value || is_same<X, half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        y = bf16_convert_rtn<Y>(x);
+    }
+};
+
 struct Scale
 {
     __host__ __device__ Scale(float scale) : scale_(scale) {}
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index 98a71a7c24..ec1cc53991 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -96,7 +96,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
     // ABDataTypeAdjusted -> ABDataType throughout this file
-#if CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
+#if CK_WORKAROUND_DENORM_FIX
     using ABDataTypeAdjusted =
         conditional_t<is_same_v<ABDataType, ck::half_t>, ck::bhalf_t, ABDataType>;
 #else
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index 1979331d07..da7ad1cacf 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -266,7 +266,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
     // FloatABAdjusted -> FloatAB throughout this file
-#if CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
+#if CK_WORKAROUND_DENORM_FIX
     using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
 #else
     using FloatABAdjusted = FloatAB;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index 775b77118c..f4504a9402 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -136,7 +136,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
     // FloatABAdjusted -> FloatAB throughout this file
-#if CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
+#if CK_WORKAROUND_DENORM_FIX
     using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
 #else
     using FloatABAdjusted = FloatAB;
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
index cba06f8e87..6665d765f8 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
@@ -6,6 +6,7 @@
 #include "ck/utility/common_header.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor/static_tensor.hpp"
 
@@ -207,15 +208,6 @@ struct ThreadwiseTensorSliceTransfer_v3r1
             auto src_vector_container = src_vector_type{
                 src_buf.template Get<src_vector_t>(src_coord_.GetOffset(), is_src_valid)};
 
-            // apply SrcElementwiseOperation on src_vector_container
-            static_for<0, SrcScalarPerVector, 1>{}([&](auto i) {
-                SrcData src_v;
-
-                src_element_op_(src_v, src_vector_container.template AsType<SrcData>()[i]);
-
-                src_vector_container.template AsType<SrcData>()(i) = src_v;
-            });
-
             // copy data from src_vector_container into src_thread_scratch_
             src_thread_scratch_tuple_(thread_scratch_id)
                 .template SetAsType<src_vector_t>(
@@ -318,7 +310,6 @@ struct ThreadwiseTensorSliceTransfer_v3r1
                 constexpr auto data_idx_seq = generate_sequence_v2(
                     [&](auto i) { return Number<data_idx[i]>{}; }, Number<nDim>{});
 
-                // TODO type_convert is not used yet!!!!!
                 using src_vector_t = vector_type_maker_t<SrcData, SrcScalarPerVector>;
                 using dst_vector_t = vector_type_maker_t<DstData, DstScalarPerVector>;
 
@@ -342,19 +333,17 @@ struct ThreadwiseTensorSliceTransfer_v3r1
                     Number<num_dst_vector>{});
 
                 // do data transpose
-                // TODO type_convert is not used yet!!!!!
                 transpose_vectors<SrcData, DstScalarPerVector, SrcScalarPerVector>{}(
                     src_vector_refs, dst_vector_refs);
             });
         }
-        else
-        {
-            static_ford<SliceLengths>{}([&](auto idx) {
-                // convert from SrcData to DstData here
-                dst_thread_scratch_(idx) =
-                    type_convert<DstData>(src_thread_scratch_tuple_[thread_scratch_id][idx]);
-            });
-        }
+
+        static_ford<SliceLengths>{}([&](auto idx) {
+            // apply the src elementwise op and convert to DstData under the hood if needed
+            DstData dst_v;
+            src_element_op_(dst_v, src_thread_scratch_tuple_[thread_scratch_id][idx]);
+            dst_thread_scratch_(idx) = dst_v;
+        });
 #endif
     }
 
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index 101061191e..d43af8a2e3 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -976,37 +976,6 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, float>(float
         uint32_t int32;
     } u = {x};
 
-    // When the exponent bits are not all 1s, then the value is zero, normal,
-    // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
-    // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
-    // This causes the bfloat16's mantissa to be incremented by 1 if the 16
-    // least significant bits of the float mantissa are greater than 0x8000,
-    // or if they are equal to 0x8000 and the least significant bit of the
-    // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
-    // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
-    // has the value 0x7f, then incrementing it causes it to become 0x00 and
-    // the exponent is incremented by one, which is the next higher FP value
-    // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
-    // with an exponent of 0x00 and a mantissa of 0x7f, it may be rounded up
-    // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
-    // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
-    // incrementing it causes it to become an exponent of 0xFF and a mantissa
-    // of 0x00, which is Inf, the next higher value to the unrounded value.
-    bool flag0 = ~u.int32 & 0x7f800000;
-
-    // When all of the exponent bits are 1, the value is Inf or NaN.
-    // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
-    // mantissa bit. Quiet NaN is indicated by the most significant mantissa
-    // bit being 1. Signaling NaN is indicated by the most significant
-    // mantissa bit being 0 but some other bit(s) being 1. If any of the
-    // lower 16 bits of the mantissa are 1, we set the least significant bit
-    // of the bfloat16 mantissa, in order to preserve signaling NaN in case
-    // the bfloat16's mantissa bits are all 0.
-    bool flag1 = !flag0 && (u.int32 & 0xffff);
-
-    u.int32 += flag0 ? 0x7fff + ((u.int32 >> 16) & 1) : 0; // Round to nearest, round to even
-    u.int32 |= flag1 ? 0x10000 : 0x0;                      // Preserve signaling NaN
-
     return uint16_t(u.int32 >> 16);
 }
 
@@ -1064,6 +1033,63 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_
     return type_convert<bhalf_t>(x_fp32);
 }
 
+// Declare a template function for bf16 conversion using RTN
+template <typename Y, typename X>
+__host__ __device__ constexpr Y bf16_convert_rtn(X x);
+
+// Convert fp32 to bf16 with RTN if higher precision is needed
+template <>
+inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, float>(float x)
+{
+    union
+    {
+        float fp32;
+        uint32_t int32;
+    } u = {x};
+
+    // When the exponent bits are not all 1s, then the value is zero, normal,
+    // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
+    // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
+    // This causes the bfloat16's mantissa to be incremented by 1 if the 16
+    // least significant bits of the float mantissa are greater than 0x8000,
+    // or if they are equal to 0x8000 and the least significant bit of the
+    // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
+    // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
+    // has the value 0x7f, then incrementing it causes it to become 0x00 and
+    // the exponent is incremented by one, which is the next higher FP value
+    // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
+    // with an exponent of 0x00 and a mantissa of 0x7f, it may be rounded up
+    // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
+    // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
+    // incrementing it causes it to become an exponent of 0xFF and a mantissa
+    // of 0x00, which is Inf, the next higher value to the unrounded value.
+    bool flag0 = ~u.int32 & 0x7f800000;
+
+    // When all of the exponent bits are 1, the value is Inf or NaN.
+    // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
+    // mantissa bit. Quiet NaN is indicated by the most significant mantissa
+    // bit being 1. Signaling NaN is indicated by the most significant
+    // mantissa bit being 0 but some other bit(s) being 1. If any of the
+    // lower 16 bits of the mantissa are 1, we set the least significant bit
+    // of the bfloat16 mantissa, in order to preserve signaling NaN in case
+    // the bfloat16's mantissa bits are all 0.
+    bool flag1 = !flag0 && (u.int32 & 0xffff);
+
+    u.int32 += flag0 ? 0x7fff + ((u.int32 >> 16) & 1) : 0; // Round to nearest, round to even
+    u.int32 |= flag1 ? 0x10000 : 0x0;                      // Preserve signaling NaN
+
+    return uint16_t(u.int32 >> 16);
+}
+
+// convert fp16 to bfp16 via fp32 with RTN if higher precision is needed
+template <>
+inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(half_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return bf16_convert_rtn<bhalf_t>(x_fp32);
+}
+
 template <typename T>
 struct NumericLimits
 {
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
index 6728bb1f47..be69f297b2 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
@@ -6,6 +6,7 @@
 #include <iostream>
 #include <sstream>
 
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/device/device_base.hpp"
 #include "ck/library/utility/host_tensor.hpp"
 
@@ -66,8 +67,26 @@ struct ReferenceGemm : public device::BaseOperator
                     ADataType v_a;
                     BDataType v_b;
 
-                    arg.a_element_op_(v_a, arg.a_m_k_(m, k));
-                    arg.b_element_op_(v_b, arg.b_k_n_(k, n));
+                    // use PassThrough instead of ConvertBF16RTN for reference calculation
+                    if constexpr(is_same_v<AElementwiseOperation,
+                                           ck::tensor_operation::element_wise::ConvertBF16RTN>)
+                    {
+                        ck::tensor_operation::element_wise::PassThrough{}(v_a, arg.a_m_k_(m, k));
+                    }
+                    else
+                    {
+                        arg.a_element_op_(v_a, arg.a_m_k_(m, k));
+                    }
+                    // same for B matrix
+                    if constexpr(is_same_v<BElementwiseOperation,
+                                           ck::tensor_operation::element_wise::ConvertBF16RTN>)
+                    {
+                        ck::tensor_operation::element_wise::PassThrough{}(v_b, arg.b_k_n_(k, n));
+                    }
+                    else
+                    {
+                        arg.b_element_op_(v_b, arg.b_k_n_(k, n));
+                    }
 
                     v_acc +=
                         ck::type_convert<AccDataType>(v_a) * ck::type_convert<AccDataType>(v_b);

From a1e344b1aef6e08298e4d8167aafdaa3a1f5f741 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Thu, 11 May 2023 20:15:02 +0800
Subject: [PATCH 013/194] Normalization/split k (#615)

---
 example/27_layernorm/CMakeLists.txt           |   3 +-
 example/27_layernorm/common.hpp               |  22 +
 example/27_layernorm/layernorm_fp16.cpp       |  39 ++
 .../27_layernorm/layernorm_splitk_fp16.cpp    |  40 ++
 ...lockwise.cpp => run_layernorm_example.inc} |  61 +-
 example/42_groupnorm/CMakeLists.txt           |   1 +
 example/42_groupnorm/common.hpp               |   1 +
 .../42_groupnorm/groupnorm_splitk_fp16.cpp    |  40 ++
 .../42_groupnorm/run_groupnorm_example.inc    |   4 +
 ...gemm_multiple_d_layernorm_xdl_cshuffle.hpp |   2 +-
 .../device/impl/device_normalization_impl.hpp |  49 +-
 .../impl/device_normalization_splitk_impl.hpp | 658 ++++++++++++++++++
 .../gridwise_normalization_naive_variance.hpp |   0
 .../gridwise_normalization_selector.hpp       |   4 +-
 .../gridwise_normalization_splitk_1st.hpp     | 252 +++++++
 .../gridwise_normalization_splitk_2nd.hpp     | 418 +++++++++++
 ...ridwise_normalization_welford_variance.hpp |   0
 17 files changed, 1513 insertions(+), 81 deletions(-)
 create mode 100644 example/27_layernorm/common.hpp
 create mode 100644 example/27_layernorm/layernorm_fp16.cpp
 create mode 100644 example/27_layernorm/layernorm_splitk_fp16.cpp
 rename example/27_layernorm/{layernorm_blockwise.cpp => run_layernorm_example.inc} (58%)
 create mode 100644 example/42_groupnorm/groupnorm_splitk_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
 rename include/ck/tensor_operation/gpu/grid/{ => normalization}/gridwise_normalization_naive_variance.hpp (100%)
 rename include/ck/tensor_operation/gpu/grid/{ => normalization}/gridwise_normalization_selector.hpp (98%)
 create mode 100644 include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
 rename include/ck/tensor_operation/gpu/grid/{ => normalization}/gridwise_normalization_welford_variance.hpp (100%)

diff --git a/example/27_layernorm/CMakeLists.txt b/example/27_layernorm/CMakeLists.txt
index d96deae45e..94c23ce774 100644
--- a/example/27_layernorm/CMakeLists.txt
+++ b/example/27_layernorm/CMakeLists.txt
@@ -1 +1,2 @@
-add_example_executable(example_layernorm_blockwise layernorm_blockwise.cpp)
+add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
+add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
diff --git a/example/27_layernorm/common.hpp b/example/27_layernorm/common.hpp
new file mode 100644
index 0000000000..8d833a3ae9
--- /dev/null
+++ b/example/27_layernorm/common.hpp
@@ -0,0 +1,22 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <getopt.h>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
diff --git a/example/27_layernorm/layernorm_fp16.cpp b/example/27_layernorm/layernorm_fp16.cpp
new file mode 100644
index 0000000000..c15ffabf50
--- /dev/null
+++ b/example/27_layernorm/layernorm_fp16.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+
+constexpr int Rank         = 2;
+constexpr int NumReduceDim = 1;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
+                                                          GammaDataType,
+                                                          BetaDataType,
+                                                          ComputeDataType,
+                                                          YDataType,
+                                                          PassThrough,
+                                                          Rank,
+                                                          NumReduceDim,
+                                                          256, // BlockSize
+                                                          8,   // ClusterM
+                                                          32,  // ClusterK
+                                                          1,   // SliceM
+                                                          8,   // SliceK
+                                                          1,   // XYVectorDim (0=M, 1=K)
+                                                          8,   // SrcScalarPerVector
+                                                          1,   // GammaVecDim (0=M, 1=K)
+                                                          8,   // GammaScalarPerVector
+                                                          1,   // BetaVecDim (0=M, 1=K)
+                                                          8,   // BetaScalarPerVector
+                                                          8>;  // OutScalarPerVector
+#include "run_layernorm_example.inc"
+
+int main() { return run_groupnorm_example<DeviceInstance>(); }
diff --git a/example/27_layernorm/layernorm_splitk_fp16.cpp b/example/27_layernorm/layernorm_splitk_fp16.cpp
new file mode 100644
index 0000000000..01ee7161eb
--- /dev/null
+++ b/example/27_layernorm/layernorm_splitk_fp16.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+
+constexpr int Rank         = 2;
+constexpr int NumReduceDim = 1;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
+                                                                GammaDataType,
+                                                                BetaDataType,
+                                                                ComputeDataType,
+                                                                YDataType,
+                                                                PassThrough,
+                                                                Rank,
+                                                                NumReduceDim,
+                                                                256, // BlockSize
+                                                                8,   // ClusterM
+                                                                32,  // ClusterK
+                                                                1,   // SliceM
+                                                                8,   // SliceK
+                                                                1,   // XYVectorDim (0=M, 1=K)
+                                                                8,   // XScalarPerVector
+                                                                1,   // GammaVecDim (0=M, 1=K)
+                                                                8,   // GammaScalarPerVector
+                                                                1,   // BetaVecDim (0=M, 1=K)
+                                                                8,   // BetaScalarPerVector
+                                                                8>;  // YScalarPerVector
+
+#include "run_layernorm_example.inc"
+
+int main() { return run_groupnorm_example<DeviceInstance>(); }
diff --git a/example/27_layernorm/layernorm_blockwise.cpp b/example/27_layernorm/run_layernorm_example.inc
similarity index 58%
rename from example/27_layernorm/layernorm_blockwise.cpp
rename to example/27_layernorm/run_layernorm_example.inc
index 7d91b69d04..678d8df281 100644
--- a/example/27_layernorm/layernorm_blockwise.cpp
+++ b/example/27_layernorm/run_layernorm_example.inc
@@ -1,58 +1,10 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-#include <getopt.h>
+#pragma once
 
-#include "ck/ck.hpp"
-#include "ck/utility/reduction_enums.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
-#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_common_util.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp"
-
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
-
-constexpr int Rank         = 2;
-constexpr int NumReduceDim = 1;
-
-using DeviceInstance =
-    ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
-                                                          GammaDataType,
-                                                          BetaDataType,
-                                                          ComputeDataType,
-                                                          YDataType,
-                                                          PassThrough,
-                                                          Rank,
-                                                          NumReduceDim,
-                                                          256, // BlockSize
-                                                          8,   // ClusterM
-                                                          32,  // ClusterK
-                                                          1,   // SliceM
-                                                          8,   // SliceK
-                                                          1,   // SrcVecDim (0=M, 1=K)
-                                                          8,   // SrcScalarPerVector
-                                                          1,   // GammaVecDim (0=M, 1=K)
-                                                          8,   // GammaScalarPerVector
-                                                          1,   // BetaVecDim (0=M, 1=K)
-                                                          8,   // BetaScalarPerVector
-                                                          8>;  // OutScalarPerVector
-
-int main()
+template <typename DeviceInstance>
+int run_groupnorm_example()
 {
     bool time_kernel = false;
 
@@ -111,6 +63,10 @@ int main()
         return 1;
     };
 
+    size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
+    DeviceMem workspace_dev(workspace_sz);
+    device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
     auto invoker_ptr = device_instance.MakeInvokerPointer();
     invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
@@ -133,7 +89,8 @@ int main()
         ref_invoker.Run(ref_argument);
 
         y_dev.FromDevice(y.mData.data());
-        pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results d1", 1e-3, 1e-3);
+        pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
     }
+
     return (pass ? 0 : 1);
 }
diff --git a/example/42_groupnorm/CMakeLists.txt b/example/42_groupnorm/CMakeLists.txt
index a9990c5d89..e8c306ac58 100644
--- a/example/42_groupnorm/CMakeLists.txt
+++ b/example/42_groupnorm/CMakeLists.txt
@@ -1,2 +1,3 @@
 add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
+add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
 add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
diff --git a/example/42_groupnorm/common.hpp b/example/42_groupnorm/common.hpp
index e159abf3e9..780154b26c 100644
--- a/example/42_groupnorm/common.hpp
+++ b/example/42_groupnorm/common.hpp
@@ -12,6 +12,7 @@
 #include "ck/ck.hpp"
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 
 #include "ck/library/utility/fill.hpp"
diff --git a/example/42_groupnorm/groupnorm_splitk_fp16.cpp b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
new file mode 100644
index 0000000000..fd4bfe3807
--- /dev/null
+++ b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+constexpr int Rank         = 5;
+constexpr int NumReduceDim = 3;
+
+using XDataType       = ck::half_t;
+using GammaDataType   = ck::half_t;
+using BetaDataType    = ck::half_t;
+using YDataType       = ck::half_t;
+using ComputeDataType = float;
+using YElementOp      = ck::tensor_operation::element_wise::Swish;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
+                                                                GammaDataType,
+                                                                BetaDataType,
+                                                                ComputeDataType,
+                                                                YDataType,
+                                                                YElementOp,
+                                                                Rank,
+                                                                NumReduceDim,
+                                                                256, // BlockSize
+                                                                1,   // ClusterM
+                                                                256, // ClusterK
+                                                                1,   // SliceM
+                                                                16,  // SliceK
+                                                                1,   // SrcVecDim (0=M, 1=K)
+                                                                2,   // SrcScalarPerVector
+                                                                1,   // GammaVecDim (0=M, 1=K)
+                                                                2,   // GammaScalarPerVector
+                                                                1,   // BetaVecDim (0=M, 1=K)
+                                                                2,   // BetaScalarPerVector
+                                                                2>;  // OutScalarPerVector
+
+#include "run_groupnorm_example.inc"
+
+int main(int argc, char* argv[]) { run_groupnorm_example(argc, argv); }
diff --git a/example/42_groupnorm/run_groupnorm_example.inc b/example/42_groupnorm/run_groupnorm_example.inc
index bd7eb98ca0..d1016a3b12 100644
--- a/example/42_groupnorm/run_groupnorm_example.inc
+++ b/example/42_groupnorm/run_groupnorm_example.inc
@@ -73,6 +73,10 @@ int run_groupnorm_example(int argc, char* argv[])
         return 1;
     };
 
+    size_t workspace_sz = device_instance.GetWorkSpaceSize(argument_ptr.get());
+    DeviceMem workspace_dev(workspace_sz);
+    device_instance.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
     auto invoker_ptr = device_instance.MakeInvokerPointer();
     float ave_time   = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true, true});
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
index a383b0bb7d..580087e00c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -807,7 +807,7 @@ struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
         // workspace for welford intermediate mean
         workspace_size += gemm_welford_size * sizeof(EMeanVarDataType) + 64;
 
-        // workspace for welford intermediate mean
+        // workspace for welford intermediate variance
         workspace_size += gemm_welford_size * sizeof(EMeanVarDataType) + 64;
 
         // workspace for welford intermediate count
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
index bb62332d1a..6a8037a324 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
@@ -10,8 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_normalization.hpp"
 #include "ck/tensor_operation/gpu/device/device_reduce.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_normalization_selector.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 
@@ -20,6 +19,10 @@ namespace tensor_operation {
 namespace device {
 
 // Y = Normalization(X, Beta, Gamma)
+// M: Invarient length
+// K: Reduce length (Calculate mean and variance along K dimension)
+// eg. Length = [N, C, H, W], reduce dim = [C, H, W]
+// Then, M = N, K = C * H * W
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
@@ -68,7 +71,6 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
     static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
                                     const std::vector<index_t>& inStrides,
-                                    int blkGroupSize,
                                     int numBlockTileIteration)
     {
         constexpr index_t NumInvariantDim  = Rank - NumReduceDim;
@@ -117,10 +119,9 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         const auto invariantLength = in_grid_desc_m_k.GetLength(Number<0>{});
         const auto reduceLength    = in_grid_desc_m_k.GetLength(Number<1>{});
 
-        const int reduceSizePerBlock = K_BlockTileSize * numBlockTileIteration;
         const auto inPad_M =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
-        const auto inPad_K = reduceSizePerBlock * blkGroupSize - reduceLength;
+        const auto inPad_K = K_BlockTileSize * numBlockTileIteration - reduceLength;
 
         auto in_grid_desc_m_k_padded = transform_tensor_descriptor(
             in_grid_desc_m_k,
@@ -132,7 +133,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         return (in_grid_desc_m_k_padded);
     };
 
-    using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1));
+    using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1));
 
     struct Argument : public BaseArgument
     {
@@ -162,26 +163,22 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
             gammaStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(gammaStrides, reduceDims);
             betaStrides_  = shuffle_tensor_dimensions<Rank, NumReduceDim>(betaStrides, reduceDims);
 
-            long_index_t invariant_total_length;
-            long_index_t reduce_total_length;
+            long_index_t invariant_length;
+            long_index_t reduce_length;
 
-            std::tie(invariant_total_length, reduce_total_length) =
+            std::tie(invariant_length, reduce_length) =
                 get_2d_lengths<Rank, NumReduceDim>(Lengths_);
 
-            blkGroupSize_          = 1;
-            numBlockTileIteration_ = (reduce_total_length + K_BlockTileSize - 1) / K_BlockTileSize;
+            numBlockTileIteration_ = math::integer_divide_ceil(reduce_length, K_BlockTileSize);
 
-            gridSize_ = math::integer_least_multiple(invariant_total_length, M_BlockTileSize) /
-                        M_BlockTileSize * blkGroupSize_;
+            gridSize_ = math::integer_divide_ceil(invariant_length, M_BlockTileSize);
 
-            x_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, xStrides_, blkGroupSize_, numBlockTileIteration_);
+            x_grid_desc_m_k_ = MakeSrc2dDescriptor(Lengths_, xStrides_, numBlockTileIteration_);
             gamma_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, gammaStrides_, blkGroupSize_, numBlockTileIteration_);
+                MakeSrc2dDescriptor(Lengths_, gammaStrides_, numBlockTileIteration_);
             beta_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, betaStrides_, blkGroupSize_, numBlockTileIteration_);
-            y_grid_desc_m_k_ =
-                MakeSrc2dDescriptor(Lengths_, yStrides_, blkGroupSize_, numBlockTileIteration_);
+                MakeSrc2dDescriptor(Lengths_, betaStrides_, numBlockTileIteration_);
+            y_grid_desc_m_k_ = MakeSrc2dDescriptor(Lengths_, yStrides_, numBlockTileIteration_);
 
             isSweeponce_ =
                 x_grid_desc_m_k_.GetLength(Number<1>{}) <= KThreadClusterSize * KThreadSliceSize;
@@ -202,7 +199,6 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
         YElementwiseOperation y_elementwise_op_;
 
-        int blkGroupSize_;
         int numBlockTileIteration_;
         size_t gridSize_;
 
@@ -286,6 +282,9 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
                 if(p_arg_->invariant_lowest_length % XSrcVectorSize != 0)
                     return false;
+
+                if(p_arg_->invariant_lowest_length % YDstVectorSize != 0)
+                    return false;
             };
         }
         else
@@ -295,12 +294,12 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
 
             if(p_arg_->Lengths_[Rank - 1] % XSrcVectorSize != 0)
                 return false;
-        };
 
-        if(p_arg_->Lengths_[Rank - 1] % YDstVectorSize != 0)
-        {
-            return false;
-        }
+            if(p_arg_->Lengths_[Rank - 1] % YDstVectorSize != 0)
+            {
+                return false;
+            }
+        };
 
         // if fastest dim is not reduced
         if constexpr(GammaSrcVectorDim == 0)
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
new file mode 100644
index 0000000000..0026a87593
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
@@ -0,0 +1,658 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/reduction_operator.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/device/device_normalization.hpp"
+#include "ck/tensor_operation/gpu/device/device_reduce.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+template <typename GridwiseWelford,
+          typename XDataType,
+          typename MeanVarDataType,
+          typename ComputeDataType,
+          typename XGridDesc_M_K,
+          typename MeanVarGridDesc_M_KBlock>
+__global__ void
+kernel_normalizationSplitK1st(const XGridDesc_M_K x_grid_desc_m_k,
+                              const MeanVarGridDesc_M_KBlock mean_var_grid_desc_m_kblock,
+                              index_t num_k_block_tile_iteration,
+                              const XDataType* const __restrict__ p_x_global,
+                              MeanVarDataType* const __restrict__ p_welford_mean,
+                              MeanVarDataType* const __restrict__ p_welford_variance,
+                              int32_t* const __restrict__ p_welford_count)
+{
+    GridwiseWelford::Run(x_grid_desc_m_k,
+                         mean_var_grid_desc_m_kblock,
+                         num_k_block_tile_iteration,
+                         p_x_global,
+                         p_welford_mean,
+                         p_welford_variance,
+                         p_welford_count);
+};
+
+template <typename GridwiseWelfordNormalization,
+          typename MeanVarDataType,
+          typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename MeanVarGridDesc_M_KBlock,
+          typename CountGridDesc_M_KBlock,
+          typename XYGammaBetaGridDesc_M_K>
+__global__ void
+kernel_normalizationSplitK2nd(const MeanVarGridDesc_M_KBlock mean_var_grid_desc_m_kblock,
+                              const CountGridDesc_M_KBlock count_grid_desc_m_kblock,
+                              const XYGammaBetaGridDesc_M_K x_grid_desc_m_k,
+                              const XYGammaBetaGridDesc_M_K gamma_grid_desc_m_k,
+                              const XYGammaBetaGridDesc_M_K beta_grid_desc_m_k,
+                              const XYGammaBetaGridDesc_M_K y_grid_desc_m_k,
+                              index_t num_k_mean_var_count_iteration,
+                              index_t num_k_block_tile_iteration,
+                              index_t k_grid_size,
+                              ComputeDataType epsilon,
+                              const MeanVarDataType* const p_mean_global,
+                              const MeanVarDataType* const p_variance_global,
+                              const int32_t* const p_welford_count_global,
+                              const XDataType* const __restrict__ p_x_global,
+                              const GammaDataType* const __restrict__ p_gamma_global,
+                              const BetaDataType* const __restrict__ p_beta_global,
+                              YDataType* const __restrict__ p_y_global,
+                              const YElementwiseOperation y_elementwise_op)
+{
+    GridwiseWelfordNormalization::Run(mean_var_grid_desc_m_kblock,
+                                      count_grid_desc_m_kblock,
+                                      x_grid_desc_m_k,
+                                      gamma_grid_desc_m_k,
+                                      beta_grid_desc_m_k,
+                                      y_grid_desc_m_k,
+                                      num_k_mean_var_count_iteration,
+                                      num_k_block_tile_iteration,
+                                      k_grid_size,
+                                      epsilon,
+                                      p_mean_global,
+                                      p_variance_global,
+                                      p_welford_count_global,
+                                      p_x_global,
+                                      p_gamma_global,
+                                      p_beta_global,
+                                      p_y_global,
+                                      y_elementwise_op);
+};
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// Y = Normalization(X, Beta, Gamma)
+// M: Invarient length
+// K: Reduce length (Calculate mean and variance along K dimension)
+// eg. Length = [N, C, H, W], reduce dim = [C, H, W]
+// Then, M = N, K = C * H * W
+template <typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename ComputeDataType,
+          typename YDataType,
+          typename YElementwiseOperation,
+          index_t Rank,
+          index_t NumReduceDim,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XYVectorDim,
+          index_t XSrcVectorSize,
+          index_t GammaSrcVectorDim,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorDim,
+          index_t BetaSrcVectorSize,
+          index_t YDstVectorSize>
+struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
+                                                                  GammaDataType,
+                                                                  BetaDataType,
+                                                                  ComputeDataType,
+                                                                  YDataType,
+                                                                  YElementwiseOperation,
+                                                                  Rank,
+                                                                  NumReduceDim>
+{
+    using MeanVarDataType = ComputeDataType;
+
+    static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize);
+    static_assert(
+        ((GammaSrcVectorDim == 0 && MThreadSliceSize % GammaSrcVectorSize == 0) ||
+         (GammaSrcVectorDim == 1 && KThreadSliceSize % GammaSrcVectorSize == 0)),
+        "Invalid thread slice sizes and/or gamma vector sizes configuration, please check!");
+
+    static_assert(
+        ((BetaSrcVectorDim == 0 && MThreadSliceSize % BetaSrcVectorSize == 0) ||
+         (BetaSrcVectorDim == 1 && KThreadSliceSize % BetaSrcVectorSize == 0)),
+        "Invalid thread slice sizes and/or beta vector sizes configuration, please check!");
+
+    using PassThrough = tensor_operation::element_wise::PassThrough;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
+                                    const std::vector<index_t>& inStrides,
+                                    int kBlockSize,
+                                    int numBlockTileIteration)
+    {
+        constexpr index_t NumInvariantDim  = Rank - NumReduceDim;
+        static constexpr index_t numSrcDim = Rank;
+        static constexpr bool reduceAllDim = (NumInvariantDim == 0);
+
+        const auto tupleSrcLengths = make_tuple_from_array(inLengths, Number<numSrcDim>{});
+        const auto tupleSrcStrides = make_tuple_from_array(inStrides, Number<numSrcDim>{});
+
+        const auto inDesc = make_naive_tensor_descriptor(tupleSrcLengths, tupleSrcStrides);
+
+        const auto in_grid_desc_m_k = [&]() {
+            if constexpr(reduceAllDim)
+            {
+                const auto one_dim_inDesc = transform_tensor_descriptor(
+                    inDesc,
+                    make_tuple(make_merge_transform(tupleSrcLengths)),
+                    make_tuple(typename arithmetic_sequence_gen<0, numSrcDim, 1>::type{}),
+                    make_tuple(Sequence<0>{}));
+
+                return transform_tensor_descriptor(one_dim_inDesc,
+                                                   make_tuple(make_unmerge_transform(make_tuple(
+                                                       1, one_dim_inDesc.GetLength(Number<0>{})))),
+                                                   make_tuple(Sequence<0>{}),
+                                                   make_tuple(Sequence<0, 1>{}));
+            }
+            else
+            {
+                using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+                using ReduceDims = typename arithmetic_sequence_gen<NumInvariantDim, Rank, 1>::type;
+
+                const auto reduceDimLengths =
+                    make_tuple_from_array_and_index_seq(inLengths, ReduceDims{});
+                const auto invariantDimLengths =
+                    make_tuple_from_array_and_index_seq(inLengths, InvariantDims{});
+
+                return transform_tensor_descriptor(
+                    inDesc,
+                    make_tuple(make_merge_transform(invariantDimLengths),
+                               make_merge_transform(reduceDimLengths)),
+                    make_tuple(InvariantDims{}, ReduceDims{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
+            }
+        }();
+
+        const auto invariantLength = in_grid_desc_m_k.GetLength(Number<0>{});
+        const auto reduceLength    = in_grid_desc_m_k.GetLength(Number<1>{});
+
+        const int reduceSizePerBlock = K_BlockTileSize * numBlockTileIteration;
+        const auto inPad_M =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto inPad_K = reduceSizePerBlock * kBlockSize - reduceLength;
+
+        auto in_grid_desc_m_k_padded = transform_tensor_descriptor(
+            in_grid_desc_m_k,
+            make_tuple(make_right_pad_transform(invariantLength, inPad_M),
+                       make_right_pad_transform(reduceLength, inPad_K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (in_grid_desc_m_k_padded);
+    };
+
+    template <typename DoPads, index_t MPerTile, index_t KPerTile>
+    static auto MakeMeanVarDescriptor_M_K(index_t M, index_t K)
+    {
+        const auto grid_desc_m_k =
+            make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(K, I1));
+        return PadTensorDescriptor(grid_desc_m_k, make_tuple(MPerTile, KPerTile), DoPads{});
+    }
+
+    template <typename DoPads, index_t MPerTile, index_t KPerTile>
+    static auto MakeCountDescriptor_M_K(index_t M, index_t K)
+    {
+        const auto grid_desc_m_k =
+            make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I0, I1));
+        return PadTensorDescriptor(grid_desc_m_k, make_tuple(MPerTile, KPerTile), DoPads{});
+    }
+
+    using SrcGridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1));
+    using Kernel1MeanVarGridDesc_M_KBlock =
+        decltype(MakeMeanVarDescriptor_M_K<Sequence<true, false>, 1, 1>(1, 1));
+
+    using Kernel2MeanVarGridDesc_M_KBlock =
+        decltype(MakeMeanVarDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+
+    using Kernel2CountGridDesc_M_KBlock =
+        decltype(MakeCountDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+
+    using GridwiseWelford = GridwiseNormalizationSplitK1st<XDataType,
+                                                           ComputeDataType,
+                                                           MeanVarDataType,
+                                                           SrcGridDesc_M_K,
+                                                           Kernel1MeanVarGridDesc_M_KBlock,
+                                                           BlockSize,
+                                                           MThreadClusterSize,
+                                                           KThreadClusterSize,
+                                                           MThreadSliceSize,
+                                                           KThreadSliceSize,
+                                                           XYVectorDim,
+                                                           XSrcVectorSize>;
+
+    using GridwiseWelfordNormalization =
+        GridwiseNormalizationSplitK2nd<MeanVarDataType,
+                                       XDataType,
+                                       GammaDataType,
+                                       BetaDataType,
+                                       YDataType,
+                                       ComputeDataType,
+                                       YElementwiseOperation,
+                                       Kernel2MeanVarGridDesc_M_KBlock,
+                                       Kernel2CountGridDesc_M_KBlock,
+                                       SrcGridDesc_M_K,
+                                       BlockSize,
+                                       MThreadClusterSize,
+                                       KThreadClusterSize,
+                                       MThreadSliceSize,
+                                       KThreadSliceSize,
+                                       XYVectorDim,
+                                       XSrcVectorSize,
+                                       GammaSrcVectorDim,
+                                       GammaSrcVectorSize,
+                                       BetaSrcVectorDim,
+                                       BetaSrcVectorSize,
+                                       XYVectorDim,
+                                       YDstVectorSize>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const std::vector<index_t> lengths,
+                 const std::vector<index_t> xStrides,
+                 const std::vector<index_t> gammaStrides,
+                 const std::vector<index_t> betaStrides,
+                 const std::vector<index_t> yStrides,
+                 const std::vector<index_t> reduceDims,
+                 YElementwiseOperation y_elementwise_op,
+                 double epsilon,
+                 const XDataType* p_x,
+                 const GammaDataType* p_gamma,
+                 const BetaDataType* p_beta,
+                 YDataType* p_y)
+            : p_x_(p_x),
+              p_gamma_(p_gamma),
+              p_beta_(p_beta),
+              p_y_(p_y),
+              p_workspace_mean_{nullptr},
+              p_workspace_var_{nullptr},
+              p_workspace_count_{nullptr},
+              y_elementwise_op_(y_elementwise_op)
+        {
+            epsilon_ = static_cast<ComputeDataType>(epsilon);
+
+            Lengths_      = shuffle_tensor_dimensions<Rank, NumReduceDim>(lengths, reduceDims);
+            xStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(xStrides, reduceDims);
+            yStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(yStrides, reduceDims);
+            gammaStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(gammaStrides, reduceDims);
+            betaStrides_  = shuffle_tensor_dimensions<Rank, NumReduceDim>(betaStrides, reduceDims);
+
+            std::tie(MRaw_, KRaw_) = get_2d_lengths<Rank, NumReduceDim>(Lengths_);
+
+            numBlockTileIteration_ = 1;
+            while(true)
+            {
+                int testKGridSize =
+                    math::integer_divide_ceil(KRaw_, K_BlockTileSize * numBlockTileIteration_);
+
+                // we want the kGridSize_ be not more than 128
+                if(testKGridSize <= 128)
+                    break;
+
+                ++numBlockTileIteration_;
+            };
+
+            kGridSize_ = math::integer_divide_ceil(KRaw_, K_BlockTileSize * numBlockTileIteration_);
+            gridSize_  = math::integer_divide_ceil(MRaw_, M_BlockTileSize) * kGridSize_;
+
+            // We do not use vector load for mean, var and count
+            static constexpr index_t K_MeanVarCountBlockTileSize = KThreadClusterSize;
+
+            numMeanVarCountIteration_ =
+                math::integer_divide_ceil(kGridSize_, K_MeanVarCountBlockTileSize);
+
+            x_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, xStrides_, kGridSize_, numBlockTileIteration_);
+            gamma_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, gammaStrides_, kGridSize_, numBlockTileIteration_);
+            beta_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, betaStrides_, kGridSize_, numBlockTileIteration_);
+            y_grid_desc_m_k_ =
+                MakeSrc2dDescriptor(Lengths_, yStrides_, kGridSize_, numBlockTileIteration_);
+
+            // We don't need to pad in K dimension for Welford1. Set KPerTile 1.
+            kernel1_mean_var_grid_desc_m_kblock_ =
+                MakeMeanVarDescriptor_M_K<Sequence<true, false>, M_BlockTileSize, 1>(MRaw_,
+                                                                                     kGridSize_);
+
+            kernel2_mean_var_grid_desc_m_kblock_ =
+                MakeMeanVarDescriptor_M_K<Sequence<true, true>,
+                                          M_BlockTileSize,
+                                          K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+
+            kernel2_count_grid_desc_m_kblock_ =
+                MakeCountDescriptor_M_K<Sequence<true, true>,
+                                        M_BlockTileSize,
+                                        K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+        }
+
+        ComputeDataType epsilon_;
+
+        const XDataType* p_x_;
+        const GammaDataType* p_gamma_;
+        const BetaDataType* p_beta_;
+        YDataType* p_y_;
+        void* p_workspace_mean_;
+        void* p_workspace_var_;
+        void* p_workspace_count_;
+
+        std::vector<index_t> Lengths_;
+        std::vector<index_t> xStrides_;
+        std::vector<index_t> gammaStrides_;
+        std::vector<index_t> betaStrides_;
+        std::vector<index_t> yStrides_;
+
+        YElementwiseOperation y_elementwise_op_;
+
+        int kGridSize_;
+        int numMeanVarCountIteration_;
+        int numBlockTileIteration_;
+        size_t gridSize_;
+
+        SrcGridDesc_M_K x_grid_desc_m_k_;
+        SrcGridDesc_M_K gamma_grid_desc_m_k_;
+        SrcGridDesc_M_K beta_grid_desc_m_k_;
+        SrcGridDesc_M_K y_grid_desc_m_k_;
+
+        Kernel1MeanVarGridDesc_M_KBlock kernel1_mean_var_grid_desc_m_kblock_;
+        Kernel2MeanVarGridDesc_M_KBlock kernel2_mean_var_grid_desc_m_kblock_;
+        Kernel2CountGridDesc_M_KBlock kernel2_count_grid_desc_m_kblock_;
+
+        index_t MRaw_; // invarient length
+        index_t KRaw_; // reduce length
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(arg.p_workspace_mean_ == nullptr || arg.p_workspace_var_ == nullptr ||
+               arg.p_workspace_count_ == nullptr)
+                throw std::runtime_error("wrong! WorkSpace pointer has not been set");
+
+            auto kernel1 = kernel_normalizationSplitK1st<GridwiseWelford,
+                                                         XDataType,
+                                                         MeanVarDataType,
+                                                         ComputeDataType,
+                                                         SrcGridDesc_M_K,
+                                                         Kernel1MeanVarGridDesc_M_KBlock>;
+
+            auto kernel2 = kernel_normalizationSplitK2nd<GridwiseWelfordNormalization,
+                                                         MeanVarDataType,
+                                                         XDataType,
+                                                         GammaDataType,
+                                                         BetaDataType,
+                                                         YDataType,
+                                                         ComputeDataType,
+                                                         YElementwiseOperation,
+                                                         Kernel2MeanVarGridDesc_M_KBlock,
+                                                         Kernel2CountGridDesc_M_KBlock,
+                                                         SrcGridDesc_M_K>;
+
+            float avg_time = 0;
+            avg_time += launch_and_time_kernel(stream_config,
+                                               kernel1,
+                                               dim3(arg.gridSize_),
+                                               dim3(BlockSize),
+                                               0,
+                                               arg.x_grid_desc_m_k_,
+                                               arg.kernel1_mean_var_grid_desc_m_kblock_,
+                                               arg.numBlockTileIteration_,
+                                               arg.p_x_,
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_mean_),
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_var_),
+                                               static_cast<int32_t*>(arg.p_workspace_count_));
+
+            avg_time += launch_and_time_kernel(stream_config,
+                                               kernel2,
+                                               dim3(arg.gridSize_),
+                                               dim3(BlockSize),
+                                               0,
+                                               arg.kernel2_mean_var_grid_desc_m_kblock_,
+                                               arg.kernel2_count_grid_desc_m_kblock_,
+                                               arg.x_grid_desc_m_k_,
+                                               arg.gamma_grid_desc_m_k_,
+                                               arg.beta_grid_desc_m_k_,
+                                               arg.y_grid_desc_m_k_,
+                                               arg.numMeanVarCountIteration_,
+                                               arg.numBlockTileIteration_,
+                                               arg.kGridSize_,
+                                               arg.epsilon_,
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_mean_),
+                                               static_cast<MeanVarDataType*>(arg.p_workspace_var_),
+                                               static_cast<int32_t*>(arg.p_workspace_count_),
+                                               arg.p_x_,
+                                               arg.p_gamma_,
+                                               arg.p_beta_,
+                                               arg.p_y_,
+                                               arg.y_elementwise_op_);
+
+            return avg_time;
+        };
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        };
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        size_t workspace_size = 0;
+
+        int welford_size = pArg_->MRaw_ * pArg_->kGridSize_;
+
+        // workspace for welford intermediate mean
+        workspace_size += welford_size * sizeof(MeanVarDataType) + 64;
+
+        // workspace for welford intermediate variance
+        workspace_size += welford_size * sizeof(MeanVarDataType) + 64;
+
+        // workspace for welford intermediate count
+        workspace_size += pArg_->kGridSize_ * sizeof(int32_t) + 64;
+
+        return (workspace_size);
+    };
+
+    void SetWorkSpacePointer(BaseArgument* pArg, void* p_workspace) const override
+    {
+        Argument* pArg_ = dynamic_cast<Argument*>(pArg);
+
+        pArg_->p_workspace_ = p_workspace;
+
+        int welford_size = pArg_->MRaw_ * pArg_->kGridSize_;
+
+        // setup buffer used for intermediate welford mean
+        pArg_->p_workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
+
+        index_t mean_space_sz = welford_size * sizeof(MeanVarDataType);
+        mean_space_sz         = math::integer_least_multiple(mean_space_sz, 64);
+
+        // setup buffer used for intermediate welford varirance
+        pArg_->p_workspace_var_ = reinterpret_cast<char*>(pArg_->p_workspace_mean_) + mean_space_sz;
+
+        index_t variance_space_sz = welford_size * sizeof(MeanVarDataType);
+        variance_space_sz         = math::integer_least_multiple(variance_space_sz, 64);
+
+        // setup buffer used for intermediate welford count
+        pArg_->p_workspace_count_ =
+            reinterpret_cast<char*>(pArg_->p_workspace_var_) + variance_space_sz;
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* p_arg_ = dynamic_cast<const Argument*>(p_arg);
+
+        constexpr index_t NumInvariantDim = Rank - NumReduceDim;
+
+        if constexpr(XYVectorDim == 0)
+        {
+            if constexpr(NumInvariantDim == 0)
+            {
+                return false;
+            }
+            else
+            {
+                if(p_arg_->xStrides_[NumInvariantDim - 1] != 1)
+                    return false;
+
+                if(p_arg_->invariant_lowest_length % XSrcVectorSize != 0)
+                    return false;
+
+                if(p_arg_->invariant_lowest_length % YDstVectorSize != 0)
+                    return false;
+            };
+        }
+        else
+        {
+            if(p_arg_->xStrides_[Rank - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % XSrcVectorSize != 0)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % YDstVectorSize != 0)
+                return false;
+        };
+
+        // if fastest dim is not reduced
+        if constexpr(GammaSrcVectorDim == 0)
+        {
+            if(p_arg_->gammaStrides_[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % GammaSrcVectorSize != 0)
+                return false;
+        }
+        else // if fastest dim is reduced
+        {
+            if(p_arg_->gammaStrides_[Rank - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % GammaSrcVectorSize != 0)
+                return false;
+        }
+
+        // if fastest dim is not reduced
+        if constexpr(BetaSrcVectorDim == 0)
+        {
+            if(p_arg_->betaStrides_[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(p_arg_->invariant_lowest_length % BetaSrcVectorSize != 0)
+                return false;
+        }
+        else // if fastest dim is reduced
+        {
+            if(p_arg_->betaStrides_[Rank - 1] != 1)
+                return false;
+
+            if(p_arg_->Lengths_[Rank - 1] % BetaSrcVectorSize != 0)
+                return false;
+        }
+
+        if(p_arg_->kGridSize_ <= 1)
+            return false;
+
+        return true;
+    };
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const std::vector<index_t> lengths,
+                        const std::vector<index_t> xStrides,
+                        const std::vector<index_t> gammaStrides,
+                        const std::vector<index_t> betaStrides,
+                        const std::vector<index_t> yStrides,
+                        const std::vector<index_t> reduceDims,
+                        double epsilon,
+                        const void* p_x,
+                        const void* p_gamma,
+                        const void* p_beta,
+                        void* p_y,
+                        void* p_saveMean,
+                        void* p_saveInvVar,
+                        YElementwiseOperation y_elementwise_op) override
+    {
+        // TODO
+        // Optional cache of the intermediate results (mean and InvVariance) during the
+        // forward pass could speedup in the backward
+        ignore = p_saveMean;
+        ignore = p_saveInvVar;
+
+        return std::make_unique<Argument>(lengths,
+                                          xStrides,
+                                          gammaStrides,
+                                          betaStrides,
+                                          yStrides,
+                                          reduceDims,
+                                          y_elementwise_op,
+                                          epsilon,
+                                          static_cast<const XDataType*>(p_x),
+                                          static_cast<const GammaDataType*>(p_gamma),
+                                          static_cast<const BetaDataType*>(p_beta),
+                                          static_cast<YDataType*>(p_y));
+    };
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>();
+    };
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceNormalizationSplitKImpl<" << BlockSize << ",";
+        str << "Cluster_MK_" << MThreadClusterSize << "_" << KThreadClusterSize << ",";
+        str << "Slice_MK_" << MThreadSliceSize << "_" << KThreadSliceSize << ",";
+        str << "XYSrcVectorDim_" << XYVectorDim  << ",";
+        str << "VectorSize_X" << XSrcVectorSize << "_Gamma" << GammaSrcVectorSize << "_Beta" << BetaSrcVectorSize << "_Y" << YDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_naive_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/grid/gridwise_normalization_naive_variance.hpp
rename to include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_selector.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
similarity index 98%
rename from include/ck/tensor_operation/gpu/grid/gridwise_normalization_selector.hpp
rename to include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
index 37795fa569..632690e1ef 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include "ck/tensor_operation/gpu/grid/gridwise_normalization_naive_variance.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp"
+#include "ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp"
 
 namespace ck {
 template <typename GridwiseReduction,
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
new file mode 100644
index 0000000000..129b4e1165
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
@@ -0,0 +1,252 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/math.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename XDataType,
+          typename ComputeDataType,
+          typename MeanVarDataType,
+          typename XGridDesc_M_K,
+          typename MeanVarGridDesc_M_KBlock,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize>
+struct GridwiseNormalizationSplitK1st
+{
+    static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_K                = Sequence<MThreadSliceSize, XSrcVectorSize>;
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
+
+    using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+    static constexpr auto thread_buffer_desc_m_1 =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, I1));
+
+    using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})));
+    using ThreadReduceDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadwiseWelford =
+        ThreadwiseWelford<ComputeDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
+
+    using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
+                                              BlockSize,
+                                              ThreadClusterLengths_M_K,
+                                              ThreadClusterArrangeOrder,
+                                              false>;
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    static constexpr index_t M_BlockTileSize     = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize     = KThreadClusterSize * KThreadSliceSize;
+    static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize;
+
+    static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
+
+    __device__ static int
+    GetKPerThread(int kRaw, int kGridSize, int block_k_cluster_id, int thread_k_cluster_id)
+    {
+        bool is_rightmost_block = block_k_cluster_id == kGridSize - 1;
+
+        if(is_rightmost_block)
+        {
+            int left_kPerBlock = math::integer_divide_ceil(kRaw, kGridSize);
+            int kPerBlock      = kRaw % kGridSize == 0 ? left_kPerBlock : kRaw % left_kPerBlock;
+            int kPerThread =
+                kPerBlock < K_BlockTileSize ? 0 : KThreadSliceSize * (kPerBlock / K_BlockTileSize);
+            int kPerBlockTail = kPerBlock - kPerThread * KThreadClusterSize;
+
+            if(kPerBlockTail > 0)
+            {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                    int thread_max_len =
+                        (thread_k_cluster_id + 1) * XSrcVectorSize + K_BlockTileStepSize * i;
+                    int delta = thread_max_len - kPerBlockTail;
+                    delta     = math::clamp(thread_max_len - kPerBlockTail, 0, XSrcVectorSize);
+                    kPerThread += XSrcVectorSize - delta;
+                });
+            }
+
+            return kPerThread;
+        }
+        else
+        {
+            int kPerBlock = math::integer_divide_ceil(kRaw, kGridSize);
+            return KThreadSliceSize * (kPerBlock / K_BlockTileSize);
+        }
+    }
+
+    // Calculate mean and variance by welford along k dimension
+    __device__ static void Run(const XGridDesc_M_K& x_grid_desc_m_k,
+                               const MeanVarGridDesc_M_KBlock& mean_var_grid_desc_m_kblock,
+                               index_t num_k_block_tile_iteration,
+                               const XDataType* const __restrict__ p_x_global,
+                               MeanVarDataType* const p_mean_global,
+                               MeanVarDataType* const p_variance_global,
+                               int32_t* const p_welford_count_global)
+    {
+        auto x_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * XSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            var_thread_buf;
+
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+
+        const index_t k_grid_size        = mean_var_grid_desc_m_kblock.GetLength(I1);
+        const index_t block_m_cluster_id = block_global_id / k_grid_size;
+        const index_t block_k_cluster_id = block_global_id % k_grid_size;
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        const index_t reduceSizePerBlock = K_BlockTileSize * num_k_block_tile_iteration;
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  ComputeDataType,
+                                                                  XGridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(
+                block_m_cluster_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                block_k_cluster_id * reduceSizePerBlock + thread_k_cluster_id * XSrcVectorSize));
+
+        auto mean_var_count_store_index = make_multi_index(
+            block_m_cluster_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+            block_k_cluster_id);
+
+        auto threadwise_welford_mean_var_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               MeanVarDataType,
+                                               decltype(thread_buffer_desc_m_1),
+                                               MeanVarGridDesc_M_KBlock,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M_1,
+                                               Sequence<0, 1>,
+                                               1,
+                                               1,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                mean_var_grid_desc_m_kblock, mean_var_count_store_index, PassThroughOp{});
+
+        constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
+
+        auto mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_mean_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        auto var_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_variance_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        auto threadwise_welford = ThreadwiseWelford();
+        int kRaw                = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
+        threadwise_welford.max_count_ =
+            GetKPerThread(kRaw, k_grid_size, block_k_cluster_id, thread_k_cluster_id);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
+            var_thread_buf(I)  = type_convert<ComputeDataType>(0.0f);
+        });
+
+        for(index_t k = 0; k < num_k_block_tile_iteration; ++k)
+        {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_x_load.Run(x_grid_desc_m_k,
+                                      x_global_val_buf,
+                                      thread_buffer_desc_m_k,
+                                      make_tuple(I0, I0),
+                                      x_thread_buf(i));
+                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+                threadwise_welford.Run(x_thread_buf[i], mean_thread_buf, var_thread_buf);
+            });
+        }
+
+        int welford_count = 0;
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            int count = threadwise_welford.cur_count_;
+            BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
+
+            // The value of count is same for all I
+            if constexpr(I == MThreadSliceSize - 1)
+                welford_count = count;
+        });
+
+        if(thread_k_cluster_id == 0)
+        {
+            threadwise_welford_mean_var_store.Run(thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  mean_thread_buf,
+                                                  mean_var_grid_desc_m_kblock,
+                                                  mean_global_val_buf);
+
+            threadwise_welford_mean_var_store.Run(thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  var_thread_buf,
+                                                  mean_var_grid_desc_m_kblock,
+                                                  var_global_val_buf);
+
+            if(block_m_cluster_id == 0 && thread_m_cluster_id == 0)
+                p_welford_count_global[block_k_cluster_id] = welford_count;
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
new file mode 100644
index 0000000000..d796d1afc0
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
@@ -0,0 +1,418 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/math.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename MeanVarDataType,
+          typename XDataType,
+          typename GammaDataType,
+          typename BetaDataType,
+          typename YDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
+          typename MeanVarGridDesc_M_KBlock,
+          typename CountGridDesc_M_KBlock,
+          typename XYGammaBetaGridDesc_M_K,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcVectorDim,
+          index_t XSrcVectorSize,
+          index_t GammaSrcVectorDim,
+          index_t GammaSrcVectorSize,
+          index_t BetaSrcVectorDim,
+          index_t BetaSrcVectorSize,
+          index_t YDstVectorDim,
+          index_t YDstVectorSize>
+struct GridwiseNormalizationSplitK2nd
+{
+    static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert((YDstVectorDim == 0 && MThreadSliceSize % YDstVectorSize == 0) ||
+                      (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert(XSrcVectorSize == YDstVectorSize);
+    static_assert(XSrcVectorSize == GammaSrcVectorSize);
+    static_assert(XSrcVectorSize == BetaSrcVectorSize);
+
+    static constexpr bool reorder_thread_cluster = (XSrcVectorDim == 0);
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadBufferLengths_M_K                = Sequence<MThreadSliceSize, XSrcVectorSize>;
+    static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
+
+    using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+    static constexpr auto thread_buffer_desc_m_1 =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, I1));
+
+    using ThreadWelfordSrcDesc_M_1 = decltype(thread_buffer_desc_m_1);
+    using ThreadWelfordDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadwiseWelford =
+        ThreadwiseWelfordMerge<ComputeDataType, ThreadWelfordSrcDesc_M_1, ThreadWelfordDstDesc_M>;
+
+    using BlockwiseWelford = BlockwiseWelford<ComputeDataType,
+                                              BlockSize,
+                                              ThreadClusterLengths_M_K,
+                                              ThreadClusterArrangeOrder>;
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    static constexpr index_t M_BlockTileSize     = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize     = KThreadClusterSize * KThreadSliceSize;
+    static constexpr index_t K_BlockTileStepSize = KThreadClusterSize * XSrcVectorSize;
+
+    static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
+
+    __device__ static void Run(const MeanVarGridDesc_M_KBlock& mean_var_grid_desc_m_kblock,
+                               const CountGridDesc_M_KBlock& count_grid_desc_m_kblock,
+                               const XYGammaBetaGridDesc_M_K& x_grid_desc_m_k,
+                               const XYGammaBetaGridDesc_M_K& gamma_grid_desc_m_k,
+                               const XYGammaBetaGridDesc_M_K& beta_grid_desc_m_k,
+                               const XYGammaBetaGridDesc_M_K& y_grid_desc_m_k,
+                               index_t num_k_mean_var_count_iteration,
+                               index_t num_k_block_tile_iteration,
+                               index_t k_grid_size,
+                               ComputeDataType epsilon,
+                               const MeanVarDataType* const p_mean_global,
+                               const MeanVarDataType* const p_variance_global,
+                               const int32_t* const p_welford_count_global,
+                               const XDataType* const __restrict__ p_x_global,
+                               const GammaDataType* const __restrict__ p_gamma_global,
+                               const BetaDataType* const __restrict__ p_beta_global,
+                               YDataType* const __restrict__ p_y_global,
+                               const YElementwiseOperation y_elementwise_op)
+    {
+        // Thread/Block id
+        const index_t thread_local_id    = get_thread_local_1d_id();
+        const index_t block_global_id    = get_block_1d_id();
+        const index_t block_m_cluster_id = block_global_id / k_grid_size;
+        const index_t block_k_cluster_id = block_global_id % k_grid_size;
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        // Global Memory
+        const auto mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_mean_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        const auto var_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_variance_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
+
+        const auto welford_count_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_welford_count_global, count_grid_desc_m_kblock.GetElementSpaceSize());
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x_global, x_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto gamma_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_gamma_global, gamma_grid_desc_m_k.GetElementSpaceSize());
+
+        const auto beta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_beta_global, beta_grid_desc_m_k.GetElementSpaceSize());
+
+        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
+
+        // VGPR
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            in_mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            in_var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
+            in_welford_count_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
+            var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
+            welford_count_thread_buf;
+
+        auto x_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * XSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto gamma_thread_buf = generate_tuple(
+            [&](auto) {
+                return StaticBuffer<AddressSpaceEnum::Vgpr,
+                                    ComputeDataType,
+                                    MThreadSliceSize * GammaSrcVectorSize,
+                                    true>{};
+            },
+            Number<ThreadBufferNumber>{});
+
+        auto& beta_thread_buf = gamma_thread_buf;
+        auto& y_thread_buf    = x_thread_buf;
+
+        // IO
+        auto threadwise_mean_var_load_m_kblock =
+            ThreadwiseTensorSliceTransfer_v2<MeanVarDataType,
+                                             ComputeDataType,
+                                             MeanVarGridDesc_M_KBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_grid_desc_m_kblock,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id));
+
+        auto threadwise_count_load_m_kblock =
+            ThreadwiseTensorSliceTransfer_v2<int32_t,
+                                             int32_t,
+                                             CountGridDesc_M_KBlock,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             1,
+                                             1,
+                                             1,
+                                             true>(
+                count_grid_desc_m_kblock,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id));
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  ComputeDataType,
+                                                                  XYGammaBetaGridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize,
+                             block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                 thread_k_cluster_id * XSrcVectorSize));
+
+        auto threadwise_gamma_load =
+            ThreadwiseTensorSliceTransfer_v2<GammaDataType,
+                                             ComputeDataType,
+                                             XYGammaBetaGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             ThreadBufferDimAccessOrder,
+                                             GammaSrcVectorDim,
+                                             GammaSrcVectorSize,
+                                             1,
+                                             true>(
+                gamma_grid_desc_m_k,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                     thread_k_cluster_id * GammaSrcVectorSize));
+
+        auto threadwise_beta_load =
+            ThreadwiseTensorSliceTransfer_v2<BetaDataType,
+                                             ComputeDataType,
+                                             XYGammaBetaGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_k),
+                                             ThreadBufferLengths_M_K,
+                                             ThreadBufferDimAccessOrder,
+                                             BetaSrcVectorDim,
+                                             BetaSrcVectorSize,
+                                             1,
+                                             true>(
+                beta_grid_desc_m_k,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                     thread_k_cluster_id * BetaSrcVectorSize));
+
+        auto threadwise_y_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               YDataType,
+                                               decltype(thread_buffer_desc_m_k),
+                                               XYGammaBetaGridDesc_M_K,
+                                               YElementwiseOperation,
+                                               ThreadBufferLengths_M_K,
+                                               ThreadBufferDimAccessOrder,
+                                               YDstVectorDim,
+                                               YDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                y_grid_desc_m_k,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_k_cluster_id * K_BlockTileSize * num_k_block_tile_iteration +
+                                     thread_k_cluster_id * YDstVectorSize),
+                y_elementwise_op);
+
+        // step1: Merge mean and variance
+        constexpr auto mean_var_count_thread_copy_step_I0_k =
+            make_multi_index(I0, KThreadClusterSize);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I)          = type_convert<ComputeDataType>(0.0f);
+            var_thread_buf(I)           = type_convert<ComputeDataType>(0.0f);
+            welford_count_thread_buf(I) = 0;
+        });
+
+        for(index_t k = 0; k < num_k_mean_var_count_iteration; ++k)
+        {
+            threadwise_mean_var_load_m_kblock.Run(mean_var_grid_desc_m_kblock,
+                                                  mean_global_val_buf,
+                                                  thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  in_mean_thread_buf);
+
+            threadwise_mean_var_load_m_kblock.Run(mean_var_grid_desc_m_kblock,
+                                                  var_global_val_buf,
+                                                  thread_buffer_desc_m_1,
+                                                  make_tuple(I0, I0),
+                                                  in_var_thread_buf);
+
+            threadwise_count_load_m_kblock.Run(count_grid_desc_m_kblock,
+                                               welford_count_global_val_buf,
+                                               thread_buffer_desc_m_1,
+                                               make_tuple(I0, I0),
+                                               in_welford_count_thread_buf);
+
+            ThreadwiseWelford::Run(in_mean_thread_buf,
+                                   in_var_thread_buf,
+                                   in_welford_count_thread_buf,
+                                   mean_thread_buf,
+                                   var_thread_buf,
+                                   welford_count_thread_buf);
+
+            threadwise_mean_var_load_m_kblock.MoveSrcSliceWindow(
+                mean_var_grid_desc_m_kblock, mean_var_count_thread_copy_step_I0_k);
+            threadwise_count_load_m_kblock.MoveSrcSliceWindow(count_grid_desc_m_kblock,
+                                                              mean_var_count_thread_copy_step_I0_k);
+        }
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            BlockwiseWelford::Run(
+                mean_thread_buf(I), var_thread_buf(I), welford_count_thread_buf(I));
+        });
+
+        // step2: normalization
+        constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
+
+        for(index_t k = 0; k < num_k_block_tile_iteration; ++k)
+        {
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_x_load.Run(x_grid_desc_m_k,
+                                      x_global_val_buf,
+                                      thread_buffer_desc_m_k,
+                                      make_tuple(I0, I0),
+                                      x_thread_buf(i));
+                threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_gamma_load.Run(gamma_grid_desc_m_k,
+                                          gamma_global_val_buf,
+                                          thread_buffer_desc_m_k,
+                                          make_tuple(I0, I0),
+                                          gamma_thread_buf(i));
+
+                threadwise_gamma_load.MoveSrcSliceWindow(gamma_grid_desc_m_k,
+                                                         thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                        // normalize
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
+                            divisor;
+
+                        // gamma
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) *
+                            gamma_thread_buf(iK0)(Number<offset_m_k>{});
+                    });
+                });
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_beta_load.Run(beta_grid_desc_m_k,
+                                         beta_global_val_buf,
+                                         thread_buffer_desc_m_k,
+                                         make_tuple(I0, I0),
+                                         beta_thread_buf(i));
+                threadwise_beta_load.MoveSrcSliceWindow(beta_grid_desc_m_k,
+                                                        thread_copy_fwd_step_m_k);
+            });
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
+                    static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
+                        constexpr auto offset_m_k =
+                            thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK1));
+
+                        // beta
+                        y_thread_buf(iK0)(Number<offset_m_k>{}) =
+                            y_thread_buf(iK0)(Number<offset_m_k>{}) +
+                            beta_thread_buf(iK0)(Number<offset_m_k>{});
+                    });
+                });
+            });
+
+            static_for<0, ThreadBufferNumber, 1>{}([&](auto i) {
+                threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                       make_tuple(I0, I0),
+                                       y_thread_buf(i),
+                                       y_grid_desc_m_k,
+                                       y_global_val_buf);
+                threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, thread_copy_fwd_step_m_k);
+            });
+        } // end for (normalization)
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/grid/gridwise_normalization_welford_variance.hpp
rename to include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp

From 642d5e9155a16c96b01eee7b8ef0e9d558fc2e16 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <bartlomiejkocot98@gmail.com>
Date: Mon, 15 May 2023 16:46:52 +0200
Subject: [PATCH 014/194] Add contraction profiler and tests (#701)

* Add contraction profiler and tests

* Build and style fixes

* Allow to use any elementwise operator for ref_contraction

* Introduce profile_contraction_scale and profile_contraction_bilinear

* Make ref_contraction generic and extend interface tests

* Stylistic minor fixes

* Extend test_contraction_interface
---
 .../contraction_bilinear_xdl_fp32.cpp         | 164 +--------
 .../contraction_bilinear_xdl_fp64.cpp         | 164 +--------
 .../contraction_scale_xdl_fp32.cpp            | 165 +--------
 .../contraction_scale_xdl_fp64.cpp            | 165 +--------
 .../cpu/reference_contraction.hpp             | 146 ++++++++
 profiler/README.md                            |  30 ++
 .../profiler/profile_contraction_impl.hpp     | 345 ++++++++++++++++++
 .../profiler/profile_contraction_utils.hpp    |  51 +++
 profiler/src/CMakeLists.txt                   |   4 +
 profiler/src/profile_contraction_bilinear.cpp | 165 +++++++++
 profiler/src/profile_contraction_scale.cpp    | 162 ++++++++
 test/CMakeLists.txt                           |   1 +
 test/contraction/CMakeLists.txt               |   4 +
 test/contraction/test_contraction.cpp         | 138 +++++++
 .../test_contraction_interface.cpp            | 195 ++++++++++
 15 files changed, 1303 insertions(+), 596 deletions(-)
 create mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
 create mode 100644 profiler/include/profiler/profile_contraction_impl.hpp
 create mode 100644 profiler/include/profiler/profile_contraction_utils.hpp
 create mode 100644 profiler/src/profile_contraction_bilinear.cpp
 create mode 100644 profiler/src/profile_contraction_scale.cpp
 create mode 100644 test/contraction/CMakeLists.txt
 create mode 100644 test/contraction/test_contraction.cpp
 create mode 100644 test/contraction/test_contraction_interface.cpp

diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
index ea105e4ff2..6004db6e0e 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
@@ -16,6 +16,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -74,141 +75,6 @@ using DeviceOpInstanceMNNN = ck::tensor_operation::device::
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-// hardcoded for NumDimM == NumDimN == NumDimK == 2
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename EDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
-struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
-{
-    // Argument
-    struct Argument : public ck::tensor_operation::device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_ms_ks,
-                 const Tensor<BDataType>& b_ns_ks,
-                 Tensor<EDataType>& e_ms_ns,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : a_ms_ks_{a_ms_ks},
-              b_ns_ks_{b_ns_ks},
-              e_ms_ns_{e_ms_ns},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_ms_ks_;
-        const Tensor<BDataType>& b_ns_ks_;
-        Tensor<EDataType>& e_ms_ns_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public ck::tensor_operation::device::BaseInvoker
-    {
-        using Argument = ReferenceContraction_M2_N2_K2::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
-                const int K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
-                const int K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
-
-                AccDataType v_acc = 0;
-
-                for(int k0 = 0; k0 < K0; ++k0)
-                {
-                    for(int k1 = 0; k1 < K1; ++k1)
-                    {
-                        AccDataType v_a;
-                        AccDataType v_b;
-
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
-
-                        v_acc += v_a * v_b;
-                    }
-                }
-
-                AccDataType v_c;
-
-                arg.cde_element_op_(v_c, v_acc);
-
-                arg.e_ms_ns_(m0, m1, n0, n1) = v_c;
-            };
-
-            make_ParallelTensorFunctor(f_ms_ns,
-                                       arg.e_ms_ns_.mDesc.GetLengths()[0],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[1],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[2],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[3])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
-    {
-        return true;
-    }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
-                             const Tensor<BDataType>& b_ns_ks,
-                             Tensor<EDataType>& e_ms_ns,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{a_ms_ks, b_ns_ks, e_ms_ns, a_element_op, b_element_op, cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceContraction_M2_N2_K2"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
@@ -385,22 +251,22 @@ int main(int argc, char* argv[])
     {
         Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
 
-        using ReferenceOpInstance = ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                  NumDimN,
-                                                                  NumDimK,
-                                                                  ADataType,
-                                                                  BDataType,
-                                                                  CShuffleDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  PassThrough>;
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
 
-        auto ref_gemm    = ReferenceOpInstance{};
-        auto ref_invoker = ref_gemm.MakeInvoker();
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
 
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
index 9a000377bb..9576ce3f2c 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
@@ -16,6 +16,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -74,141 +75,6 @@ using DeviceOpInstanceMNNN = ck::tensor_operation::device::
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-// hardcoded for NumDimM == NumDimN == NumDimK == 2
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename EDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
-struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
-{
-    // Argument
-    struct Argument : public ck::tensor_operation::device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_ms_ks,
-                 const Tensor<BDataType>& b_ns_ks,
-                 Tensor<EDataType>& e_ms_ns,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : a_ms_ks_{a_ms_ks},
-              b_ns_ks_{b_ns_ks},
-              e_ms_ns_{e_ms_ns},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_ms_ks_;
-        const Tensor<BDataType>& b_ns_ks_;
-        Tensor<EDataType>& e_ms_ns_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public ck::tensor_operation::device::BaseInvoker
-    {
-        using Argument = ReferenceContraction_M2_N2_K2::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
-                const int K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
-                const int K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
-
-                AccDataType v_acc = 0;
-
-                for(int k0 = 0; k0 < K0; ++k0)
-                {
-                    for(int k1 = 0; k1 < K1; ++k1)
-                    {
-                        AccDataType v_a;
-                        AccDataType v_b;
-
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
-
-                        v_acc += v_a * v_b;
-                    }
-                }
-
-                AccDataType v_c;
-
-                arg.cde_element_op_(v_c, v_acc);
-
-                arg.e_ms_ns_(m0, m1, n0, n1) = v_c;
-            };
-
-            make_ParallelTensorFunctor(f_ms_ns,
-                                       arg.e_ms_ns_.mDesc.GetLengths()[0],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[1],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[2],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[3])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
-    {
-        return true;
-    }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
-                             const Tensor<BDataType>& b_ns_ks,
-                             Tensor<EDataType>& e_ms_ns,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{a_ms_ks, b_ns_ks, e_ms_ns, a_element_op, b_element_op, cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceContraction_M2_N2_K2"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
@@ -385,22 +251,22 @@ int main(int argc, char* argv[])
     {
         Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
 
-        using ReferenceOpInstance = ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                  NumDimN,
-                                                                  NumDimK,
-                                                                  ADataType,
-                                                                  BDataType,
-                                                                  CShuffleDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  PassThrough>;
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
 
-        auto ref_gemm    = ReferenceOpInstance{};
-        auto ref_invoker = ref_gemm.MakeInvoker();
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
 
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
diff --git a/example/26_contraction/contraction_scale_xdl_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp32.cpp
index 26f176b059..3aa2a7ba93 100644
--- a/example/26_contraction/contraction_scale_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp32.cpp
@@ -16,6 +16,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -73,141 +74,6 @@ using DeviceOpInstanceMNN = ck::tensor_operation::device::
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-// hardcoded for NumDimM == NumDimN == NumDimK == 2
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename EDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
-struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
-{
-    // Argument
-    struct Argument : public ck::tensor_operation::device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_ms_ks,
-                 const Tensor<BDataType>& b_ns_ks,
-                 Tensor<EDataType>& e_ms_ns,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : a_ms_ks_{a_ms_ks},
-              b_ns_ks_{b_ns_ks},
-              e_ms_ns_{e_ms_ns},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_ms_ks_;
-        const Tensor<BDataType>& b_ns_ks_;
-        Tensor<EDataType>& e_ms_ns_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public ck::tensor_operation::device::BaseInvoker
-    {
-        using Argument = ReferenceContraction_M2_N2_K2::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
-                const int K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
-                const int K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
-
-                AccDataType v_acc = 0;
-
-                for(int k0 = 0; k0 < K0; ++k0)
-                {
-                    for(int k1 = 0; k1 < K1; ++k1)
-                    {
-                        AccDataType v_a;
-                        AccDataType v_b;
-
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
-
-                        v_acc += v_a * v_b;
-                    }
-                }
-
-                AccDataType v_c;
-
-                arg.cde_element_op_(v_c, v_acc);
-
-                arg.e_ms_ns_(m0, m1, n0, n1) = v_c;
-            };
-
-            make_ParallelTensorFunctor(f_ms_ns,
-                                       arg.e_ms_ns_.mDesc.GetLengths()[0],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[1],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[2],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[3])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
-    {
-        return true;
-    }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
-                             const Tensor<BDataType>& b_ns_ks,
-                             Tensor<EDataType>& e_ms_ns,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{a_ms_ks, b_ns_ks, e_ms_ns, a_element_op, b_element_op, cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceContraction_M2_N2_K2"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
@@ -368,22 +234,23 @@ int main(int argc, char* argv[])
     {
         Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
 
-        using ReferenceOpInstance = ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                  NumDimN,
-                                                                  NumDimK,
-                                                                  ADataType,
-                                                                  BDataType,
-                                                                  CShuffleDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  PassThrough>;
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
 
-        auto ref_gemm    = ReferenceOpInstance{};
-        auto ref_invoker = ref_gemm.MakeInvoker();
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
 
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
diff --git a/example/26_contraction/contraction_scale_xdl_fp64.cpp b/example/26_contraction/contraction_scale_xdl_fp64.cpp
index 38ed60266d..cccf6505c6 100644
--- a/example/26_contraction/contraction_scale_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp64.cpp
@@ -16,6 +16,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -73,141 +74,6 @@ using DeviceOpInstanceMNN = ck::tensor_operation::device::
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-// hardcoded for NumDimM == NumDimN == NumDimK == 2
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename EDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
-struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
-{
-    // Argument
-    struct Argument : public ck::tensor_operation::device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_ms_ks,
-                 const Tensor<BDataType>& b_ns_ks,
-                 Tensor<EDataType>& e_ms_ns,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : a_ms_ks_{a_ms_ks},
-              b_ns_ks_{b_ns_ks},
-              e_ms_ns_{e_ms_ns},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_ms_ks_;
-        const Tensor<BDataType>& b_ns_ks_;
-        Tensor<EDataType>& e_ms_ns_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public ck::tensor_operation::device::BaseInvoker
-    {
-        using Argument = ReferenceContraction_M2_N2_K2::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
-                const int K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
-                const int K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
-
-                AccDataType v_acc = 0;
-
-                for(int k0 = 0; k0 < K0; ++k0)
-                {
-                    for(int k1 = 0; k1 < K1; ++k1)
-                    {
-                        AccDataType v_a;
-                        AccDataType v_b;
-
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
-
-                        v_acc += v_a * v_b;
-                    }
-                }
-
-                AccDataType v_c;
-
-                arg.cde_element_op_(v_c, v_acc);
-
-                arg.e_ms_ns_(m0, m1, n0, n1) = v_c;
-            };
-
-            make_ParallelTensorFunctor(f_ms_ns,
-                                       arg.e_ms_ns_.mDesc.GetLengths()[0],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[1],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[2],
-                                       arg.e_ms_ns_.mDesc.GetLengths()[3])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
-    {
-        return true;
-    }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
-                             const Tensor<BDataType>& b_ns_ks,
-                             Tensor<EDataType>& e_ms_ns,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{a_ms_ks, b_ns_ks, e_ms_ns, a_element_op, b_element_op, cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceContraction_M2_N2_K2"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
@@ -368,22 +234,23 @@ int main(int argc, char* argv[])
     {
         Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
 
-        using ReferenceOpInstance = ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                  NumDimN,
-                                                                  NumDimK,
-                                                                  ADataType,
-                                                                  BDataType,
-                                                                  CShuffleDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  PassThrough>;
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
 
-        auto ref_gemm    = ReferenceOpInstance{};
-        auto ref_invoker = ref_gemm.MakeInvoker();
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
 
-        auto ref_argument = ref_gemm.MakeArgument(
-            a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op, PassThrough{});
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
new file mode 100644
index 0000000000..92a8c82a6e
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+// hardcoded for NumDimM == NumDimN == NumDimK == 2
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
+struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::BaseOperator
+{
+    // Argument
+    struct Argument : public ck::tensor_operation::device::BaseArgument
+    {
+        Argument(const Tensor<ADataType>& a_ms_ks,
+                 const Tensor<BDataType>& b_ns_ks,
+                 Tensor<CDataType>& c_ms_ns,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op)
+            : a_ms_ks_{a_ms_ks},
+              b_ns_ks_{b_ns_ks},
+              c_ms_ns_{c_ms_ns},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op}
+        {
+        }
+
+        const Tensor<ADataType>& a_ms_ks_;
+        const Tensor<BDataType>& b_ns_ks_;
+        Tensor<CDataType>& c_ms_ns_;
+
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+    };
+
+    // Invoker
+    struct Invoker : public ck::tensor_operation::device::BaseInvoker
+    {
+        using Argument = ReferenceContraction_M2_N2_K2::Argument;
+
+        float Run(const Argument& arg)
+        {
+            auto f_ms_ns = [&](auto m0, auto m1, auto n0, auto n1) {
+                const ck::index_t K0 = arg.a_ms_ks_.mDesc.GetLengths()[2];
+                const ck::index_t K1 = arg.a_ms_ks_.mDesc.GetLengths()[3];
+
+                AccDataType v_acc = 0;
+
+                for(ck::index_t k0 = 0; k0 < K0; ++k0)
+                {
+                    for(ck::index_t k1 = 0; k1 < K1; ++k1)
+                    {
+                        AccDataType v_a;
+                        AccDataType v_b;
+
+                        arg.a_element_op_(
+                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
+                        arg.b_element_op_(
+                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
+
+                        v_acc += v_a * v_b;
+                    }
+                }
+
+                arg.c_ms_ns_(m0, m1, n0, n1) = v_acc;
+            };
+
+            make_ParallelTensorFunctor(f_ms_ns,
+                                       arg.c_ms_ns_.mDesc.GetLengths()[0],
+                                       arg.c_ms_ns_.mDesc.GetLengths()[1],
+                                       arg.c_ms_ns_.mDesc.GetLengths()[2],
+                                       arg.c_ms_ns_.mDesc.GetLengths()[3])(
+                std::thread::hardware_concurrency());
+
+            return 0;
+        }
+
+        float Run(const ck::tensor_operation::device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    bool IsSupportedArgument(const ck::tensor_operation::device::BaseArgument*) override
+    {
+        return true;
+    }
+
+    static auto MakeArgument(const Tensor<ADataType>& a_ms_ks,
+                             const Tensor<BDataType>& b_ns_ks,
+                             Tensor<CDataType>& c_ms_ns,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op)
+    {
+        return Argument{a_ms_ks, b_ns_ks, c_ms_ns, a_element_op, b_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<ck::tensor_operation::device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceContraction_M2_N2_K2"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/README.md b/profiler/README.md
index bfd6a3a53b..400a64a396 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -46,3 +46,33 @@ out_n_k_ho_wo: dim 4, lengths {128, 256, 36, 36}, strides {331776, 1, 9216, 256}
 ....
 Best Perf: 1.42509 ms, 102.988 TFlops, 234.086 GB/s
 ```
+
+## Profile contraction kernels
+```bash
+#arg1: tensor operation (contraction_bilinear=CONTRACTION+Bilinear)
+#arg2: data type (0: fp32; 1: f64)\n"
+#arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1])
+#arg4: verification (0: no; 1: yes)
+#arg5: initialization (0: no init; 1: integer value; 2: decimal value)
+#arg6: print tensor value (0: no; 1: yes)
+#arg7: time kernel (0: no, 1: yes)
+#arg8 and arg9: alpha and beta
+#arg10 to 15: M0, M1, N0, N1, K0, K1
+#arg16 to 31: Strides for A, B, D and E (skip for default)
+
+################                   op  datatype  layout  verify  init  log  time  alpha  beta  M0  M1  N0  N1  K0  K1
+./bin/ckProfiler contraction_bilinear         0       1       0     0    0     1    1.0   1.0 128 128 128 128 128 128
+```
+
+Result (MI100)
+```bash
+a_m_k: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
+b_k_n: dim 4, lengths {128, 128, 128, 128}, strides {128, 1, 2097152, 16384}
+d_m_n: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
+e_m_n: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
+....
+Best Perf: 211.405 ms, 41.6077 TFlops, 15.2372 GB/s
+```
diff --git a/profiler/include/profiler/profile_contraction_impl.hpp b/profiler/include/profiler/profile_contraction_impl.hpp
new file mode 100644
index 0000000000..660cc3f9e5
--- /dev/null
+++ b/profiler/include/profiler/profile_contraction_impl.hpp
@@ -0,0 +1,345 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+#include <iostream>
+#include <typeinfo>
+#include <limits>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction_scale.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+#include "ck/host_utility/io.hpp"
+
+namespace ck {
+namespace profiler {
+
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+using Scale    = ck::tensor_operation::element_wise::Scale;
+
+template <typename ALayout,
+          typename BLayout,
+          typename CDELayout,
+          typename DataType,
+          typename DTupleDataType,
+          typename CDElementOp>
+int profile_contraction_impl(ck::index_t do_verification,
+                             ck::index_t init_method,
+                             bool do_log,
+                             bool time_kernel,
+                             CDElementOp cde_element_op,
+                             const std::vector<ck::index_t>& M,
+                             const std::vector<ck::index_t>& N,
+                             const std::vector<ck::index_t>& K,
+                             const std::vector<ck::index_t>& StridesA,
+                             const std::vector<ck::index_t>& StridesB,
+                             const std::vector<ck::index_t>& StridesE,
+                             const std::vector<ck::index_t>& StridesD)
+{
+    bool pass = true;
+
+    auto f_host_tensor_descriptor = [](const std::vector<ck::index_t>& dims01,
+                                       const std::vector<ck::index_t>& dims23,
+                                       const std::vector<ck::index_t>& strides) {
+        std::vector<std::size_t> dims_szt(dims01.begin(), dims01.end());
+        dims_szt.insert(dims_szt.end(), dims23.begin(), dims23.end());
+        std::vector<std::size_t> strides_szt(strides.begin(), strides.end());
+
+        return HostTensorDescriptor(dims_szt, strides);
+    };
+
+    Tensor<DataType> a_m_k(f_host_tensor_descriptor(M, K, StridesA));
+    Tensor<DataType> b_k_n(f_host_tensor_descriptor(K, N, StridesB));
+    Tensor<DataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
+    Tensor<DataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StridesE));
+    Tensor<DataType> d_m_n(f_host_tensor_descriptor(M, N, StridesD));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_device_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<DataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
+    }
+
+    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+    DeviceMem a_device_buf(sizeof(DataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(DataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(DataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DataType) * d_m_n.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    e_device_buf.SetZero();
+    d_device_buf.ToDevice(d_m_n.mData.data());
+
+    const std::vector<index_t> a_ms_ks_lengths = {M[0], M[1], K[0], K[1]};
+    const std::vector<index_t> b_ns_ks_lengths = {N[0], N[1], K[0], K[1]};
+    const std::vector<index_t> e_ms_ns_lengths = {M[0], M[1], N[0], N[1]};
+    const std::vector<index_t> d_m_n_lengths   = {M[0], M[1], N[0], N[1]};
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+
+    constexpr ck::index_t NumDim = 2;
+    using DeviceOp               = ck::tensor_operation::device::DeviceContractionMultipleD<NumDim,
+                                                                              NumDim,
+                                                                              NumDim,
+                                                                              DataType,
+                                                                              DataType,
+                                                                              DTupleDataType,
+                                                                              DataType,
+                                                                              AElementOp,
+                                                                              BElementOp,
+                                                                              CDElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    // Run reference op
+    if(do_verification)
+    {
+        using ReferenceGemmInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDim,
+                                                                      NumDim,
+                                                                      NumDim,
+                                                                      DataType,
+                                                                      DataType,
+                                                                      DataType,
+                                                                      DataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceGemmInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<DataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_m_n_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_m_n_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_m_n_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_m_n_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        if constexpr(is_same<CDElementOp, Bilinear>::value)
+                        {
+                            cde_element_op(e_m_n_host_result(m0, m1, n0, n1),
+                                           c_m_n_host_result(m0, m1, n0, n1),
+                                           d_m_n(m0, m1, n0, n1));
+                        }
+                        else if constexpr(is_same<CDElementOp, Scale>::value)
+                        {
+                            cde_element_op(e_m_n_host_result(m0, m1, n0, n1),
+                                           c_m_n_host_result(m0, m1, n0, n1));
+                        }
+                        else
+                        {
+                            static_assert("Unsupported CDElementOp in contraction profiler.");
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    std::string best_op_name;
+    float best_avg_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        std::unique_ptr<tensor_operation::device::BaseArgument> argument_ptr;
+        if constexpr(is_same<CDElementOp, Bilinear>::value)
+        {
+            argument_ptr = op_ptr->MakeArgumentPointer(
+                static_cast<DataType*>(a_device_buf.GetDeviceBuffer()),
+                static_cast<DataType*>(b_device_buf.GetDeviceBuffer()),
+                std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                static_cast<DataType*>(e_device_buf.GetDeviceBuffer()),
+                a_ms_ks_lengths,
+                StridesA,
+                b_ns_ks_lengths,
+                StridesB,
+                std::array<std::vector<ck::index_t>, 1>{d_m_n_lengths},
+                std::array<std::vector<ck::index_t>, 1>{StridesD},
+                e_ms_ns_lengths,
+                StridesE,
+                a_element_op,
+                b_element_op,
+                cde_element_op);
+        }
+        else if constexpr(is_same<CDElementOp, Scale>::value)
+        {
+            argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<DataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<DataType*>(b_device_buf.GetDeviceBuffer()),
+                                            std::array<const void*, 0>{},
+                                            static_cast<DataType*>(e_device_buf.GetDeviceBuffer()),
+                                            a_ms_ks_lengths,
+                                            StridesA,
+                                            b_ns_ks_lengths,
+                                            StridesB,
+                                            std::array<std::vector<ck::index_t>, 0>{},
+                                            std::array<std::vector<ck::index_t>, 0>{},
+                                            e_ms_ns_lengths,
+                                            StridesE,
+                                            a_element_op,
+                                            b_element_op,
+                                            cde_element_op);
+        }
+        else
+        {
+            static_assert("Unsupported CDElementOp in contraction profiler.");
+        }
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        auto nelems_m = M[0] * M[1];
+        auto nelems_n = N[0] * N[1];
+        auto nelems_k = K[0] * K[1];
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init C to zero before profiling next kernel
+            e_device_buf.SetZero();
+
+            std::string op_name = op_ptr->GetTypeString();
+
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop = std::size_t(2) * nelems_m * nelems_n * nelems_k;
+
+            std::size_t num_btype = sizeof(DataType) * nelems_m * nelems_k +
+                                    sizeof(DataType) * nelems_k * nelems_n +
+                                    sizeof(DataType) * nelems_m * nelems_n;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
+
+            float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+                float threshold =
+                    static_cast<DataType>(nelems_k) * std::numeric_limits<DataType>::epsilon();
+                pass = pass & ck::utils::check_err(e_m_n_device_result,
+                                                   e_m_n_host_result,
+                                                   "Error: incorrect results!",
+                                                   threshold,
+                                                   threshold);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_host  : ", e_m_n_host_result.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_device: ", e_m_n_device_result.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    }
+
+    if constexpr(is_same<DataType, float>::value)
+    {
+        std::cout << "Best Perf for datatype = f32";
+    }
+    else if constexpr(is_same<DataType, double>::value)
+    {
+        std::cout << "Best Perf for datatype = f64";
+    }
+
+    if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " ALayout =  RowMajor";
+    }
+    else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " ALayout =  ColumnMajor";
+    }
+
+    if constexpr(is_same<BLayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " BLayout =  RowMajor";
+    }
+    else if constexpr(is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " BLayout =  ColumnMajor";
+    }
+
+    if constexpr(is_same<CDELayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " CDELayout =  RowMajor";
+    }
+    else if constexpr(is_same<CDELayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " CDELayout =  ColumnMajor";
+    }
+
+    std::cout << " M = " << M << " N = " << N << " K = " << K << " StridesA = " << StridesA
+              << " StridesB = " << StridesB << " StridesE = " << StridesE << " : " << best_avg_time
+              << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec << " GB/s, "
+              << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_contraction_utils.hpp b/profiler/include/profiler/profile_contraction_utils.hpp
new file mode 100644
index 0000000000..076bbd4559
--- /dev/null
+++ b/profiler/include/profiler/profile_contraction_utils.hpp
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/ck.hpp"
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+using Scale    = ck::tensor_operation::element_wise::Scale;
+
+enum struct ContractionMatrixLayout
+{
+    MK_KN_MN_MN, // 0
+    MK_NK_MN_MN, // 1
+    KM_KN_MN_MN, // 2
+    KM_NK_MN_MN, // 3
+};
+
+enum struct ContractionDataType
+{
+    F32_F32_F32_F32, // 0
+    F64_F64_F64_F64, // 1
+};
+
+inline void collect_index_params(char* argv[],
+                                 std::vector<ck::index_t>& params,
+                                 const ck::index_t from,
+                                 const ck::index_t num)
+{
+    for(ck::index_t p = from; p < from + num; p++)
+        params.push_back(std::stoi(argv[p]));
+}
+
+// Defualt strides for row-major: {Dim1 * Dim2 * Dim3, Dim2 * Dim3, Dim3, 1}
+// Defualt strides for column-major: {Dim1, 1, Dim0 * Dim1 * Dim3, Dim0 * Dim1}
+inline void
+assign_default_strides(Row, std::vector<ck::index_t>& strides, std::vector<ck::index_t> dims)
+{
+    strides = {dims[1] * dims[2] * dims[3], dims[2] * dims[3], dims[3], 1};
+}
+
+inline void
+assign_default_strides(Col, std::vector<ck::index_t>& strides, std::vector<ck::index_t> dims)
+{
+    strides = {dims[1], 1, dims[0] * dims[1] * dims[3], dims[0] * dims[1]};
+}
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index c21fff7de9..0a50eedb73 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -30,6 +30,8 @@ set(PROFILER_SOURCES
     profile_batchnorm_bwd.cpp
     profile_batchnorm_infer.cpp
     profile_grouped_gemm_fastgelu.cpp
+    profile_contraction_bilinear.cpp
+    profile_contraction_scale.cpp
 )
 
 set(PROFILER_EXECUTABLE ckProfiler)
@@ -70,4 +72,6 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_contraction_bilinear.cpp b/profiler/src/profile_contraction_bilinear.cpp
new file mode 100644
index 0000000000..6ed1841204
--- /dev/null
+++ b/profiler/src/profile_contraction_bilinear.cpp
@@ -0,0 +1,165 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <vector>
+
+#include "profiler/profile_contraction_impl.hpp"
+#include "profiler/profile_contraction_utils.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "contraction_bilinear"
+#define OP_DESC "CONTRACTION+Bilinear"
+
+static void print_helper_msg()
+{
+    std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+              << "arg2: data type (0: fp32; 1: f64)\n"
+              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
+              << "arg4: verification (0: no; 1: yes)\n"
+              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
+              << "value)\n"
+              << "arg6: print tensor value (0: no; 1: yes)\n"
+              << "arg7: time kernel (0: no, 1: yes)\n"
+              << "arg8 and arg9: alpha and beta\n"
+              << "arg10 to 15: M0, M1, N0, N1, K0, K1\n"
+              << "arg16 to 31: Strides for A, B, D and E (skip for default)\n"
+              << std::endl;
+}
+
+int profile_contraction_bilinear(int argc, char* argv[])
+{
+    const bool default_strides = argc == 16;
+
+    if(argc != 32 && argc != 16)
+    {
+        print_helper_msg();
+        exit(1);
+    }
+
+    const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification    = std::stoi(argv[4]);
+    const ck::index_t init_method = std::stoi(argv[5]);
+    const bool do_log             = std::stoi(argv[6]);
+    const bool time_kernel        = std::stoi(argv[7]);
+    const float alpha             = std::stof(argv[8]);
+    const float beta              = std::stof(argv[9]);
+
+    std::vector<ck::index_t> M;
+    std::vector<ck::index_t> N;
+    std::vector<ck::index_t> K;
+    const ck::index_t dims_arg_num = 10;
+    collect_index_params(argv, M, dims_arg_num, 2);
+    collect_index_params(argv, N, dims_arg_num + 2, 2);
+    collect_index_params(argv, K, dims_arg_num + 4, 2);
+
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesE;
+    std::vector<ck::index_t> StridesD;
+    if(!default_strides)
+    {
+        collect_index_params(argv, StridesA, dims_arg_num + 6, 4);
+        collect_index_params(argv, StridesB, dims_arg_num + 10, 4);
+        collect_index_params(argv, StridesE, dims_arg_num + 14, 4);
+        collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
+    }
+
+    using F32 = float;
+    using F64 = double;
+
+    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
+        using ALayout   = decltype(a_layout);
+        using BLayout   = decltype(b_layout);
+        using CDELayout = decltype(cde_layout);
+
+        using DataType = decltype(type);
+
+        if(default_strides)
+        {
+            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+        }
+        bool pass = ck::profiler::profile_contraction_impl<ALayout,
+                                                           BLayout,
+                                                           CDELayout,
+                                                           DataType,
+                                                           ck::Tuple<DataType>,
+                                                           Bilinear>(do_verification,
+                                                                     init_method,
+                                                                     do_log,
+                                                                     time_kernel,
+                                                                     Bilinear{alpha, beta},
+                                                                     M,
+                                                                     N,
+                                                                     K,
+                                                                     StridesA,
+                                                                     StridesB,
+                                                                     StridesE,
+                                                                     StridesD);
+
+        return pass;
+    };
+
+    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+       layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F64{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_bilinear);
diff --git a/profiler/src/profile_contraction_scale.cpp b/profiler/src/profile_contraction_scale.cpp
new file mode 100644
index 0000000000..6784b916f6
--- /dev/null
+++ b/profiler/src/profile_contraction_scale.cpp
@@ -0,0 +1,162 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+#include <vector>
+
+#include "profiler/profile_contraction_impl.hpp"
+#include "profiler/profile_contraction_utils.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "contraction_scale"
+#define OP_DESC "CONTRACTION+Scale"
+
+static void print_helper_msg()
+{
+    std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+              << "arg2: data type (0: fp32; 1: f64)\n"
+              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
+              << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
+                 "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
+              << "arg4: verification (0: no; 1: yes)\n"
+              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
+              << "value)\n"
+              << "arg6: print tensor value (0: no; 1: yes)\n"
+              << "arg7: time kernel (0: no, 1: yes)\n"
+              << "arg8: alpha\n"
+              << "arg9 to 14: M0, M1, N0, N1, K0, K1\n"
+              << "arg15 to 30: Strides for A, B, D and E (skip for default)\n"
+              << std::endl;
+}
+
+int profile_contraction_scale(int argc, char* argv[])
+{
+    const bool default_strides = argc == 15;
+
+    if(argc != 31 && argc != 15)
+    {
+        print_helper_msg();
+        exit(1);
+    }
+
+    const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification    = std::stoi(argv[4]);
+    const ck::index_t init_method = std::stoi(argv[5]);
+    const bool do_log             = std::stoi(argv[6]);
+    const bool time_kernel        = std::stoi(argv[7]);
+    const float alpha             = std::stof(argv[8]);
+
+    std::vector<ck::index_t> M;
+    std::vector<ck::index_t> N;
+    std::vector<ck::index_t> K;
+    const ck::index_t dims_arg_num = 9;
+    collect_index_params(argv, M, dims_arg_num, 2);
+    collect_index_params(argv, N, dims_arg_num + 2, 2);
+    collect_index_params(argv, K, dims_arg_num + 4, 2);
+
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesE;
+    std::vector<ck::index_t> StridesD;
+    if(!default_strides)
+    {
+        collect_index_params(argv, StridesA, dims_arg_num + 6, 4);
+        collect_index_params(argv, StridesB, dims_arg_num + 10, 4);
+        collect_index_params(argv, StridesE, dims_arg_num + 14, 4);
+        collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
+    }
+
+    using F32 = float;
+    using F64 = double;
+
+    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
+        using ALayout   = decltype(a_layout);
+        using BLayout   = decltype(b_layout);
+        using CDELayout = decltype(cde_layout);
+
+        using DataType = decltype(type);
+
+        if(default_strides)
+        {
+            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+        }
+
+        bool pass = ck::profiler::
+            profile_contraction_impl<ALayout, BLayout, CDELayout, DataType, ck::Tuple<>, Scale>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                Scale{alpha},
+                M,
+                N,
+                K,
+                StridesA,
+                StridesB,
+                StridesE,
+                StridesD);
+
+        return pass;
+    };
+
+    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+       layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F64{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_scale);
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 6f43e52355..4f212d53aa 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -56,6 +56,7 @@ add_subdirectory(normalization)
 add_subdirectory(data_type)
 add_subdirectory(elementwise_normalization)
 add_subdirectory(batchnorm)
+add_subdirectory(contraction)
 if(GPU_TARGETS MATCHES "gfx1100")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/contraction/CMakeLists.txt b/test/contraction/CMakeLists.txt
new file mode 100644
index 0000000000..4817170004
--- /dev/null
+++ b/test/contraction/CMakeLists.txt
@@ -0,0 +1,4 @@
+add_gtest_executable(test_contraction test_contraction.cpp)
+add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
+target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
diff --git a/test/contraction/test_contraction.cpp b/test/contraction/test_contraction.cpp
new file mode 100644
index 0000000000..c86b849235
--- /dev/null
+++ b/test/contraction/test_contraction.cpp
@@ -0,0 +1,138 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <memory>
+#include <initializer_list>
+#include <vector>
+#include <tuple>
+#include <gtest/gtest.h>
+
+#include "profiler/profile_contraction_impl.hpp"
+
+using F32 = float;
+using F64 = double;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+using Scale    = ck::tensor_operation::element_wise::Scale;
+
+struct MemoryParams
+{
+    std::vector<ck::index_t> M;
+    std::vector<ck::index_t> N;
+    std::vector<ck::index_t> K;
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesC;
+    std::vector<ck::index_t> StridesD;
+};
+
+template <typename Tuple>
+class TestContraction : public ::testing::Test
+{
+    protected:
+    using ALayout        = std::tuple_element_t<0, Tuple>;
+    using BLayout        = std::tuple_element_t<1, Tuple>;
+    using CDLayout       = std::tuple_element_t<2, Tuple>;
+    using DataType       = std::tuple_element_t<3, Tuple>;
+    using DTupleDataType = std::tuple_element_t<4, Tuple>;
+    using CDElementOp    = std::tuple_element_t<5, Tuple>;
+
+    std::vector<MemoryParams> list_of_memory_params = {{{32, 32},
+                                                        {32, 32},
+                                                        {32, 32},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1}},
+                                                       {{16, 16},
+                                                        {32, 32},
+                                                        {16, 16},
+                                                        {4096, 256, 16, 1},
+                                                        {16, 1, 8192, 256},
+                                                        {16384, 1024, 32, 1},
+                                                        {16384, 1024, 32, 1}}};
+
+    std::vector<ck::index_t> init_methods = {0, 1, 2};
+    std::unique_ptr<CDElementOp> p_cd_element_op;
+    void Run()
+    {
+        for(auto& memory_params : list_of_memory_params)
+        {
+            for(const ck::index_t init_method : init_methods)
+            {
+                bool pass =
+                    ck::profiler::profile_contraction_impl<ALayout,
+                                                           BLayout,
+                                                           CDLayout,
+                                                           DataType,
+                                                           DTupleDataType,
+                                                           CDElementOp>(true /*do_verification*/,
+                                                                        init_method,
+                                                                        false /*do_logs*/,
+                                                                        false /*time_kernel*/,
+                                                                        *p_cd_element_op,
+                                                                        memory_params.M,
+                                                                        memory_params.N,
+                                                                        memory_params.K,
+                                                                        memory_params.StridesA,
+                                                                        memory_params.StridesB,
+                                                                        memory_params.StridesC,
+                                                                        memory_params.StridesD);
+                EXPECT_TRUE(pass);
+            }
+        }
+    }
+};
+
+template <typename Tuple>
+class TestContractionScale : public TestContraction<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestContractionBilinear : public TestContraction<Tuple>
+{
+};
+
+using BilinearKernelTypes =
+    ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Col, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Col, Row, F64, ck::Tuple<F32>, Bilinear>>;
+
+using ScaleKernelTypes = ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Col, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Row, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Col, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Row, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Col, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Row, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Col, Row, F64, ck::Tuple<>, Scale>>;
+
+TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes);
+TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
+
+TYPED_TEST(TestContractionBilinear, bilinear)
+{
+    this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
+    this->Run();
+    this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
+    this->Run();
+}
+
+TYPED_TEST(TestContractionScale, scale)
+{
+    this->p_cd_element_op = std::make_unique<Scale>(1.f);
+    this->Run();
+    this->p_cd_element_op = std::make_unique<Scale>(0.5f);
+    this->Run();
+}
diff --git a/test/contraction/test_contraction_interface.cpp b/test/contraction/test_contraction_interface.cpp
new file mode 100644
index 0000000000..c9e720c597
--- /dev/null
+++ b/test/contraction/test_contraction_interface.cpp
@@ -0,0 +1,195 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <stdexcept>
+#include <vector>
+
+#include "gtest/gtest.h"
+
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+
+using Pass     = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear = ck::tensor_operation::element_wise::Bilinear;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F32 = float;
+using F64 = double;
+
+template <ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t CDEBlockTransferScalarPerVector>
+class ContractionInstanceWrapper
+{
+    public:
+    static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+    static constexpr ck::index_t NumDim = 2;
+    // clang-format off
+    using ContractionDeviceInstance = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|         DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer|             ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer|              BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|                  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |               |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |               |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
+        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,         Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
+    // clang-format on
+
+    bool isSupported(std::vector<ck::index_t>& ADims,
+                     std::vector<ck::index_t>& BDims,
+                     std::vector<ck::index_t>& DDims,
+                     std::vector<ck::index_t>& EDims,
+                     std::vector<ck::index_t>& AStrides,
+                     std::vector<ck::index_t>& BStrides,
+                     std::vector<ck::index_t>& DStrides,
+                     std::vector<ck::index_t>& EStrides) const
+    {
+        auto contraction = ContractionDeviceInstance{};
+
+        auto argument = contraction.MakeArgument(nullptr,
+                                                 nullptr,
+                                                 std::array<const void*, 1>{nullptr},
+                                                 nullptr,
+                                                 ADims,
+                                                 AStrides,
+                                                 BDims,
+                                                 BStrides,
+                                                 std::array<std::vector<ck::index_t>, 1>{DDims},
+                                                 std::array<std::vector<ck::index_t>, 1>{DStrides},
+                                                 EDims,
+                                                 EStrides,
+                                                 Pass{},
+                                                 Pass{},
+                                                 Bilinear{1.f, 1.f});
+        return contraction.IsSupportedArgument(argument);
+    }
+};
+
+template <typename DataTypeA,
+          typename DataTypeB,
+          typename DataTypeC,
+          typename DataTypeD,
+          ck::index_t NumDim>
+class ContractionDeviceOpWrapper
+{
+
+    protected:
+    using DeviceOp = ck::tensor_operation::device::DeviceContractionMultipleD<NumDim,
+                                                                              NumDim,
+                                                                              NumDim,
+                                                                              DataTypeA,
+                                                                              DataTypeB,
+                                                                              ck::Tuple<DataTypeC>,
+                                                                              DataTypeD,
+                                                                              Pass,
+                                                                              Pass,
+                                                                              Bilinear>;
+
+    public:
+    bool IsSupportedInstance(std::vector<ck::index_t>& Dims,
+                             std::vector<ck::index_t>& Strides) const
+    {
+
+        bool supported     = false;
+        const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+        for(auto& op_ptr : op_ptrs)
+        {
+            auto argument_ptr =
+                op_ptr->MakeArgumentPointer(nullptr,
+                                            nullptr,
+                                            std::array<const void*, 1>{nullptr},
+                                            nullptr,
+                                            Dims,
+                                            Strides,
+                                            Dims,
+                                            Strides,
+                                            std::array<std::vector<ck::index_t>, 1>{Dims},
+                                            std::array<std::vector<ck::index_t>, 1>{Strides},
+                                            Dims,
+                                            Strides,
+                                            Pass{},
+                                            Pass{},
+                                            Bilinear{1.f, 1.f});
+
+            supported = supported || op_ptr->IsSupportedArgument(argument_ptr.get());
+        }
+        return supported;
+    }
+};
+
+TEST(TestContractionInterface, IncorrectNumDims)
+{
+    std::vector<std::vector<ck::index_t>> Dims    = {{4, 4}, {4, 4, 4, 4}, {4, 4, 4, 4, 4, 4}};
+    std::vector<std::vector<ck::index_t>> Strides = {{1, 1}, {1, 1, 1, 1}, {1, 1, 1, 1, 1, 1}};
+    ContractionDeviceOpWrapper<F32, F32, F32, F32, 1> wrapper_1d;
+    ContractionDeviceOpWrapper<F32, F32, F32, F32, 2> wrapper_2d;
+    ContractionDeviceOpWrapper<F32, F32, F32, F32, 3> wrapper_3d;
+    EXPECT_FALSE(wrapper_1d.IsSupportedInstance(Dims[0], Strides[0]));
+    EXPECT_TRUE(wrapper_2d.IsSupportedInstance(Dims[1], Strides[1]));
+    EXPECT_FALSE(wrapper_3d.IsSupportedInstance(Dims[2], Strides[2]));
+}
+
+TEST(TestContractionInterface, IncorrectDataTypes)
+{
+    std::vector<ck::index_t> Dims    = {4, 4, 4, 4};
+    std::vector<ck::index_t> Strides = {64, 16, 4, 1};
+    ContractionDeviceOpWrapper<F32, F32, F64, F64, 2> wrapper_1;
+    ContractionDeviceOpWrapper<F64, F64, F32, F32, 2> wrapper_2;
+    EXPECT_FALSE(wrapper_1.IsSupportedInstance(Dims, Strides));
+    EXPECT_FALSE(wrapper_2.IsSupportedInstance(Dims, Strides));
+}
+
+TEST(TestContractionSupportedArgs, ABMemoryAccess)
+{
+    std::vector<ck::index_t> Dims           = {4, 4, 4, 4};
+    std::vector<ck::index_t> Strides        = {64, 16, 4, 1};
+    std::vector<ck::index_t> StridesM1      = {4, 1, 64, 16};
+    std::vector<ck::index_t> StridesK1      = {64, 16, 4, 1};
+    std::vector<ck::index_t> InvalidStrides = {4, 4, 4, 4};
+    // Memory access to A
+    ContractionInstanceWrapper<1, 2, 4> wrapperA1;
+    ContractionInstanceWrapper<2, 2, 4> wrapperA2;
+    EXPECT_FALSE(
+        wrapperA1.isSupported(Dims, Dims, Dims, Dims, InvalidStrides, Strides, Strides, Strides));
+    EXPECT_FALSE(
+        wrapperA2.isSupported(Dims, Dims, Dims, Dims, InvalidStrides, Strides, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperA1.isSupported(Dims, Dims, Dims, Dims, StridesM1, Strides, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperA2.isSupported(Dims, Dims, Dims, Dims, StridesK1, Strides, Strides, Strides));
+    // Memory access to B
+    ContractionInstanceWrapper<2, 1, 4> wrapperB1;
+    ContractionInstanceWrapper<2, 2, 4> wrapperB2;
+    EXPECT_FALSE(
+        wrapperB1.isSupported(Dims, Dims, Dims, Dims, Strides, InvalidStrides, Strides, Strides));
+    EXPECT_FALSE(
+        wrapperB2.isSupported(Dims, Dims, Dims, Dims, Strides, InvalidStrides, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperB1.isSupported(Dims, Dims, Dims, Dims, Strides, StridesM1, Strides, Strides));
+    EXPECT_TRUE(
+        wrapperB2.isSupported(Dims, Dims, Dims, Dims, Strides, StridesK1, Strides, Strides));
+}
+
+TEST(TestContractionSupportedArgs, DEMemoryAccess)
+{
+    std::vector<ck::index_t> Dims           = {4, 4, 4, 4};
+    std::vector<ck::index_t> Strides        = {64, 16, 4, 1};
+    std::vector<ck::index_t> InvalidStrides = {64, 16, 1, 4};
+    ContractionInstanceWrapper<2, 2, 4> wrapper;
+    // Memory access to D
+    EXPECT_FALSE(
+        wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, InvalidStrides, Strides));
+    EXPECT_TRUE(wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, Strides, Strides));
+    // Memory access to E
+    EXPECT_FALSE(
+        wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, Strides, InvalidStrides));
+    EXPECT_TRUE(wrapper.isSupported(Dims, Dims, Dims, Dims, Strides, Strides, Strides, Strides));
+}

From 3cff34042385ee7dadf9252fc667b1285d4cee38 Mon Sep 17 00:00:00 2001
From: Sam Wu <sam.wu2@amd.com>
Date: Thu, 18 May 2023 11:08:38 -0600
Subject: [PATCH 015/194] Documentation Updates (#710)

* update documentation dependencies

add version number to docs

rename doc config directories

enable more doc formats on rtd

add license section in docs
---
 .github/dependabot.yml                    |   2 +-
 .gitignore                                |   2 +-
 .readthedocs.yaml                         |   4 +-
 README.md                                 |  16 +-
 docs/.sphinx/_toc.yml.in                  |   1 -
 docs/conf.py                              |  15 +-
 docs/{.doxygen => doxygen}/Doxyfile       |   0
 docs/license.rst                          |   6 +
 docs/sphinx/_toc.yml.in                   |  10 ++
 docs/{.sphinx => sphinx}/requirements.in  |   2 +-
 docs/{.sphinx => sphinx}/requirements.txt | 171 ++++------------------
 11 files changed, 79 insertions(+), 150 deletions(-)
 delete mode 100644 docs/.sphinx/_toc.yml.in
 rename docs/{.doxygen => doxygen}/Doxyfile (100%)
 create mode 100644 docs/license.rst
 create mode 100644 docs/sphinx/_toc.yml.in
 rename docs/{.sphinx => sphinx}/requirements.in (54%)
 rename docs/{.sphinx => sphinx}/requirements.txt (51%)

diff --git a/.github/dependabot.yml b/.github/dependabot.yml
index 9cdf2d670c..276690bd4f 100644
--- a/.github/dependabot.yml
+++ b/.github/dependabot.yml
@@ -6,7 +6,7 @@
 version: 2
 updates:
   - package-ecosystem: "pip" # See documentation for possible values
-    directory: "/docs/.sphinx" # Location of package manifests
+    directory: "/docs/sphinx" # Location of package manifests
     open-pull-requests-limit: 10
     schedule:
       interval: "daily"
diff --git a/.gitignore b/.gitignore
index 362fb9e2ef..7af066c82d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -49,10 +49,10 @@ build*
 install.dir*
 
 # documentation artifacts
-build/
 _build/
 _images/
 _static/
 _templates/
 _toc.yml
 docBin/
+_doxygen/
diff --git a/.readthedocs.yaml b/.readthedocs.yaml
index b739536837..5f50df2525 100644
--- a/.readthedocs.yaml
+++ b/.readthedocs.yaml
@@ -11,8 +11,8 @@ build:
 sphinx:
    configuration: docs/conf.py
 
-formats: [htmlzip]
+formats: [htmlzip, pdf, epub]
 
 python:
    install:
-   - requirements: docs/.sphinx/requirements.txt
+   - requirements: docs/sphinx/requirements.txt
diff --git a/README.md b/README.md
index 04199f11bf..a45f61a37d 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,7 @@
 # Composable Kernel
 
 ## Methodology
+
 Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
 
 CK utilizes two concepts to achieve performance portability and code maintainability:
@@ -10,6 +11,7 @@ CK utilizes two concepts to achieve performance portability and code maintainabi
 ![ALT](/docs/data/ck_component.png "CK Components")
 
 ## Code Structure
+
 Current CK library are structured into 4 layers:
 * "Templated Tile Operators" layer
 * "Templated Kernel and Invoker" layer
@@ -24,30 +26,35 @@ Run the steps below to build documentation locally.
 
 ```
 cd docs
-pip3 install -r .sphinx/requirements.txt
+pip3 install -r sphinx/requirements.txt
 python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
 ```
 
 ## Contributors
+
 The list of developers and contributors is here: [Contributors](/CONTRIBUTORS.md)
 
 ## Citation
+
 If you use CK, please use following citations:
 * CK paper will be freely available on arXiv soon: [Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming](???)
 * [CITATION.cff](/CITATION.cff)
 
 ## License
+
 CK is released under the MIT license. [License File](/LICENSE)
 
 
 # Build CK
 
 ## Build docker image
+
 ```bash
 DOCKER_BUILDKIT=1 docker build -t ck:latest -f Dockerfile .
 ```
 
 ## Launch docker
+
 ```bash
 docker run                                     \
 -it                                            \
@@ -60,10 +67,12 @@ ck:latest                                      \
 ```
 
 ## Build CK
+
 ```bash
 mkdir build && cd build
 
 # Need to specify target ID, example below is for gfx908 and gfx90a
+
 cmake                                                                                             \
 -D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
@@ -74,6 +83,7 @@ cmake
 ```
 
 ### Build examples and tests
+
 ```bash
  make -j examples tests
  make test
@@ -83,21 +93,25 @@ Instructions for running each individual examples are under [example](/example)
 
 
 ## Build ckProfiler
+
 ```bash
  make -j ckProfiler
 ```
 Instructions for running ckProfiler are under [profiler](/profiler)
 
 ## Install CK
+
 ```bash
 make install
 ```
 
 ## Using CK as pre-built kernel library
+
 Instructions for using CK as a pre-built kernel library are under [client_example](/client_example)
 
 ## Caveat
 ### Kernel Timing and Verification
+
 CK's own kernel timer will warn up kernel once, and then run it multiple times
 to get average kernel time. For some kernels that use atomic add, this will cause
 output buffer to be accumulated multiple times, causing verification failure.
diff --git a/docs/.sphinx/_toc.yml.in b/docs/.sphinx/_toc.yml.in
deleted file mode 100644
index ff21248873..0000000000
--- a/docs/.sphinx/_toc.yml.in
+++ /dev/null
@@ -1 +0,0 @@
-root: index
diff --git a/docs/conf.py b/docs/conf.py
index 3ec81ee9df..0de590da1a 100644
--- a/docs/conf.py
+++ b/docs/conf.py
@@ -4,10 +4,21 @@
 # list see the documentation:
 # https://www.sphinx-doc.org/en/master/usage/configuration.html
 
+import subprocess
+
 from rocm_docs import ROCmDocs
 
-docs_core = ROCmDocs("Composable Kernel Documentation")
-docs_core.run_doxygen()
+
+name = "Composable Kernel"
+get_version = r'sed -n -e "s/^rocm_setup_version(.* \([0-9\.]\{1,\}\).*/\1/p" ../CMakeLists.txt'
+version = subprocess.getoutput(get_version)
+if len(version) > 0:
+    name = f"{name} {version}"
+
+external_toc_path = "./sphinx/_toc.yml"
+
+docs_core = ROCmDocs(f"{name} Documentation")
+docs_core.run_doxygen(doxygen_root="doxygen", doxygen_path="doxygen/docBin/xml")
 docs_core.setup()
 
 mathjax3_config = {
diff --git a/docs/.doxygen/Doxyfile b/docs/doxygen/Doxyfile
similarity index 100%
rename from docs/.doxygen/Doxyfile
rename to docs/doxygen/Doxyfile
diff --git a/docs/license.rst b/docs/license.rst
new file mode 100644
index 0000000000..ddb544496e
--- /dev/null
+++ b/docs/license.rst
@@ -0,0 +1,6 @@
+=======
+License
+=======
+
+.. include:: ../LICENSE
+   :literal:
diff --git a/docs/sphinx/_toc.yml.in b/docs/sphinx/_toc.yml.in
new file mode 100644
index 0000000000..83dd1e7b1a
--- /dev/null
+++ b/docs/sphinx/_toc.yml.in
@@ -0,0 +1,10 @@
+# Anywhere {branch} is used, the branch name will be substituted.
+# These comments will also be removed.
+defaults:
+  numbered: False
+  maxdepth: 6
+root: index
+subtrees:
+  - caption: About
+    entries:
+      - file: license
diff --git a/docs/.sphinx/requirements.in b/docs/sphinx/requirements.in
similarity index 54%
rename from docs/.sphinx/requirements.in
rename to docs/sphinx/requirements.in
index 1905de6e6c..4bdf41b959 100644
--- a/docs/.sphinx/requirements.in
+++ b/docs/sphinx/requirements.in
@@ -1,2 +1,2 @@
-rocm-docs-core==0.2.0
+rocm-docs-core==0.10.3
 sphinxcontrib-bibtex==2.5.0
diff --git a/docs/.sphinx/requirements.txt b/docs/sphinx/requirements.txt
similarity index 51%
rename from docs/.sphinx/requirements.txt
rename to docs/sphinx/requirements.txt
index d1698b2855..097acba225 100644
--- a/docs/.sphinx/requirements.txt
+++ b/docs/sphinx/requirements.txt
@@ -1,25 +1,17 @@
 #
-# This file is autogenerated by pip-compile with Python 3.10
+# This file is autogenerated by pip-compile with Python 3.8
 # by the following command:
 #
-#    pip-compile .sphinx/requirements.in
+#    pip-compile requirements.in
 #
 accessible-pygments==0.0.3
     # via pydata-sphinx-theme
 alabaster==0.7.13
     # via sphinx
-asttokens==2.2.1
-    # via stack-data
-attrs==22.2.0
-    # via
-    #   jsonschema
-    #   jupyter-cache
 babel==2.12.1
     # via
     #   pydata-sphinx-theme
     #   sphinx
-backcall==0.2.0
-    # via ipython
 beautifulsoup4==4.11.2
     # via pydata-sphinx-theme
 breathe==4.34.0
@@ -27,19 +19,15 @@ breathe==4.34.0
 certifi==2022.12.7
     # via requests
 cffi==1.15.1
-    # via pynacl
+    # via
+    #   cryptography
+    #   pynacl
 charset-normalizer==3.1.0
     # via requests
 click==8.1.3
-    # via
-    #   jupyter-cache
-    #   sphinx-external-toc
-comm==0.1.2
-    # via ipykernel
-debugpy==1.6.6
-    # via ipykernel
-decorator==5.1.1
-    # via ipython
+    # via sphinx-external-toc
+cryptography==40.0.2
+    # via pyjwt
 deprecated==1.2.13
     # via pygithub
 docutils==0.16
@@ -48,52 +36,26 @@ docutils==0.16
     #   myst-parser
     #   pybtex-docutils
     #   pydata-sphinx-theme
-    #   rocm-docs-core
     #   sphinx
     #   sphinxcontrib-bibtex
-executing==1.2.0
-    # via stack-data
-fastjsonschema==2.16.3
-    # via nbformat
 gitdb==4.0.10
     # via gitpython
 gitpython==3.1.31
     # via rocm-docs-core
-greenlet==2.0.2
-    # via sqlalchemy
 idna==3.4
     # via requests
 imagesize==1.4.1
     # via sphinx
 importlib-metadata==6.0.0
     # via
-    #   jupyter-cache
-    #   myst-nb
-ipykernel==6.21.3
-    # via myst-nb
-ipython==8.11.0
-    # via
-    #   ipykernel
-    #   myst-nb
-jedi==0.18.2
-    # via ipython
+    #   sphinx
+    #   sphinxcontrib-bibtex
+importlib-resources==5.12.0
+    # via rocm-docs-core
 jinja2==3.1.2
     # via
     #   myst-parser
     #   sphinx
-jsonschema==4.17.3
-    # via nbformat
-jupyter-cache==0.5.0
-    # via myst-nb
-jupyter-client==8.0.3
-    # via
-    #   ipykernel
-    #   nbclient
-jupyter-core==5.3.0
-    # via
-    #   ipykernel
-    #   jupyter-client
-    #   nbformat
 latexcodec==2.0.1
     # via pybtex
 linkify-it-py==1.0.3
@@ -104,54 +66,16 @@ markdown-it-py==2.2.0
     #   myst-parser
 markupsafe==2.1.2
     # via jinja2
-matplotlib-inline==0.1.6
-    # via
-    #   ipykernel
-    #   ipython
 mdit-py-plugins==0.3.5
     # via myst-parser
 mdurl==0.1.2
     # via markdown-it-py
-myst-nb==0.17.1
+myst-parser[linkify]==1.0.0
     # via rocm-docs-core
-myst-parser[linkify]==0.18.1
-    # via
-    #   myst-nb
-    #   rocm-docs-core
-nbclient==0.5.13
-    # via
-    #   jupyter-cache
-    #   myst-nb
-nbformat==5.7.3
-    # via
-    #   jupyter-cache
-    #   myst-nb
-    #   nbclient
-nest-asyncio==1.5.6
-    # via
-    #   ipykernel
-    #   nbclient
 packaging==23.0
     # via
-    #   ipykernel
     #   pydata-sphinx-theme
     #   sphinx
-parso==0.8.3
-    # via jedi
-pexpect==4.8.0
-    # via ipython
-pickleshare==0.7.5
-    # via ipython
-platformdirs==3.1.1
-    # via jupyter-core
-prompt-toolkit==3.0.38
-    # via ipython
-psutil==5.9.4
-    # via ipykernel
-ptyprocess==0.7.0
-    # via pexpect
-pure-eval==0.2.2
-    # via stack-data
 pybtex==0.24.0
     # via
     #   pybtex-docutils
@@ -160,57 +84,47 @@ pybtex-docutils==1.0.2
     # via sphinxcontrib-bibtex
 pycparser==2.21
     # via cffi
-pydata-sphinx-theme==0.13.1
-    # via sphinx-book-theme
-pygithub==1.57
+pydata-sphinx-theme==0.13.3
+    # via
+    #   rocm-docs-core
+    #   sphinx-book-theme
+pygithub==1.58.2
     # via rocm-docs-core
 pygments==2.14.0
     # via
     #   accessible-pygments
-    #   ipython
     #   pydata-sphinx-theme
     #   sphinx
-pyjwt==2.6.0
+pyjwt[crypto]==2.6.0
     # via pygithub
 pynacl==1.5.0
     # via pygithub
-pyrsistent==0.19.3
-    # via jsonschema
-python-dateutil==2.8.2
-    # via jupyter-client
+pytz==2023.3
+    # via babel
 pyyaml==6.0
     # via
-    #   jupyter-cache
-    #   myst-nb
     #   myst-parser
     #   pybtex
     #   sphinx-external-toc
-pyzmq==25.0.1
-    # via
-    #   ipykernel
-    #   jupyter-client
 requests==2.28.2
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==0.2.0
-    # via -r .sphinx/requirements.in
+rocm-docs-core==0.10.3
+    # via -r requirements.in
 six==1.16.0
     # via
-    #   asttokens
     #   latexcodec
     #   pybtex
-    #   python-dateutil
 smmap==5.0.0
     # via gitdb
 snowballstemmer==2.2.0
     # via sphinx
 soupsieve==2.4
     # via beautifulsoup4
-sphinx==4.3.1
+sphinx==5.3.0
     # via
     #   breathe
-    #   myst-nb
     #   myst-parser
     #   pydata-sphinx-theme
     #   rocm-docs-core
@@ -220,7 +134,7 @@ sphinx==4.3.1
     #   sphinx-external-toc
     #   sphinx-notfound-page
     #   sphinxcontrib-bibtex
-sphinx-book-theme==1.0.0rc2
+sphinx-book-theme==1.0.1
     # via rocm-docs-core
 sphinx-copybutton==0.5.1
     # via rocm-docs-core
@@ -233,7 +147,7 @@ sphinx-notfound-page==0.8.3
 sphinxcontrib-applehelp==1.0.4
     # via sphinx
 sphinxcontrib-bibtex==2.5.0
-    # via -r .sphinx/requirements.in
+    # via -r requirements.in
 sphinxcontrib-devhelp==1.0.2
     # via sphinx
 sphinxcontrib-htmlhelp==2.0.1
@@ -244,40 +158,15 @@ sphinxcontrib-qthelp==1.0.3
     # via sphinx
 sphinxcontrib-serializinghtml==1.1.5
     # via sphinx
-sqlalchemy==1.4.46
-    # via jupyter-cache
-stack-data==0.6.2
-    # via ipython
-tabulate==0.9.0
-    # via jupyter-cache
-tornado==6.2
-    # via
-    #   ipykernel
-    #   jupyter-client
-traitlets==5.9.0
-    # via
-    #   comm
-    #   ipykernel
-    #   ipython
-    #   jupyter-client
-    #   jupyter-core
-    #   matplotlib-inline
-    #   nbclient
-    #   nbformat
 typing-extensions==4.5.0
-    # via
-    #   myst-nb
-    #   myst-parser
+    # via pydata-sphinx-theme
 uc-micro-py==1.0.1
     # via linkify-it-py
 urllib3==1.26.15
     # via requests
-wcwidth==0.2.6
-    # via prompt-toolkit
 wrapt==1.15.0
     # via deprecated
 zipp==3.15.0
-    # via importlib-metadata
-
-# The following packages are considered to be unsafe in a requirements file:
-# setuptools
+    # via
+    #   importlib-metadata
+    #   importlib-resources

From d821d1e54f6ce8131070a1253dfc4dd6662d3d85 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 23 May 2023 09:23:16 -0700
Subject: [PATCH 016/194] Enable gemm_dl and other kernels on Navi3x. (#714)

* enable dl kernels on navi3

* do not build xdl tests and examples on Navi

* run tests before building everything on jenkins

* disable gemm_bilinear on gfx1030

* add gpu targets to installer on Navi

* put tests in the same order as before

* reduce the number of navi targets in CI

* build CI installed for gfx940 as well

* only build for MI300 during QA runs
---
 Jenkinsfile                                   | 32 +++++++++++---
 example/02_gemm_bilinear/CMakeLists.txt       |  4 +-
 example/03_gemm_bias_relu/CMakeLists.txt      |  4 +-
 .../04_gemm_add_add_fastgelu/CMakeLists.txt   | 32 +++++++-------
 example/09_convnd_fwd/CMakeLists.txt          | 15 +++----
 .../CMakeLists.txt                            | 31 +++++++-------
 example/14_gemm_quantization/CMakeLists.txt   |  6 ++-
 .../CMakeLists.txt                            | 42 ++++++++++---------
 example/17_convnd_bwd_data/CMakeLists.txt     |  7 ++--
 example/18_batched_gemm_reduce/CMakeLists.txt |  4 +-
 .../20_grouped_conv_bwd_weight/CMakeLists.txt | 11 ++---
 .../run_grouped_conv_bwd_weight_example.inc   |  4 +-
 example/21_gemm_layernorm/CMakeLists.txt      | 10 +++--
 .../CMakeLists.txt                            | 36 ++++++++--------
 example/31_batched_gemm_gemm/CMakeLists.txt   | 20 +++++----
 example/35_splitK_gemm/CMakeLists.txt         | 21 +++++-----
 .../CMakeLists.txt                            | 13 +++---
 .../40_conv2d_fwd_quantization/CMakeLists.txt | 12 +++---
 .../41_grouped_conv_conv_fwd/CMakeLists.txt   | 18 ++++----
 .../CMakeLists.txt                            |  4 +-
 ..._conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp |  8 ++--
 ...ice_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp |  7 +++-
 .../device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp |  4 +-
 .../gpu/device/impl/device_gemm_dl.hpp        |  4 +-
 .../device/impl/device_gemm_multiple_d_dl.hpp |  8 ++--
 ...uped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp |  4 +-
 .../device_grouped_gemm_multiple_d_dl.hpp     |  5 ++-
 test/batched_gemm/CMakeLists.txt              | 26 ++++++------
 test/batched_gemm_gemm/CMakeLists.txt         | 10 +++--
 test/batched_gemm_reduce/CMakeLists.txt       |  8 ++--
 test/batched_gemm_softmax_gemm/CMakeLists.txt | 10 +++--
 .../CMakeLists.txt                            | 28 +++++++------
 test/contraction/CMakeLists.txt               |  6 ++-
 test/convnd_bwd_data/CMakeLists.txt           |  6 ++-
 test/convnd_fwd/CMakeLists.txt                |  6 ++-
 test/gemm_layernorm/CMakeLists.txt            | 13 +++---
 test/gemm_split_k/CMakeLists.txt              |  8 ++--
 test/grouped_convnd_bwd_weight/CMakeLists.txt |  6 ++-
 test/grouped_gemm/CMakeLists.txt              |  8 ++--
 39 files changed, 291 insertions(+), 210 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 83559c223b..fbff349fc3 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -493,10 +493,11 @@ def Build_CK(Map conf=[:]){
                 {
                     cmake_build(conf)
                     dir("build"){
+                        //run tests and examples 	
+                        sh 'make -j\$(( \$(nproc) / 2 )) check'
                         if (navi_node == 0 ){
-                           //run tests and examples on all nodes except Navi
-                           sh 'make -j check'
-                           //we only need the ckProfiler to run the performance tests, so we pack and stash it
+                            //we only need the ckProfiler to run the performance tests, so we pack and stash it
+                            //do not stash profiler on Navi nodes
                            sh 'tar -zcvf ckProfiler.tar.gz bin/ckProfiler'
                            stash "ckProfiler.tar.gz"
                         }
@@ -686,12 +687,31 @@ pipeline {
         {
             parallel
             {
+                stage("Build CK and run Tests on MI100/MI200/MI300")
+                {
+                    when {
+                        beforeAgent true
+                        expression { params.RUN_FULL_QA.toBoolean() }
+                    }
+                    agent{ label rocmnode("gfx908 || gfx90a") }
+                    environment{
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                    }
+                    steps{
+                        Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                    }
+                }
                 stage("Build CK and run Tests on MI100/MI200")
                 {
+                    when {
+                        beforeAgent true
+                        expression { !params.RUN_FULL_QA.toBoolean() }
+                    }
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
                         setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908,gfx90a" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
@@ -705,8 +725,8 @@ pipeline {
                     }
                     agent{ label rocmnode("navi21") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install """ 
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1030;gfx1100;gfx1101;gfx1102" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" """ 
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1030" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
 
                     }
                     steps{
diff --git a/example/02_gemm_bilinear/CMakeLists.txt b/example/02_gemm_bilinear/CMakeLists.txt
index 16a8211027..eecec2437c 100644
--- a/example/02_gemm_bilinear/CMakeLists.txt
+++ b/example/02_gemm_bilinear/CMakeLists.txt
@@ -1,4 +1,6 @@
-add_example_executable(example_gemm_bilinear_xdl_fp16 gemm_bilinear_xdl_fp16.cpp)
 if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
     add_example_executable(example_gemm_bilinear_wmma_fp16 gemm_bilinear_wmma_fp16.cpp)
 endif()
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_example_executable(example_gemm_bilinear_xdl_fp16 gemm_bilinear_xdl_fp16.cpp)
+endif()
diff --git a/example/03_gemm_bias_relu/CMakeLists.txt b/example/03_gemm_bias_relu/CMakeLists.txt
index 35c54abac0..8834a910f0 100644
--- a/example/03_gemm_bias_relu/CMakeLists.txt
+++ b/example/03_gemm_bias_relu/CMakeLists.txt
@@ -1 +1,3 @@
-add_example_executable(example_gemm_bias_relu_xdl_fp16 gemm_bias_relu_xdl_fp16.cpp)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_example_executable(example_gemm_bias_relu_xdl_fp16 gemm_bias_relu_xdl_fp16.cpp)
+endif()
\ No newline at end of file
diff --git a/example/04_gemm_add_add_fastgelu/CMakeLists.txt b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
index c75c5ba51e..a706830b68 100644
--- a/example/04_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
@@ -1,17 +1,19 @@
-add_custom_target(example_gemm_add_add_fastgelu_xdl)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_custom_target(example_gemm_add_add_fastgelu_xdl)
 
-add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
-add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
-add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
-endif(USE_BITINT_EXTENSION_INT4)
-add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    if(USE_BITINT_EXTENSION_INT4)
+       add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
+    endif(USE_BITINT_EXTENSION_INT4)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
 
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
-if(USE_BITINT_EXTENSION_INT4)
-  add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
-endif(USE_BITINT_EXTENSION_INT4)
-add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
+    if(USE_BITINT_EXTENSION_INT4)
+        add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
+    endif(USE_BITINT_EXTENSION_INT4)
+    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+endif()
\ No newline at end of file
diff --git a/example/09_convnd_fwd/CMakeLists.txt b/example/09_convnd_fwd/CMakeLists.txt
index e0a53005b8..1bcf2d148d 100644
--- a/example/09_convnd_fwd/CMakeLists.txt
+++ b/example/09_convnd_fwd/CMakeLists.txt
@@ -1,10 +1,11 @@
-add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
-add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
-add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
-add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
-# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
-
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
+    add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
+    add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
+    add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
+    # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
+    add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
+endif()
 add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
 add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
 add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
index 98941b4db5..de26462f69 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
@@ -1,16 +1,15 @@
-add_custom_target(example_convnd_fwd_reduce_xdl)
-
-add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
-add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
-add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
-add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
-
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
-add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
-
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
-  add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
-endif(USE_BITINT_EXTENSION_INT4)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(example_convnd_fwd_reduce_xdl)
+   add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+   add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+   add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+   add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
+   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
+      add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
+   endif(USE_BITINT_EXTENSION_INT4)
+endif()
\ No newline at end of file
diff --git a/example/14_gemm_quantization/CMakeLists.txt b/example/14_gemm_quantization/CMakeLists.txt
index 8ea11df9ce..584333e7bb 100644
--- a/example/14_gemm_quantization/CMakeLists.txt
+++ b/example/14_gemm_quantization/CMakeLists.txt
@@ -2,5 +2,7 @@
 add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
 
 # xdlops
-add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
-add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
\ No newline at end of file
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
+   add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
+endif()
\ No newline at end of file
diff --git a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
index 226656a736..0f1ca777c5 100644
--- a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+++ b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
@@ -1,40 +1,42 @@
-add_custom_target(example_gemm_reduce_xdl)
-add_custom_target(example_gemm_reduce_xdl_max)
-add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
-add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(example_gemm_reduce_xdl)
+   add_custom_target(example_gemm_reduce_xdl_max)
+   add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
+   add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
 
-add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
-add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
-add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
-add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
+   add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
+   add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
+   add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
+   add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
 
-add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
+   add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
 
-add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
-add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
-add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
-add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
+   add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+   add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
+   add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
+   add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
 
-add_dependencies(example_gemm_reduce_xdl_max
+   add_dependencies(example_gemm_reduce_xdl_max
                  example_gemm_max_xdl_bf16
                  example_gemm_max_xdl_fp16
                  example_gemm_max_xdl_fp32
                  example_gemm_max_xdl_int8)
 
-add_dependencies(example_gemm_reduce_xdl_mean_meansquare
+   add_dependencies(example_gemm_reduce_xdl_mean_meansquare
                  example_gemm_mean_meansquare_xdl_fp16
                  example_gemm_mean_meansquare_xdl_fp32
                  example_gemm_mean_meansquare_xdl_bf16
                  example_gemm_add_addsquare_xdl_int8)
 
-add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
+   add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
 
-add_dependencies(example_gemm_reduce_xdl
+   add_dependencies(example_gemm_reduce_xdl
                  example_gemm_reduce_xdl_mean_meansquare
                  example_gemm_reduce_xdl_max
                  example_gemm_add_add_mean_meansquare_xdl)
 
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
-  add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
+      add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
+   endif()
 endif()
diff --git a/example/17_convnd_bwd_data/CMakeLists.txt b/example/17_convnd_bwd_data/CMakeLists.txt
index fa4e65d965..ed95946c42 100644
--- a/example/17_convnd_bwd_data/CMakeLists.txt
+++ b/example/17_convnd_bwd_data/CMakeLists.txt
@@ -1,5 +1,6 @@
-add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
-target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
-
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
+   target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
+endif()
 add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
 target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
diff --git a/example/18_batched_gemm_reduce/CMakeLists.txt b/example/18_batched_gemm_reduce/CMakeLists.txt
index 99fc0043d2..0c3648dbfd 100644
--- a/example/18_batched_gemm_reduce/CMakeLists.txt
+++ b/example/18_batched_gemm_reduce/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_example_executable(example_batched_gemm_reduce_xdl_fp16 batched_gemm_reduce_xdl_fp16.cpp)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_batched_gemm_reduce_xdl_fp16 batched_gemm_reduce_xdl_fp16.cpp)
+endif()
 
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index cbe4f5f486..0ee39ac84e 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -1,11 +1,12 @@
-add_custom_target(example_grouped_conv_bwd_weight)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(example_grouped_conv_bwd_weight)
 
-add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
-add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
+   add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
 
-
-add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
+   add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
                                                  example_grouped_conv_bwd_weight_xdl_bf16)
+endif()
 
 add_custom_target(example_grouped_conv_bwd_weight_dl)
 
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 7891812375..d46b374766 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -18,7 +18,9 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     // Set split_k = 2 for xdl op, split_k = 1 for dl
     // Dl op doesn't support split_k > 1
     // TODO: Add Dl op split_k > 1 support
-    if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+    if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+         ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+         ck::get_device_name() == "gfx1102"))
     {
         split_k = 2;
     }
diff --git a/example/21_gemm_layernorm/CMakeLists.txt b/example/21_gemm_layernorm/CMakeLists.txt
index 2eb7052e1e..7f974221b7 100644
--- a/example/21_gemm_layernorm/CMakeLists.txt
+++ b/example/21_gemm_layernorm/CMakeLists.txt
@@ -1,4 +1,6 @@
-add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
-add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
-add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
-add_example_executable(example_gemm_xdl_layernorm_naive_single_kernel_fp16 gemm_xdl_layernorm_naive_single_kernel_fp16.cpp)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
+   add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
+   add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
+   add_example_executable(example_gemm_xdl_layernorm_naive_single_kernel_fp16 gemm_xdl_layernorm_naive_single_kernel_fp16.cpp)
+endif()
diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index 4b0ea4f157..0472996858 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -1,25 +1,23 @@
-add_custom_target(example_grouped_conv_fwd_multiple_d)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(example_grouped_conv_fwd_multiple_d)
 
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
-add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
 
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
-
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
-
-  add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
-endif() # USE_BITINT_EXTENSION_INT4
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
 
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
+   endif() # USE_BITINT_EXTENSION_INT4
+   add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
+   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
+endif()
 if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
 endif()
-
-add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
-
-add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index ad40c96b41..dd9aef94a0 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -1,10 +1,12 @@
-add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
-add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
-add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
-if(NOT GPU_TARGETS MATCHES "gfx940")
-	add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
-endif()
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
+   add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
+   add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
+   if(NOT GPU_TARGETS MATCHES "gfx940")
+      add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+   endif()
 
-if(USE_BITINT_EXTENSION_INT4)
-add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
-endif(USE_BITINT_EXTENSION_INT4)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
+   endif(USE_BITINT_EXTENSION_INT4)
+endif()
\ No newline at end of file
diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index 7945839546..f5a6ccb241 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -1,17 +1,18 @@
-add_custom_target(example_splitK_gemm_xdl)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(example_splitK_gemm_xdl)
+   add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
+   add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
+   add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
+   add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
 
-add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
-add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
-add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
-add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
-
-add_dependencies(example_splitK_gemm_xdl
+   add_dependencies(example_splitK_gemm_xdl
                  example_splitK_gemm_xdl_fp32
                  example_splitK_gemm_xdl_fp16
                  example_splitK_gemm_xdl_bfp16
                  example_splitK_gemm_xdl_int8)
 
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
-  add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
+      add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
+   endif()
 endif()
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index 9cf960c501..472d59c77d 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -1,7 +1,8 @@
-add_custom_target(example_grouped_conv_bwd_data)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(example_grouped_conv_bwd_data)
+   add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
 
-add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
-add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
-
-add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
-add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+   add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
+   add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+endif()
\ No newline at end of file
diff --git a/example/40_conv2d_fwd_quantization/CMakeLists.txt b/example/40_conv2d_fwd_quantization/CMakeLists.txt
index 0a314cd74c..c12ab7a340 100644
--- a/example/40_conv2d_fwd_quantization/CMakeLists.txt
+++ b/example/40_conv2d_fwd_quantization/CMakeLists.txt
@@ -1,21 +1,23 @@
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
+   add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
+   add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
+   add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
+endif()
 # Conv perlayer quantization
 add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
-add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
 
 # Conv perchannel quantization
 add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
-add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
 
 # Conv + bias + relu perlayer quantization
 add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
-add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
 
 # Conv + bias + relu perchannel quantization
 add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
-add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
 
 # Conv + bias + tanh perlayer quantization
 add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
 
 # Conv + bias + tanh perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
+add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
\ No newline at end of file
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index 4eb79371a7..14f5c284aa 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -1,9 +1,11 @@
-add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
-if(NOT GPU_TARGETS MATCHES "gfx940")
-	add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
+   if(NOT GPU_TARGETS MATCHES "gfx940")
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+   endif()
+   if(USE_BITINT_EXTENSION_INT4)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
+   endif(USE_BITINT_EXTENSION_INT4)
 endif()
-if(USE_BITINT_EXTENSION_INT4)
-add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
-endif(USE_BITINT_EXTENSION_INT4)
diff --git a/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt b/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
index d1b3dd4be2..b60789ea37 100644
--- a/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
+++ b/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
@@ -1 +1,3 @@
-add_example_executable(example_gemm_bias_softmax_gemm_permute gemm_bias_softmax_gemm_permute.cpp)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_example_executable(example_gemm_bias_softmax_gemm_permute gemm_bias_softmax_gemm_permute.cpp)
+endif()
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 5880f5f601..6180333350 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -134,8 +134,9 @@ __global__ void
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) || \
-    defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) ||           \
+    defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -711,7 +712,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
         // check device
         if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
              ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
+             ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
index c77772faa4..80c864c836 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -106,7 +106,8 @@ __global__ void
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) || \
+    defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -600,7 +601,9 @@ struct DeviceGroupedConvFwdDl_NHWC_KYXC_NHWK : public DeviceGroupedConvFwd<NDimS
         namespace ctc = tensor_layout::convolution;
 
         // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+             ck::get_device_name() == "gfx1102"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
index 2a2edc29ba..aff25aa7c6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
@@ -1393,7 +1393,9 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Dl
     static bool IsSupportedArgument(const Argument& arg)
     {
         // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+             ck::get_device_name() == "gfx1102"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
index af1989fc4a..36366a7631 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
@@ -485,7 +485,9 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030")
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+           ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
index 4397b6f99a..c91328ff71 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
@@ -50,8 +50,9 @@ __global__ void
             const CGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11,
             const Block2CTileMap block_2_ctile_map)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__))
 
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
@@ -553,7 +554,8 @@ struct DeviceGemmMultipleD_Dl : public DeviceGemmMultipleD<ALayout,
     {
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
            ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030" ||
-           ck::get_device_name() == "gfx940")
+           ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
+           ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
index 9529744f31..0473eaf762 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
@@ -1027,7 +1027,9 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     static bool IsSupportedArgument(const Argument& arg)
     {
         // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030"))
+        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
+             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+             ck::get_device_name() == "gfx1102"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
index d424f29920..e88bf8ed75 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -39,8 +39,9 @@ __global__ void
                                           const BElementwiseOperation b_element_op,
                                           const CDEElementwiseOperation cde_element_op)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx90a__) || defined(__gfx1030__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||              \
+    defined(__gfx90a__) || defined(__gfx1030__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__) || defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
diff --git a/test/batched_gemm/CMakeLists.txt b/test/batched_gemm/CMakeLists.txt
index 0574f98e87..5552a6c98e 100644
--- a/test/batched_gemm/CMakeLists.txt
+++ b/test/batched_gemm/CMakeLists.txt
@@ -1,15 +1,17 @@
-add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
-target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
-target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
+   target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
 
-add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
-target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
-target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
+   target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
 
-add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
-target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
-target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
+   target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
 
-add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
-target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
-target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
+   target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+endif()
\ No newline at end of file
diff --git a/test/batched_gemm_gemm/CMakeLists.txt b/test/batched_gemm_gemm/CMakeLists.txt
index 386809717f..eff3391750 100644
--- a/test/batched_gemm_gemm/CMakeLists.txt
+++ b/test/batched_gemm_gemm/CMakeLists.txt
@@ -1,5 +1,7 @@
-add_custom_target(test_batched_gemm_gemm)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(test_batched_gemm_gemm)
 
-add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
-target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
-add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
\ No newline at end of file
+   add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
+   target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
+   add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
+endif()
\ No newline at end of file
diff --git a/test/batched_gemm_reduce/CMakeLists.txt b/test/batched_gemm_reduce/CMakeLists.txt
index 4dc0b08257..1a8d7112a3 100644
--- a/test/batched_gemm_reduce/CMakeLists.txt
+++ b/test/batched_gemm_reduce/CMakeLists.txt
@@ -1,3 +1,5 @@
-add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
-target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
-target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
+   target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
+   target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
+endif()
diff --git a/test/batched_gemm_softmax_gemm/CMakeLists.txt b/test/batched_gemm_softmax_gemm/CMakeLists.txt
index 1ceecefb5f..dee0bb56ae 100644
--- a/test/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -1,5 +1,7 @@
-add_custom_target(test_batched_gemm_softmax_gemm)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(test_batched_gemm_softmax_gemm)
 
-add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
-target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
-add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
\ No newline at end of file
+   add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
+   target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
+   add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
+endif()
\ No newline at end of file
diff --git a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index 79af2b0d3a..cb68af06c6 100644
--- a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -1,15 +1,17 @@
-add_custom_target(test_batched_gemm_softmax_gemm_permute)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(test_batched_gemm_softmax_gemm_permute)
 
-add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
-add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
-target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
-add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
+   add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
+   add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
+   target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
 
-add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
-add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
-target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
-add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
\ No newline at end of file
+   add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
+   add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
+   target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
+   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
+endif()
\ No newline at end of file
diff --git a/test/contraction/CMakeLists.txt b/test/contraction/CMakeLists.txt
index 4817170004..ec44151f53 100644
--- a/test/contraction/CMakeLists.txt
+++ b/test/contraction/CMakeLists.txt
@@ -1,4 +1,6 @@
 add_gtest_executable(test_contraction test_contraction.cpp)
-add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
 target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
-target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
+    target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+endif()
diff --git a/test/convnd_bwd_data/CMakeLists.txt b/test/convnd_bwd_data/CMakeLists.txt
index 16ca4de872..4bfd21945d 100644
--- a/test/convnd_bwd_data/CMakeLists.txt
+++ b/test/convnd_bwd_data/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_gtest_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
-target_link_libraries(test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_gtest_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
+   target_link_libraries(test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance)
+endif()
\ No newline at end of file
diff --git a/test/convnd_fwd/CMakeLists.txt b/test/convnd_fwd/CMakeLists.txt
index 97e170d851..058f990266 100644
--- a/test/convnd_fwd/CMakeLists.txt
+++ b/test/convnd_fwd/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_gtest_executable(test_convnd_fwd convnd_fwd.cpp)
-target_link_libraries(test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_gtest_executable(test_convnd_fwd convnd_fwd.cpp)
+   target_link_libraries(test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance)
+endif()
diff --git a/test/gemm_layernorm/CMakeLists.txt b/test/gemm_layernorm/CMakeLists.txt
index c4feb5c564..b2a5178ac4 100644
--- a/test/gemm_layernorm/CMakeLists.txt
+++ b/test/gemm_layernorm/CMakeLists.txt
@@ -1,7 +1,6 @@
-add_custom_target(test_gemm_layernorm)
-
-add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
-
-target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
-
-add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_custom_target(test_gemm_layernorm)
+   add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
+   target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
+   add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
+endif()
diff --git a/test/gemm_split_k/CMakeLists.txt b/test/gemm_split_k/CMakeLists.txt
index 793091e53c..09bbf79389 100644
--- a/test/gemm_split_k/CMakeLists.txt
+++ b/test/gemm_split_k/CMakeLists.txt
@@ -1,3 +1,5 @@
-add_test_executable(test_gemm_split_k gemm_split_k.cpp)
-target_link_libraries(test_gemm_split_k PRIVATE utility)
-target_link_libraries(test_gemm_split_k PRIVATE device_gemm_splitk_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_test_executable(test_gemm_split_k gemm_split_k.cpp)
+   target_link_libraries(test_gemm_split_k PRIVATE utility)
+   target_link_libraries(test_gemm_split_k PRIVATE device_gemm_splitk_instance)
+endif()
diff --git a/test/grouped_convnd_bwd_weight/CMakeLists.txt b/test/grouped_convnd_bwd_weight/CMakeLists.txt
index e2f0790c8b..da554f677a 100644
--- a/test/grouped_convnd_bwd_weight/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_weight/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_gtest_executable(test_grouped_convnd_bwd_weight grouped_convnd_bwd_weight.cpp) 
-target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_gtest_executable(test_grouped_convnd_bwd_weight grouped_convnd_bwd_weight.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+endif()
\ No newline at end of file
diff --git a/test/grouped_gemm/CMakeLists.txt b/test/grouped_gemm/CMakeLists.txt
index 31a78733d3..a7619eac6e 100644
--- a/test/grouped_gemm/CMakeLists.txt
+++ b/test/grouped_gemm/CMakeLists.txt
@@ -1,3 +1,5 @@
-add_test_executable(test_grouped_gemm_fp16 grouped_gemm_fp16.cpp)
-target_link_libraries(test_grouped_gemm_fp16 PRIVATE utility)
-target_link_libraries(test_grouped_gemm_fp16 PRIVATE device_grouped_gemm_instance)
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   add_test_executable(test_grouped_gemm_fp16 grouped_gemm_fp16.cpp)
+   target_link_libraries(test_grouped_gemm_fp16 PRIVATE utility)
+   target_link_libraries(test_grouped_gemm_fp16 PRIVATE device_grouped_gemm_instance)
+endif()

From 76ec0089fb254c221138ec8cb2962ba5056f7fa9 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Wed, 24 May 2023 22:05:04 +0800
Subject: [PATCH 017/194] Pool3d fwd (#697)

* Expand the base class of pool2d, prepare to share base class with pool3d

* Add pool3d device op

* Add pool3d f16 example

* Refactor the base class. implement generic pooling in the future

* clang format

* get original index in max pooling

* Add outputindex to base class

* Fix dimension

* Add pooling instance

* Use indexType instead

* Remove useless header

* Extract IndexDataType to template

* Extract pooling reference code

* clang format

* clang format

* Fix typo

* Add tensor stride

* Add missing header

* Add index stride and output stride

* Refine naming

* Add type to base class

* Rename file

* Use proper size

* Fix typo

* Refine naming

* Modify the argument into vector.

* Add max pool profiler

* Refine naming

* Support f32 pool

* Fix typo

* Add avg pool2d fwd in profiler

* clang format

* Rename AccDatatype to ComputeDatatype

* Fix init

* test pool

* Extract variable

* Add client example

* Check the pooling dim

* clang format

* Connect argv and arg_parser

* Add found check

* Remove useless header

* Refine naming

* Adjust the order of device_pool_fwd
---
 .../18_groupnorm/groupnorm_swish.cpp          |   7 +-
 client_example/19_pool_fwd/CMakeLists.txt     |   5 +
 client_example/19_pool_fwd/avg_pool3d_fwd.cpp | 199 ++++++++++
 client_example/19_pool_fwd/max_pool2d_fwd.cpp | 193 ++++++++++
 example/13_pool2d_fwd/pool2d_fwd_common.hpp   | 172 ++-------
 example/13_pool2d_fwd/pool2d_fwd_fp16.cpp     |   9 +-
 example/13_pool2d_fwd/pool2d_fwd_fp32.cpp     |   9 +-
 example/48_pool3d_fwd/CMakeLists.txt          |   2 +
 example/48_pool3d_fwd/pool3d_fwd_common.hpp   | 187 +++++++++
 example/48_pool3d_fwd/pool3d_fwd_fp16.cpp     |  83 ++++
 .../gpu/device/device_pool2d_fwd.hpp          |  40 --
 .../gpu/device/device_pool_fwd.hpp            |  44 +++
 .../impl/device_pool2d_fwd_nhwc_nhwc.hpp      | 111 +++---
 .../impl/device_pool3d_fwd_ndhwc_ndhwc.hpp    | 357 ++++++++++++++++++
 .../device/impl/device_reduce_threadwise.hpp  |   2 +
 .../grid/gridwise_2d_reduction_threadwise.hpp |  36 +-
 .../cpu/reference_pool_fwd.hpp                | 345 +++++++++++++++++
 .../gpu/pool2d_fwd.hpp                        | 111 ++++++
 .../gpu/pool3d_fwd.hpp                        | 111 ++++++
 .../device_reduce_instance_threadwise.hpp     |   1 +
 .../ck/library/utility/host_tensor.hpp        |   6 +
 .../gpu/pool_fwd/CMakeLists.txt               |  10 +
 ...evice_avg_pool2d_fwd_nhwc_f16_instance.cpp |  23 ++
 ...evice_avg_pool2d_fwd_nhwc_f32_instance.cpp |  23 ++
 ...vice_avg_pool3d_fwd_ndhwc_f16_instance.cpp |  23 ++
 ...vice_avg_pool3d_fwd_ndhwc_f32_instance.cpp |  23 ++
 ...evice_max_pool2d_fwd_nhwc_f16_instance.cpp |  30 ++
 ...evice_max_pool2d_fwd_nhwc_f32_instance.cpp |  30 ++
 ...vice_max_pool3d_fwd_ndhwc_f16_instance.cpp |  30 ++
 ...vice_max_pool3d_fwd_ndhwc_f32_instance.cpp |  30 ++
 .../gpu/pool_fwd/pool_fwd_instance_common.hpp |  55 +++
 .../profiler/profile_pool2d_fwd_impl.hpp      | 264 +++++++++++++
 .../profiler/profile_pool3d_fwd_impl.hpp      | 271 +++++++++++++
 profiler/src/CMakeLists.txt                   |   4 +
 profiler/src/profile_avg_pool2d_fwd.cpp       | 141 +++++++
 profiler/src/profile_groupnorm.cpp            |   2 +-
 profiler/src/profile_max_pool3d_fwd.cpp       | 168 +++++++++
 test/CMakeLists.txt                           |   1 +
 test/pool_fwd/CMakeLists.txt                  |  16 +
 test/pool_fwd/test_avg_pool2d_fwd.cpp         |  56 +++
 test/pool_fwd/test_avg_pool3d_fwd.cpp         |  56 +++
 test/pool_fwd/test_max_pool2d_fwd.cpp         |  75 ++++
 test/pool_fwd/test_max_pool3d_fwd.cpp         |  75 ++++
 test/pool_fwd/test_pool_fwd_common.hpp        |  31 ++
 44 files changed, 3226 insertions(+), 241 deletions(-)
 create mode 100644 client_example/19_pool_fwd/CMakeLists.txt
 create mode 100644 client_example/19_pool_fwd/avg_pool3d_fwd.cpp
 create mode 100644 client_example/19_pool_fwd/max_pool2d_fwd.cpp
 create mode 100644 example/48_pool3d_fwd/CMakeLists.txt
 create mode 100644 example/48_pool3d_fwd/pool3d_fwd_common.hpp
 create mode 100644 example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
 delete mode 100644 include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
 create mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
 create mode 100644 profiler/include/profiler/profile_pool2d_fwd_impl.hpp
 create mode 100644 profiler/include/profiler/profile_pool3d_fwd_impl.hpp
 create mode 100644 profiler/src/profile_avg_pool2d_fwd.cpp
 create mode 100644 profiler/src/profile_max_pool3d_fwd.cpp
 create mode 100644 test/pool_fwd/CMakeLists.txt
 create mode 100644 test/pool_fwd/test_avg_pool2d_fwd.cpp
 create mode 100644 test/pool_fwd/test_avg_pool3d_fwd.cpp
 create mode 100644 test/pool_fwd/test_max_pool2d_fwd.cpp
 create mode 100644 test/pool_fwd/test_max_pool3d_fwd.cpp
 create mode 100644 test/pool_fwd/test_pool_fwd_common.hpp

diff --git a/client_example/18_groupnorm/groupnorm_swish.cpp b/client_example/18_groupnorm/groupnorm_swish.cpp
index a79630c237..84f62ceac4 100644
--- a/client_example/18_groupnorm/groupnorm_swish.cpp
+++ b/client_example/18_groupnorm/groupnorm_swish.cpp
@@ -131,11 +131,12 @@ int main(int argc, char* argv[])
         }
     }
 
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
-              << best_op_name << std::endl;
-
     // run the best intance
+    if(found)
     {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
         auto& op_ptr = op_ptrs[best_op_id];
         std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
                   << std::endl;
diff --git a/client_example/19_pool_fwd/CMakeLists.txt b/client_example/19_pool_fwd/CMakeLists.txt
new file mode 100644
index 0000000000..13f9f73c83
--- /dev/null
+++ b/client_example/19_pool_fwd/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp)
+target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_operations)
+
+add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp)
+target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_operations)
\ No newline at end of file
diff --git a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
new file mode 100644
index 0000000000..47bd7738f6
--- /dev/null
+++ b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
@@ -0,0 +1,199 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+
+using InDataType    = ck::half_t;
+using OutDataType   = ck::half_t;
+using IndexDataType = int32_t;
+
+constexpr ck::index_t InOutRank  = 5;
+constexpr ck::index_t WindowRank = 3;
+#if 0
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::MAX;
+constexpr bool OutputIndex = false;
+#else
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::AVG;
+constexpr bool OutputIndex = false;
+#endif
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N               = 2;
+    ck::index_t C               = 32;
+    ck::index_t Z               = 2;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Di              = 30;
+    ck::index_t Hi              = 30;
+    ck::index_t Wi              = 30;
+    ck::index_t window_stride_d = 2;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_d   = 1;
+    ck::index_t in_left_pad_h   = 1;
+    ck::index_t in_left_pad_w   = 1;
+    ck::index_t in_right_pad_d  = 1;
+    ck::index_t in_right_pad_h  = 1;
+    ck::index_t in_right_pad_w  = 1;
+
+    ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
+    ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    // Pool API only support the order of NCDHW
+    std::vector<ck::index_t> in_length              = {N, C, Di, Hi, Wi};
+    std::vector<ck::index_t> out_length             = {N, C, Do, Ho, Wo};
+    std::vector<ck::index_t> window_spatial_lengths = {Z, Y, X};
+    std::vector<ck::index_t> window_strides   = {window_stride_d, window_stride_h, window_stride_w};
+    std::vector<ck::index_t> input_left_pads  = {in_left_pad_d, in_left_pad_h, in_left_pad_w};
+    std::vector<ck::index_t> input_right_pads = {in_right_pad_d, in_right_pad_h, in_right_pad_w};
+
+    std::size_t in_tensor_size  = N * C * Di * Hi * Wi;
+    std::size_t out_tensor_size = N * C * Do * Ho * Wo;
+
+    // tensor layout = NDHWC
+    std::vector<ck::index_t> in_tensor_stride  = {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C};
+    std::vector<ck::index_t> out_tensor_stride = {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C};
+
+    SimpleDeviceMem in_device_buf(sizeof(InDataType) * in_tensor_size);
+    SimpleDeviceMem out_device_buf(sizeof(OutDataType) * out_tensor_size);
+    SimpleDeviceMem out_indices_device_buf(sizeof(IndexDataType) * out_tensor_size);
+
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3, 4});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes =
+                in_tensor_size * sizeof(InDataType) + out_tensor_size * sizeof(OutDataType);
+
+            if constexpr(OutputIndex)
+                num_bytes += out_tensor_size * sizeof(IndexDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3, 4});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/19_pool_fwd/max_pool2d_fwd.cpp b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
new file mode 100644
index 0000000000..12ee619201
--- /dev/null
+++ b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
@@ -0,0 +1,193 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp"
+
+using InDataType    = ck::half_t;
+using OutDataType   = ck::half_t;
+using IndexDataType = int32_t;
+
+constexpr ck::index_t InOutRank  = 4;
+constexpr ck::index_t WindowRank = 2;
+#if 1
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::MAX;
+constexpr bool OutputIndex = true;
+#else
+constexpr auto ReduceOpId  = ck::ReduceTensorOp::AVG;
+constexpr bool OutputIndex = false;
+#endif
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N               = 2;
+    ck::index_t C               = 32;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Hi              = 30;
+    ck::index_t Wi              = 30;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_h   = 1;
+    ck::index_t in_left_pad_w   = 1;
+    ck::index_t in_right_pad_h  = 1;
+    ck::index_t in_right_pad_w  = 1;
+
+    ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    // Pool API only support the order of NCHW
+    std::vector<ck::index_t> in_length              = {N, C, Hi, Wi};
+    std::vector<ck::index_t> out_length             = {N, C, Ho, Wo};
+    std::vector<ck::index_t> window_spatial_lengths = {Y, X};
+    std::vector<ck::index_t> window_strides         = {window_stride_h, window_stride_w};
+    std::vector<ck::index_t> input_left_pads        = {in_left_pad_h, in_left_pad_w};
+    std::vector<ck::index_t> input_right_pads       = {in_right_pad_h, in_right_pad_w};
+
+    std::size_t in_tensor_size  = N * C * Hi * Wi;
+    std::size_t out_tensor_size = N * C * Ho * Wo;
+
+    // tensor layout = NHWC
+    std::vector<ck::index_t> in_tensor_stride  = {C * Hi * Wi, 1, Wi * C, C};
+    std::vector<ck::index_t> out_tensor_stride = {C * Ho * Wo, 1, Wo * C, C};
+
+    SimpleDeviceMem in_device_buf(sizeof(InDataType) * in_tensor_size);
+    SimpleDeviceMem out_device_buf(sizeof(OutDataType) * out_tensor_size);
+    SimpleDeviceMem out_indices_device_buf(sizeof(IndexDataType) * out_tensor_size);
+
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes =
+                in_tensor_size * sizeof(InDataType) + out_tensor_size * sizeof(OutDataType);
+
+            if constexpr(OutputIndex)
+                num_bytes += out_tensor_size * sizeof(IndexDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/example/13_pool2d_fwd/pool2d_fwd_common.hpp b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
index b83cb6a96f..9abc986713 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_common.hpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
@@ -17,115 +17,11 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
 
 template <typename InDataType,
           typename OutDataType,
-          typename AccDataType,
-          typename IndexDataType,
-          ck::ReduceTensorOp ReduceOpId,
-          bool PropagateNan,
-          bool OutputIndex>
-static void pool_host_verify(const Tensor<InDataType>& in,
-                             Tensor<OutDataType>& out,
-                             Tensor<IndexDataType>& out_indices,
-                             const std::array<ck::index_t, 2>& window_spatial_lengths,
-                             const std::array<ck::index_t, 2>& window_strides,
-                             const std::array<ck::index_t, 2>& in_left_pads,
-                             const std::array<ck::index_t, 2>& /*in_right_pads*/)
-{
-    const int32_t reduceLength = window_spatial_lengths[0] * window_spatial_lengths[1];
-
-    using ReduceOperation = typename ck::reduce_binary_operator<ReduceOpId>::opType;
-
-    auto elementwise_ops =
-        ck::reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(reduceLength);
-
-    auto in_elementwise_op  = std::get<0>(elementwise_ops);
-    auto acc_elementwise_op = std::get<1>(elementwise_ops);
-
-    if constexpr(!OutputIndex)
-    {
-        using Accumulation =
-            ck::detail::AccumulateWithNanCheck<PropagateNan, ReduceOperation, AccDataType>;
-
-        auto f_nchw = [&](auto n, auto c, auto ho, auto wo) {
-            auto accuVal = ReduceOperation::template GetIdentityValue<AccDataType>();
-
-            for(ck::index_t y = 0; y < window_spatial_lengths[0]; ++y)
-            {
-                ck::index_t hi = ho * window_strides[0] + y - in_left_pads[0];
-                for(ck::index_t x = 0; x < window_spatial_lengths[1]; ++x)
-                {
-                    ck::index_t wi = wo * window_strides[1] + x - in_left_pads[1];
-                    if(hi >= 0 && hi < static_cast<ck::index_t>(in.mDesc.GetLengths()[2]) &&
-                       wi >= 0 && wi < static_cast<ck::index_t>(in.mDesc.GetLengths()[3]))
-                    {
-                        AccDataType currVal = static_cast<AccDataType>(in(n, c, hi, wi));
-
-                        in_elementwise_op(currVal, currVal);
-
-                        Accumulation::Calculate(accuVal, currVal);
-                    }
-                }
-            }
-
-            acc_elementwise_op(accuVal, accuVal);
-
-            out(n, c, ho, wo) = accuVal;
-        };
-
-        make_ParallelTensorFunctor(f_nchw,
-                                   out.mDesc.GetLengths()[0],
-                                   out.mDesc.GetLengths()[1],
-                                   out.mDesc.GetLengths()[2],
-                                   out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-    }
-    else
-    {
-        using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
-                                                                        ReduceOperation,
-                                                                        AccDataType,
-                                                                        IndexDataType>;
-        auto f_nchw        = [&](auto n, auto c, auto ho, auto wo) {
-            auto accuVal            = ReduceOperation::template GetIdentityValue<AccDataType>();
-            IndexDataType accuIndex = 0;
-
-            for(ck::index_t y = 0; y < window_spatial_lengths[0]; ++y)
-            {
-                ck::index_t hi = ho * window_strides[0] + y - in_left_pads[0];
-                for(ck::index_t x = 0; x < window_spatial_lengths[1]; ++x)
-                {
-                    ck::index_t wi = wo * window_strides[1] + x - in_left_pads[1];
-                    if(hi >= 0 && hi < in.mDesc.GetLengths()[2] && wi >= 0 &&
-                       wi < in.mDesc.GetLengths()[3])
-                    {
-                        AccDataType currVal     = static_cast<AccDataType>(in(n, c, hi, wi));
-                        IndexDataType currIndex = y * window_spatial_lengths[1] + x;
-
-                        in_elementwise_op(currVal, currVal);
-
-                        Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
-                    }
-                }
-            }
-
-            acc_elementwise_op(accuVal, accuVal);
-
-            out(n, c, ho, wo)         = accuVal;
-            out_indices(n, c, ho, wo) = accuIndex;
-        };
-
-        make_ParallelTensorFunctor(f_nchw,
-                                   out.mDesc.GetLengths()[0],
-                                   out.mDesc.GetLengths()[1],
-                                   out.mDesc.GetLengths()[2],
-                                   out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-    };
-}
-
-template <typename InDataType,
-          typename OutDataType,
-          typename AccDataType,
+          typename ComputeDataType,
           typename IndexDataType,
           typename InLayout,
           typename OutLayout,
@@ -150,9 +46,10 @@ bool pool_test(bool do_verification,
 {
     using DevicePoolFwdInstance =
         ck::tensor_operation::device::DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<
-            InDataType,  // InDataType
-            OutDataType, // OutDataType
-            AccDataType, // AccDataType
+            InDataType,      // InDataType
+            OutDataType,     // OutDataType
+            IndexDataType,   // IndexDataType
+            ComputeDataType, // ComputeDataType
             ReduceOpId,
             OutputIndex,
             64, // BlockSize
@@ -165,10 +62,10 @@ bool pool_test(bool do_verification,
     const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
     const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
 
-    const std::array<ck::index_t, 2> window_spatial_lengths{{Y, X}};
-    const std::array<ck::index_t, 2> window_strides{{window_stride_h, window_stride_w}};
-    const std::array<ck::index_t, 2> input_left_pads{{in_left_pad_h, in_left_pad_w}};
-    const std::array<ck::index_t, 2> input_right_pads{{in_right_pad_h, in_right_pad_w}};
+    const std::vector<ck::index_t> window_spatial_lengths{Y, X};
+    const std::vector<ck::index_t> window_strides{window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
+    const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
 
     // tensor layout
     auto f_host_tensor_descriptor =
@@ -219,14 +116,16 @@ bool pool_test(bool do_verification,
         static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
         static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
         static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
-        N,
-        C,
-        std::array<ck::index_t, 2>{{Hi, Wi}},
-        std::array<ck::index_t, 2>{{Y, X}},
-        std::array<ck::index_t, 2>{{Ho, Wo}},
+        {N, C, Hi, Wi},
+        {Y, X},
+        {N, C, Ho, Wo},
+        {C * Hi * Wi, 1, Wi * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
         window_strides,
         input_left_pads,
-        input_right_pads);
+        input_right_pads,
+        {2, 3});
 
     if(!pool.IsSupportedArgument(argument_ptr.get()))
     {
@@ -252,19 +151,28 @@ bool pool_test(bool do_verification,
 
     if(do_verification)
     {
-        pool_host_verify<InDataType,
-                         OutDataType,
-                         AccDataType,
-                         IndexDataType,
-                         ReduceOpId,
-                         PropagateNan,
-                         OutputIndex>(in_n_c_hi_wi,
-                                      out_n_c_ho_wo_host,
-                                      out_indices_n_c_ho_wo_host,
-                                      window_spatial_lengths,
-                                      window_strides,
-                                      input_left_pads,
-                                      input_right_pads);
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<4,
+                                                            2,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ReduceOpId,
+                                                            PropagateNan,
+                                                            OutputIndex>;
+
+        auto ref_pooling          = ReferencePoolingFwdInstance{};
+        auto ref_pooling_invoker  = ref_pooling.MakeInvoker();
+        auto ref_pooling_argument = ref_pooling.MakeArgument(in_n_c_hi_wi,
+                                                             out_n_c_ho_wo_host,
+                                                             out_indices_n_c_ho_wo_host,
+                                                             window_spatial_lengths,
+                                                             window_strides,
+                                                             input_left_pads,
+                                                             input_right_pads);
+
+        ref_pooling_invoker.Run(ref_pooling_argument);
 
         out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
index 659f3251dc..20c3e47017 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
-#include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
@@ -10,9 +9,9 @@
 
 #include "pool2d_fwd_common.hpp"
 
-using InDataType  = ck::half_t;
-using OutDataType = ck::half_t;
-using AccDataType = float;
+using InDataType      = ck::half_t;
+using OutDataType     = ck::half_t;
+using ComputeDataType = float;
 
 using IndexDataType = int32_t;
 
@@ -91,7 +90,7 @@ int main(int argc, char* argv[])
 
     bool pass = pool_test<InDataType,
                           OutDataType,
-                          AccDataType,
+                          ComputeDataType,
                           IndexDataType,
                           InLayout,
                           OutLayout,
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
index f47c7ff151..34ff6f435e 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
-#include <cstdlib>
 
 #include "ck/ck.hpp"
 #include "ck/utility/reduction_enums.hpp"
@@ -10,9 +9,9 @@
 
 #include "pool2d_fwd_common.hpp"
 
-using InDataType  = float;
-using OutDataType = float;
-using AccDataType = float;
+using InDataType      = float;
+using OutDataType     = float;
+using ComputeDataType = float;
 
 using IndexDataType = int32_t;
 
@@ -91,7 +90,7 @@ int main(int argc, char* argv[])
 
     bool pass = pool_test<InDataType,
                           OutDataType,
-                          AccDataType,
+                          ComputeDataType,
                           IndexDataType,
                           InLayout,
                           OutLayout,
diff --git a/example/48_pool3d_fwd/CMakeLists.txt b/example/48_pool3d_fwd/CMakeLists.txt
new file mode 100644
index 0000000000..5d58d6a0b1
--- /dev/null
+++ b/example/48_pool3d_fwd/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
+
diff --git a/example/48_pool3d_fwd/pool3d_fwd_common.hpp b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
new file mode 100644
index 0000000000..5706deb6d1
--- /dev/null
+++ b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
@@ -0,0 +1,187 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+#pragma once
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/utility/reduction_enums.hpp"
+#include "ck/utility/reduction_functions_accumulate.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+
+template <typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          typename InLayout,
+          typename OutLayout,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+bool pool3d_test(bool do_verification,
+                 bool time_kernel,
+                 ck::index_t N,
+                 ck::index_t C,
+                 ck::index_t Z,
+                 ck::index_t Y,
+                 ck::index_t X,
+                 ck::index_t Di,
+                 ck::index_t Hi,
+                 ck::index_t Wi,
+                 ck::index_t window_stride_d,
+                 ck::index_t window_stride_h,
+                 ck::index_t window_stride_w,
+                 ck::index_t in_left_pad_d,
+                 ck::index_t in_left_pad_h,
+                 ck::index_t in_left_pad_w,
+                 ck::index_t in_right_pad_d,
+                 ck::index_t in_right_pad_h,
+                 ck::index_t in_right_pad_w)
+{
+    using DevicePoolFwdInstance =
+        ck::tensor_operation::device::DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<
+            InDataType,      // InDataType
+            OutDataType,     // OutDataType
+            IndexDataType,   // IndexDataType
+            ComputeDataType, // ComputeDataType
+            ReduceOpId,
+            OutputIndex,
+            64, // BlockSize
+            64, // ReduceMThreadClusterSize
+            1,  // ReduceKThreadClusterSize
+            4,  // ReduceMThreadSliceSize
+            1,  // ReduceKThreadSliceSize
+            4>; // InSrcOutDstVectorSize
+
+    const ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    const std::vector<ck::index_t> window_spatial_lengths{Z, Y, X};
+    const std::vector<ck::index_t> window_strides{
+        window_stride_d, window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> input_left_pads{in_left_pad_d, in_left_pad_h, in_left_pad_w};
+    const std::vector<ck::index_t> input_right_pads{in_right_pad_d, in_right_pad_h, in_right_pad_w};
+
+    // tensor layout
+    auto f_host_tensor_descriptor = [](std::size_t N_,
+                                       std::size_t C_,
+                                       std::size_t D,
+                                       std::size_t H,
+                                       std::size_t W,
+                                       auto layout) {
+        using namespace ck::literals;
+
+        if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+        {
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
+        }
+        else if constexpr(ck::is_same<decltype(layout),
+                                      ck::tensor_layout::convolution::NDHWC>::value)
+        {
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+        }
+    };
+
+    Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi, InLayout{}));
+    Tensor<OutDataType> out_n_c_do_ho_wo_host(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_host(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+    Tensor<OutDataType> out_n_c_do_ho_wo_device(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_device(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
+
+    std::cout << "in_n_c_di_hi_wi: " << in_n_c_di_hi_wi.mDesc << std::endl;
+    std::cout << "out_n_c_do_ho_wo: " << out_n_c_do_ho_wo_host.mDesc << std::endl;
+
+    in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_di_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
+                                     out_indices_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_di_hi_wi.mData.data());
+
+    auto pool         = DevicePoolFwdInstance{};
+    auto invoker_ptr  = pool.MakeInvokerPointer();
+    auto argument_ptr = pool.MakeArgumentPointer(
+        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+        static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+        {N, C, Di, Hi, Wi},
+        {Z, Y, X},
+        {N, C, Do, Ho, Wo},
+        {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C},
+        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+        window_strides,
+        input_left_pads,
+        input_right_pads,
+        {2, 3, 4});
+
+    if(!pool.IsSupportedArgument(argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! device_op with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+    std::cout << "Perf: " << ave_time << std::endl;
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<5,
+                                                            3,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ReduceOpId,
+                                                            PropagateNan,
+                                                            OutputIndex>;
+
+        auto ref_pooling          = ReferencePoolingFwdInstance{};
+        auto ref_pooling_invoker  = ref_pooling.MakeInvoker();
+        auto ref_pooling_argument = ref_pooling.MakeArgument(in_n_c_di_hi_wi,
+                                                             out_n_c_do_ho_wo_host,
+                                                             out_indices_n_c_do_ho_wo_host,
+                                                             window_spatial_lengths,
+                                                             window_strides,
+                                                             input_left_pads,
+                                                             input_right_pads);
+
+        ref_pooling_invoker.Run(ref_pooling_argument);
+
+        out_device_buf.FromDevice(out_n_c_do_ho_wo_device.mData.data());
+
+        pass = pass && ck::utils::check_err(out_n_c_do_ho_wo_device, out_n_c_do_ho_wo_host);
+
+        if constexpr(OutputIndex)
+        {
+            out_indices_device_buf.FromDevice(out_indices_n_c_do_ho_wo_device.mData.data());
+
+            pass = pass && ck::utils::check_err(out_indices_n_c_do_ho_wo_device,
+                                                out_indices_n_c_do_ho_wo_host);
+        };
+    }
+
+    return (pass);
+};
diff --git a/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
new file mode 100644
index 0000000000..4d3686bcb0
--- /dev/null
+++ b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "pool3d_fwd_common.hpp"
+
+using InDataType      = ck::half_t;
+using OutDataType     = ck::half_t;
+using ComputeDataType = float;
+
+using IndexDataType = int32_t;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using OutLayout = ck::tensor_layout::convolution::NDHWC;
+
+#if 1
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+#else
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+#endif
+
+static constexpr bool OutputIndex  = false;
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 2;
+    ck::index_t C               = 32;
+    ck::index_t Z               = 2;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Di              = 30;
+    ck::index_t Hi              = 30;
+    ck::index_t Wi              = 30;
+    ck::index_t window_stride_d = 2;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_d   = 1;
+    ck::index_t in_left_pad_h   = 1;
+    ck::index_t in_left_pad_w   = 1;
+    ck::index_t in_right_pad_d  = 1;
+    ck::index_t in_right_pad_h  = 1;
+    ck::index_t in_right_pad_w  = 1;
+
+    bool pass = pool3d_test<InDataType,
+                            OutDataType,
+                            ComputeDataType,
+                            IndexDataType,
+                            InLayout,
+                            OutLayout,
+                            ReduceOpId,
+                            PropagateNan,
+                            OutputIndex>(do_verification,
+                                         time_kernel,
+                                         N,
+                                         C,
+                                         Z,
+                                         Y,
+                                         X,
+                                         Di,
+                                         Hi,
+                                         Wi,
+                                         window_stride_d,
+                                         window_stride_h,
+                                         window_stride_w,
+                                         in_left_pad_d,
+                                         in_left_pad_h,
+                                         in_left_pad_w,
+                                         in_right_pad_d,
+                                         in_right_pad_h,
+                                         in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp
deleted file mode 100644
index 3b376c6f73..0000000000
--- a/include/ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <array>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-#include "ck/utility/reduction_enums.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-template <ck::ReduceTensorOp ReduceOpId>
-struct DevicePool2dFwd : public BaseOperator
-{
-    virtual std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(const void* in_dev,
-                        void* out_dev,
-                        void* out_indices_dev,
-                        ck::index_t N,
-                        ck::index_t C,
-                        std::array<ck::index_t, 2> input_spatial_lengths,
-                        std::array<ck::index_t, 2> window_spatial_lengths,
-                        std::array<ck::index_t, 2> output_spatial_lengths,
-                        std::array<ck::index_t, 2> window_strides,
-                        std::array<ck::index_t, 2> input_left_pads,
-                        std::array<ck::index_t, 2> input_right_pads) = 0;
-
-    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
-};
-
-template <ck::ReduceTensorOp ReduceOpId>
-using DevicePool2dFwdPtr = std::unique_ptr<DevicePool2dFwd<ReduceOpId>>;
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
new file mode 100644
index 0000000000..e801e98a25
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <index_t InOutRank,
+          index_t WindowRank,
+          typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+struct DevicePoolFwd : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in_dev,
+                        void* p_out_dev,
+                        void* p_out_indices_dev,
+                        std::vector<ck::index_t> input_lengths,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> output_lengths,
+                        std::vector<ck::index_t> input_stride,
+                        std::vector<ck::index_t> output_stride,
+                        std::vector<ck::index_t> indices_stride,
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> pooling_dims) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
index bfde40cda2..6933db68dc 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
@@ -9,7 +9,7 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
-#include "ck/tensor_operation/gpu/device/device_pool2d_fwd.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
@@ -20,16 +20,18 @@ namespace device {
 
 template <typename InDataType,
           typename OutDataType,
-          typename AccDataType,
+          typename IndexDataType, // enable if OutputIndex == true
+          typename ComputeDataType,
           ck::ReduceTensorOp ReduceOpId,
-          bool OuputIndex,
+          bool OutputIndex,
           ck::index_t BlockSize,
           ck::index_t ReduceMThreadClusterSize,
           ck::index_t ReduceKThreadClusterSize,
           ck::index_t ReduceMThreadSliceSize,
           ck::index_t ReduceKThreadSliceSize,
           ck::index_t InSrcOutDstVectorSize>
-struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd<ReduceOpId>
+struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C
+    : public DevicePoolFwd<4, 2, InDataType, OutDataType, IndexDataType, ReduceOpId, OutputIndex>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
@@ -38,7 +40,8 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
     static constexpr auto I4 = Number<4>{};
     static constexpr auto I5 = Number<5>{};
 
-    using IndexDataType = int32_t;
+    static constexpr index_t InOutRank  = 4;
+    static constexpr index_t WindowRank = 2;
 
     using ReduceOperation = typename reduce_binary_operator<ReduceOpId>::opType;
 
@@ -59,12 +62,12 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
 
     static auto MakeABGridDescriptor_A_M_K_B_M(ck::index_t N,
                                                ck::index_t C,
-                                               std::array<ck::index_t, 2> input_spatial_lengths,
-                                               std::array<ck::index_t, 2> window_spatial_lengths,
-                                               std::array<ck::index_t, 2> output_spatial_lengths,
-                                               std::array<ck::index_t, 2> window_strides,
-                                               std::array<ck::index_t, 2> input_left_pads,
-                                               std::array<ck::index_t, 2> input_right_pads)
+                                               std::vector<ck::index_t> input_spatial_lengths,
+                                               std::vector<ck::index_t> window_spatial_lengths,
+                                               std::vector<ck::index_t> output_spatial_lengths,
+                                               std::vector<ck::index_t> window_strides,
+                                               std::vector<ck::index_t> input_left_pads,
+                                               std::vector<ck::index_t> input_right_pads)
     {
         const index_t Hi = input_spatial_lengths[0];
         const index_t Wi = input_spatial_lengths[1];
@@ -141,9 +144,7 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
         return make_tuple(in_grid_desc_reducem_reducek, out_grid_desc_reducem);
     }
 
-    using ABGridDescs = decltype(
-        MakeABGridDescriptor_A_M_K_B_M(1, 1, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}));
-
+    using ABGridDescs   = decltype(MakeABGridDescriptor_A_M_K_B_M(1, 1, {}, {}, {}, {}, {}, {}));
     using AGridDesc_M_K = remove_cvref_t<decltype(ABGridDescs{}[I0])>;
     using BGridDesc_M   = remove_cvref_t<decltype(ABGridDescs{}[I1])>;
 
@@ -152,15 +153,15 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
     {
         Argument(const InDataType* p_in_dev,
                  OutDataType* p_out_dev,
-                 int* p_out_indices_dev,
+                 IndexDataType* p_out_indices_dev,
                  ck::index_t N,
                  ck::index_t C,
-                 std::array<ck::index_t, 2>& input_spatial_lengths,
-                 std::array<ck::index_t, 2>& window_spatial_lengths,
-                 std::array<ck::index_t, 2>& output_spatial_lengths,
-                 std::array<ck::index_t, 2>& window_strides,
-                 std::array<ck::index_t, 2>& input_left_pads,
-                 std::array<ck::index_t, 2>& input_right_pads)
+                 std::vector<ck::index_t>& input_spatial_lengths,
+                 std::vector<ck::index_t>& window_spatial_lengths,
+                 std::vector<ck::index_t>& output_spatial_lengths,
+                 std::vector<ck::index_t>& window_strides,
+                 std::vector<ck::index_t>& input_left_pads,
+                 std::vector<ck::index_t>& input_right_pads)
             : p_in_dev_{p_in_dev},
               p_out_dev_{p_out_dev},
               p_out_indices_dev_{p_out_indices_dev},
@@ -190,7 +191,7 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
 
         const InDataType* p_in_dev_;
         OutDataType* p_out_dev_;
-        int* p_out_indices_dev_;
+        IndexDataType* p_out_indices_dev_;
         AGridDesc_M_K a_grid_desc_m_k_;
         BGridDesc_M b_grid_desc_m_;
         InElementwiseOperation in_element_op_;
@@ -208,7 +209,7 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
             using gridwise_reduce =
                 GridwiseReduction_mk_to_m_threadwise<InDataType,
                                                      OutDataType,
-                                                     AccDataType,
+                                                     ComputeDataType,
                                                      IndexDataType,
                                                      AGridDesc_M_K,
                                                      BGridDesc_M,
@@ -224,17 +225,19 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
                                                      InSrcOutDstVectorSize,
                                                      InSrcOutDstVectorSize>;
 
-            const auto kernel = kernel_reduce_threadwise<gridwise_reduce,
-                                                         OuputIndex,
-                                                         false, // don't have index input
-                                                         InDataType,
-                                                         OutDataType,
-                                                         AccDataType,
-                                                         IndexDataType,
-                                                         AGridDesc_M_K,
-                                                         BGridDesc_M,
-                                                         InElementwiseOperation,
-                                                         AccElementwiseOperation>;
+            const auto kernel =
+                kernel_reduce_threadwise<gridwise_reduce,
+                                         OutputIndex,
+                                         true,  // pooling need to return global index
+                                         false, // don't have index input
+                                         InDataType,
+                                         OutDataType,
+                                         ComputeDataType,
+                                         IndexDataType,
+                                         AGridDesc_M_K,
+                                         BGridDesc_M,
+                                         InElementwiseOperation,
+                                         AccElementwiseOperation>;
 
             ck::index_t ReduceM = arg.a_grid_desc_m_k_.GetLength(I0);
 
@@ -280,22 +283,42 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C : public DevicePool2dFwd
     MakeArgumentPointer(const void* p_in_dev,
                         void* p_out_dev,
                         void* p_out_indices_dev,
-                        ck::index_t N,
-                        ck::index_t C,
-                        std::array<ck::index_t, 2> input_spatial_lengths,
-                        std::array<ck::index_t, 2> window_spatial_lengths,
-                        std::array<ck::index_t, 2> output_spatial_lengths,
-                        std::array<ck::index_t, 2> window_strides,
-                        std::array<ck::index_t, 2> input_left_pads,
-                        std::array<ck::index_t, 2> input_right_pads) override
+                        std::vector<ck::index_t> input_lengths,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> output_lengths,
+                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> pooling_dims) override
     {
+        if(input_lengths.size() != InOutRank || window_lengths.size() != WindowRank ||
+           input_lengths.size() != InOutRank || window_strides.size() != WindowRank ||
+           input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+            throw std::runtime_error("dimension is incorrect");
+
+        if(pooling_dims != std::vector<ck::index_t>{2, 3})
+            throw std::runtime_error("pooling_dims only support {2, 3} in pool2d so far");
+
+        index_t N  = input_lengths[0];
+        index_t C  = input_lengths[1];
+        index_t Hi = input_lengths[2];
+        index_t Wi = input_lengths[3];
+        index_t Ho = output_lengths[2];
+        index_t Wo = output_lengths[3];
+
+        std::vector<ck::index_t> input_spatial_lengths  = {Hi, Wi};
+        std::vector<ck::index_t> output_spatial_lengths = {Ho, Wo};
+
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_dev),
                                           static_cast<OutDataType*>(p_out_dev),
-                                          static_cast<int*>(p_out_indices_dev),
+                                          static_cast<IndexDataType*>(p_out_indices_dev),
                                           N,
                                           C,
                                           input_spatial_lengths,
-                                          window_spatial_lengths,
+                                          window_lengths,
                                           output_spatial_lengths,
                                           window_strides,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
new file mode 100644
index 0000000000..d330fda8ce
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
@@ -0,0 +1,357 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType, // enable if OutputIndex == true
+          typename ComputeDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool OutputIndex,
+          ck::index_t BlockSize,
+          ck::index_t MThreadClusterSize,
+          ck::index_t KThreadClusterSize,
+          ck::index_t MThreadSliceSize,
+          ck::index_t KThreadSliceSize,
+          ck::index_t InSrcOutDstVectorSize>
+struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
+    : public DevicePoolFwd<5, 3, InDataType, OutDataType, IndexDataType, ReduceOpId, OutputIndex>
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr index_t InOutRank  = 5;
+    static constexpr index_t WindowRank = 3;
+
+    using ReduceOperation = typename reduce_binary_operator<ReduceOpId>::opType;
+
+    using InElementwiseOperation =
+        typename reduce_unary_operator<ReduceOpId, true, true>::InElementwiseOperation;
+
+    using AccElementwiseOperation =
+        typename reduce_unary_operator<ReduceOpId, true, true>::AccElementwiseOperation;
+
+    // for NDHWC, the dim C is the vector Dim for both input and output in memory, which is not
+    // reduced.
+    static constexpr index_t InSrcOutDstVectorDim = 0;
+
+    static constexpr ck::index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr ck::index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto MakeABGridDescriptor_A_M_K_B_M(ck::index_t N,
+                                               ck::index_t C,
+                                               std::vector<ck::index_t> input_spatial_lengths,
+                                               std::vector<ck::index_t> window_spatial_lengths,
+                                               std::vector<ck::index_t> output_spatial_lengths,
+                                               std::vector<ck::index_t> window_strides,
+                                               std::vector<ck::index_t> input_left_pads,
+                                               std::vector<ck::index_t> input_right_pads)
+    {
+        const index_t Di = input_spatial_lengths[0];
+        const index_t Hi = input_spatial_lengths[1];
+        const index_t Wi = input_spatial_lengths[2];
+
+        const index_t Do = output_spatial_lengths[0];
+        const index_t Ho = output_spatial_lengths[1];
+        const index_t Wo = output_spatial_lengths[2];
+
+        const index_t Z = window_spatial_lengths[0];
+        const index_t Y = window_spatial_lengths[1];
+        const index_t X = window_spatial_lengths[2];
+
+        const index_t ConvStrideD = window_strides[0];
+        const index_t ConvStrideH = window_strides[1];
+        const index_t ConvStrideW = window_strides[2];
+
+        const index_t InLeftPadD = input_left_pads[0];
+        const index_t InLeftPadH = input_left_pads[1];
+        const index_t InLeftPadW = input_left_pads[2];
+
+        const index_t InRightPadD = input_right_pads[0];
+        const index_t InRightPadH = input_right_pads[1];
+        const index_t InRightPadW = input_right_pads[2];
+
+        const index_t MRaw = N * Do * Ho * Wo * C;
+        const index_t MPad = math::integer_least_multiple(MRaw, M_BlockTileSize) - MRaw;
+
+        const index_t KRaw = Z * Y * X;
+        const index_t KPad = math::integer_least_multiple(KRaw, K_BlockTileSize) - KRaw;
+
+        // A[ReduceM, ReduceK]
+        const auto in_grid_desc_n_di_hi_wi_c =
+            make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
+
+        const auto in_grid_desc_n_dip_hip_wip_c = transform_tensor_descriptor(
+            in_grid_desc_n_di_hi_wi_c,
+            make_tuple(make_pass_through_transform(N),
+                       make_pad_transform(Di, InLeftPadD, InRightPadD),
+                       make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                       make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+        const auto in_grid_desc_n_z_do_y_ho_x_wo_c = transform_tensor_descriptor(
+            in_grid_desc_n_dip_hip_wip_c,
+            make_tuple(make_pass_through_transform(N),
+                       make_embed_transform(make_tuple(Z, Do), make_tuple(I1, ConvStrideD)),
+                       make_embed_transform(make_tuple(Y, Ho), make_tuple(I1, ConvStrideH)),
+                       make_embed_transform(make_tuple(X, Wo), make_tuple(I1, ConvStrideW)),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+            make_tuple(Sequence<0>{},
+                       Sequence<1, 2>{},
+                       Sequence<3, 4>{},
+                       Sequence<5, 6>{},
+                       Sequence<7>{}));
+
+        const auto in_grid_desc_reducemraw_reducekraw = transform_tensor_descriptor(
+            in_grid_desc_n_z_do_y_ho_x_wo_c,
+            make_tuple(make_merge_transform(make_tuple(N, Do, Ho, Wo, C)),
+                       make_merge_transform(make_tuple(Z, Y, X))),
+            make_tuple(Sequence<0, 2, 4, 6, 7>{}, Sequence<1, 3, 5>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto in_grid_desc_reducem_reducek = transform_tensor_descriptor(
+            in_grid_desc_reducemraw_reducekraw,
+            make_tuple(make_right_pad_transform(MRaw, MPad), make_right_pad_transform(KRaw, KPad)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        // B[ReduceM]
+        const auto out_grid_desc_reducemraw =
+            make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo * C));
+
+        const auto out_grid_desc_reducem =
+            transform_tensor_descriptor(out_grid_desc_reducemraw,
+                                        make_tuple(make_right_pad_transform(MRaw, MPad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+
+        return make_tuple(in_grid_desc_reducem_reducek, out_grid_desc_reducem);
+    }
+
+    using ABGridDescs   = decltype(MakeABGridDescriptor_A_M_K_B_M(1, 1, {}, {}, {}, {}, {}, {}));
+    using AGridDesc_M_K = remove_cvref_t<decltype(ABGridDescs{}[I0])>;
+    using BGridDesc_M   = remove_cvref_t<decltype(ABGridDescs{}[I1])>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const InDataType* p_in_dev,
+                 OutDataType* p_out_dev,
+                 IndexDataType* p_out_indices_dev,
+                 ck::index_t N,
+                 ck::index_t C,
+                 std::vector<ck::index_t>& input_spatial_lengths,
+                 std::vector<ck::index_t>& window_spatial_lengths,
+                 std::vector<ck::index_t>& output_spatial_lengths,
+                 std::vector<ck::index_t>& window_strides,
+                 std::vector<ck::index_t>& input_left_pads,
+                 std::vector<ck::index_t>& input_right_pads)
+            : p_in_dev_{p_in_dev},
+              p_out_dev_{p_out_dev},
+              p_out_indices_dev_{p_out_indices_dev},
+              a_grid_desc_m_k_{},
+              b_grid_desc_m_{}
+        {
+            const auto descs = MakeABGridDescriptor_A_M_K_B_M(N,
+                                                              C,
+                                                              input_spatial_lengths,
+                                                              window_spatial_lengths,
+                                                              output_spatial_lengths,
+                                                              window_strides,
+                                                              input_left_pads,
+                                                              input_right_pads);
+
+            a_grid_desc_m_k_ = descs[I0];
+            b_grid_desc_m_   = descs[I1];
+
+            invariant_lowest_length_ = C;
+
+            int32_t reduceLength =
+                window_spatial_lengths[0] * window_spatial_lengths[1] * window_spatial_lengths[2];
+
+            std::tie(in_element_op_, acc_element_op_) =
+                reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(reduceLength);
+        }
+
+        const InDataType* p_in_dev_;
+        OutDataType* p_out_dev_;
+        IndexDataType* p_out_indices_dev_;
+        AGridDesc_M_K a_grid_desc_m_k_;
+        BGridDesc_M b_grid_desc_m_;
+        InElementwiseOperation in_element_op_;
+        AccElementwiseOperation acc_element_op_;
+
+        // for checking vector load/store
+        ck::index_t invariant_lowest_length_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            using gridwise_reduce =
+                GridwiseReduction_mk_to_m_threadwise<InDataType,
+                                                     OutDataType,
+                                                     ComputeDataType,
+                                                     IndexDataType,
+                                                     AGridDesc_M_K,
+                                                     BGridDesc_M,
+                                                     ReduceOperation,
+                                                     InElementwiseOperation,
+                                                     AccElementwiseOperation,
+                                                     InMemoryDataOperationEnum::Set,
+                                                     false, // propagate_nan
+                                                     BlockSize,
+                                                     MThreadSliceSize,
+                                                     KThreadSliceSize,
+                                                     InSrcOutDstVectorDim,
+                                                     InSrcOutDstVectorSize,
+                                                     InSrcOutDstVectorSize>;
+
+            const auto kernel =
+                kernel_reduce_threadwise<gridwise_reduce,
+                                         OutputIndex,
+                                         true,  // pooling need to return global index
+                                         false, // don't have index input
+                                         InDataType,
+                                         OutDataType,
+                                         ComputeDataType,
+                                         IndexDataType,
+                                         AGridDesc_M_K,
+                                         BGridDesc_M,
+                                         InElementwiseOperation,
+                                         AccElementwiseOperation>;
+
+            ck::index_t M = arg.a_grid_desc_m_k_.GetLength(I0);
+
+            const index_t grid_size = (M / M_BlockTileSize);
+
+            return launch_and_time_kernel(stream_config,
+                                          kernel,
+                                          dim3(grid_size),
+                                          dim3(BlockSize),
+                                          0,
+                                          arg.a_grid_desc_m_k_,
+                                          arg.b_grid_desc_m_,
+                                          arg.in_element_op_,
+                                          arg.acc_element_op_,
+                                          float(1),
+                                          arg.p_in_dev_,
+                                          nullptr,
+                                          float(0),
+                                          arg.p_out_dev_,
+                                          arg.p_out_indices_dev_);
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
+
+        if(pArg->invariant_lowest_length_ % InSrcOutDstVectorSize != 0)
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in_dev,
+                        void* p_out_dev,
+                        void* p_out_indices_dev,
+                        std::vector<ck::index_t> input_lengths,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> output_lengths,
+                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
+                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> pooling_dims) override
+    {
+        if(input_lengths.size() != InOutRank || window_lengths.size() != WindowRank ||
+           input_lengths.size() != InOutRank || window_strides.size() != WindowRank ||
+           input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+            throw std::runtime_error("dimension is incorrect");
+
+        if(pooling_dims != std::vector<ck::index_t>{2, 3, 4})
+            throw std::runtime_error("pooling_dims only support {2, 3, 4} in pool3d so far");
+
+        index_t N  = input_lengths[0];
+        index_t C  = input_lengths[1];
+        index_t Di = input_lengths[2];
+        index_t Hi = input_lengths[3];
+        index_t Wi = input_lengths[4];
+        index_t Do = output_lengths[2];
+        index_t Ho = output_lengths[3];
+        index_t Wo = output_lengths[4];
+
+        std::vector<ck::index_t> input_spatial_lengths  = {Di, Hi, Wi};
+        std::vector<ck::index_t> output_spatial_lengths = {Do, Ho, Wo};
+
+        return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_dev),
+                                          static_cast<OutDataType*>(p_out_dev),
+                                          static_cast<IndexDataType*>(p_out_indices_dev),
+                                          N,
+                                          C,
+                                          input_spatial_lengths,
+                                          window_lengths,
+                                          output_spatial_lengths,
+                                          window_strides,
+                                          input_left_pads,
+                                          input_right_pads);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<" << BlockSize << ",";
+        str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
+        str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
+        str <<"InSrcOutDstVectorSize_" << InSrcOutDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
index a1d976f1a1..aa255da645 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
@@ -28,6 +28,7 @@ template <typename InDataType,
           typename AccElementwiseOperation,
           bool PropagateNan,
           bool OutputIndex,
+          bool TransformIndexKtoGlobal,
           bool HaveIndexInputIfOutputIndex,
           index_t BlockSize,
           index_t MThreadSliceSize,
@@ -260,6 +261,7 @@ struct DeviceReduceThreadWise : public DeviceReduce<InDataType,
 
             const auto kernel = kernel_reduce_threadwise<GridwiseReduce,
                                                          OutputIndex,
+                                                         TransformIndexKtoGlobal,
                                                          HaveIndexInput,
                                                          InDataType,
                                                          OutDataType,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
index 6c5bd29f9b..5986641c6f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
@@ -15,6 +15,7 @@ namespace ck {
 
 template <typename GridwiseReduction,
           bool OutputIndex,
+          bool TransformIndexKtoGlobal,
           bool HaveIndexInput,
           typename InDataType,
           typename OutDataType,
@@ -48,16 +49,17 @@ __global__ void kernel_reduce_threadwise(const InGridDesc_M_K in_grid_desc_m_k,
     }
     else
     {
-        GridwiseReduction::template RunWithIndex<HaveIndexInput>(in_grid_desc_m_k,
-                                                                 out_grid_desc_m,
-                                                                 in_elementwise_op,
-                                                                 acc_elementwise_op,
-                                                                 alpha,
-                                                                 p_in_value_global,
-                                                                 p_in_index_global,
-                                                                 beta,
-                                                                 p_out_value_global,
-                                                                 p_out_index_global);
+        GridwiseReduction::template RunWithIndex<TransformIndexKtoGlobal, HaveIndexInput>(
+            in_grid_desc_m_k,
+            out_grid_desc_m,
+            in_elementwise_op,
+            acc_elementwise_op,
+            alpha,
+            p_in_value_global,
+            p_in_index_global,
+            beta,
+            p_out_value_global,
+            p_out_index_global);
     };
 };
 
@@ -232,7 +234,7 @@ struct GridwiseReduction_mk_to_m_threadwise
             reduced_data_desc, make_tuple(I0), accu_value_buf, out_grid_desc_m, dst_global_buf);
     };
 
-    template <bool HaveIndexInput>
+    template <bool TransformIndexKtoGlobal, bool HaveIndexInput>
     __device__ static void RunWithIndex(const InGridDesc_M_K& in_grid_desc_m_k,
                                         const OutGridDesc_M& out_grid_desc_m,
                                         const InElementwiseOperation& in_elementwise_op,
@@ -390,6 +392,18 @@ struct GridwiseReduction_mk_to_m_threadwise
                 indexStart += KThreadSliceSize;
                 reducedLength += KThreadSliceSize;
             } while(reducedLength < toReduceLength);
+
+            if constexpr(TransformIndexKtoGlobal)
+            {
+                static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                    const auto coord = make_tensor_coordinate(
+                        in_grid_desc_m_k,
+                        make_multi_index(thread_global_1d_id * MThreadSliceSize + I,
+                                         accu_index_buf(I)));
+
+                    accu_index_buf(I) = coord.GetOffset();
+                });
+            }
         };
 
         // for indiced operation, acc_elementwise_op shoud do nothing
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
new file mode 100644
index 0000000000..3fc35a83c7
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
@@ -0,0 +1,345 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+#include <algorithm>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/utility/reduction_functions_accumulate.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+template <index_t InOutRank,
+          index_t WindowRank,
+          typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+struct ReferencePoolingFwd : public device::BaseOperator
+{
+    using ReduceOperation = typename ck::reduce_binary_operator<ReduceOpId>::opType;
+
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const Tensor<InDataType>& in,
+                 Tensor<OutDataType>& out,
+                 Tensor<IndexDataType>& out_indices,
+                 const std::vector<ck::index_t>& window_spatial_lengths,
+                 const std::vector<ck::index_t>& window_strides,
+                 const std::vector<ck::index_t>& in_left_pads,
+                 const std::vector<ck::index_t>& /*in_right_pads*/)
+            : in_(in),
+              out_(out),
+              out_indices_(out_indices),
+              window_spatial_lengths_(window_spatial_lengths),
+              window_strides_(window_strides),
+              in_left_pads_(in_left_pads),
+              reduceLength_(1)
+        {
+            static_for<0, WindowRank, 1>{}(
+                [&](auto I) { reduceLength_ *= window_spatial_lengths[I]; });
+        }
+
+        const Tensor<InDataType>& in_;
+        Tensor<OutDataType>& out_;
+        Tensor<IndexDataType>& out_indices_;
+        const std::vector<ck::index_t>& window_spatial_lengths_;
+        const std::vector<ck::index_t>& window_strides_;
+        const std::vector<ck::index_t>& in_left_pads_;
+        int reduceLength_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        float RunPooling3dFwd(const Argument& arg)
+        {
+
+            auto elementwise_ops =
+                ck::reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
+                    arg.reduceLength_);
+
+            auto in_elementwise_op  = std::get<0>(elementwise_ops);
+            auto acc_elementwise_op = std::get<1>(elementwise_ops);
+
+            if constexpr(!OutputIndex)
+            {
+                using Accumulation = ck::detail::
+                    AccumulateWithNanCheck<PropagateNan, ReduceOperation, ComputeDataType>;
+
+                auto f_ncdhw = [&](auto n, auto c, auto do_, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+
+                    for(ck::index_t z = 0; z < arg.window_spatial_lengths_[0]; ++z)
+                    {
+                        ck::index_t di = do_ * arg.window_strides_[0] + z - arg.in_left_pads_[0];
+                        for(ck::index_t y = 0; y < arg.window_spatial_lengths_[1]; ++y)
+                        {
+                            ck::index_t hi = ho * arg.window_strides_[1] + y - arg.in_left_pads_[1];
+                            for(ck::index_t x = 0; x < arg.window_spatial_lengths_[2]; ++x)
+                            {
+                                ck::index_t wi =
+                                    wo * arg.window_strides_[2] + x - arg.in_left_pads_[2];
+                                if(di >= 0 &&
+                                   di < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                                   hi >= 0 &&
+                                   hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]) &&
+                                   wi >= 0 &&
+                                   wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[4]))
+                                {
+                                    ComputeDataType currVal =
+                                        static_cast<ComputeDataType>(arg.in_(n, c, di, hi, wi));
+
+                                    in_elementwise_op(currVal, currVal);
+
+                                    Accumulation::Calculate(accuVal, currVal);
+                                }
+                            }
+                        }
+                    }
+                    acc_elementwise_op(accuVal, accuVal);
+
+                    arg.out_(n, c, do_, ho, wo) = accuVal;
+                };
+
+                make_ParallelTensorFunctor(f_ncdhw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3],
+                                           arg.out_.mDesc.GetLengths()[4])(
+                    std::thread::hardware_concurrency());
+            }
+            else
+            {
+                using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
+                                                                                ReduceOperation,
+                                                                                ComputeDataType,
+                                                                                IndexDataType>;
+
+                auto f_ncdhw = [&](auto n, auto c, auto do_, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+                    IndexDataType accuIndex = 0;
+
+                    for(ck::index_t z = 0; z < arg.window_spatial_lengths_[0]; ++z)
+                    {
+                        ck::index_t di = do_ * arg.window_strides_[0] + z - arg.in_left_pads_[0];
+                        for(ck::index_t y = 0; y < arg.window_spatial_lengths_[1]; ++y)
+                        {
+                            ck::index_t hi = ho * arg.window_strides_[1] + y - arg.in_left_pads_[1];
+                            for(ck::index_t x = 0; x < arg.window_spatial_lengths_[2]; ++x)
+                            {
+                                ck::index_t wi =
+                                    wo * arg.window_strides_[2] + x - arg.in_left_pads_[2];
+                                if(di >= 0 &&
+                                   di < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                                   hi >= 0 &&
+                                   hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]) &&
+                                   wi >= 0 &&
+                                   wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[4]))
+                                {
+                                    ComputeDataType currVal =
+                                        static_cast<ComputeDataType>(arg.in_(n, c, di, hi, wi));
+                                    IndexDataType currIndex =
+                                        arg.in_.GetOffsetFromMultiIndex(n, c, di, hi, wi);
+
+                                    in_elementwise_op(currVal, currVal);
+
+                                    Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
+                                }
+                            }
+                        }
+                    }
+
+                    acc_elementwise_op(accuVal, accuVal);
+
+                    arg.out_(n, c, do_, ho, wo)         = accuVal;
+                    arg.out_indices_(n, c, do_, ho, wo) = accuIndex;
+                };
+
+                make_ParallelTensorFunctor(f_ncdhw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3],
+                                           arg.out_.mDesc.GetLengths()[4])(
+                    std::thread::hardware_concurrency());
+            };
+
+            return 0;
+        }
+
+        float RunPooling2dFwd(const Argument& arg)
+        {
+
+            auto elementwise_ops =
+                ck::reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(
+                    arg.reduceLength_);
+
+            auto in_elementwise_op  = std::get<0>(elementwise_ops);
+            auto acc_elementwise_op = std::get<1>(elementwise_ops);
+
+            if constexpr(!OutputIndex)
+            {
+                using Accumulation = ck::detail::
+                    AccumulateWithNanCheck<PropagateNan, ReduceOperation, ComputeDataType>;
+
+                auto f_nchw = [&](auto n, auto c, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+
+                    for(ck::index_t y = 0; y < arg.window_spatial_lengths_[0]; ++y)
+                    {
+                        ck::index_t hi = ho * arg.window_strides_[0] + y - arg.in_left_pads_[0];
+                        for(ck::index_t x = 0; x < arg.window_spatial_lengths_[1]; ++x)
+                        {
+                            ck::index_t wi = wo * arg.window_strides_[1] + x - arg.in_left_pads_[1];
+                            if(hi >= 0 &&
+                               hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                               wi >= 0 &&
+                               wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]))
+                            {
+                                ComputeDataType currVal =
+                                    static_cast<ComputeDataType>(arg.in_(n, c, hi, wi));
+
+                                in_elementwise_op(currVal, currVal);
+
+                                Accumulation::Calculate(accuVal, currVal);
+                            }
+                        }
+                    }
+
+                    acc_elementwise_op(accuVal, accuVal);
+                    arg.out_(n, c, ho, wo) = accuVal;
+                };
+
+                make_ParallelTensorFunctor(f_nchw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3])(
+                    std::thread::hardware_concurrency());
+            }
+            else
+            {
+                using Accumulation = ck::detail::AccumulateWithIndexAndNanCheck<PropagateNan,
+                                                                                ReduceOperation,
+                                                                                ComputeDataType,
+                                                                                IndexDataType>;
+
+                auto f_nchw = [&](auto n, auto c, auto ho, auto wo) {
+                    auto accuVal = ReduceOperation::template GetIdentityValue<ComputeDataType>();
+                    IndexDataType accuIndex = 0;
+
+                    for(ck::index_t y = 0; y < arg.window_spatial_lengths_[0]; ++y)
+                    {
+                        ck::index_t hi = ho * arg.window_strides_[0] + y - arg.in_left_pads_[0];
+                        for(ck::index_t x = 0; x < arg.window_spatial_lengths_[1]; ++x)
+                        {
+                            ck::index_t wi = wo * arg.window_strides_[1] + x - arg.in_left_pads_[1];
+                            if(hi >= 0 &&
+                               hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
+                               wi >= 0 &&
+                               wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]))
+                            {
+                                ComputeDataType currVal =
+                                    static_cast<ComputeDataType>(arg.in_(n, c, hi, wi));
+
+                                IndexDataType currIndex =
+                                    arg.in_.GetOffsetFromMultiIndex(n, c, hi, wi);
+
+                                in_elementwise_op(currVal, currVal);
+
+                                Accumulation::Calculate(accuVal, currVal, accuIndex, currIndex);
+                            }
+                        }
+                    }
+
+                    acc_elementwise_op(accuVal, accuVal);
+                    arg.out_(n, c, ho, wo)         = accuVal;
+                    arg.out_indices_(n, c, ho, wo) = accuIndex;
+                };
+
+                make_ParallelTensorFunctor(f_nchw,
+                                           arg.out_.mDesc.GetLengths()[0],
+                                           arg.out_.mDesc.GetLengths()[1],
+                                           arg.out_.mDesc.GetLengths()[2],
+                                           arg.out_.mDesc.GetLengths()[3])(
+                    std::thread::hardware_concurrency());
+            };
+
+            return 0;
+        }
+
+        float Run(const Argument& arg)
+        {
+            // TODO - support generic pooling
+            if constexpr(InOutRank == 5 && WindowRank == 3)
+                return RunPooling3dFwd(arg);
+            else if constexpr(InOutRank == 4 && WindowRank == 2)
+                return RunPooling2dFwd(arg);
+            else
+                throw std::runtime_error("Only support pooling3d or pooling2d so far");
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(const Tensor<InDataType>& in,
+                             Tensor<OutDataType>& out,
+                             Tensor<IndexDataType>& out_indices,
+                             const std::vector<ck::index_t>& window_spatial_lengths,
+                             const std::vector<ck::index_t>& window_strides,
+                             const std::vector<ck::index_t>& in_left_pads,
+                             const std::vector<ck::index_t>& in_right_pads)
+    {
+        return Argument{in,
+                        out,
+                        out_indices,
+                        window_spatial_lengths,
+                        window_strides,
+                        in_left_pads,
+                        in_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferencePoolingFwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
new file mode 100644
index 0000000000..44d89cf361
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
@@ -0,0 +1,111 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto InOutRank  = 4;
+static constexpr auto WindowRank = 2;
+
+static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
+static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
+
+// FP16
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
+
+// FP16 - return index
+void add_device_pool2d_fwd_nhwc_index_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
+
+// FP32
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
+
+// FP32 - return index
+void add_device_pool2d_fwd_nhwc_index_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  OutputIndex>>
+{
+    using DeviceOp = DevicePoolFwd<InOutRank,
+                                   WindowRank,
+                                   InDataType,
+                                   OutDataType,
+                                   IndexDataType,
+                                   ReduceOpId,
+                                   OutputIndex>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
+                     is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool2d_fwd_nhwc_index_f16_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool2d_fwd_nhwc_f16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                          is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool2d_fwd_nhwc_index_f32_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool2d_fwd_nhwc_f32_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
new file mode 100644
index 0000000000..88523c7039
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -0,0 +1,111 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto InOutRank  = 5;
+static constexpr auto WindowRank = 3;
+
+static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
+static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
+
+// FP16
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
+
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
+
+// FP16 - return index
+void add_device_pool3d_fwd_ndhwc_index_f16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
+
+// FP32
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
+
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
+
+// FP32 - return index
+void add_device_pool3d_fwd_ndhwc_index_f32_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  OutputIndex>>
+{
+    using DeviceOp = DevicePoolFwd<InOutRank,
+                                   WindowRank,
+                                   InDataType,
+                                   OutDataType,
+                                   IndexDataType,
+                                   ReduceOpId,
+                                   OutputIndex>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
+                     is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool3d_fwd_ndhwc_index_f16_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                          is_same_v<IndexDataType, I32>)
+        {
+            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            {
+                add_device_pool3d_fwd_ndhwc_index_f32_instances(op_ptrs);
+            }
+            else
+            {
+                add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
index f77c50a8e8..325ed1e6d4 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
@@ -90,6 +90,7 @@ void add_device_reduce_instance_threadwise(
                                                             AccElementwiseOp,
                                                             PropagateNan,
                                                             OutputIndex,
+                                                            false,
                                                             false, // HaveIndexInputIfOutputIndex
                                                             cfg1::BlockSize_,
                                                             cfg2::MThreadSliceSize_,
diff --git a/library/include/ck/library/utility/host_tensor.hpp b/library/include/ck/library/utility/host_tensor.hpp
index 29d94b0036..844c29ed14 100644
--- a/library/include/ck/library/utility/host_tensor.hpp
+++ b/library/include/ck/library/utility/host_tensor.hpp
@@ -411,6 +411,12 @@ struct Tensor
         }
     }
 
+    template <typename... Is>
+    std::size_t GetOffsetFromMultiIndex(Is... is) const
+    {
+        return mDesc.GetOffsetFromMultiIndex(is...);
+    }
+
     template <typename... Is>
     T& operator()(Is... is)
     {
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
new file mode 100644
index 0000000000..0d0f896c8d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
@@ -0,0 +1,10 @@
+add_instance_library(device_pool_fwd_instance
+    device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+    device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+    device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+    device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+    device_max_pool2d_fwd_nhwc_f16_instance.cpp
+    device_max_pool2d_fwd_nhwc_f32_instance.cpp
+    device_max_pool3d_fwd_ndhwc_f16_instance.cpp
+    device_max_pool3d_fwd_ndhwc_f32_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
new file mode 100644
index 0000000000..38338ff997
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
new file mode 100644
index 0000000000..0f4a35deef
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
new file mode 100644
index 0000000000..6fcb519a9b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
new file mode 100644
index 0000000000..67ffd47085
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -0,0 +1,23 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
new file mode 100644
index 0000000000..a41cd0094b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F16, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F16, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
new file mode 100644
index 0000000000..fa70569bab
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool2d_fwd_nhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool2d_fwd_nhwc_index_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
new file mode 100644
index 0000000000..f3367b9460
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool3d_fwd_ndhwc_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F16, ReduceOpId, false>{});
+}
+
+void add_device_pool3d_fwd_ndhwc_index_f16_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F16, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
new file mode 100644
index 0000000000..8477a884da
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -0,0 +1,30 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool3d_fwd_ndhwc_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, false>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
+}
+
+void add_device_pool3d_fwd_ndhwc_index_f32_instances(
+    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, true>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
new file mode 100644
index 0000000000..cd508b55b0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I32 = int32_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+using device_pool2d_fwd_nhwc_instances =
+    // clang-format off
+    std::tuple <
+        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
+        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
+        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
+               // clang-format on
+               >;
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+using device_pool3d_fwd_ndhwc_instances =
+    // clang-format off
+    std::tuple <
+        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
+        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
+        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
+               // clang-format on
+               >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_pool2d_fwd_impl.hpp b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
new file mode 100644
index 0000000000..c313a00beb
--- /dev/null
+++ b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
@@ -0,0 +1,264 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+bool profile_pool2d_fwd_impl(int do_verification,
+                             int init_method,
+                             bool do_log,
+                             bool time_kernel,
+                             std::vector<index_t> in_length, // NCHW
+                             std::vector<index_t> window_spatial_lengths,
+                             std::vector<index_t> window_strides,
+                             std::vector<index_t> input_left_pads,
+                             std::vector<index_t> input_right_pads)
+{
+    constexpr index_t InOutRank  = 4;
+    constexpr index_t WindowRank = 2;
+
+    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
+       window_strides.size() != WindowRank || input_left_pads.size() != WindowRank ||
+       input_right_pads.size() != WindowRank)
+        return false;
+
+    std::vector<index_t> out_length(InOutRank);
+
+    int N = in_length[0];
+    int C = in_length[1];
+
+    out_length[0] = N;
+    out_length[1] = C;
+
+    // Calculate Ho, Wo
+    for(int i = 2; i < InOutRank; ++i)
+    {
+        auto pad1           = input_left_pads[i - 2];
+        auto pad2           = input_right_pads[i - 2];
+        auto windows_size   = window_spatial_lengths[i - 2];
+        auto windows_stride = window_strides[i - 2];
+        out_length[i]       = (in_length[i] + pad1 + pad2 - windows_size) / windows_stride + 1;
+    }
+
+    int Hi = in_length[2];
+    int Wi = in_length[3];
+    int Ho = out_length[2];
+    int Wo = out_length[3];
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+            return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
+        };
+
+    Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
+    Tensor<OutDataType> out_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    Tensor<OutDataType> out_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    switch(init_method)
+    {
+    case 0: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{}); break;
+    case 1: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5}); break;
+    default: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
+                                     out_indices_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
+
+    // add device normalization instances
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using ReferenceInstance = ck::tensor_operation::host::ReferencePoolingFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  ComputeDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  PropagateNan,
+                                                                                  OutputIndex>;
+
+        ReferenceInstance ref;
+        auto ref_argument = ref.MakeArgument(in_n_c_hi_wi,
+                                             out_n_c_ho_wo_host,
+                                             out_indices_n_c_ho_wo_host,
+                                             window_spatial_lengths,
+                                             window_strides,
+                                             input_left_pads,
+                                             input_right_pads);
+        auto ref_invoker  = ref.MakeInvoker();
+        ref_invoker.Run(ref_argument);
+    }
+
+    int num_kernel = 0;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            {C * Hi * Wi, 1, Wi * C, C},
+            {C * Ho * Wo, 1, Wo * C, C},
+            {C * Ho * Wo, 1, Wo * C, C},
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3});
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
+                LogRange(std::cout << "input lengths = ", in_length, ", ") << std::endl;
+            }
+
+            continue;
+        }
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+
+        float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        std::size_t num_bytes = in_n_c_hi_wi.mDesc.GetElementSize() * sizeof(InDataType) +
+                                out_n_c_ho_wo_host.mDesc.GetElementSize() * sizeof(OutDataType);
+
+        if constexpr(OutputIndex)
+            num_bytes += out_indices_n_c_ho_wo_host.mDesc.GetElementSize() * sizeof(IndexDataType);
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
+
+            bool pass = ck::utils::check_err(out_n_c_ho_wo_device.mData,
+                                             out_n_c_ho_wo_host.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+
+            if constexpr(OutputIndex)
+            {
+                out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data());
+
+                pass = pass && ck::utils::check_err(out_indices_n_c_ho_wo_device,
+                                                    out_indices_n_c_ho_wo_host);
+            }
+
+            if(do_log)
+            {
+                LogRangeAsType<float>(std::cout << "in_n_c_hi_wi  : ", in_n_c_hi_wi.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_ho_wo_host  : ", out_n_c_ho_wo_host.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_ho_wo_device  : ", out_n_c_ho_wo_device.mData, ",")
+                    << std::endl;
+
+                if constexpr(OutputIndex)
+                    LogRangeAsType<float>(std::cout << "out_indices_n_c_ho_wo_device  : ",
+                                          out_indices_n_c_ho_wo_device.mData,
+                                          ",")
+                        << std::endl;
+            }
+
+            if(!pass)
+            {
+                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
+                LogRange(std::cout << "lengths = [", in_length, ", ") << "]." << std::endl;
+                return false;
+            }
+            else
+            {
+                if(time_kernel)
+                    std::cout << "pass" << std::endl;
+            }
+        }
+    }
+
+    if(time_kernel)
+    {
+        LogRange(std::cout << "length = ", in_length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_pool3d_fwd_impl.hpp b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
new file mode 100644
index 0000000000..c9e4c193f8
--- /dev/null
+++ b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
@@ -0,0 +1,271 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename ComputeDataType,
+          typename IndexDataType,
+          ck::ReduceTensorOp ReduceOpId,
+          bool PropagateNan,
+          bool OutputIndex>
+bool profile_pool3d_fwd_impl(int do_verification,
+                             int init_method,
+                             bool do_log,
+                             bool time_kernel,
+                             std::vector<index_t> in_length, // NCDHW
+                             std::vector<index_t> window_spatial_lengths,
+                             std::vector<index_t> window_strides,
+                             std::vector<index_t> input_left_pads,
+                             std::vector<index_t> input_right_pads)
+{
+    constexpr index_t InOutRank  = 5;
+    constexpr index_t WindowRank = 3;
+
+    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
+       window_strides.size() != WindowRank || input_left_pads.size() != WindowRank ||
+       input_right_pads.size() != WindowRank)
+        return false;
+
+    std::vector<index_t> out_length(InOutRank);
+
+    int N = in_length[0];
+    int C = in_length[1];
+
+    out_length[0] = N;
+    out_length[1] = C;
+
+    // Calculate Do, Ho, Wo
+    for(int i = 2; i < InOutRank; ++i)
+    {
+        auto pad1           = input_left_pads[i - 2];
+        auto pad2           = input_right_pads[i - 2];
+        auto windows_size   = window_spatial_lengths[i - 2];
+        auto windows_stride = window_strides[i - 2];
+        out_length[i]       = (in_length[i] + pad1 + pad2 - windows_size) / windows_stride + 1;
+    }
+
+    int Di = in_length[2];
+    int Hi = in_length[3];
+    int Wi = in_length[4];
+    int Do = out_length[2];
+    int Ho = out_length[3];
+    int Wo = out_length[4];
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t D, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+        };
+
+    Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+    Tensor<OutDataType> out_n_c_do_ho_wo_host(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_host(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+
+    Tensor<OutDataType> out_n_c_do_ho_wo_device(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo_device(
+        f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+
+    switch(init_method)
+    {
+    case 0: in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{}); break;
+    case 1: in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5}); break;
+    default: in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_di_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
+                                     out_indices_n_c_do_ho_wo_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_di_hi_wi.mData.data());
+
+    // add device normalization instances
+    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                                 WindowRank,
+                                                                 InDataType,
+                                                                 OutDataType,
+                                                                 IndexDataType,
+                                                                 ReduceOpId,
+                                                                 OutputIndex>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using ReferenceInstance = ck::tensor_operation::host::ReferencePoolingFwd<InOutRank,
+                                                                                  WindowRank,
+                                                                                  InDataType,
+                                                                                  OutDataType,
+                                                                                  ComputeDataType,
+                                                                                  IndexDataType,
+                                                                                  ReduceOpId,
+                                                                                  PropagateNan,
+                                                                                  OutputIndex>;
+
+        ReferenceInstance ref;
+        auto ref_argument = ref.MakeArgument(in_n_c_di_hi_wi,
+                                             out_n_c_do_ho_wo_host,
+                                             out_indices_n_c_do_ho_wo_host,
+                                             window_spatial_lengths,
+                                             window_strides,
+                                             input_left_pads,
+                                             input_right_pads);
+        auto ref_invoker  = ref.MakeInvoker();
+        ref_invoker.Run(ref_argument);
+    }
+
+    int num_kernel = 0;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C},
+            {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+            {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+            window_strides,
+            input_left_pads,
+            input_right_pads,
+            {2, 3, 4});
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
+                LogRange(std::cout << "input lengths = ", in_length, ", ") << std::endl;
+            }
+
+            continue;
+        }
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+
+        float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        std::size_t num_bytes = in_n_c_di_hi_wi.mDesc.GetElementSize() * sizeof(InDataType) +
+                                out_n_c_do_ho_wo_host.mDesc.GetElementSize() * sizeof(OutDataType);
+
+        if constexpr(OutputIndex)
+            num_bytes +=
+                out_indices_n_c_do_ho_wo_host.mDesc.GetElementSize() * sizeof(IndexDataType);
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            out_device_buf.FromDevice(out_n_c_do_ho_wo_device.mData.data());
+
+            bool pass = ck::utils::check_err(out_n_c_do_ho_wo_device.mData,
+                                             out_n_c_do_ho_wo_host.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+
+            if constexpr(OutputIndex)
+            {
+                out_indices_device_buf.FromDevice(out_indices_n_c_do_ho_wo_device.mData.data());
+
+                pass = pass && ck::utils::check_err(out_indices_n_c_do_ho_wo_device,
+                                                    out_indices_n_c_do_ho_wo_host);
+            }
+
+            if(do_log)
+            {
+                LogRangeAsType<float>(
+                    std::cout << "in_n_c_di_hi_wi  : ", in_n_c_di_hi_wi.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_do_ho_wo_host  : ", out_n_c_do_ho_wo_host.mData, ",")
+                    << std::endl;
+                LogRangeAsType<float>(
+                    std::cout << "out_n_c_do_ho_wo_device  : ", out_n_c_do_ho_wo_device.mData, ",")
+                    << std::endl;
+
+                if constexpr(OutputIndex)
+                    LogRangeAsType<float>(std::cout << "out_indices_n_c_do_ho_wo_device  : ",
+                                          out_indices_n_c_do_ho_wo_device.mData,
+                                          ",")
+                        << std::endl;
+            }
+
+            if(!pass)
+            {
+                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
+                LogRange(std::cout << "lengths = [", in_length, ", ") << "]." << std::endl;
+                return false;
+            }
+            else
+            {
+                if(time_kernel)
+                    std::cout << "pass" << std::endl;
+            }
+        }
+    }
+
+    if(time_kernel)
+    {
+        LogRange(std::cout << "length = ", in_length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 0a50eedb73..c9fccc2588 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -25,6 +25,8 @@ set(PROFILER_SOURCES
     profile_reduce.cpp
     profile_groupnorm.cpp
     profile_layernorm.cpp
+    profile_avg_pool2d_fwd.cpp
+    profile_max_pool3d_fwd.cpp
     profile_softmax.cpp
     profile_batchnorm_fwd.cpp
     profile_batchnorm_bwd.cpp
@@ -74,4 +76,6 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
+
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_avg_pool2d_fwd.cpp b/profiler/src/profile_avg_pool2d_fwd.cpp
new file mode 100644
index 0000000000..b92288096d
--- /dev/null
+++ b/profiler/src/profile_avg_pool2d_fwd.cpp
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <unordered_map>
+
+#include "profiler/data_type_enum.hpp"
+#include "profiler/profile_pool2d_fwd_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+struct avgPoolFwdArgParser
+{
+    std::unordered_map<std::string, std::vector<int>> long_opts = {
+        {"length", {}}, {"wsize", {}}, {"wstride", {}}, {"pad1", {}}, {"pad2", {}}};
+
+    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
+    {
+        if(std::string("--") + key == argv[i])
+        {
+            int pos = i;
+            while(++i < argc && argv[i][0] != '-') {}
+            int end = i;
+            for(int j = pos + 1; j < end; j++)
+            {
+                long_opts[key].push_back(std::stoi(argv[j]));
+            }
+            return true;
+        }
+        return false;
+    }
+
+    void operator()(int argc, char* argv[])
+    {
+        for(auto& kv : long_opts)
+        {
+            for(int i = 1; i < argc; i++)
+            {
+                if(parse_opt(argc, argv, kv.first, i))
+                    break;
+            }
+        }
+    }
+};
+
+void print_help_avg_pool2d_fwd()
+{
+    std::cout << "arg1: data type (0: fp16; 1: fp32)\n"
+              << "arg2: verification (0: no; 1: yes)\n"
+              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+              << "arg4: print tensor value (0: no; 1: yes)\n"
+              << "arg5: time kernel (0=no, 1=yes)\n"
+              << "--length: input tensor length for NDHW(e.g, --length 2 32 30 30) \n"
+              << "--wsize: window size for YX (e.g, --wsize 2 2) \n"
+              << "--wstride: window stride for HW (e.g, --wstride 2 2) \n"
+              << "--pad1: left side of padding in HW (e.g, --pad1 1 1) \n"
+              << "--pad2: right side of padding in HW (e.g, --pad2 1 1) \n"
+              << "eg: ckProfiler avg_pool2d_fwd 0 1 2 0 1 0 --length 2 32 30 30 --wsize 2 2 "
+                 "--wstride 2 2 --pad1 1 1 --pad2 1 1"
+              << std::endl;
+}
+
+int profile_avg_pool2d_fwd(int argc, char* argv[])
+{
+    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
+    bool do_verification       = true;
+    int init_method            = 0;
+    bool do_log                = false;
+    bool time_kernel           = true;
+
+    std::vector<index_t> in_length = {2, 32, 30, 30};
+    std::vector<index_t> wsize     = {2, 2};
+    std::vector<index_t> wstride   = {2, 2};
+    std::vector<index_t> pad1      = {1, 1};
+    std::vector<index_t> pad2      = {1, 1};
+
+    if(argc != 2 && argc != 25)
+    {
+        print_help_avg_pool2d_fwd();
+        return 0;
+    }
+    else if(argc == 25)
+    {
+        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
+        do_verification = std::stoi(argv[3]);
+        init_method     = std::stoi(argv[4]);
+        do_log          = std::stoi(argv[5]);
+        time_kernel     = std::stoi(argv[6]);
+
+        // parse the long options
+        avgPoolFwdArgParser arg_parser;
+        arg_parser(argc, argv);
+        in_length = arg_parser.long_opts["length"];
+        wsize     = arg_parser.long_opts["wsize"];
+        wstride   = arg_parser.long_opts["wstride"];
+        pad1      = arg_parser.long_opts["pad1"];
+        pad2      = arg_parser.long_opts["pad2"];
+    }
+
+    using F16                 = ck::half_t;
+    using F32                 = float;
+    using I32                 = int32_t;
+    constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+    if(data_type == ck::DataTypeEnum::Half)
+    {
+        ck::profiler::profile_pool2d_fwd_impl<F16, F16, F32, I32, ReduceOpId, false, false>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            in_length,
+            wsize,
+            wstride,
+            pad1,
+            pad2);
+    }
+    else if(data_type == ck::DataTypeEnum::Float)
+    {
+        ck::profiler::profile_pool2d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, false>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            in_length,
+            wsize,
+            wstride,
+            pad1,
+            pad2);
+    }
+    else
+    {
+        throw std::runtime_error("not implemented yet");
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("avg_pool2d_fwd", "avg_pool2d fwd", profile_avg_pool2d_fwd);
diff --git a/profiler/src/profile_groupnorm.cpp b/profiler/src/profile_groupnorm.cpp
index 2741f52717..d55529a0f0 100644
--- a/profiler/src/profile_groupnorm.cpp
+++ b/profiler/src/profile_groupnorm.cpp
@@ -64,7 +64,7 @@ int profile_groupnorm(int argc, char* argv[])
     ck::DataTypeEnum data_type  = ck::DataTypeEnum::Half;
     bool do_verification        = false;
     int init_method             = 0;
-    bool do_log                 = 0;
+    bool do_log                 = false;
     bool time_kernel            = 1;
     std::vector<index_t> length = {64, 16, 16, 32, 40};
 
diff --git a/profiler/src/profile_max_pool3d_fwd.cpp b/profiler/src/profile_max_pool3d_fwd.cpp
new file mode 100644
index 0000000000..90c6e4e2b9
--- /dev/null
+++ b/profiler/src/profile_max_pool3d_fwd.cpp
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <unordered_map>
+
+#include "profiler/data_type_enum.hpp"
+#include "profiler/profile_pool3d_fwd_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+struct maxPoolFwdArgParser
+{
+    std::unordered_map<std::string, std::vector<int>> long_opts = {
+        {"length", {}}, {"wsize", {}}, {"wstride", {}}, {"pad1", {}}, {"pad2", {}}};
+
+    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
+    {
+        if(std::string("--") + key == argv[i])
+        {
+            int pos = i;
+            while(++i < argc && argv[i][0] != '-') {}
+            int end = i;
+            for(int j = pos + 1; j < end; j++)
+            {
+                long_opts[key].push_back(std::stoi(argv[j]));
+            }
+            return true;
+        }
+        return false;
+    }
+
+    void operator()(int argc, char* argv[])
+    {
+        for(auto& kv : long_opts)
+        {
+            for(int i = 1; i < argc; i++)
+            {
+                if(parse_opt(argc, argv, kv.first, i))
+                    break;
+            }
+        }
+    }
+};
+
+void print_help_max_pool3d_fwd()
+{
+    std::cout << "arg1: data type (0: fp16; 1: fp32)\n"
+              << "arg2: verification (0: no; 1: yes)\n"
+              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+              << "arg4: print tensor value (0: no; 1: yes)\n"
+              << "arg5: time kernel (0=no, 1=yes)\n"
+              << "arg6: return index (0=no, 1=yes)\n"
+              << "--length: input tensor length for NCDHW(e.g, --length 2 32 30 30 30) \n"
+              << "--wsize: window size for ZYX (e.g, --wsize 2 2 2) \n"
+              << "--wstride: window stride for DHW (e.g, --wstride 2 2 2) \n"
+              << "--pad1: left side of padding in DHW (e.g, --pad1 1 1 1) \n"
+              << "--pad2: right side of padding in DHW (e.g, --pad2 1 1 1) \n"
+              << "eg: ckProfiler max_pool3d_fwd 0 1 2 0 1 0 --length 2 32 30 30 30 --wsize 2 2 2 "
+                 "--wstride 2 2 2 --pad1 1 1 1 --pad2 1 1 1"
+              << std::endl;
+}
+
+int profile_max_pool3d_fwd(int argc, char* argv[])
+{
+    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
+    bool do_verification       = true;
+    int init_method            = 0;
+    bool do_log                = false;
+    bool time_kernel           = true;
+    bool return_index          = false;
+
+    std::vector<index_t> in_length = {2, 32, 30, 30, 30};
+    std::vector<index_t> wsize     = {2, 2, 2};
+    std::vector<index_t> wstride   = {2, 2, 2};
+    std::vector<index_t> pad1      = {1, 1, 1};
+    std::vector<index_t> pad2      = {1, 1, 1};
+
+    if(argc != 2 && argc != 30)
+    {
+        print_help_max_pool3d_fwd();
+        return 0;
+    }
+    else if(argc == 30)
+    {
+        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
+        do_verification = std::stoi(argv[3]);
+        init_method     = std::stoi(argv[4]);
+        do_log          = std::stoi(argv[5]);
+        time_kernel     = std::stoi(argv[6]);
+        return_index    = std::stoi(argv[7]);
+
+        // parse the long options
+        maxPoolFwdArgParser arg_parser;
+        arg_parser(argc, argv);
+        in_length = arg_parser.long_opts["length"];
+        wsize     = arg_parser.long_opts["wsize"];
+        wstride   = arg_parser.long_opts["wstride"];
+        pad1      = arg_parser.long_opts["pad1"];
+        pad2      = arg_parser.long_opts["pad2"];
+    }
+
+    using F16                 = ck::half_t;
+    using F32                 = float;
+    using I32                 = int32_t;
+    constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+    if(data_type == ck::DataTypeEnum::Half)
+    {
+        if(return_index)
+            ck::profiler::profile_pool3d_fwd_impl<F16, F16, F16, I32, ReduceOpId, false, true>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+        else
+            ck::profiler::profile_pool3d_fwd_impl<F16, F16, F16, I32, ReduceOpId, false, false>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+    }
+    else if(data_type == ck::DataTypeEnum::Float)
+    {
+        if(return_index)
+            ck::profiler::profile_pool3d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, true>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+        else
+            ck::profiler::profile_pool3d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, false>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                in_length,
+                wsize,
+                wstride,
+                pad1,
+                pad2);
+    }
+    else
+    {
+        throw std::runtime_error("not implemented yet");
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("max_pool3d_fwd", "max_pool3d fwd", profile_max_pool3d_fwd);
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 4f212d53aa..dad9b53cec 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -57,6 +57,7 @@ add_subdirectory(data_type)
 add_subdirectory(elementwise_normalization)
 add_subdirectory(batchnorm)
 add_subdirectory(contraction)
+add_subdirectory(pool_fwd)
 if(GPU_TARGETS MATCHES "gfx1100")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/pool_fwd/CMakeLists.txt b/test/pool_fwd/CMakeLists.txt
new file mode 100644
index 0000000000..6f59b95f6f
--- /dev/null
+++ b/test/pool_fwd/CMakeLists.txt
@@ -0,0 +1,16 @@
+add_custom_target(test_pool_fwd)
+
+add_gtest_executable(test_avg_pool2d_fwd test_avg_pool2d_fwd.cpp)
+add_gtest_executable(test_avg_pool3d_fwd test_avg_pool3d_fwd.cpp)
+add_gtest_executable(test_max_pool2d_fwd test_max_pool2d_fwd.cpp)
+add_gtest_executable(test_max_pool3d_fwd test_max_pool3d_fwd.cpp)
+
+target_link_libraries(test_avg_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_max_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
+
+add_dependencies(test_pool_fwd test_avg_pool2d_fwd)
+add_dependencies(test_pool_fwd test_avg_pool3d_fwd)
+add_dependencies(test_pool_fwd test_max_pool2d_fwd)
+add_dependencies(test_pool_fwd test_max_pool3d_fwd)
diff --git a/test/pool_fwd/test_avg_pool2d_fwd.cpp b/test/pool_fwd/test_avg_pool2d_fwd.cpp
new file mode 100644
index 0000000000..4e5f1e0e91
--- /dev/null
+++ b/test/pool_fwd/test_avg_pool2d_fwd.cpp
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool2d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestAvgPool2dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            bool success =
+                ck::profiler::profile_pool2d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::AVG,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestAvgPool2dFwd, KernelTypes);
+TYPED_TEST(TestAvgPool2dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool_fwd/test_avg_pool3d_fwd.cpp
new file mode 100644
index 0000000000..0d6b105b1b
--- /dev/null
+++ b/test/pool_fwd/test_avg_pool3d_fwd.cpp
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool3d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestAvgPool3dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            bool success =
+                ck::profiler::profile_pool3d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::AVG,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestAvgPool3dFwd, KernelTypes);
+TYPED_TEST(TestAvgPool3dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_max_pool2d_fwd.cpp b/test/pool_fwd/test_max_pool2d_fwd.cpp
new file mode 100644
index 0000000000..d16ac7faba
--- /dev/null
+++ b/test/pool_fwd/test_max_pool2d_fwd.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool2d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestMaxPool2dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            // max pool
+            bool success =
+                ck::profiler::profile_pool2d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::MAX,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+
+            // max pool + index
+            success = ck::profiler::profile_pool2d_fwd_impl<InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            false,
+                                                            true>(true,
+                                                                  2,
+                                                                  false,
+                                                                  false,
+                                                                  param.length_,
+                                                                  param.window_spatial_lengths_,
+                                                                  param.window_strides_,
+                                                                  param.input_left_pads_,
+                                                                  param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestMaxPool2dFwd, KernelTypes);
+TYPED_TEST(TestMaxPool2dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
+                    {{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool_fwd/test_max_pool3d_fwd.cpp
new file mode 100644
index 0000000000..f084dd9cbf
--- /dev/null
+++ b/test/pool_fwd/test_max_pool3d_fwd.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_pool3d_fwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestMaxPool3dFwd : public ::testing::Test
+{
+    protected:
+    using InDataType      = std::tuple_element_t<0, Tuple>;
+    using OutDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using IndexDataType   = std::tuple_element_t<3, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            // max pool
+            bool success =
+                ck::profiler::profile_pool3d_fwd_impl<InDataType,
+                                                      OutDataType,
+                                                      ComputeDataType,
+                                                      IndexDataType,
+                                                      ck::ReduceTensorOp::MAX,
+                                                      false,
+                                                      false>(true,
+                                                             2,
+                                                             false,
+                                                             false,
+                                                             param.length_,
+                                                             param.window_spatial_lengths_,
+                                                             param.window_strides_,
+                                                             param.input_left_pads_,
+                                                             param.input_right_pads_);
+            EXPECT_TRUE(success);
+
+            // max pool + index
+            success = ck::profiler::profile_pool3d_fwd_impl<InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            false,
+                                                            true>(true,
+                                                                  2,
+                                                                  false,
+                                                                  false,
+                                                                  param.length_,
+                                                                  param.window_spatial_lengths_,
+                                                                  param.window_strides_,
+                                                                  param.input_left_pads_,
+                                                                  param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+using KernelTypes =
+    ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
+
+TYPED_TEST_SUITE(TestMaxPool3dFwd, KernelTypes);
+TYPED_TEST(TestMaxPool3dFwd, Test_Pool)
+{
+    // length, window_length, window_stride, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_pool_fwd_common.hpp b/test/pool_fwd/test_pool_fwd_common.hpp
new file mode 100644
index 0000000000..a78785171d
--- /dev/null
+++ b/test/pool_fwd/test_pool_fwd_common.hpp
@@ -0,0 +1,31 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/ck.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+using I32 = int32_t;
+using ck::index_t;
+
+struct PoolingParam
+{
+    PoolingParam(const std::vector<index_t>& length,
+                 const std::vector<index_t>& window_spatial_lengths,
+                 const std::vector<index_t>& window_strides,
+                 const std::vector<index_t>& input_left_pads,
+                 const std::vector<index_t>& input_right_pads)
+        : length_(length),
+          window_spatial_lengths_(window_spatial_lengths),
+          window_strides_(window_strides),
+          input_left_pads_(input_left_pads),
+          input_right_pads_(input_right_pads)
+    {
+    }
+    std::vector<index_t> length_;
+    std::vector<index_t> window_spatial_lengths_;
+    std::vector<index_t> window_strides_;
+    std::vector<index_t> input_left_pads_;
+    std::vector<index_t> input_right_pads_;
+};

From ac9e01e2cc3721be24619807adc444e1f59a9d25 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 24 May 2023 08:11:25 -0700
Subject: [PATCH 018/194] Clean-up the headers (#713)

* fix headers for gpu instances

* remove unused headers

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 ...volution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp |  275 ---
 ...lution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp |  355 ----
 ...into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp |  150 --
 ...lution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp |  132 --
 ...into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp |  150 --
 ...lution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp |  135 --
 ...lution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp |  147 --
 ...volution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp |  260 ---
 ...volution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp |  179 --
 ...lution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp |  132 --
 ...lution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp |  132 --
 ...lution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp |  134 --
 ...volution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp |  135 --
 .../impl/device_gemm_bias_e_permute_xdl.hpp   |  586 ------
 .../grid/gridwise_contraction_dlops_v1r2.hpp  |  662 -------
 .../gpu/grid/gridwise_gemm_dlops_v1r2.hpp     |  608 -------
 .../gpu/grid/gridwise_gemm_dlops_v2.hpp       |  461 -----
 .../gpu/grid/gridwise_gemm_dlops_v3.hpp       | 1597 -----------------
 .../threadwise_tensor_slice_transfer_v3r3.hpp |  886 ---------
 include/ck/utility/amd_llvm_intrinsic.hpp     |   14 -
 include/ck/utility/print.hpp                  |   25 -
 .../cpu/reference_gemm_bias_2d.hpp            |  136 --
 .../cpu/reference_gemm_bias_activation.hpp    |  140 --
 .../reference_gemm_bias_activation_add.hpp    |  148 --
 .../gpu/batched_gemm.hpp                      |    4 +-
 .../gpu/batched_gemm_add_relu_gemm_add.hpp    |    4 +-
 ...batched_gemm_bias_softmax_gemm_permute.hpp |    4 +-
 .../gpu/batched_gemm_gemm.hpp                 |    4 +-
 .../gpu/batched_gemm_softmax_gemm_permute.hpp |    4 +-
 .../gpu/contraction_bilinear.hpp              |    2 -
 .../gpu/contraction_scale.hpp                 |    2 -
 .../gpu/convolution_backward_data.hpp         |    4 +-
 .../gpu/convolution_forward.hpp               |    4 +-
 .../gpu/device_elementwise_instance.hpp       |    3 +-
 .../device_gemm_mean_squaremean_instance.hpp  |    2 +-
 .../tensor_operation_instance/gpu/gemm.hpp    |    2 -
 .../gpu/gemm_add_add_fastgelu.hpp             |    2 -
 .../gpu/gemm_bilinear.hpp                     |    2 -
 .../gpu/gemm_splitk.hpp                       |    4 +-
 .../gpu/grouped_convolution_forward.hpp       |    2 +-
 .../gpu/grouped_gemm.hpp                      |    4 +-
 .../gpu/normalization.hpp                     |    4 +-
 .../include/ck/library/utility/host_conv.hpp  |  152 --
 .../ck/library/utility/op_instance_engine.hpp |  249 ---
 .../profiler/data_type_enum_helper.hpp        |   77 -
 .../profiler/profile_convnd_bwd_data_impl.hpp |  486 -----
 .../profile_convnd_bwd_weight_impl.hpp        |  474 -----
 47 files changed, 23 insertions(+), 9051 deletions(-)
 delete mode 100644 include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp
 delete mode 100644 include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
 delete mode 100644 include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp
 delete mode 100644 include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
 delete mode 100644 include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
 delete mode 100644 include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp
 delete mode 100644 include/ck/utility/amd_llvm_intrinsic.hpp
 delete mode 100644 include/ck/utility/print.hpp
 delete mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp
 delete mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp
 delete mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp
 delete mode 100644 library/include/ck/library/utility/host_conv.hpp
 delete mode 100644 library/include/ck/library/utility/op_instance_engine.hpp
 delete mode 100644 profiler/include/profiler/data_type_enum_helper.hpp
 delete mode 100644 profiler/include/profiler/profile_convnd_bwd_data_impl.hpp
 delete mode 100644 profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp

diff --git a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index db8e48df6d..0000000000
--- a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,275 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// Number of GEMMs = YTilde * XTilde
-// GemmM = C
-// GemmN = N * HTildeSlice * WTildeSlice
-// GemmK = K * YDotSlice * XDotSlice
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t IYTildeValue,
-          index_t IXTildeValue,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_data_convolution_into_gemm_v4r1_nhwc_kyxc_nhwk(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<IYTildeValue>,
-    Number<IXTildeValue>,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1  = Number<GemmK1Value>{};
-    constexpr auto IYTilde = Number<IYTildeValue>{};
-    constexpr auto IXTilde = Number<IXTildeValue>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
-    const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
-
-    const auto YTilde = ConvStrideH / GcdStrideDilationH;
-    const auto XTilde = ConvStrideW / GcdStrideDilationW;
-
-    const auto YDot = math::integer_divide_ceil(Y, YTilde);
-    const auto XDot = math::integer_divide_ceil(X, XTilde);
-
-    const auto HTilde = Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
-    const auto WTilde = Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
-
-    // only work on HTilde and WTilde that contribute to non-padding area of input tensor
-    const auto IHTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
-    const auto IWTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
-
-    const auto IHTildeSliceEnd =
-        math::min(HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
-    const auto IWTildeSliceEnd =
-        math::min(WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
-
-    const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
-    const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
-
-    // GemmK is different for each GEMM
-    const auto YDotSlice = math::integer_divide_ceil(Y - IYTilde, YTilde);
-    const auto XDotSlice = math::integer_divide_ceil(X - IXTilde, XTilde);
-
-    const auto K1 = GemmK1;
-    const auto K0 = K / K1;
-
-    // weight tensor
-    const auto wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc = transform_tensor_descriptor(
-        wei_k_y_x_c_grid_desc,
-        make_tuple(make_pass_through_transform(K),
-                   make_embed_transform(make_tuple(YDot, YTilde),
-                                        make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, XTilde),
-                                        make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto wei_k0_k1_ydotslice_xdotslice_c_grid_desc =
-        transform_tensor_descriptor(wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                                               make_slice_transform(YDot, I0, YDotSlice),
-                                               make_slice_transform(XDot, I0, XDotSlice),
-                                               make_freeze_transform(IYTilde),
-                                               make_freeze_transform(IXTilde),
-                                               make_pass_through_transform(C)),
-                                    make_tuple(Sequence<0>{},
-                                               Sequence<1>{},
-                                               Sequence<3>{},
-                                               Sequence<2>{},
-                                               Sequence<4>{},
-                                               Sequence<5>{}),
-                                    make_tuple(Sequence<0, 1>{},
-                                               Sequence<2>{},
-                                               Sequence<3>{},
-                                               Sequence<>{},
-                                               Sequence<>{},
-                                               Sequence<4>{}));
-
-#if 1
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<0, 2, 3>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // output tensor
-    // this add padding check
-    const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
-        out_n_ho_wo_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Ho, I0, I0),
-                   make_pad_transform(Wo, I0, I0),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto out_n_ydot_htilde_xdot_wtilde_k_grid_desc = transform_tensor_descriptor(
-        out_n_hop_wop_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YDot, HTilde),
-                                        make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, WTilde),
-                                        make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc =
-        transform_tensor_descriptor(
-            out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
-            make_tuple(make_pass_through_transform(N),
-                       make_slice_transform(YDot, I0, YDotSlice),
-                       make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                       make_slice_transform(XDot, I0, XDotSlice),
-                       make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                       make_unmerge_transform(make_tuple(K0, K1))),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5>{}),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5, 6>{}));
-
-#if 1
-    const auto out_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto out_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<5, 1, 3>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YTilde, HTilde),
-                                        make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(XTilde, WTilde),
-                                        make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_n_htildeslice_wtildeslice_c_grid_desc = transform_tensor_descriptor(
-        in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_freeze_transform(IYTilde),
-                   make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                   make_freeze_transform(IXTilde),
-                   make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{},
-                   Sequence<1>{},
-                   Sequence<2>{},
-                   Sequence<3>{},
-                   Sequence<4>{},
-                   Sequence<5>{}),
-        make_tuple(Sequence<0>{},
-                   Sequence<>{},
-                   Sequence<1>{},
-                   Sequence<>{},
-                   Sequence<2>{},
-                   Sequence<3>{}));
-
-    const auto in_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        in_n_htildeslice_wtildeslice_c_grid_desc,
-        make_tuple(make_pass_through_transform(C),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice))),
-        make_tuple(Sequence<3>{}, Sequence<0, 1, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(wei_gemmk0_gemmm_gemmk1_grid_desc,
-                      out_gemmk0_gemmn_gemmk1_grid_desc,
-                      in_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 5391b595b5..0000000000
--- a/include/ck/problem_transform/transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,355 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_DATA_CONVOLUTION_INTO_GEMM_V4R1R2_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: out
-// B: wei
-// C: in
-// Number of GEMMs = YTilde * XTilde
-// GemmM = N * HTildeSlice * WTildeSlice
-// GemmN = C
-// GemmK = K * YDotSlice * XDotSlice
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          typename IYTilde,
-          typename IXTilde,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk(
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    IYTilde i_ytilde,
-    IXTilde i_xtilde,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
-    const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
-
-    const auto YTilde = ConvStrideH / GcdStrideDilationH;
-    const auto XTilde = ConvStrideW / GcdStrideDilationW;
-
-    const auto YDot = math::integer_divide_ceil(Y, YTilde);
-    const auto XDot = math::integer_divide_ceil(X, XTilde);
-
-    const auto HTilde = Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
-    const auto WTilde = Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
-
-    // only work on HTilde and WTilde that contribute to non-padding area of input tensor
-    const auto IHTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
-    const auto IWTildeSliceBegin = math::integer_divide_floor(
-        math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
-
-    const auto IHTildeSliceEnd =
-        math::min(HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
-    const auto IWTildeSliceEnd =
-        math::min(WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
-
-    const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
-    const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
-
-    // GemmK is different for each GEMM
-    const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
-    const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
-
-    const auto K1 = GemmK1;
-    const auto K0 = K / K1;
-
-    // A: output tensor
-    // this add padding check
-    const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
-        out_n_ho_wo_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Ho, I0, I0),
-                   make_pad_transform(Wo, I0, I0),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto out_n_ydot_htilde_xdot_wtilde_k_grid_desc = transform_tensor_descriptor(
-        out_n_hop_wop_k_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YDot, HTilde),
-                                        make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, WTilde),
-                                        make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc =
-        transform_tensor_descriptor(
-            out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
-            make_tuple(make_pass_through_transform(N),
-                       make_slice_transform(YDot, I0, YDotSlice),
-                       make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                       make_slice_transform(XDot, I0, XDotSlice),
-                       make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                       make_unmerge_transform(make_tuple(K0, K1))),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5>{}),
-            make_tuple(Sequence<0>{},
-                       Sequence<1>{},
-                       Sequence<2>{},
-                       Sequence<3>{},
-                       Sequence<4>{},
-                       Sequence<5, 6>{}));
-
-#if 1
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k0_k1_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<5, 1, 3>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // B: weight tensor
-    const auto wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc = transform_tensor_descriptor(
-        wei_k_y_x_c_grid_desc,
-        make_tuple(make_pass_through_transform(K),
-                   make_embed_transform(make_tuple(YDot, YTilde),
-                                        make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
-                   make_embed_transform(make_tuple(XDot, XTilde),
-                                        make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto wei_k0_k1_ydotslice_xdotslice_c_grid_desc =
-        transform_tensor_descriptor(wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                                               make_slice_transform(YDot, I0, YDotSlice),
-                                               make_slice_transform(XDot, I0, XDotSlice),
-                                               make_freeze_transform(i_ytilde),
-                                               make_freeze_transform(i_xtilde),
-                                               make_pass_through_transform(C)),
-                                    make_tuple(Sequence<0>{},
-                                               Sequence<1>{},
-                                               Sequence<3>{},
-                                               Sequence<2>{},
-                                               Sequence<4>{},
-                                               Sequence<5>{}),
-                                    make_tuple(Sequence<0, 1>{},
-                                               Sequence<2>{},
-                                               Sequence<3>{},
-                                               Sequence<>{},
-                                               Sequence<>{},
-                                               Sequence<4>{}));
-
-#if 1
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K0)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#else
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        wei_k0_k1_ydotslice_xdotslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(K0, YDotSlice, XDotSlice)),
-                   make_pass_through_transform(C),
-                   make_pass_through_transform(K1)),
-        make_tuple(Sequence<0, 2, 3>{}, Sequence<4>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-#endif
-
-    // C: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(YTilde, HTilde),
-                                        make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(XTilde, WTilde),
-                                        make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_n_htildeslice_wtildeslice_c_grid_desc = transform_tensor_descriptor(
-        in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_freeze_transform(i_ytilde),
-                   make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                   make_freeze_transform(i_xtilde),
-                   make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{},
-                   Sequence<1>{},
-                   Sequence<2>{},
-                   Sequence<3>{},
-                   Sequence<4>{},
-                   Sequence<5>{}),
-        make_tuple(Sequence<0>{},
-                   Sequence<>{},
-                   Sequence<1>{},
-                   Sequence<>{},
-                   Sequence<2>{},
-                   Sequence<3>{}));
-
-    const auto in_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        in_n_htildeslice_wtildeslice_c_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0, 1, 2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmk0_gemmm_gemmk1_grid_desc,
-                      wei_gemmk0_gemmn_gemmk1_grid_desc,
-                      in_gemmm_gemmn_grid_desc);
-}
-
-// A: out
-// B: wei
-// C: in
-// Number of GEMMs = 1
-// GemmM = N * Ho * Wo
-// GemmN = C
-// GemmK = K
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_data_convolution_into_gemm_v4r1r2_nhwc_kyxc_nhwk_1x1(
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const TensorDescriptor<Wei...>& /* wei_k_y_x_c_grid_desc */,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const ConvStrides& conv_strides,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto K1 = GemmK1;
-    const auto K0 = K / K1;
-
-    // A: output tensor
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-                                    make_tuple(make_pass_through_transform(N * Ho * Wo),
-                                               make_unmerge_transform(make_tuple(K0, K1))),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0, 2>{}));
-
-    // B: weight tensor
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C)),
-        make_tuple(make_unmerge_transform(make_tuple(K0, K1)), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // C: input tensor
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(I1, Ho), make_tuple(I1, ConvStrideH)),
-                   make_embed_transform(make_tuple(I1, Wo), make_tuple(I1, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        in_n_y_ho_x_wo_c_grid_desc,
-        make_tuple(make_freeze_transform(I0),
-                   make_freeze_transform(I0),
-                   make_merge_transform(make_tuple(N, Ho, Wo)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<1>{}, Sequence<3>{}, Sequence<0, 2, 4>{}, Sequence<5>{}),
-        make_tuple(Sequence<>{}, Sequence<>{}, Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmk0_gemmm_gemmk1_grid_desc,
-                      wei_gemmk0_gemmn_gemmk1_grid_desc,
-                      in_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index bb1dc239f4..0000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,150 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_ATOMIC_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmK = N * Ho * Wo
-// GemmN = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value,
-          typename GemmKBatchType,
-          typename GemmKPadType>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r2_atomic_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>,
-    GemmKBatchType GemmKBatch,
-    GemmKPadType GemmKPad)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM      = K;
-    const auto GemmN      = C * Y * X;
-    const auto GemmKTotal = N * Ho * Wo;
-    const index_t GemmK0  = GemmKPad / (GemmKBatch * GemmK1);
-
-    // A: output tensor
-    const auto out_gemmktotal_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-        out_gemmktotal_gemmm_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_gemmkpad_gemmm_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // B: input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmktotal_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-        in_gemmktotal_gemmn_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        in_gemmkpad_gemmn_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index ca530934e4..0000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,132 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmK = N * Ho * Wo
-// GemmN = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = K;
-    const auto GemmN  = C * Y * X;
-    const auto GemmK  = N * Ho * Wo;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto out_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(out_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    return make_tuple(out_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index e960f90c4b..0000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,150 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_ATOMIC_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: in
-// B: wei
-// C: out
-// GemmM = N * Ho * Wo
-// GemmN = K
-// GemmK = Y * X * C
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value,
-          typename GemmKBatchType,
-          typename GemmKPadType>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r4_atomic_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>,
-    GemmKBatchType GemmKBatch,
-    GemmKPadType GemmKPad)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM      = Y * X * C;
-    const auto GemmN      = K;
-    const auto GemmKTotal = N * Ho * Wo;
-    const index_t GemmK0  = GemmKPad / (GemmKBatch * GemmK1);
-
-    // A: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmktotal_gemmm_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-        in_gemmktotal_gemmm_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        in_gemmkpad_gemmm_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // B: output tensor
-    const auto out_gemmktotal_gemmn_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-
-    const auto out_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-        out_gemmktotal_gemmn_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_gemmkpad_gemmn_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(in_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                      out_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 052bab423d..0000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,135 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: in
-// B: wei
-// C: out
-// GemmM = N * Ho * Wo
-// GemmN = K
-// GemmK = Y * X * C
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = Y * X * C;
-    const auto GemmN  = K;
-    const auto GemmK  = N * Ho * Wo;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // A: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmm_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // B: output tensor
-    const auto out_gemmk_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(out_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(in_gemmk0_gemmm_gemmk1_grid_desc,
-                      out_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index c301a9e0c6..0000000000
--- a/include/ck/problem_transform/transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,147 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_BACKWARD_WEIGHT_CONVOLUTION_INTO_GEMM_V4R4R5_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: out
-// B: in
-// C: wei
-// GemmM = K
-// GemmN = Y * X * C
-// GemmKTotal = N * Ho * Wo
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value,
-          typename GemmKBatchType,
-          typename GemmKPadType>
-__host__ __device__ constexpr auto
-transform_backward_weight_convolution_into_gemm_v4r4r5_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>,
-    GemmKBatchType GemmKBatch,
-    GemmKPadType GemmKPad)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM      = K;
-    const auto GemmN      = Y * X * C;
-    const auto GemmKTotal = N * Ho * Wo;
-    const index_t GemmK0  = GemmKPad / (GemmKBatch * GemmK1);
-
-    // A: output tensor
-    const auto out_gemmktotal_gemmm_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-
-    const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-        out_gemmktotal_gemmm_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-        out_gemmkpad_gemmm_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmM)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // B: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmktotal_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-        in_gemmktotal_gemmn_grid_desc,
-        make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-        in_gemmkpad_gemmn_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1)),
-                   make_pass_through_transform(GemmN)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-    // C: weight tensor
-    const auto wei_gemmm_gemmn_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
-
-    return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                      wei_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index 381f9ac9d6..0000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,260 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_global_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_global_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_global_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_global_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_global_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_global_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_global_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_global_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_global_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_global_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_global_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_global_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_global_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_global_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_global_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_global_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_global_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_global_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_global_desc, in_gemmk_gemmn_global_desc, out_gemmm_gemmn_global_desc);
-}
-
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw_no_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_global_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_global_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_global_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_global_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_global_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_global_desc.GetLength(I1);
-
-    const auto Ho = out_n_k_ho_wo_global_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_global_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_global_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_global_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    assert(InLeftPadH == 0 && InLeftPadW == 0 && InRightPadH == 0 && InRightPadW == 0);
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_n_c_y_ho_x_wo_global_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_global_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_global_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_global_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_global_desc, in_gemmk_gemmn_global_desc, out_gemmm_gemmn_global_desc);
-}
-
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nchw_kcyx_nkhw_1x1(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_global_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_global_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_global_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_global_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_global_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_global_desc.GetLength(I1);
-
-    const auto Ho = out_n_k_ho_wo_global_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_global_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_global_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_global_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    assert(Y == 1 && X == 1 && ConvStrideH == 1 && ConvStrideW == 1 && ConvDilationH == 1 &&
-           ConvDilationW == 1 && InLeftPadH == 0 && InLeftPadW == 0 && InRightPadH == 0 &&
-           InRightPadW == 0);
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_gemmk_gemmn_global_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_global_desc,
-        make_tuple(make_pass_through_transform(C), make_merge_transform(make_tuple(N, Ho, Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2, 3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_global_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_global_desc, in_gemmk_gemmn_global_desc, out_gemmm_gemmn_global_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index ebfaabb03e..0000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,179 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_grid_desc, in_gemmk_gemmn_grid_desc, out_gemmm_gemmn_grid_desc);
-}
-
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4_nhwc_kyxc_nhwk_1x1(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    assert(Y == 1 && X == 1 && ConvStrideH == 1 && ConvStrideW == 1 && ConvDilationH == 1 &&
-           ConvDilationW == 1 && InLeftPadH == 0 && InLeftPadW == 0 && InRightPadH == 0 &&
-           InRightPadW == 0);
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // input tensor
-    const auto in_gemmk_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, C)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(
-        wei_gemmk_gemmm_grid_desc, in_gemmk_gemmn_grid_desc, out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index 6e576d69f5..0000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,132 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_gemm_v4r4r2_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = K;
-    const auto GemmN  = N * Ho * Wo;
-    const auto GemmK  = C * Y * X;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(C * Y * X)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(wei_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_c_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_c_y_ho_x_wo_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(C, Y, X)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 2, 4>{}, Sequence<0, 3, 5>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo)),
-        make_tuple(make_pass_through_transform(K), make_merge_transform(make_tuple(N, Ho * Wo))),
-        make_tuple(Sequence<1>{}, Sequence<0, 2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(wei_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmk0_gemmn_gemmk1_grid_desc,
-                      out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 13e1bf251a..0000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,132 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R2_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM = K
-// GemmN = N * Ho * Wo
-// GemmK = C * Y * X
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_gemm_v4r4r2_nhwc_kyxc_nhwk_pad(
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = K;
-    const auto GemmN  = N * Ho * Wo;
-    const auto GemmK  = C * Y * X;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // weight tensor
-    const auto wei_gemmk_gemmm_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto wei_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(wei_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmn_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    return make_tuple(wei_gemmk0_gemmm_gemmk1_grid_desc,
-                      in_gemmk0_gemmn_gemmk1_grid_desc,
-                      out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
deleted file mode 100644
index 088d14b2ee..0000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk.hpp
+++ /dev/null
@@ -1,134 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// A: in
-// B: wei
-// C: out
-// GemmM = N * Ho * Wo
-// GemmN = K
-// GemmK = Y * X * C
-template <typename... In,
-          typename... Wei,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          index_t GemmK1Value>
-__host__ __device__ constexpr auto transform_forward_convolution_into_gemm_v4r4r4_nhwc_kyxc_nhwk(
-    const TensorDescriptor<In...>& in_n_hi_wi_c_grid_desc,
-    const TensorDescriptor<Wei...>& wei_k_y_x_c_grid_desc,
-    const TensorDescriptor<Out...>& out_n_ho_wo_k_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    Number<GemmK1Value>)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    constexpr auto GemmK1 = Number<GemmK1Value>{};
-
-    const auto N = in_n_hi_wi_c_grid_desc.GetLength(I0);
-    const auto C = in_n_hi_wi_c_grid_desc.GetLength(I3);
-    const auto K = out_n_ho_wo_k_grid_desc.GetLength(I3);
-
-    const auto Hi = in_n_hi_wi_c_grid_desc.GetLength(I1);
-    const auto Wi = in_n_hi_wi_c_grid_desc.GetLength(I2);
-
-    const auto Ho = out_n_ho_wo_k_grid_desc.GetLength(I1);
-    const auto Wo = out_n_ho_wo_k_grid_desc.GetLength(I2);
-
-    const auto Y = wei_k_y_x_c_grid_desc.GetLength(I1);
-    const auto X = wei_k_y_x_c_grid_desc.GetLength(I2);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto GemmM  = N * Ho * Wo;
-    const auto GemmN  = K;
-    const auto GemmK  = Y * X * C;
-    const auto GemmK0 = GemmK / GemmK1;
-
-    // A: input tensor
-    const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-        in_n_hi_wi_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-        in_n_hip_wip_c_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                   make_pass_through_transform(C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-    const auto in_gemmk_gemmm_grid_desc =
-        transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                    make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                               make_merge_transform(make_tuple(N, Ho, Wo))),
-                                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    const auto in_gemmk0_gemmm_gemmk1_grid_desc =
-        transform_tensor_descriptor(in_gemmk_gemmm_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmM)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // B: weight tensor
-    const auto wei_gemmk_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C)),
-        make_tuple(make_pass_through_transform(K), make_pass_through_transform(Y * X * C)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-    const auto wei_gemmk0_gemmn_gemmk1_grid_desc =
-        transform_tensor_descriptor(wei_gemmk_gemmn_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1)),
-                                               make_pass_through_transform(GemmN)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-    // C: output tensor
-    const auto out_gemmm_gemmn_grid_desc = transform_tensor_descriptor(
-        make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
-        make_tuple(make_pass_through_transform(N * Ho * Wo), make_pass_through_transform(K)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-    return make_tuple(in_gemmk0_gemmm_gemmk1_grid_desc,
-                      wei_gemmk0_gemmn_gemmk1_grid_desc,
-                      out_gemmm_gemmn_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp b/include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp
deleted file mode 100644
index a6785d56df..0000000000
--- a/include/ck/problem_transform/transform_forward_convolution_into_gemm_v6r1_nchw_kcyx_nkhw.hpp
+++ /dev/null
@@ -1,135 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP
-#define CK_TRANSFORM_FORWARD_CONVOLUTION_INTO_CONTRACTION_V6R1_NCHW_KCYX_NKHW_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-
-namespace ck {
-
-// GemmM0 = 1
-// GemmM1 = K
-// GemmN0 = N0
-// GemmN1 = (N / N0) * Ho * Wo
-// GemmK0 = (C / C0) * Y * X
-// GemmK1 = C0
-template <typename... Wei,
-          typename... In,
-          typename... Out,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads,
-          typename N0Type,
-          typename C0Type>
-__host__ __device__ constexpr auto
-transform_forward_convolution_into_contraction_v6r1_nchw_kcyx_nkhw_pad(
-    const TensorDescriptor<Wei...>& wei_k_c_y_x_grid_desc,
-    const TensorDescriptor<In...>& in_n_c_hi_wi_grid_desc,
-    const TensorDescriptor<Out...>& out_n_k_ho_wo_grid_desc,
-    const ConvStrides& conv_strides,
-    const ConvDilations& conv_dilations,
-    const InLeftPads& in_left_pads,
-    const InRightPads& in_right_pads,
-    const N0Type& N0,
-    const C0Type& C0)
-{
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    constexpr auto I3 = Number<3>{};
-
-    const auto N = in_n_c_hi_wi_grid_desc.GetLength(I0);
-    const auto C = in_n_c_hi_wi_grid_desc.GetLength(I1);
-    const auto K = out_n_k_ho_wo_grid_desc.GetLength(I1);
-
-    const auto Hi = in_n_c_hi_wi_grid_desc.GetLength(I2);
-    const auto Wi = in_n_c_hi_wi_grid_desc.GetLength(I3);
-
-    const auto Ho = out_n_k_ho_wo_grid_desc.GetLength(I2);
-    const auto Wo = out_n_k_ho_wo_grid_desc.GetLength(I3);
-
-    const auto Y = wei_k_c_y_x_grid_desc.GetLength(I2);
-    const auto X = wei_k_c_y_x_grid_desc.GetLength(I3);
-
-    const auto ConvStrideH = conv_strides[I0];
-    const auto ConvStrideW = conv_strides[I1];
-
-    const auto ConvDilationH = conv_dilations[I0];
-    const auto ConvDilationW = conv_dilations[I1];
-
-    const auto InLeftPadH = in_left_pads[I0];
-    const auto InLeftPadW = in_left_pads[I1];
-
-    const auto InRightPadH = in_right_pads[I0];
-    const auto InRightPadW = in_right_pads[I1];
-
-    const auto N1 = N / N0;
-    const auto C1 = C / C0;
-
-    // weight tensor
-    const auto wei_gk0_gm0_gm1_gk1_grid_desc =
-        transform_tensor_descriptor(make_naive_tensor_descriptor_packed(make_tuple(K, C * Y * X)),
-                                    make_tuple(make_unmerge_transform(make_tuple(I1, K)),
-                                               make_unmerge_transform(make_tuple(C0, C1 * Y * X))),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                    make_tuple(Sequence<1, 2>{}, Sequence<3, 0>{}));
-
-    // input tensor
-    const auto in_n_c_hip_wip_grid_desc = transform_tensor_descriptor(
-        in_n_c_hi_wi_grid_desc,
-        make_tuple(make_pass_through_transform(N),
-                   make_pass_through_transform(C),
-                   make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                   make_pad_transform(Wi, InLeftPadW, InRightPadW)),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    const auto in_n0_n1_c0_c1_y_ho_x_wo_grid_desc = transform_tensor_descriptor(
-        in_n_c_hip_wip_grid_desc,
-        make_tuple(make_unmerge_transform(make_tuple(N0, N1)),
-                   make_unmerge_transform(make_tuple(C0, C1)),
-                   make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                   make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW))),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-        make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}, Sequence<4, 5>{}, Sequence<6, 7>{}));
-
-    const auto in_gk0_gn0_gn1_gk1_grid_desc = transform_tensor_descriptor(
-        in_n0_n1_c0_c1_y_ho_x_wo_grid_desc,
-        make_tuple(make_merge_transform(make_tuple(C1, Y, X)),
-                   make_pass_through_transform(N0),
-                   make_merge_transform(make_tuple(N1, Ho, Wo)),
-                   make_pass_through_transform(C0)),
-        make_tuple(Sequence<3, 4, 6>{}, Sequence<0>{}, Sequence<1, 5, 7>{}, Sequence<2>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    // output tensor
-    const auto out_n_k_howo_grid_desc =
-        make_naive_tensor_descriptor_packed(make_tuple(N, K, Ho * Wo));
-
-    const auto out_n0_n1_1_k_howo_grid_desc =
-        transform_tensor_descriptor(out_n_k_howo_grid_desc,
-                                    make_tuple(make_unmerge_transform(make_tuple(N0, N1)),
-                                               make_unmerge_transform(make_tuple(I1, K)),
-                                               make_pass_through_transform(Ho * Wo)),
-                                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                                    make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-    const auto out_gm0_gm1_gn0_gn1_grid_desc = transform_tensor_descriptor(
-        out_n0_n1_1_k_howo_grid_desc,
-        make_tuple(make_pass_through_transform(I1),
-                   make_pass_through_transform(K),
-                   make_pass_through_transform(N0),
-                   make_merge_transform_v2_magic_division(make_tuple(N1, Ho * Wo))),
-        make_tuple(Sequence<2>{}, Sequence<3>{}, Sequence<0>{}, Sequence<1, 4>{}),
-        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-    return make_tuple(
-        wei_gk0_gm0_gm1_gk1_grid_desc, in_gk0_gn0_gn1_gk1_grid_desc, out_gm0_gm1_gn0_gn1_grid_desc);
-}
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
deleted file mode 100644
index 9f9fe0f1c9..0000000000
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_e_permute_xdl.hpp
+++ /dev/null
@@ -1,586 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/utility/common_header.hpp"
-#include "ck/tensor_description/tensor_descriptor.hpp"
-#include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp"
-#include "ck/host_utility/device_prop.hpp"
-#include "ck/host_utility/kernel_launch.hpp"
-
-namespace ck {
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatDsPointer,
-          typename FloatE,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          typename AGridDesc_AK0_M_AK1,
-          typename BGridDesc_BK0_N_BK1,
-          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename Block2ETileMap,
-          bool HasMainKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_bias_e_permute(const FloatAB* __restrict__ p_a_grid,
-                                   const FloatAB* __restrict__ p_b_grid,
-                                   FloatDsPointer p_ds_grid,
-                                   FloatE* __restrict__ p_e_grid,
-                                   const AElementwiseOperation a_element_op,
-                                   const BElementwiseOperation b_element_op,
-                                   const CDEElementwiseOperation cde_element_op,
-                                   const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
-                                   const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
-                                   const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                                       ds_grid_desc_mblock_mperblock_nblock_nperblock,
-                                   const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                                       e_grid_desc_mblock_mperblock_nblock_nperblock,
-                                   const Block2ETileMap block_2_etile_map)
-{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
-    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
-
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
-                                                  p_b_grid,
-                                                  p_ds_grid,
-                                                  p_e_grid,
-                                                  p_shared,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  cde_element_op,
-                                                  a_grid_desc_ak0_m_ak1,
-                                                  b_grid_desc_bk0_n_bk1,
-                                                  ds_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  e_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  block_2_etile_map);
-#else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_ds_grid;
-    ignore = p_e_grid;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = cde_element_op;
-    ignore = a_grid_desc_ak0_m_ak1;
-    ignore = b_grid_desc_bk0_n_bk1;
-    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = block_2_etile_map;
-#endif
-}
-
-} // namespace ck
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-// input : A[M, K], or A[K, N]
-// input : B[K, N], or A[N, K]
-// input : D0[M, N], D1[M, N], ...
-// output : E[M, N]
-// C = a_op(A) * b_op(B)
-// E = cde_op(C, D0, D1, ...)
-template <typename ALayout,
-          typename BLayout,
-          typename CDELayout,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DDataType,
-          typename EDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          GemmSpecialization GemmSpec,
-          index_t NumGemmKPrefetchStage,
-          index_t BlockSize,
-          index_t MPerBlock,
-          index_t NPerBlock,
-          index_t KPerBlock,
-          index_t AK1,
-          index_t BK1,
-          index_t MPerXDL,
-          index_t NPerXDL,
-          index_t MXdlPerWave,
-          index_t NXdlPerWave,
-          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_AK1,
-          index_t ABlockLdsExtraM,
-          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          index_t BBlockTransferDstScalarPerVector_BK1,
-          index_t BBlockLdsExtraN,
-          index_t CShuffleMXdlPerWavePerShuffle,
-          index_t CShuffleNXdlPerWavePerShuffle,
-          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-          index_t CDEBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched = make_default_loop_scheduler()>
-struct DeviceGemmBiasEPermute_Xdl : public DeviceGemmBiasCPermute<AElementwiseOperation,
-                                                                  BElementwiseOperation,
-                                                                  CDEElementwiseOperation>
-{
-    using DeviceOp = DeviceGemmBiasEPermute_Xdl;
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    static constexpr auto matrix_padder =
-        MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock};
-
-    static constexpr index_t NumDTensor = 1;
-
-    static auto MakeAGridDescriptor_M_K(index_t MRaw, index_t KRaw, index_t StrideA)
-    {
-        const auto a_grid_desc_mraw_kraw = [&]() {
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(I1, StrideA));
-            }
-        }();
-
-        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
-    }
-
-    static auto MakeBGridDescriptor_N_K(index_t KRaw, index_t NRaw, index_t StrideB)
-    {
-        const auto b_grid_desc_nraw_kraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(I1, StrideB));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(StrideB, I1));
-            }
-        }();
-
-        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
-    }
-
-    static auto MakeEGridDescriptor_M_N(DEGridDesc_M0_M1_M2_N0_N1 d_e_grid_desc)
-    {
-        index_t M0 = d_e_grid_desc.M0_;
-        index_t M1 = d_e_grid_desc.M1_;
-        index_t M2 = d_e_grid_desc.M2_;
-        index_t N0 = d_e_grid_desc.N0_;
-        index_t N1 = d_e_grid_desc.N1_;
-
-        index_t stride_M0 = d_e_grid_desc.stride_M0_;
-        index_t stride_M1 = d_e_grid_desc.stride_M1_;
-        index_t stride_M2 = d_e_grid_desc.stride_M2_;
-        index_t stride_N0 = d_e_grid_desc.stride_N0_;
-        index_t stride_N1 = d_e_grid_desc.stride_N1_;
-
-        const auto e_grid_desc_mraw_nraw = [&]() {
-            const auto e_grid_desc_m0_m1_m2_n0_n1 = make_naive_tensor_descriptor(
-                make_tuple(M0, M1, M2, N0, N1),
-                make_tuple(stride_M0, stride_M1, stride_M2, stride_N0, stride_N1));
-
-            return transform_tensor_descriptor(
-                e_grid_desc_m0_m1_m2_n0_n1,
-                make_tuple(make_merge_transform(make_tuple(M0, M1, M2)),
-                           make_merge_transform(make_tuple(N0, N1))),
-                make_tuple(Sequence<0, 1, 2>{}, Sequence<3, 4>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }();
-
-        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
-    }
-
-    using AGridDesc_M_K = decltype(MakeAGridDescriptor_M_K(1, 1, 1));
-    using BGridDesc_N_K = decltype(MakeBGridDescriptor_N_K(1, 1, 1));
-    using EGridDesc_M_N = decltype(MakeEGridDescriptor_M_N(DEGridDesc_M0_M1_M2_N0_N1{}));
-
-    using DsGridDesc_M_N = Tuple<EGridDesc_M_N>;
-
-    // GridwiseGemm
-    using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
-        ADataType, // TODO: distinguish A/B datatype
-        AccDataType,
-        CShuffleDataType,
-        ck::Tuple<DDataType>,
-        EDataType,
-        AElementwiseOperation,
-        BElementwiseOperation,
-        CDEElementwiseOperation,
-        InMemoryDataOperationEnum::Set,
-        NumGemmKPrefetchStage,
-        BlockSize,
-        MPerBlock,
-        NPerBlock,
-        KPerBlock,
-        AK1,
-        BK1,
-        MPerXDL,
-        NPerXDL,
-        MXdlPerWave,
-        NXdlPerWave,
-        ABlockTransferThreadClusterLengths_AK0_M_AK1,
-        ABlockTransferThreadClusterArrangeOrder,
-        ABlockTransferSrcAccessOrder,
-        ABlockTransferSrcVectorDim,
-        ABlockTransferSrcScalarPerVector,
-        ABlockTransferDstScalarPerVector_AK1,
-        false,
-        ABlockLdsExtraM,
-        BBlockTransferThreadClusterLengths_BK0_N_BK1,
-        BBlockTransferThreadClusterArrangeOrder,
-        BBlockTransferSrcAccessOrder,
-        BBlockTransferSrcVectorDim,
-        BBlockTransferSrcScalarPerVector,
-        BBlockTransferDstScalarPerVector_BK1,
-        false,
-        BBlockLdsExtraN,
-        CShuffleMXdlPerWavePerShuffle,
-        CShuffleNXdlPerWavePerShuffle,
-        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-        CDEBlockTransferScalarPerVector_NPerBlock,
-        LoopSched>;
-
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-
-    using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;
-
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const void* p_a_grid,
-                 const void* p_b_grid,
-                 const void* p_d_grid,
-                 void* p_e_grid,
-                 index_t MRaw,
-                 index_t NRaw,
-                 index_t KRaw,
-                 index_t StrideA,
-                 index_t StrideB,
-                 DEGridDesc_M0_M1_M2_N0_N1 d_grid_desc,
-                 DEGridDesc_M0_M1_M2_N0_N1 e_grid_desc,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
-              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
-              p_ds_grid_{},
-              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
-              a_grid_desc_m_k_{DeviceOp::MakeAGridDescriptor_M_K(MRaw, KRaw, StrideA)},
-              b_grid_desc_n_k_{DeviceOp::MakeBGridDescriptor_N_K(KRaw, NRaw, StrideB)},
-              ds_grid_desc_m_n_{},
-              e_grid_desc_m_n_{DeviceOp::MakeEGridDescriptor_M_N(e_grid_desc)},
-              a_grid_desc_ak0_m_ak1_{
-                  GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(a_grid_desc_m_k_)},
-              b_grid_desc_bk0_n_bk1_{
-                  GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(b_grid_desc_n_k_)},
-              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              e_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_etile_map_{GridwiseGemm::MakeDefaultBlock2ETileMap(e_grid_desc_m_n_)},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-
-            if(MRaw != d_grid_desc.M0_ * d_grid_desc.M1_ * d_grid_desc.M2_)
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            if(NRaw != d_grid_desc.N0_ * d_grid_desc.N1_)
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            // populate pointer, desc for Ds
-            // D pointer
-            p_ds_grid_(I0) = static_cast<const DDataType*>(p_d_grid);
-
-            // D desc
-            ds_grid_desc_m_n_(I0) = DeviceOp::MakeEGridDescriptor_M_N(d_grid_desc);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_m_k_,
-                                           b_grid_desc_n_k_,
-                                           ds_grid_desc_m_n_,
-                                           e_grid_desc_m_n_,
-                                           block_2_etile_map_))
-            {
-                e_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        e_grid_desc_m_n_);
-
-                ds_grid_desc_mblock_mperblock_nblock_nperblock_(I0) =
-                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        ds_grid_desc_m_n_[I0]);
-            }
-        }
-
-        //  private:
-        // pointers
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        typename GridwiseGemm::DsGridPointer p_ds_grid_;
-        EDataType* p_e_grid_;
-
-        // tensor descriptors for problem definiton
-        AGridDesc_M_K a_grid_desc_m_k_;
-        BGridDesc_N_K b_grid_desc_n_k_;
-        DsGridDesc_M_N ds_grid_desc_m_n_;
-        EGridDesc_M_N e_grid_desc_m_n_;
-
-        // tensor descriptors for block/thread-wise copy
-        AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
-        BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
-        typename GridwiseGemm::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
-        typename GridwiseGemm::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            e_grid_desc_mblock_mperblock_nblock_nperblock_;
-
-        // block-to-e-tile map
-        Block2ETileMap block_2_etile_map_;
-
-        // element-wise op
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public BaseInvoker
-    {
-        using Argument = DeviceOp::Argument;
-
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
-        {
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_,
-                                            arg.b_grid_desc_n_k_,
-                                            arg.ds_grid_desc_m_n_,
-                                            arg.e_grid_desc_m_n_,
-                                            arg.block_2_etile_map_))
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            const index_t grid_size =
-                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
-
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
-
-            auto launch_kernel = [&](auto has_main_k_block_loop) {
-                constexpr bool has_main_loop = has_main_k_block_loop.value;
-
-                const auto kernel = kernel_gemm_bias_e_permute<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    typename GridwiseGemm::DsGridPointer,
-                    EDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CDEElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2ETileMap,
-                    has_main_loop>;
-
-                return launch_and_time_kernel(stream_config,
-                                              kernel,
-                                              dim3(grid_size),
-                                              dim3(BlockSize),
-                                              0,
-                                              arg.p_a_grid_,
-                                              arg.p_b_grid_,
-                                              arg.p_ds_grid_,
-                                              arg.p_e_grid_,
-                                              arg.a_element_op_,
-                                              arg.b_element_op_,
-                                              arg.cde_element_op_,
-                                              arg.a_grid_desc_ak0_m_ak1_,
-                                              arg.b_grid_desc_bk0_n_bk1_,
-                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                              arg.block_2_etile_map_);
-            };
-
-            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
-            {
-                return launch_kernel(integral_constant<bool, true>{});
-            }
-            else
-            {
-                return launch_kernel(integral_constant<bool, false>{});
-            }
-        }
-
-        // polymorphic
-        float Run(const BaseArgument* p_arg,
-                  const StreamConfig& stream_config = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
-        }
-    };
-
-    static bool IsSupportedArgument(const Argument& arg)
-    {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
-        {
-            return false;
-        }
-
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_,
-                                           arg.b_grid_desc_n_k_,
-                                           arg.ds_grid_desc_m_n_,
-                                           arg.e_grid_desc_m_n_,
-                                           arg.block_2_etile_map_);
-    }
-
-    // polymorphic
-    bool IsSupportedArgument(const BaseArgument* p_arg) override
-    {
-        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
-    }
-
-    static auto MakeArgument(const void* p_a,
-                             const void* p_b,
-                             const void* p_d,
-                             void* p_e,
-                             index_t MRaw,
-                             index_t NRaw,
-                             index_t KRaw,
-                             index_t StrideA,
-                             index_t StrideB,
-                             DEGridDesc_M0_M1_M2_N0_N1 d_grid_desc,
-                             DEGridDesc_M0_M1_M2_N0_N1 e_grid_desc,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{p_a,
-                        p_b,
-                        p_d,
-                        p_e,
-                        MRaw,
-                        NRaw,
-                        KRaw,
-                        StrideA,
-                        StrideB,
-                        d_grid_desc,
-                        e_grid_desc,
-                        a_element_op,
-                        b_element_op,
-                        cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    // polymorphic
-    std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(const void* p_a,
-                        const void* p_b,
-                        const void* p_d,
-                        void* p_e,
-                        index_t MRaw,
-                        index_t NRaw,
-                        index_t KRaw,
-                        index_t StrideA,
-                        index_t StrideB,
-                        DEGridDesc_M0_M1_M2_N0_N1 d_grid_desc,
-                        DEGridDesc_M0_M1_M2_N0_N1 e_grid_desc,
-                        AElementwiseOperation a_element_op,
-                        BElementwiseOperation b_element_op,
-                        CDEElementwiseOperation cde_element_op) override
-    {
-        return std::make_unique<Argument>(p_a,
-                                          p_b,
-                                          p_d,
-                                          p_e,
-                                          MRaw,
-                                          NRaw,
-                                          KRaw,
-                                          StrideA,
-                                          StrideB,
-                                          d_grid_desc,
-                                          e_grid_desc,
-                                          a_element_op,
-                                          b_element_op,
-                                          cde_element_op);
-    }
-
-    // polymorphic
-    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    // polymorphic
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "DeviceGemmBiasEPermute_Xdl"
-            << "<"
-            << BlockSize << ", "
-            << MPerBlock << ", "
-            << NPerBlock << ", "
-            << KPerBlock << ", "
-            << AK1 << ", "
-            << BK1 << ", "
-            << K1 << ", "
-            << MPerXDL << ", "
-            << NPerXDL << ", "
-            << MXdlPerWave << ", "
-            << NXdlPerWave << ", "
-            << ABlockTransferSrcScalarPerVector << ", "
-            << ABlockTransferDstScalarPerVector_K1 << ", "
-            << BBlockTransferSrcScalarPerVector << ", "
-            << BBlockTransferDstScalarPerVector_K1 << ", "
-            << CShuffleMXdlPerWavePerShuffle << ", "
-            << CShuffleNXdlPerWavePerShuffle << ", "
-            << CBlockTransferScalarPerVector_NWaveNPerXdl
-            << ">";
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp
deleted file mode 100644
index 2369f51795..0000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_contraction_dlops_v1r2.hpp
+++ /dev/null
@@ -1,662 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_CONTRACTION_DLOPS_V1R2_HPP
-#define CK_GRIDWISE_CONTRACTION_DLOPS_V1R2_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_gemm_dlops_v2r3.hpp"
-#include "blockwise_tensor_slice_transfer_v2.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_set.hpp"
-
-namespace ck {
-
-template <typename GridwiseContraction,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_GK0_GM0_GM10_GM11_GK1,
-          typename BGridDesc_GK0_GN0_GN10_GN11_GK1,
-          typename CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1,
-          typename CGridBlockCluster_BlockId_To_GM10_GN10,
-          bool HasMainKBlockLoop,
-          bool HasDoubleTailKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_contraction_dlops_v1r2(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_GK0_GM0_GM10_GM11_GK1 a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-            const BGridDesc_GK0_GN0_GN10_GN11_GK1 b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-            const CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1 c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-            const CGridBlockCluster_BlockId_To_GM10_GN10 c_grid_block_cluster_blockid_to_gm10_gn10)
-{
-    constexpr index_t shared_block_size =
-        GridwiseContraction::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseContraction::Run(p_a_grid,
-                             p_b_grid,
-                             p_c_grid,
-                             p_shared_block,
-                             a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                             b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                             c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                             c_grid_block_cluster_blockid_to_gm10_gn10,
-                             integral_constant<bool, HasMainKBlockLoop>{},
-                             integral_constant<bool, HasDoubleTailKBlockLoop>{});
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_GK0_GM0_GM1_GK1,
-          typename BGridDesc_GK0_GN0_GN1_GK1,
-          typename CGridDesc_GM0_GM1_GN0_GN1,
-          index_t GM1PerBlockGM11,
-          index_t GN1PerBlockGN11,
-          index_t GK0PerBlock,
-          index_t BM1PerThreadBM11,
-          index_t BN1PerThreadBN11,
-          index_t BK0PerThread,
-          typename BM10BN10ThreadClusterBM10Xs,
-          typename BM10BN10ThreadClusterBN10Xs,
-          typename ABlockTransferThreadSliceLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferThreadClusterLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          typename ABlockTransferSrcVectorTensorLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferDstVectorTensorLengths_GK0_GM0_GM10_GM11_GK1,
-          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
-          typename BBlockTransferThreadSliceLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferThreadClusterLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          typename BBlockTransferSrcVectorTensorLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferDstVectorTensorLengths_GK0_GN0_GN10_GN11_GK1,
-          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGridStepHacks,
-          typename BGridStepHacks,
-          typename CGridStepHacks,
-          typename AGridMoveSliceWindowStepHacks,
-          typename BGridMoveSliceWindowStepHacks>
-struct GridwiseContractionDlops_A_GK0_GM0_GM1_GK1_B_GK0_GN0_GN1_GK1_C_GM0_GM1_GN0_GN1
-{
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    // GM0 and GN0 need to known at compile-time
-    static constexpr auto GM0 = CGridDesc_GM0_GM1_GN0_GN1{}.GetLength(I0);
-    static constexpr auto GN0 = CGridDesc_GM0_GM1_GN0_GN1{}.GetLength(I2);
-    static constexpr auto GK1 = AGridDesc_GK0_GM0_GM1_GK1{}.GetLength(I3);
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        // lds max alignment
-        // TODO: part of them should be moved into blockwise-gemm
-        // TODO: change this. I think it needs multi-dimensional alignment
-        constexpr auto max_lds_align = GK1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_block_desc_gk0_gm0_gm10_gm11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GM0, I1, Number<GM1PerBlockGM11>{}, GK1),
-            max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_block_desc_gk0_gn0_gn10_gn11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GN0, I1, Number<GN1PerBlockGN11>{}, GK1),
-            max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size = math::integer_least_multiple(
-            a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
-            b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        return 2 * (a_block_aligned_space_size + b_block_aligned_space_size) * sizeof(FloatAB);
-    }
-
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_GK0_GM0_GM1_GK1& a_grid_desc_gk0_gm0_gm1_gk1,
-                  const BGridDesc_GK0_GN0_GN1_GK1& b_grid_desc_gk0_gn0_gn1_gk1,
-                  const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        static_assert(is_known_at_compile_time<remove_cv_t<decltype(GM0)>>::value &&
-                          is_known_at_compile_time<remove_cv_t<decltype(GN0)>>::value,
-                      "wrong! GM0 and GN0 need to be known at compile-time");
-
-        const auto GM1 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I2);
-        const auto GN1 = b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I2);
-        const auto GK0 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I0);
-
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
-
-        return (
-            (GM0 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I0) &&
-             GM1 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1) &&
-             GN0 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I2) &&
-             GN1 == c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3) &&
-             GM0 == a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I1) &&
-             GM1 == a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I2) &&
-             GN0 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I1) &&
-             GN1 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I2) &&
-             GK0 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I0) &&
-             GK1 == b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I3)) &&
-            (GM1 % GM1PerBlockGM11 == 0 && GN1 % GN1PerBlockGN11 == 0 && GK0 % GK0PerBlock == 0));
-    }
-
-    __host__ __device__ static constexpr index_t
-    CalculateGridSize(const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        const auto GM1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1);
-        const auto GN1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3);
-
-        constexpr index_t GM11 = GM1PerBlockGM11;
-        constexpr index_t GN11 = GN1PerBlockGN11;
-
-        const index_t GM10 = GM1 / GM11;
-        const index_t GN10 = GN1 / GN11;
-
-        const index_t grid_size = GM10 * GN10;
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t GK0)
-    {
-        const bool has_main_k_block_loop = (GK0 + GK0PerBlock) / (2 * GK0PerBlock) > 1;
-
-        return has_main_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasDoubleTailKBlockLoop(index_t GK0)
-    {
-        const bool has_double_tail_k_block_loop = (GK0 / GK0PerBlock) % 2 == 0;
-
-        return has_double_tail_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto MakeAGridDescriptor_GK0_GM0_GM10_GM11_GK1(
-        const AGridDesc_GK0_GM0_GM1_GK1& a_grid_desc_gk0_gm0_gm1_gk1)
-    {
-        const auto GK0 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I0);
-        const auto GM1 = a_grid_desc_gk0_gm0_gm1_gk1.GetLength(I2);
-
-        const auto GM11 = Number<GM1PerBlockGM11>{};
-        const auto GM10 = GM1 / GM11;
-
-        const auto a_grid_desc_gk0_gm0_gm10_gm11_gk1 = transform_tensor_descriptor(
-            a_grid_desc_gk0_gm0_gm1_gk1,
-            make_tuple(make_pass_through_transform(GK0),
-                       make_pass_through_transform(GM0),
-                       make_unmerge_transform(make_tuple(GM10, GM11)),
-                       make_pass_through_transform(GK1)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-        return a_grid_desc_gk0_gm0_gm10_gm11_gk1;
-    }
-
-    __host__ __device__ static constexpr auto MakeBGridDescriptor_GK0_GN0_GN10_GN11_GK1(
-        const BGridDesc_GK0_GN0_GN1_GK1& b_grid_desc_gk0_gn0_gn1_gk1)
-    {
-        const auto GK0 = b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I0);
-        const auto GN1 = b_grid_desc_gk0_gn0_gn1_gk1.GetLength(I2);
-
-        const auto GN11 = Number<GN1PerBlockGN11>{};
-        const auto GN10 = GN1 / GN11;
-
-        const auto b_grid_desc_gk0_gn0_gn10_gn11_gk1 = transform_tensor_descriptor(
-            b_grid_desc_gk0_gn0_gn1_gk1,
-            make_tuple(make_pass_through_transform(GK0),
-                       make_pass_through_transform(GN0),
-                       make_unmerge_transform(make_tuple(GN10, GN11)),
-                       make_pass_through_transform(GK1)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-        return b_grid_desc_gk0_gn0_gn10_gn11_gk1;
-    }
-
-    __host__ __device__ static constexpr auto MakeCGridDescriptor_GM10_BM0_BM1_GN10_BN0_BN1(
-        const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        const auto GM1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1);
-        const auto GN1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3);
-
-        constexpr auto GM11 = Number<GM1PerBlockGM11>{};
-        constexpr auto GN11 = Number<GN1PerBlockGN11>{};
-
-        const auto GM10 = GM1 / GM11;
-        const auto GN10 = GN1 / GN11;
-
-        constexpr auto BM = GM0 * GM11;
-        constexpr auto BN = GN0 * GN11;
-
-        constexpr auto BM1 =
-            Number<container_reduce(BM10BN10ThreadClusterBM10Xs{}, math::multiplies{}, I1) *
-                   BM1PerThreadBM11>{};
-        constexpr auto BN1 =
-            Number<container_reduce(BM10BN10ThreadClusterBN10Xs{}, math::multiplies{}, I1) *
-                   BN1PerThreadBN11>{};
-
-        constexpr auto BM0 = BM / BM1;
-        constexpr auto BN0 = BN / BN1;
-
-        const auto c_gm0_gm10_gm11_gn0_gn10_gn11_grid_desc = transform_tensor_descriptor(
-            c_grid_desc_gm0_gm1_gn0_gn1,
-            make_tuple(make_pass_through_transform(GM0),
-                       make_unmerge_transform(make_tuple(GM10, GM11)),
-                       make_pass_through_transform(GN0),
-                       make_unmerge_transform(make_tuple(GN10, GN11))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}, Sequence<4, 5>{}));
-
-        const auto c_gm10_bm_gn10_bn_grid_desc = transform_tensor_descriptor(
-            c_gm0_gm10_gm11_gn0_gn10_gn11_grid_desc,
-            make_tuple(make_pass_through_transform(GM10),
-                       make_merge_transform(make_tuple(GM0, GM11)),
-                       make_pass_through_transform(GN10),
-                       make_merge_transform(make_tuple(GN0, GN11))),
-            make_tuple(Sequence<1>{}, Sequence<0, 2>{}, Sequence<4>{}, Sequence<3, 5>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-        const auto c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1 = transform_tensor_descriptor(
-            c_gm10_bm_gn10_bn_grid_desc,
-            make_tuple(make_pass_through_transform(GM10),
-                       make_unmerge_transform(make_tuple(BM0, BM1)),
-                       make_pass_through_transform(GN10),
-                       make_unmerge_transform(make_tuple(BN0, BN1))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}, Sequence<4, 5>{}));
-
-        return c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1;
-    }
-
-    __host__ __device__ static constexpr auto MakeCGridBlockCluster_BlockId_To_GM10_GN10(
-        const CGridDesc_GM0_GM1_GN0_GN1& c_grid_desc_gm0_gm1_gn0_gn1)
-    {
-        const auto GM1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I1);
-        const auto GN1 = c_grid_desc_gm0_gm1_gn0_gn1.GetLength(I3);
-
-        constexpr auto GM11 = Number<GM1PerBlockGM11>{};
-        constexpr auto GN11 = Number<GN1PerBlockGN11>{};
-
-        const auto GM10 = GM1 / GM11;
-        const auto GN10 = GN1 / GN11;
-
-        const auto c_grid_block_cluster_blockid_to_gm10_gn10 = make_single_stage_tensor_adaptor(
-            make_tuple(make_merge_transform(make_tuple(GM10, GN10))),
-            make_tuple(Sequence<0, 1>{}),
-            make_tuple(Sequence<0>{}));
-
-        return c_grid_block_cluster_blockid_to_gm10_gn10;
-    }
-
-    using AGridDesc_GK0_GM0_GM10_GM11_GK1 =
-        decltype(MakeAGridDescriptor_GK0_GM0_GM10_GM11_GK1(AGridDesc_GK0_GM0_GM1_GK1{}));
-    using BGridDesc_GK0_GN0_GN10_GN11_GK1 =
-        decltype(MakeBGridDescriptor_GK0_GN0_GN10_GN11_GK1(BGridDesc_GK0_GN0_GN1_GK1{}));
-    using CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1 =
-        decltype(MakeCGridDescriptor_GM10_BM0_BM1_GN10_BN0_BN1(CGridDesc_GM0_GM1_GN0_GN1{}));
-    using CGridBlockCluster_BlockId_To_GM10_GN10 =
-        decltype(MakeCGridBlockCluster_BlockId_To_GM10_GN10(CGridDesc_GM0_GM1_GN0_GN1{}));
-
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_GK0_GM0_GM10_GM11_GK1& a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-        const BGridDesc_GK0_GN0_GN10_GN11_GK1& b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-        const CGridDesc_GM10_BM0_BM1_GN10_BN0_BN1& c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-        const CGridBlockCluster_BlockId_To_GM10_GN10& c_grid_block_cluster_blockid_to_gm10_gn10,
-        integral_constant<bool, HasMainKBlockLoop>,
-        integral_constant<bool, HasDoubleTailKBlockLoop>)
-    {
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_grid, a_grid_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_grid, b_grid_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize());
-        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_grid, c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1.GetElementSpaceSize());
-
-        const auto GK0 = a_grid_desc_gk0_gm0_gm10_gm11_gk1.GetLength(I0);
-
-        // divide block work by [GM10, GN10]
-        const auto c_gm10_gn10_block_cluster_idx =
-            c_grid_block_cluster_blockid_to_gm10_gn10.CalculateBottomIndex(
-                make_multi_index(get_block_1d_id()));
-
-        // HACK: this force index data into SGPR
-        const index_t igm10 = __builtin_amdgcn_readfirstlane(c_gm10_gn10_block_cluster_idx[I0]);
-        const index_t ign10 = __builtin_amdgcn_readfirstlane(c_gm10_gn10_block_cluster_idx[I1]);
-
-        // lds max alignment
-        // TODO: part of them should be moved into blockwise-gemm
-        // TODO: change this. I think it needs multi-dimensional alignment
-        constexpr auto max_lds_align = GK1;
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_block_desc_gk0_gm0_gm10_gm11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GM0, I1, Number<GM1PerBlockGM11>{}, GK1),
-            max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_block_desc_gk0_gn0_gn10_gn11_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GN0, I1, Number<GN1PerBlockGN11>{}, GK1),
-            max_lds_align);
-
-        // A matrix in LDS memory for blockwise GEMM
-        //   be careful of LDS alignment
-        constexpr auto a_block_desc_gk0_bm_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GM0 * Number<GM1PerBlockGM11>{}, GK1), max_lds_align);
-
-        // B matrix in LDS memory for blockwise GEMM
-        //   be careful of LDS alignment
-        constexpr auto b_block_desc_gk0_bn_gk1 = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<GK0PerBlock>{}, GN0 * Number<GN1PerBlockGN11>{}, GK1), max_lds_align);
-
-        static_assert(a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize() ==
-                              a_block_desc_gk0_bm_gk1.GetElementSpaceSize() &&
-                          b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize() ==
-                              b_block_desc_gk0_bn_gk1.GetElementSpaceSize(),
-                      "wrong!");
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy = BlockwiseTensorSliceTransfer_v5r1<
-            BlockSize,
-            InMemoryDataOperationEnum::Set,
-            Sequence<GK0PerBlock, GM0, 1, GM1PerBlockGM11, GK1.value>,
-            ABlockTransferThreadSliceLengths_GK0_GM0_GM10_GM11_GK1,
-            ABlockTransferThreadClusterLengths_GK0_GM0_GM10_GM11_GK1,
-            ABlockTransferThreadClusterArrangeOrder,
-            FloatAB,
-            FloatAB,
-            decltype(a_grid_desc_gk0_gm0_gm10_gm11_gk1),
-            decltype(a_block_desc_gk0_gm0_gm10_gm11_gk1),
-            ABlockTransferSrcAccessOrder,
-            Sequence<0, 1, 2, 3, 4>,
-            ABlockTransferSrcVectorTensorLengths_GK0_GM0_GM10_GM11_GK1, // SrcVectorTensorLengths
-            ABlockTransferDstVectorTensorLengths_GK0_GM0_GM10_GM11_GK1, // DstVectorTensorLengths
-            ABlockTransferSrcVectorTensorContiguousDimOrder, // SrcVectorTensorContiguousDimOrder
-            Sequence<0, 1, 2, 3, 4>,                         // DstVectorTensorContiguousDimOrder
-            false,
-            true>(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                  make_multi_index(0, 0, igm10, 0, 0),
-                  a_block_desc_gk0_gm0_gm10_gm11_gk1,
-                  make_multi_index(0, 0, 0, 0, 0));
-
-        // B matrix blockwise copy
-        auto b_blockwise_copy = BlockwiseTensorSliceTransfer_v5r1<
-            BlockSize,
-            InMemoryDataOperationEnum::Set,
-            Sequence<GK0PerBlock, GN0, 1, GN1PerBlockGN11, GK1.value>,
-            BBlockTransferThreadSliceLengths_GK0_GN0_GN10_GN11_GK1,
-            BBlockTransferThreadClusterLengths_GK0_GN0_GN10_GN11_GK1,
-            BBlockTransferThreadClusterArrangeOrder,
-            FloatAB,
-            FloatAB,
-            decltype(b_grid_desc_gk0_gn0_gn10_gn11_gk1),
-            decltype(b_block_desc_gk0_gn0_gn10_gn11_gk1),
-            BBlockTransferSrcAccessOrder,
-            Sequence<0, 1, 2, 3, 4>,
-            BBlockTransferSrcVectorTensorLengths_GK0_GN0_GN10_GN11_GK1, // SrcVectorTensorLengths
-            BBlockTransferDstVectorTensorLengths_GK0_GN0_GN10_GN11_GK1, // DstVectorTensorLengths
-            BBlockTransferSrcVectorTensorContiguousDimOrder, // SrcVectorTensorContiguousDimOrder
-            Sequence<0, 1, 2, 3, 4>,                         // DstVectorTensorContiguousDimOrder
-            false,
-            true>(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                  make_multi_index(0, 0, ign10, 0, 0),
-                  b_block_desc_gk0_gn0_gn10_gn11_gk1,
-                  make_multi_index(0, 0, 0, 0, 0));
-
-        // GEMM definition
-        //   c_mtx += transpose(a_mtx) * b_mtx
-        //     a_mtx[GK0PerBlock, GM1PerBlockGM11] is in LDS
-        //     b_mtx[KPerBlocl, GN1PerBlockGN11] is in LDS
-        //     c_mtx[GM1PerBlockGM11, GN1PerBlockGN11] is distributed among threads, and saved in
-        //       register
-        const auto blockwise_gemm =
-            BlockwiseGemmDlops_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
-                BlockSize,
-                FloatAB,
-                FloatAB,
-                FloatAcc,
-                decltype(a_block_desc_gk0_bm_gk1),
-                decltype(b_block_desc_gk0_bn_gk1),
-                BM1PerThreadBM11,
-                BN1PerThreadBN11,
-                BK0PerThread,
-                BM10BN10ThreadClusterBM10Xs,
-                BM10BN10ThreadClusterBN10Xs,
-                BM1PerThreadBM11,
-                BN1PerThreadBN11>{};
-
-        constexpr auto c_thread_tensor_lengths_bm0_bm1_bn0_bn1 =
-            decltype(blockwise_gemm)::GetCThreadTensorLengths_BM0_BM1_BN0_BN1();
-
-        constexpr auto c_thread_desc_bm0_bm1_bn0_bn1 = make_naive_tensor_descriptor_packed(
-            sequence_to_tuple_of_number(c_thread_tensor_lengths_bm0_bm1_bn0_bn1));
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size = math::integer_least_multiple(
-            a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size = math::integer_least_multiple(
-            b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize(), max_lds_align);
-
-        FloatAB* p_a_block_double = p_shared_block;
-        FloatAB* p_b_block_double = p_shared_block + 2 * a_block_aligned_space_size;
-
-        // register allocation for output
-        auto c_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAcc>(
-            c_thread_desc_bm0_bm1_bn0_bn1.GetElementSpaceSize());
-
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_thread_desc_bm0_bm1_bn0_bn1),
-                                    decltype(c_thread_tensor_lengths_bm0_bm1_bn0_bn1)>{}
-            .Run(c_thread_desc_bm0_bm1_bn0_bn1,
-                 make_tuple(I0, I0, I0, I0),
-                 c_thread_buf,
-                 FloatAcc{0});
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(GK0PerBlock, 0, 0, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(GK0PerBlock, 0, 0, 0, 0);
-
-        auto a_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double, a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize());
-        auto b_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double, b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize());
-
-        auto a_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double + a_block_aligned_space_size,
-            a_block_desc_gk0_gm0_gm10_gm11_gk1.GetElementSpaceSize());
-        auto b_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double + b_block_aligned_space_size,
-            b_block_desc_gk0_gn0_gn10_gn11_gk1.GetElementSpaceSize());
-
-        // LDS double buffer: preload data into LDS
-        {
-            a_blockwise_copy.RunRead(
-                a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-            b_blockwise_copy.RunRead(
-                b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-            a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_even_buf);
-            b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_even_buf);
-        }
-
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t gk0_block_on_grid = 0;
-
-            // LDS double buffer: main body
-            // use Do-While loop instead of For loop to simplify control flow
-            do
-            {
-                // even iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                                                    a_block_slice_copy_step,
-                                                    AGridMoveSliceWindowStepHacks{});
-                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                                                    b_block_slice_copy_step,
-                                                    BGridMoveSliceWindowStepHacks{});
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-                b_blockwise_copy.RunRead(
-                    b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(c_thread_desc_bm0_bm1_bn0_bn1,
-                                   a_block_even_buf,
-                                   b_block_even_buf,
-                                   c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_odd_buf);
-                b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_odd_buf);
-
-                // odd iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                                                    a_block_slice_copy_step,
-                                                    AGridMoveSliceWindowStepHacks{});
-                b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                                                    b_block_slice_copy_step,
-                                                    BGridMoveSliceWindowStepHacks{});
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-                b_blockwise_copy.RunRead(
-                    b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(
-                    c_thread_desc_bm0_bm1_bn0_bn1, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_even_buf);
-                b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_even_buf);
-
-                gk0_block_on_grid += 2 * GK0PerBlock;
-            } while(gk0_block_on_grid < GK0 - 2 * GK0PerBlock);
-        }
-
-        // LDS double buffer: tail
-        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
-        {
-            a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc_gk0_gm0_gm10_gm11_gk1,
-                                                a_block_slice_copy_step,
-                                                AGridMoveSliceWindowStepHacks{});
-            b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_gk0_gn0_gn10_gn11_gk1,
-                                                b_block_slice_copy_step,
-                                                BGridMoveSliceWindowStepHacks{});
-
-            __syncthreads();
-
-            // LDS double buffer: load last data from device mem
-            a_blockwise_copy.RunRead(
-                a_grid_desc_gk0_gm0_gm10_gm11_gk1, a_global_buf, AGridStepHacks{});
-            b_blockwise_copy.RunRead(
-                b_grid_desc_gk0_gn0_gn10_gn11_gk1, b_global_buf, BGridStepHacks{});
-
-            // LDS double buffer: GEMM on 2nd-last data
-            blockwise_gemm.Run(
-                c_thread_desc_bm0_bm1_bn0_bn1, a_block_even_buf, b_block_even_buf, c_thread_buf);
-
-            // LDS double buffer: store last data to LDS
-            a_blockwise_copy.RunWrite(a_block_desc_gk0_gm0_gm10_gm11_gk1, a_block_odd_buf);
-            b_blockwise_copy.RunWrite(b_block_desc_gk0_gn0_gn10_gn11_gk1, b_block_odd_buf);
-
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_thread_desc_bm0_bm1_bn0_bn1, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-        }
-        else // if has 1 iteration left
-        {
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_thread_desc_bm0_bm1_bn0_bn1, a_block_even_buf, b_block_even_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            constexpr auto c_thread_desc_gm10_bm0_bm1_gn10_bn0_bn1 =
-                make_naive_tensor_descriptor_packed(
-                    make_tuple(I1,
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I0]>{},
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I1]>{},
-                               I1,
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I2]>{},
-                               Number<c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I3]>{}));
-
-            const auto c_thread_origin_on_block_bm0_bm1_bn0_bn1 =
-                blockwise_gemm.CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(
-                    get_thread_local_1d_id());
-
-            ThreadwiseTensorSliceTransfer_v1r3<
-                FloatAcc,
-                FloatC,
-                decltype(c_thread_desc_gm10_bm0_bm1_gn10_bn0_bn1),
-                decltype(c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1),
-                Sequence<1,
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I0],
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I1],
-                         1,
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I2],
-                         c_thread_tensor_lengths_bm0_bm1_bn0_bn1[I3]>,
-                CThreadTransferSrcDstAccessOrder,
-                CThreadTransferSrcDstVectorDim,
-                CThreadTransferDstScalarPerVector,
-                CGlobalMemoryDataOperation,
-                1,
-                false>{c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                       make_multi_index(igm10,
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I0],
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I1],
-                                        ign10,
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I2],
-                                        c_thread_origin_on_block_bm0_bm1_bn0_bn1[I3])}
-                .Run(c_thread_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                     make_tuple(I0, I0, I0, I0, I0, I0),
-                     c_thread_buf,
-                     c_grid_desc_gm10_bm0_bm1_gn10_bn0_bn1,
-                     c_grid_buf,
-                     CGridStepHacks{});
-        }
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp
deleted file mode 100644
index 84e033e1e9..0000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v1r2.hpp
+++ /dev/null
@@ -1,608 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_GEMM_DLOPS_V1R2_HPP
-#define CK_GRIDWISE_GEMM_DLOPS_V1R2_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_gemm_dlops_v2r2.hpp"
-#include "blockwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_set.hpp"
-
-namespace ck {
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AKM0M1GridDesc,
-          typename BKN0N1GridDesc,
-          typename CM0M10M11N0N10N11GridDesc,
-          typename CBlockIdToM0N0BlockClusterAdaptor,
-          bool HasMainKBlockLoop,
-          bool HasDoubleTailKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v1r2(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AKM0M1GridDesc a_k_m0_m1_grid_desc,
-            const BKN0N1GridDesc b_k_n0_n1_grid_desc,
-            const CM0M10M11N0N10N11GridDesc c_m0_m10_m11_n0_n10_n11_grid_desc,
-            const CBlockIdToM0N0BlockClusterAdaptor cblockid_to_m0_n0_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::Run(p_a_grid,
-                      p_b_grid,
-                      p_c_grid,
-                      p_shared_block,
-                      a_k_m0_m1_grid_desc,
-                      b_k_n0_n1_grid_desc,
-                      c_m0_m10_m11_n0_n10_n11_grid_desc,
-                      cblockid_to_m0_n0_block_cluster_adaptor,
-                      integral_constant<bool, HasMainKBlockLoop>{},
-                      integral_constant<bool, HasDoubleTailKBlockLoop>{});
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AKMGridDesc,
-          typename BKNGridDesc,
-          typename CMNGridDesc,
-          index_t MPerBlockM1,
-          index_t NPerBlockN1,
-          index_t KPerBlock,
-          index_t M1PerThreadM111,
-          index_t N1PerThreadN111,
-          index_t KPerThread,
-          index_t M11N11ThreadClusterM1100,
-          index_t M11N11ThreadClusterN1100,
-          index_t M11N11ThreadClusterM1101,
-          index_t M11N11ThreadClusterN1101,
-          typename ABlockTransferThreadSliceLengths_K_M0_M1,
-          typename ABlockTransferThreadClusterLengths_K_M0_M1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_M1,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          typename BBlockTransferThreadSliceLengths_K_N0_N1,
-          typename BBlockTransferThreadClusterLengths_K_N0_N1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          index_t BBlockTransferDstScalarPerVector_N1,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGridStepHacks,
-          typename BGridStepHacks,
-          typename CGridStepHacks,
-          typename AGridMoveSliceWindowStepHacks,
-          typename BGridMoveSliceWindowStepHacks>
-struct GridwiseGemmDlops_km_kn_mn_v1r2
-{
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        constexpr auto max_lds_align = math::lcm(Number<ABlockTransferDstScalarPerVector_M1>{},
-                                                 Number<BBlockTransferDstScalarPerVector_N1>{},
-                                                 Number<M1PerThreadM111>{},
-                                                 Number<N1PerThreadN111>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_k_m_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<MPerBlockM1>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_k_n_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<NPerBlockN1>{}), max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size =
-            math::integer_least_multiple(a_k_m_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size =
-            math::integer_least_multiple(b_k_n_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        return 2 * (a_block_aligned_space_size + b_block_aligned_space_size) * sizeof(FloatAB);
-    }
-
-    __host__ __device__ static constexpr bool CheckValidity(const AKMGridDesc& a_k_m_grid_desc,
-                                                            const BKNGridDesc& b_k_n_grid_desc,
-                                                            const CMNGridDesc& c_m_n_grid_desc)
-    {
-        const auto M = a_k_m_grid_desc.GetLength(I1);
-        const auto N = b_k_n_grid_desc.GetLength(I1);
-        const auto K = a_k_m_grid_desc.GetLength(I0);
-
-        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
-
-        return (M == c_m_n_grid_desc.GetLength(I0) && N == c_m_n_grid_desc.GetLength(I1) &&
-                K == b_k_n_grid_desc.GetLength(I0)) &&
-               (M % MPerBlockM1 == 0 && N % NPerBlockN1 == 0 && K % KPerBlock == 0);
-    }
-
-    __host__ __device__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
-    {
-        const index_t grid_size = (M / MPerBlockM1) * (N / NPerBlockN1);
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
-    {
-        const bool has_main_k_block_loop = (K + KPerBlock) / (2 * KPerBlock) > 1;
-
-        return has_main_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasDoubleTailKBlockLoop(index_t K)
-    {
-        const bool has_double_tail_k_block_loop = (K / KPerBlock) % 2 == 0;
-
-        return has_double_tail_k_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeAKM0M1GridDescriptor(const AKMGridDesc& a_k_m_grid_desc)
-    {
-        const auto K = a_k_m_grid_desc.GetLength(I0);
-        const auto M = a_k_m_grid_desc.GetLength(I1);
-
-        const auto M1 = Number<MPerBlockM1>{};
-        const auto M0 = M / M1;
-
-        const auto a_k_m0_m1_grid_desc = transform_tensor_descriptor(
-            a_k_m_grid_desc,
-            make_tuple(make_pass_through_transform(K), make_unmerge_transform(make_tuple(M0, M1))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}));
-
-        return a_k_m0_m1_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeBKN0N1GridDescriptor(const BKNGridDesc& b_k_n_grid_desc)
-    {
-        const auto K = b_k_n_grid_desc.GetLength(I0);
-        const auto N = b_k_n_grid_desc.GetLength(I1);
-
-        const auto N1 = Number<NPerBlockN1>{};
-        const auto N0 = N / N1;
-
-        const auto b_k_n0_n1_grid_desc = transform_tensor_descriptor(
-            b_k_n_grid_desc,
-            make_tuple(make_pass_through_transform(K), make_unmerge_transform(make_tuple(N0, N1))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}));
-
-        return b_k_n0_n1_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCM0M10M11N0N10N11GridDescriptor(const CMNGridDesc& c_m_n_grid_desc)
-    {
-        const auto M = c_m_n_grid_desc.GetLength(I0);
-        const auto N = c_m_n_grid_desc.GetLength(I1);
-
-        constexpr auto M1 = Number<MPerBlockM1>{};
-        constexpr auto N1 = Number<NPerBlockN1>{};
-
-        const auto M0 = M / M1;
-        const auto N0 = N / N1;
-
-        constexpr auto M11 =
-            Number<M11N11ThreadClusterM1100 * M11N11ThreadClusterM1101 * M1PerThreadM111>{};
-        constexpr auto N11 =
-            Number<M11N11ThreadClusterN1100 * M11N11ThreadClusterN1101 * N1PerThreadN111>{};
-
-        constexpr auto M10 = M1 / M11;
-        constexpr auto N10 = N1 / N11;
-
-        const auto c_m0_m10_m11_n0_n10_n11_grid_desc = transform_tensor_descriptor(
-            c_m_n_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(M0, M10, M11)),
-                       make_unmerge_transform(make_tuple(N0, N10, N11))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0, 1, 2>{}, Sequence<3, 4, 5>{}));
-
-        return c_m0_m10_m11_n0_n10_n11_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCBlockIdToM0N0BlockClusterAdaptor(const CMNGridDesc& c_m_n_grid_desc)
-    {
-        const auto M = c_m_n_grid_desc.GetLength(I0);
-        const auto N = c_m_n_grid_desc.GetLength(I1);
-
-        constexpr auto M1 = Number<MPerBlockM1>{};
-        constexpr auto N1 = Number<NPerBlockN1>{};
-
-        const auto M0 = M / M1;
-        const auto N0 = N / N1;
-
-        const auto cblockid_to_m0_n0_block_cluster_adaptor =
-            make_single_stage_tensor_adaptor(make_tuple(make_merge_transform(make_tuple(M0, N0))),
-                                             make_tuple(Sequence<0, 1>{}),
-                                             make_tuple(Sequence<0>{}));
-
-        return cblockid_to_m0_n0_block_cluster_adaptor;
-    }
-
-    using AKM0M1GridDesc            = decltype(MakeAKM0M1GridDescriptor(AKMGridDesc{}));
-    using BKN0N1GridDesc            = decltype(MakeBKN0N1GridDescriptor(BKNGridDesc{}));
-    using CM0M10M11N0N10N11GridDesc = decltype(MakeCM0M10M11N0N10N11GridDescriptor(CMNGridDesc{}));
-    using CBlockIdToM0N0BlockClusterAdaptor =
-        decltype(MakeCBlockIdToM0N0BlockClusterAdaptor(CMNGridDesc{}));
-
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        FloatAB* __restrict__ p_shared_block,
-        const AKM0M1GridDesc& a_k_m0_m1_grid_desc,
-        const BKN0N1GridDesc& b_k_n0_n1_grid_desc,
-        const CM0M10M11N0N10N11GridDesc& c_m0_m10_m11_n0_n10_n11_grid_desc,
-        const CBlockIdToM0N0BlockClusterAdaptor& cblockid_to_m0_n0_block_cluster_adaptor,
-        integral_constant<bool, HasMainKBlockLoop>,
-        integral_constant<bool, HasDoubleTailKBlockLoop>)
-    {
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_grid, a_k_m0_m1_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_grid, b_k_n0_n1_grid_desc.GetElementSpaceSize());
-        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_grid, c_m0_m10_m11_n0_n10_n11_grid_desc.GetElementSpaceSize());
-
-        const auto K = a_k_m0_m1_grid_desc.GetLength(I0);
-
-        // divide block work by [M, N]
-        const auto c_m0_n0_block_cluster_idx =
-            cblockid_to_m0_n0_block_cluster_adaptor.CalculateBottomIndex(
-                make_multi_index(get_block_1d_id()));
-
-        // HACK: this force index data into SGPR
-        const index_t im0 = __builtin_amdgcn_readfirstlane(c_m0_n0_block_cluster_idx[I0]);
-        const index_t in0 = __builtin_amdgcn_readfirstlane(c_m0_n0_block_cluster_idx[I1]);
-
-        // lds max alignment
-        constexpr auto max_lds_align = math::lcm(Number<ABlockTransferDstScalarPerVector_M1>{},
-                                                 Number<BBlockTransferDstScalarPerVector_N1>{},
-                                                 Number<M1PerThreadM111>{},
-                                                 Number<N1PerThreadN111>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_k_m_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<MPerBlockM1>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_k_n_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, Number<NPerBlockN1>{}), max_lds_align);
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_k_m0_m1_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, I1, Number<MPerBlockM1>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_k_n0_n1_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<KPerBlock>{}, I1, Number<NPerBlockN1>{}), max_lds_align);
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<KPerBlock, 1, MPerBlockM1>,
-                                            ABlockTransferThreadSliceLengths_K_M0_M1,
-                                            ABlockTransferThreadClusterLengths_K_M0_M1,
-                                            ABlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(a_k_m0_m1_grid_desc),
-                                            decltype(a_k_m0_m1_block_desc),
-                                            ABlockTransferSrcAccessOrder,
-                                            Sequence<0, 1, 2>,
-                                            ABlockTransferSrcVectorDim,
-                                            2,
-                                            ABlockTransferSrcScalarPerVector,
-                                            ABlockTransferDstScalarPerVector_M1,
-                                            1,
-                                            1,
-                                            AThreadTransferSrcResetCoordinateAfterRun,
-                                            true>(a_k_m0_m1_grid_desc,
-                                                  make_multi_index(0, im0, 0),
-                                                  a_k_m0_m1_block_desc,
-                                                  make_multi_index(0, 0, 0));
-
-        // B matrix blockwise copy
-        auto b_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<KPerBlock, 1, NPerBlockN1>,
-                                            BBlockTransferThreadSliceLengths_K_N0_N1,
-                                            BBlockTransferThreadClusterLengths_K_N0_N1,
-                                            BBlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(b_k_n0_n1_grid_desc),
-                                            decltype(b_k_n0_n1_block_desc),
-                                            BBlockTransferSrcAccessOrder,
-                                            Sequence<0, 1, 2>,
-                                            BBlockTransferSrcVectorDim,
-                                            2,
-                                            BBlockTransferSrcScalarPerVector,
-                                            BBlockTransferDstScalarPerVector_N1,
-                                            1,
-                                            1,
-                                            BThreadTransferSrcResetCoordinateAfterRun,
-                                            true>(b_k_n0_n1_grid_desc,
-                                                  make_multi_index(0, in0, 0),
-                                                  b_k_n0_n1_block_desc,
-                                                  make_multi_index(0, 0, 0));
-
-        // GEMM definition
-        //   c_mtx += transpose(a_mtx) * b_mtx
-        //     a_mtx[KPerBlock, MPerBlockM1] is in LDS
-        //     b_mtx[KPerBlocl, NPerBlockN1] is in LDS
-        //     c_mtx[MPerBlockM1, NPerBlockN1] is distributed among threads, and saved in
-        //       register
-        const auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v2r2_pipeline_2x2<BlockSize,
-                                                                FloatAB,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_k_m_block_desc),
-                                                                decltype(b_k_n_block_desc),
-                                                                M1PerThreadM111,
-                                                                N1PerThreadN111,
-                                                                KPerThread,
-                                                                M11N11ThreadClusterM1100,
-                                                                M11N11ThreadClusterN1100,
-                                                                M11N11ThreadClusterM1101,
-                                                                M11N11ThreadClusterN1101,
-                                                                M1PerThreadM111,
-                                                                N1PerThreadN111>{};
-        constexpr auto c_m10_m11_n10_n11_thread_tensor_lengths =
-            decltype(blockwise_gemm)::GetCM0M1N0N1ThreadTensorLengths();
-
-        constexpr auto c_m10_m11_n10_n11_thread_desc = make_naive_tensor_descriptor_packed(
-            sequence_to_tuple_of_number(c_m10_m11_n10_n11_thread_tensor_lengths));
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_aligned_space_size =
-            math::integer_least_multiple(a_k_m0_m1_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        constexpr auto b_block_aligned_space_size =
-            math::integer_least_multiple(b_k_n0_n1_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        FloatAB* p_a_block_double = p_shared_block;
-        FloatAB* p_b_block_double = p_shared_block + 2 * a_block_aligned_space_size;
-
-        // register allocation for output
-        auto c_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAcc>(
-            c_m10_m11_n10_n11_thread_desc.GetElementSpaceSize());
-
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_m10_m11_n10_n11_thread_desc),
-                                    decltype(c_m10_m11_n10_n11_thread_tensor_lengths)>{}
-            .Run(c_m10_m11_n10_n11_thread_desc,
-                 make_tuple(I0, I0, I0, I0),
-                 c_thread_buf,
-                 FloatAcc{0});
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock, 0, 0);
-
-        // hack to control index calculation when iterating over A and B matrix for threadwise copy
-        constexpr auto a_k_m0_m1_global_step_hacks = AGridStepHacks{};
-        constexpr auto b_k_n0_n1_global_step_hacks = BGridStepHacks{};
-
-        // hack to control index calculation when move slice window for A and B matrix for
-        // threadwise copy
-        constexpr auto a_k_m0_m1_global_move_slice_window_step_hack =
-            AGridMoveSliceWindowStepHacks{};
-        constexpr auto b_k_n0_n1_global_move_slice_window_step_hack =
-            BGridMoveSliceWindowStepHacks{};
-
-        auto a_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double, a_k_m0_m1_block_desc.GetElementSpaceSize());
-        auto b_block_even_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double, b_k_n0_n1_block_desc.GetElementSpaceSize());
-
-        auto a_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_a_block_double + a_block_aligned_space_size,
-            a_k_m0_m1_block_desc.GetElementSpaceSize());
-        auto b_block_odd_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_b_block_double + b_block_aligned_space_size,
-            b_k_n0_n1_block_desc.GetElementSpaceSize());
-
-        // LDS double buffer: preload data into LDS
-        {
-            a_blockwise_copy.RunRead(
-                a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-            b_blockwise_copy.RunRead(
-                b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-            a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_even_buf);
-            b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_even_buf);
-        }
-
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t k_block_data_begin = 0;
-
-            // LDS double buffer: main body
-            // use Do-While loop instead of For loop to simplify control flow
-            do
-            {
-                // even iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_k_m0_m1_grid_desc,
-                                                    a_block_slice_copy_step,
-                                                    a_k_m0_m1_global_move_slice_window_step_hack);
-                b_blockwise_copy.MoveSrcSliceWindow(b_k_n0_n1_grid_desc,
-                                                    b_block_slice_copy_step,
-                                                    b_k_n0_n1_global_move_slice_window_step_hack);
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-                b_blockwise_copy.RunRead(
-                    b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(c_m10_m11_n10_n11_thread_desc,
-                                   a_block_even_buf,
-                                   b_block_even_buf,
-                                   c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_odd_buf);
-                b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_odd_buf);
-
-                // odd iteration
-                a_blockwise_copy.MoveSrcSliceWindow(a_k_m0_m1_grid_desc,
-                                                    a_block_slice_copy_step,
-                                                    a_k_m0_m1_global_move_slice_window_step_hack);
-                b_blockwise_copy.MoveSrcSliceWindow(b_k_n0_n1_grid_desc,
-                                                    b_block_slice_copy_step,
-                                                    b_k_n0_n1_global_move_slice_window_step_hack);
-
-                __syncthreads();
-
-                // LDS doubel buffer: load next data from device mem
-                a_blockwise_copy.RunRead(
-                    a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-                b_blockwise_copy.RunRead(
-                    b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(
-                    c_m10_m11_n10_n11_thread_desc, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-
-                // LDS double buffer: store next data to LDS
-                a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_even_buf);
-                b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_even_buf);
-
-                k_block_data_begin += 2 * KPerBlock;
-            } while(k_block_data_begin < K - 2 * KPerBlock);
-        }
-
-        // LDS double buffer: tail
-        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
-        {
-            a_blockwise_copy.MoveSrcSliceWindow(a_k_m0_m1_grid_desc,
-                                                a_block_slice_copy_step,
-                                                a_k_m0_m1_global_move_slice_window_step_hack);
-            b_blockwise_copy.MoveSrcSliceWindow(b_k_n0_n1_grid_desc,
-                                                b_block_slice_copy_step,
-                                                b_k_n0_n1_global_move_slice_window_step_hack);
-
-            __syncthreads();
-
-            // LDS double buffer: load last data from device mem
-            a_blockwise_copy.RunRead(
-                a_k_m0_m1_grid_desc, a_global_buf, a_k_m0_m1_global_step_hacks);
-            b_blockwise_copy.RunRead(
-                b_k_n0_n1_grid_desc, b_global_buf, b_k_n0_n1_global_step_hacks);
-
-            // LDS double buffer: GEMM on 2nd-last data
-            blockwise_gemm.Run(
-                c_m10_m11_n10_n11_thread_desc, a_block_even_buf, b_block_even_buf, c_thread_buf);
-
-            // LDS double buffer: store last data to LDS
-            a_blockwise_copy.RunWrite(a_k_m0_m1_block_desc, a_block_odd_buf);
-            b_blockwise_copy.RunWrite(b_k_n0_n1_block_desc, b_block_odd_buf);
-
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_m10_m11_n10_n11_thread_desc, a_block_odd_buf, b_block_odd_buf, c_thread_buf);
-        }
-        else // if has 1 iteration left
-        {
-            __syncthreads();
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(
-                c_m10_m11_n10_n11_thread_desc, a_block_even_buf, b_block_even_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            constexpr auto c_m0_m10_m11_n0_n10_n11_thread_desc =
-                make_naive_tensor_descriptor_packed(
-                    make_tuple(I1,
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I0]>{},
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I1]>{},
-                               I1,
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I2]>{},
-                               Number<c_m10_m11_n10_n11_thread_tensor_lengths[I3]>{}));
-
-            const auto c_m10_m11_n10_n11_thread_origin_idx_on_block =
-                blockwise_gemm.CalculateCM0M1N0N1ThreadOriginOnBlock(get_thread_local_1d_id());
-
-            ThreadwiseTensorSliceTransfer_v1r3<
-                FloatAcc,
-                FloatC,
-                decltype(c_m0_m10_m11_n0_n10_n11_thread_desc),
-                decltype(c_m0_m10_m11_n0_n10_n11_grid_desc),
-                Sequence<1,
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I0],
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I1],
-                         1,
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I2],
-                         c_m10_m11_n10_n11_thread_tensor_lengths[I3]>,
-                CThreadTransferSrcDstAccessOrder,
-                CThreadTransferSrcDstVectorDim,
-                CThreadTransferDstScalarPerVector,
-                CGlobalMemoryDataOperation,
-                1,
-                true>{c_m0_m10_m11_n0_n10_n11_grid_desc,
-                      make_multi_index(im0,
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I0],
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I1],
-                                       in0,
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I2],
-                                       c_m10_m11_n10_n11_thread_origin_idx_on_block[I3])}
-                .Run(c_m0_m10_m11_n0_n10_n11_thread_desc,
-                     make_tuple(I0, I0, I0, I0, I0, I0),
-                     c_thread_buf,
-                     c_m0_m10_m11_n0_n10_n11_grid_desc,
-                     c_grid_buf,
-                     CGridStepHacks{});
-        }
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp
deleted file mode 100644
index b1dfb0c73f..0000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v2.hpp
+++ /dev/null
@@ -1,461 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_GEMM_V2_HPP
-#define CK_GRIDWISE_GEMM_V2_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "blockwise_gemm_dlops_v3.hpp"
-
-namespace ck {
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGlobalDesc,
-          typename BGlobalDesc,
-          typename CGlobalDesc,
-          index_t KPerBlock,
-          index_t HoPerBlock,
-          index_t WoPerBlock,
-          index_t EPerBlock,
-          index_t KPerThread,
-          index_t HoPerThread,
-          index_t WoPerThread,
-          index_t EPerThread,
-          typename ABlockTransferThreadSliceLengths_E_K,
-          typename ABlockTransferThreadClusterLengths_E_K,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_K,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGlobalStepHacks,
-          typename BGlobalStepHacks,
-          typename CGlobalStepHacks,
-          typename AGlobalMoveSliceWindowStepHacks,
-          typename BGlobalMoveSliceWindowStepHacks>
-struct GridwiseGemmDlops_km_kn_mn_v3
-{
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        constexpr auto E = EPerBlock * 3 * 3;
-
-        constexpr auto max_lds_align =
-            math::lcm(Number<ABlockTransferDstScalarPerVector_K>{}, Number<KPerBlock>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_e_k_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E>{}, Number<KPerBlock>{}), max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_space_size =
-            math::integer_least_multiple(a_e_k_desc.GetElementSpaceSize(), max_lds_align);
-
-        return a_block_space_size * sizeof(FloatAB);
-    }
-
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const AGlobalDesc& a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const BGlobalDesc& b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const CGlobalDesc& c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        FloatAB* __restrict__ p_shared_block,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-        constexpr auto I2 = Number<2>{};
-        constexpr auto I3 = Number<3>{};
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e_k_global_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e_n_ho_wo_global_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k_n_ho_wo_global_desc.GetElementSpaceSize());
-
-        constexpr auto E = EPerBlock * 3 * 3;
-
-        // const auto E = a_e_k_global_desc.GetLength(I0);
-        const auto K = a_e_k_global_desc.GetLength(I1);
-
-        const auto N  = b_e_n_ho_wo_global_desc.GetLength(I1);
-        const auto Ho = b_e_n_ho_wo_global_desc.GetLength(I2);
-        const auto Wo = b_e_n_ho_wo_global_desc.GetLength(I3);
-
-// divide block work by [M, N]
-#if 0
-        const auto ho_block_work_num  = Ho / Number<HoPerBlock>{};
-        const auto wo_block_work_num  = Wo / Number<WoPerBlock>{};
-        const auto hwo_block_work_num = ho_block_work_num * wo_block_work_num;
-
-        const index_t k_block_work_id   = get_block_1d_id() / hwo_block_work_num;
-        const index_t hwo_block_work_id = get_block_1d_id() - k_block_work_id * hwo_block_work_num;
-
-        const index_t ho_block_work_id = hwo_block_work_id / wo_block_work_num;
-        const index_t wo_block_work_id = hwo_block_work_id - ho_block_work_id * wo_block_work_num;
-#else
-        // Hack: this force result into SGPR
-        const index_t ho_block_work_num  = __builtin_amdgcn_readfirstlane(Ho / HoPerBlock);
-        const index_t wo_block_work_num  = __builtin_amdgcn_readfirstlane(Wo / WoPerBlock);
-        const index_t hwo_block_work_num = ho_block_work_num * wo_block_work_num;
-
-        const index_t k_block_work_id =
-            __builtin_amdgcn_readfirstlane(get_block_1d_id() / hwo_block_work_num);
-        const index_t hwo_block_work_id = get_block_1d_id() - k_block_work_id * hwo_block_work_num;
-
-        const index_t ho_block_work_id =
-            __builtin_amdgcn_readfirstlane(hwo_block_work_id / wo_block_work_num);
-        const index_t wo_block_work_id = hwo_block_work_id - ho_block_work_id * wo_block_work_num;
-#endif
-
-        // lds max alignment
-        constexpr auto max_lds_align =
-            math::lcm(Number<ABlockTransferDstScalarPerVector_K>{}, Number<KPerBlock>{});
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_e_k_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<EPerBlock>{}, Number<KPerBlock>{}), max_lds_align);
-
-        constexpr auto a_e_k_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E>{}, Number<KPerBlock>{}), max_lds_align);
-
-        // B matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto b_e_n_ho_wo_block_desc = make_naive_tensor_descriptor_packed(make_tuple(
-            Number<EPerBlock>{}, Number<1>{}, Number<HoPerBlock>{}, Number<WoPerBlock>{}));
-
-        // c_thread_mtx definition: this is a mess
-        // TODO:: more elegent way of defining c_thread_mtx
-        constexpr auto c_k_n_ho_wo_thread_desc = make_naive_tensor_descriptor_packed(make_tuple(
-            Number<KPerThread>{}, Number<1>{}, Number<HoPerThread>{}, Number<WoPerThread>{}));
-
-        auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v3<BlockSize,
-                                                 FloatAB,
-                                                 FloatAB,
-                                                 FloatAcc,
-                                                 decltype(a_e_k_block_desc),
-                                                 decltype(b_e_n_ho_wo_block_desc),
-                                                 decltype(c_k_n_ho_wo_thread_desc),
-                                                 KPerThread,
-                                                 HoPerThread,
-                                                 WoPerThread,
-                                                 EPerThread,
-                                                 ABlockTransferSrcScalarPerVector,
-                                                 ABlockTransferDstScalarPerVector_K>{};
-
-        auto c_thread_mtx_index = blockwise_gemm.GetBeginOfThreadMatrixC(get_thread_local_1d_id());
-
-        const auto k_thread_id  = c_thread_mtx_index.k;
-        const auto ho_thread_id = c_thread_mtx_index.h;
-        const auto wo_thread_id = c_thread_mtx_index.w;
-
-        const index_t k_block_data_on_global  = k_block_work_id * KPerBlock;
-        const index_t ho_block_data_on_global = ho_block_work_id * HoPerBlock;
-        const index_t wo_block_data_on_global = wo_block_work_id * WoPerBlock;
-
-        const index_t ho_thread_data_on_global =
-            ho_block_data_on_global + ho_thread_id * HoPerThread;
-        const index_t wo_thread_data_on_global =
-            wo_block_data_on_global + wo_thread_id * WoPerThread;
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<E, KPerBlock>,
-                                            ABlockTransferThreadSliceLengths_E_K,
-                                            ABlockTransferThreadClusterLengths_E_K,
-                                            ABlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(a_e_k_global_desc),
-                                            decltype(a_e_k_desc),
-                                            ABlockTransferSrcAccessOrder,
-                                            Sequence<0, 1>,
-                                            ABlockTransferSrcVectorDim,
-                                            1,
-                                            ABlockTransferSrcScalarPerVector,
-                                            ABlockTransferDstScalarPerVector_K,
-                                            1,
-                                            1,
-                                            AThreadTransferSrcResetCoordinateAfterRun,
-                                            true>(a_e_k_global_desc,
-                                                  make_multi_index(0, k_block_data_on_global),
-                                                  a_e_k_desc,
-                                                  make_multi_index(0, 0));
-
-        constexpr auto b_e_n_ho_wo_thread_desc = make_naive_tensor_descriptor_packed(make_tuple(
-            Number<EPerBlock>{}, Number<1>{}, Number<HoPerThread>{}, Number<WoPerThread>{}));
-
-        auto b_threadwise_transfer =
-            ThreadwiseTensorSliceTransfer_v2<FloatAB,
-                                             FloatAB,
-                                             decltype(b_e_n_ho_wo_global_desc),
-                                             decltype(b_e_n_ho_wo_thread_desc),
-                                             Sequence<EPerBlock, 1, HoPerThread, WoPerThread>,
-                                             BBlockTransferSrcAccessOrder,
-                                             BBlockTransferSrcVectorDim,
-                                             BBlockTransferSrcScalarPerVector,
-                                             1,
-                                             true>(
-                b_e_n_ho_wo_global_desc,
-                make_multi_index(0, 0, ho_thread_data_on_global, wo_thread_data_on_global));
-
-        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_shared_block, a_e_k_desc.GetElementSpaceSize());
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k_n_ho_wo_thread_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        // initialize output thread tensor
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_k_n_ho_wo_thread_desc),
-                                    Sequence<KPerThread, 1, HoPerThread, WoPerThread>>{}
-            .Run(c_k_n_ho_wo_thread_desc, make_tuple(I0, I0, I0, I0), c_thread_buf, FloatAcc{0});
-
-        constexpr auto b_thread_slice_copy_step = make_multi_index(EPerBlock, 0, 0, 0);
-
-        // hack to control index calculation when iterating over A and B matrix for threadwise copy
-        constexpr auto a_e_k_global_step_hacks       = AGlobalStepHacks{};
-        constexpr auto b_e_n_ho_wo_global_step_hacks = BGlobalStepHacks{};
-
-        // hack to control index calculation when move slice window for A and B matrix for
-        // threadwise copy
-        constexpr auto a_e_k_global_move_slice_window_step_hack = AGlobalMoveSliceWindowStepHacks{};
-        constexpr auto b_e_n_ho_wo_global_move_slice_window_step_hack =
-            BGlobalMoveSliceWindowStepHacks{};
-
-        // double regsiter buffer for b
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAB,
-                     b_e_n_ho_wo_thread_desc.GetElementSpaceSize(),
-                     true>
-            b_thread_even_buf, b_thread_odd_buf;
-
-        // LDS double buffer: preload data
-        {
-            a_blockwise_copy.RunRead(a_e_k_global_desc, a_global_buf, a_e_k_global_step_hacks);
-
-            b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                      b_global_buf,
-                                      b_e_n_ho_wo_thread_desc,
-                                      make_tuple(I0, I0, I0, I0),
-                                      b_thread_even_buf,
-                                      b_e_n_ho_wo_global_step_hacks);
-
-            a_blockwise_copy.RunWrite(a_e_k_desc, a_block_buf);
-        }
-
-        __syncthreads();
-
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t e_block_data_begin = 0;
-
-            // LDS double buffer: main body
-            // use Do-While loop instead of For loop to simplify control flow
-            do
-            {
-                // even iteration
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e_n_ho_wo_global_desc,
-                                                         b_thread_slice_copy_step);
-
-                b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                          b_global_buf,
-                                          b_e_n_ho_wo_thread_desc,
-                                          make_tuple(I0, I0, I0, I0),
-                                          b_thread_odd_buf,
-                                          b_e_n_ho_wo_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                // TODO: @Zhang Jing: blockwise gemm should be able to move slice window
-                blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                blockwise_gemm.MoveASliceWindow(a_e_k_block_desc, make_tuple(EPerBlock, 0));
-
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e_n_ho_wo_global_desc,
-                                                         b_thread_slice_copy_step);
-
-                b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                          b_global_buf,
-                                          b_e_n_ho_wo_thread_desc,
-                                          make_tuple(I0, I0, I0, I0),
-                                          b_thread_even_buf,
-                                          b_e_n_ho_wo_global_step_hacks);
-
-                // LDS double buffer: GEMM on current data
-                blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-
-                blockwise_gemm.MoveASliceWindow(a_e_k_block_desc, make_tuple(EPerBlock, 0));
-
-                e_block_data_begin += 2 * EPerBlock;
-
-            } while(e_block_data_begin < E - 2 * EPerBlock);
-        }
-
-        // LDS double buffer: tail
-        if constexpr(HasDoubleTailKBlockLoop) // if has 2 iteration left
-        {
-            b_threadwise_transfer.MoveSrcSliceWindow(b_e_n_ho_wo_global_desc,
-                                                     b_thread_slice_copy_step);
-
-            b_threadwise_transfer.Run(b_e_n_ho_wo_global_desc,
-                                      b_global_buf,
-                                      b_e_n_ho_wo_thread_desc,
-                                      make_tuple(I0, I0, I0, I0),
-                                      b_thread_odd_buf,
-                                      b_e_n_ho_wo_global_step_hacks);
-
-            // LDS double buffer: GEMM on 2nd-last data
-            blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-            blockwise_gemm.MoveASliceWindow(a_e_k_block_desc, make_tuple(EPerBlock, 0));
-
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-        }
-        else // if has 1 iteration left
-        {
-            // LDS double buffer: GEMM on last data
-            blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-        }
-
-        // output: register to global memory
-        {
-            // hack to control index calculation when iterating over c_k_n_ho_wo_global tensor
-            constexpr auto c_k_n_ho_wo_global_tensor_step_hacks = CGlobalStepHacks{};
-
-            const index_t k_thread_data_on_global =
-                k_block_data_on_global + k_thread_id * KPerThread;
-
-            ThreadwiseTensorSliceTransfer_v1r3<FloatAcc,
-                                               FloatC,
-                                               decltype(c_k_n_ho_wo_thread_desc),
-                                               decltype(c_k_n_ho_wo_global_desc),
-                                               Sequence<KPerThread, 1, HoPerThread, WoPerThread>,
-                                               CThreadTransferSrcDstAccessOrder,
-                                               CThreadTransferSrcDstVectorDim,
-                                               CThreadTransferDstScalarPerVector,
-                                               CGlobalMemoryDataOperation,
-                                               1,
-                                               true>(
-                c_k_n_ho_wo_global_desc,
-                make_multi_index(
-                    k_thread_data_on_global, 0, ho_thread_data_on_global, wo_thread_data_on_global))
-                .Run(c_k_n_ho_wo_thread_desc,
-                     make_tuple(I0, I0, I0, I0),
-                     c_thread_buf,
-                     c_k_n_ho_wo_global_desc,
-                     c_global_buf,
-                     c_k_n_ho_wo_global_tensor_step_hacks);
-        }
-    }
-
-    // pass tensor descriptor by reference
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const AGlobalDesc& a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const BGlobalDesc& b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const CGlobalDesc& c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        constexpr index_t shared_block_size = GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-        __shared__ FloatAB p_shared_block[shared_block_size];
-
-        Run(a_e_k_global_desc,
-            p_a_global,
-            b_e_n_ho_wo_global_desc,
-            p_b_global,
-            c_k_n_ho_wo_global_desc,
-            p_c_global,
-            p_shared_block,
-            integral_constant<bool, HasMainKBlockLoop>{},
-            integral_constant<bool, HasDoubleTailKBlockLoop>{});
-    }
-
-    // pass tensor descriptors by their pointers
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const AGlobalDesc* p_a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const BGlobalDesc* p_b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const CGlobalDesc* p_c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        const auto a_e_k_global_desc       = *p_a_e_k_global_desc;
-        const auto b_e_n_ho_wo_global_desc = *p_b_e_n_ho_wo_global_desc;
-        const auto c_k_n_ho_wo_global_desc = *p_c_k_n_ho_wo_global_desc;
-
-        Run(a_e_k_global_desc,
-            p_a_global,
-            b_e_n_ho_wo_global_desc,
-            p_b_global,
-            c_k_n_ho_wo_global_desc,
-            p_c_global,
-            integral_constant<bool, HasMainKBlockLoop>{},
-            integral_constant<bool, HasDoubleTailKBlockLoop>{});
-    }
-
-    // pass tensor descriptors by void*
-    template <bool HasMainKBlockLoop, bool HasDoubleTailKBlockLoop>
-    __device__ void Run(const void* p_a_e_k_global_desc,
-                        const FloatAB* __restrict__ p_a_global,
-                        const void* p_b_e_n_ho_wo_global_desc,
-                        const FloatAB* __restrict__ p_b_global,
-                        const void* p_c_k_n_ho_wo_global_desc,
-                        FloatC* __restrict__ p_c_global,
-                        integral_constant<bool, HasMainKBlockLoop>,
-                        integral_constant<bool, HasDoubleTailKBlockLoop>) const
-    {
-        const auto a_e_k_global_desc = *reinterpret_cast<const AGlobalDesc*>(p_a_e_k_global_desc);
-        const auto b_e_n_ho_wo_global_desc =
-            *reinterpret_cast<const BGlobalDesc*>(p_b_e_n_ho_wo_global_desc);
-        const auto c_k_n_ho_wo_global_desc =
-            *reinterpret_cast<const CGlobalDesc*>(p_c_k_n_ho_wo_global_desc);
-
-        Run(a_e_k_global_desc,
-            p_a_global,
-            b_e_n_ho_wo_global_desc,
-            p_b_global,
-            c_k_n_ho_wo_global_desc,
-            p_c_global,
-            integral_constant<bool, HasMainKBlockLoop>{},
-            integral_constant<bool, HasDoubleTailKBlockLoop>{});
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp
deleted file mode 100644
index ace8443384..0000000000
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dlops_v3.hpp
+++ /dev/null
@@ -1,1597 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_GRIDWISE_GEMM_V3_HPP
-#define CK_GRIDWISE_GEMM_V3_HPP
-
-#include "common_header.hpp"
-#include "multi_index_transform_helper.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "blockwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_transfer.hpp"
-#include "threadwise_tensor_slice_set.hpp"
-#include "blockwise_gemm_dlops_v3.hpp"
-
-namespace ck {
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_E0_E1_K0_K1_E2,
-          typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-          typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-          typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-          bool HasMainE0BlockLoop,
-          ActivTypeEnum ActivType>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            const FloatC* __restrict__ p_bias_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_E0_E1_K0_K1_E2 a_e0_e1_k0_k1_e2_grid_desc,
-            const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-            const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-            const CBlockIdToBlockClusterAdaptor_K_N_H_W cblockid_to_k_n_h_w_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::ConvBiasActiv(p_a_grid,
-                                p_b_grid,
-                                p_bias_grid,
-                                p_c_grid,
-                                p_shared_block,
-                                a_e0_e1_k0_k1_e2_grid_desc,
-                                b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                                cblockid_to_k_n_h_w_block_cluster_adaptor,
-                                integral_constant<bool, HasMainE0BlockLoop>{},
-                                integral_constant<ActivTypeEnum, ActivType>{});
-}
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_E0_E1_K0_K1_E2,
-          typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-          typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-          typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-          typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-          bool HasMainE0BlockLoop,
-          ActivTypeEnum ActivType>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v3_resize_add(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            const FloatC* __restrict__ p_bias_grid,
-            FloatC* __restrict__ p_d_grid,
-            const AGridDesc_E0_E1_K0_K1_E2 a_e0_e1_k0_k1_e2_grid_desc,
-            const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-            const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-            const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-            const CBlockIdToBlockClusterAdaptor_K_N_H_W cblockid_to_k_n_h_w_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::ConvBiasActivResizeAdd(p_a_grid,
-                                         p_b_grid,
-                                         p_bias_grid,
-                                         p_d_grid,
-                                         p_shared_block,
-                                         a_e0_e1_k0_k1_e2_grid_desc,
-                                         b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                         c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                                         d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                                         cblockid_to_k_n_h_w_block_cluster_adaptor,
-                                         integral_constant<bool, HasMainE0BlockLoop>{},
-                                         integral_constant<ActivTypeEnum, ActivType>{});
-}
-
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_E0_E1_K0_K1_E2,
-          typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-          typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-          typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-          typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-          bool HasMainE0BlockLoop,
-          ActivTypeEnum ActivType>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_gemm_dlops_v3_maxpool(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            const FloatC* __restrict__ p_bias_grid,
-            FloatC* __restrict__ p_c_grid,
-            FloatC* __restrict__ p_d_grid,
-            const AGridDesc_E0_E1_K0_K1_E2 a_e0_e1_k0_k1_e2_grid_desc,
-            const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-            const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-            const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-            const CBlockIdToBlockClusterAdaptor_K_N_H_W cblockid_to_k_n_h_w_block_cluster_adaptor)
-{
-    constexpr index_t shared_block_size =
-        GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
-
-    __shared__ FloatAB p_shared_block[shared_block_size];
-
-    GridwiseGemm::ConvBiasActivMaxpool(p_a_grid,
-                                       p_b_grid,
-                                       p_bias_grid,
-                                       p_c_grid,
-                                       p_d_grid,
-                                       p_shared_block,
-                                       a_e0_e1_k0_k1_e2_grid_desc,
-                                       b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                       c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                                       d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                                       cblockid_to_k_n_h_w_block_cluster_adaptor,
-                                       integral_constant<bool, HasMainE0BlockLoop>{},
-                                       integral_constant<ActivTypeEnum, ActivType>{});
-}
-
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename FloatC,
-          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_E0_E1_K_E2,
-          typename BGridDesc_E0_E1_N_Ho_Wo_E2,
-          typename CGridDesc_K_N_Ho_Wo,
-          typename DGridDesc_K_N_Hx_Wx,
-          index_t E1_,
-          index_t E2_,
-          index_t K2_,
-          index_t KPerBlock,
-          index_t HoPerBlock,
-          index_t WoPerBlock,
-          index_t E1PerBlock,
-          index_t KPerThread,
-          index_t HoPerThread,
-          index_t WoPerThread,
-          index_t EPerThread,
-          typename ABlockTransferThreadSliceLengths_E0_E1_K0_K1_E2,
-          typename ABlockTransferThreadClusterLengths_E0_E1_K0_K1_E2,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_E2,
-          bool AThreadTransferSrcResetCoordinateAfterRun,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          bool BThreadTransferSrcResetCoordinateAfterRun,
-          typename CThreadTransferSrcDstAccessOrder,
-          index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          typename AGlobalStepHacks,
-          typename BGlobalStepHacks,
-          typename CGlobalStepHacks,
-          typename DGlobalStepHacks,
-          typename AGlobalMoveSliceWindowStepHacks,
-          typename BGlobalMoveSliceWindowStepHacks>
-struct GridwiseGemmDlops_km_kn_mn_v3
-{
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-    static constexpr auto I4 = Number<4>{};
-    static constexpr auto I5 = Number<5>{};
-
-    static constexpr auto E1 = Number<E1_>{};
-    static constexpr auto E2 = Number<E2_>{};
-    static constexpr auto K2 = Number<K2_>{};
-
-    static constexpr auto NPerBlock = I1;
-
-    static constexpr FloatAcc alpha = 0.3;
-
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
-    {
-        constexpr auto max_lds_align = Number<ABlockTransferDstScalarPerVector_E2>{};
-
-        // A matrix in LDS memory, dst of blockwise copy
-        //   be careful of LDS alignment
-        constexpr auto a_e0_e1_k1_e2_block_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(I1, Number<E1>{}, Number<KPerBlock>{}, Number<E2>{}), max_lds_align);
-
-        // LDS allocation for A and B: be careful of alignment
-        constexpr auto a_block_space_size = math::integer_least_multiple(
-            a_e0_e1_k1_e2_block_desc.GetElementSpaceSize(), max_lds_align);
-
-        return a_block_space_size * sizeof(FloatAB);
-    }
-
-    __host__ __device__ static constexpr index_t
-    CalculateGridSize(const CGridDesc_K_N_Ho_Wo& c_k_n_ho_wo_grid_desc)
-    {
-        const auto K  = c_k_n_ho_wo_grid_desc.GetLength(I0);
-        const auto N  = c_k_n_ho_wo_grid_desc.GetLength(I1);
-        const auto Ho = c_k_n_ho_wo_grid_desc.GetLength(I2);
-        const auto Wo = c_k_n_ho_wo_grid_desc.GetLength(I3);
-
-        const auto K0 = K / KPerBlock;
-        const auto N0 = N / NPerBlock;
-        const auto H0 = Ho / HoPerBlock;
-        const auto W0 = Wo / WoPerBlock;
-
-        const index_t grid_size = K0 * N0 * H0 * W0;
-
-        return grid_size;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainE0BlockLoop(const index_t E0)
-    {
-        const bool has_main_e0_block_loop = E0 > 1;
-
-        return has_main_e0_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasMainE1BlockLoop()
-    {
-        const bool has_main_e1_block_loop = ((E1 + E1PerBlock) / (2 * E1PerBlock)) > 1;
-
-        return has_main_e1_block_loop;
-    }
-
-    __host__ __device__ static constexpr bool CalculateHasDoubleTailE1BlockLoop()
-    {
-        const bool has_double_tail_e1_block_loop = (E1 / E1PerBlock) % 2 == 0;
-
-        return has_double_tail_e1_block_loop;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeAE0E1K0K1E2GridDescriptor(const AGridDesc_E0_E1_K_E2& a_e0_e1_k_e2_grid_desc)
-    {
-        const auto E0 = a_e0_e1_k_e2_grid_desc.GetLength(I0);
-        const auto K  = a_e0_e1_k_e2_grid_desc.GetLength(I2);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-        const auto a_e0_e1_k0_k1_e2_grid_desc = transform_tensor_descriptor(
-            a_e0_e1_k_e2_grid_desc,
-            make_tuple(make_pass_through_transform(E0),
-                       make_pass_through_transform(E1),
-                       make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(E2)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2, 3>{}, Sequence<4>{}));
-
-        return a_e0_e1_k0_k1_e2_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto MakeBE0E1NH0H1H2W0W1W2E2GridDescriptor(
-        const BGridDesc_E0_E1_N_Ho_Wo_E2& b_e0_e1_n_ho_wo_e2_grid_desc)
-    {
-        const auto E0 = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I0);
-        // const auto E1 = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I1);
-        const auto N  = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I2);
-        const auto Ho = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I3);
-        const auto Wo = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I4);
-        // const auto E2 = b_e0_e1_n_ho_wo_e2_grid_desc.GetLength(I5);
-
-        const auto H2 = Number<HoPerThread>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-        const auto H0 = Ho / (H1 * H2);
-
-        const auto W2 = Number<WoPerThread>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-        const auto W0 = Wo / (W1 * W2);
-
-        const auto b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc =
-            transform_tensor_descriptor(b_e0_e1_n_ho_wo_e2_grid_desc,
-                                        make_tuple(make_pass_through_transform(E0),
-                                                   make_pass_through_transform(E1),
-                                                   make_pass_through_transform(N),
-                                                   make_unmerge_transform(make_tuple(H0, H1, H2)),
-                                                   make_unmerge_transform(make_tuple(W0, W1, W2)),
-                                                   make_pass_through_transform(E2)),
-                                        make_tuple(Sequence<0>{},
-                                                   Sequence<1>{},
-                                                   Sequence<2>{},
-                                                   Sequence<3>{},
-                                                   Sequence<4>{},
-                                                   Sequence<5>{}),
-                                        make_tuple(Sequence<0>{},
-                                                   Sequence<1>{},
-                                                   Sequence<2>{},
-                                                   Sequence<3, 4, 5>{},
-                                                   Sequence<6, 7, 8>{},
-                                                   Sequence<9>{}));
-
-        return b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCK0K1NH0H1H2W0W1W2GridDescriptor(const CGridDesc_K_N_Ho_Wo& c_k_n_ho_wo_grid_desc)
-    {
-        const auto K  = c_k_n_ho_wo_grid_desc.GetLength(I0);
-        const auto N  = c_k_n_ho_wo_grid_desc.GetLength(I1);
-        const auto Ho = c_k_n_ho_wo_grid_desc.GetLength(I2);
-        const auto Wo = c_k_n_ho_wo_grid_desc.GetLength(I3);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-        const auto H2 = Number<HoPerThread>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-        const auto H0 = Ho / (H1 * H2);
-
-        const auto W2 = Number<WoPerThread>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-        const auto W0 = Wo / (W1 * W2);
-
-        const auto c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc = transform_tensor_descriptor(
-            c_k_n_ho_wo_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(N),
-                       make_unmerge_transform(make_tuple(H0, H1, H2)),
-                       make_unmerge_transform(make_tuple(W0, W1, W2))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0, 1>{}, Sequence<2>{}, Sequence<3, 4, 5>{}, Sequence<6, 7, 8>{}));
-
-        return c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeDK0K1NH0H1HxW0W1WxGridDescriptorMaxPool(const DGridDesc_K_N_Hx_Wx& d_k_n_hx_wx_grid_desc)
-    {
-        const auto K  = d_k_n_hx_wx_grid_desc.GetLength(I0);
-        const auto N  = d_k_n_hx_wx_grid_desc.GetLength(I1);
-        const auto Hx = d_k_n_hx_wx_grid_desc.GetLength(I2);
-        const auto Wx = d_k_n_hx_wx_grid_desc.GetLength(I3);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-#if CK_EXPERIMENTAL_STATIC_TENSOR_DESCRIPTOR
-        const auto H2 = Number<HoPerThread / 2>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-        const auto H0 = Number<Hx / (H1 * H2)>{};
-
-        const auto W2 = Number<WoPerThread / 2>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-        const auto W0 = Number<Wx / (W1 * W2)>{};
-#else
-        const auto H2 = HoPerThread / 2;
-        const auto H1 = HoPerBlock / HoPerThread;
-        const auto H0 = Hx / (H1 * H2);
-
-        const auto W2 = WoPerThread / 2;
-        const auto W1 = WoPerBlock / WoPerThread;
-        const auto W0 = Wx / (W1 * W2);
-#endif
-
-        const auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc = transform_tensor_descriptor(
-            d_k_n_hx_wx_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(N),
-                       make_unmerge_transform(make_tuple(H0, H1, H2)),
-                       make_unmerge_transform(make_tuple(W0, W1, W2))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0, 1>{}, Sequence<2>{}, Sequence<3, 4, 5>{}, Sequence<6, 7, 8>{}));
-
-        return d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeDK0K1NH0H1HxW0W1WxGridDescriptorResizeAdd(const DGridDesc_K_N_Hx_Wx& d_k_n_hx_wx_grid_desc)
-    {
-        const auto K  = d_k_n_hx_wx_grid_desc.GetLength(I0);
-        const auto N  = d_k_n_hx_wx_grid_desc.GetLength(I1);
-        const auto Hx = d_k_n_hx_wx_grid_desc.GetLength(I2);
-        const auto Wx = d_k_n_hx_wx_grid_desc.GetLength(I3);
-
-        const auto K1 = Number<KPerBlock>{};
-        const auto K0 = K / K1;
-
-        const auto H2 = Number<HoPerThread * 2>{};
-        const auto H1 = Number<HoPerBlock / HoPerThread>{};
-
-        const auto W2 = Number<WoPerThread * 2>{};
-        const auto W1 = Number<WoPerBlock / WoPerThread>{};
-
-#if CK_EXPERIMENTAL_STATIC_TENSOR_DESCRIPTOR
-        const auto H0 = Number<Hx / (H1 * H2)>{};
-        const auto W0 = Number<Wx / (W1 * W2)>{};
-#else
-        const auto H0 = Hx / (H1 * H2);
-        const auto W0 = Wx / (W1 * W2);
-#endif
-
-        const auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc = transform_tensor_descriptor(
-            d_k_n_hx_wx_grid_desc,
-            make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
-                       make_pass_through_transform(N),
-                       make_unmerge_transform(make_tuple(H0, H1, H2)),
-                       make_unmerge_transform(make_tuple(W0, W1, W2))),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0, 1>{}, Sequence<2>{}, Sequence<3, 4, 5>{}, Sequence<6, 7, 8>{}));
-
-        return d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCBlockIdToKNHoWoBlockClusterAdaptor(const CGridDesc_K_N_Ho_Wo& c_k_n_ho_wo_grid_desc)
-    {
-        const auto K  = c_k_n_ho_wo_grid_desc.GetLength(I0);
-        const auto N  = c_k_n_ho_wo_grid_desc.GetLength(I1);
-        const auto Ho = c_k_n_ho_wo_grid_desc.GetLength(I2);
-        const auto Wo = c_k_n_ho_wo_grid_desc.GetLength(I3);
-
-#if CK_EXPERIMENTAL_STATIC_TENSOR_DESCRIPTOR
-        const auto K0 = Number<K / KPerBlock>{};
-        const auto N0 = Number<N / NPerBlock>{};
-        const auto H0 = Number<Ho / HoPerBlock>{};
-        const auto W0 = Number<Wo / WoPerBlock>{};
-#else
-        const auto K0 = K / KPerBlock;
-        const auto N0 = N / NPerBlock;
-        const auto H0 = Ho / HoPerBlock;
-        const auto W0 = Wo / WoPerBlock;
-#endif
-
-        const auto cblockid_to_k_n_ho_wo_block_cluster_adaptor = make_single_stage_tensor_adaptor(
-            make_tuple(make_merge_transform(make_tuple(K0, N0, H0, W0))),
-            make_tuple(Sequence<0, 1, 2, 3>{}),
-            make_tuple(Sequence<0>{}));
-
-        return cblockid_to_k_n_ho_wo_block_cluster_adaptor;
-    }
-
-    // using AGridDesc_E0_E1_K0_K1_E2 =
-    // decltype(MakeAE0E1K0K1E2GridDescriptor(AGridDesc_E0_E1_K_E2{}));
-    // using BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2 =
-    // decltype(MakeBE0E1NH0H1H2W0W1W2E2GridDescriptor(BGridDesc_E0_E1_N_Ho_Wo_E2{}));
-    // using CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2 =
-    // decltype(MakeCK0K1NH0H1H2W0W1W2GridDescriptor(CGridDesc_K_N_Ho_Wo{}));
-    // using DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx =
-    // decltype(MakeDK0K1NH0H1HxW0W1WxGridDescriptor(DGridDesc_K_N_Hx_Wx{}));
-
-    using CBlockIdToBlockClusterAdaptor_K_N_H_W =
-        decltype(MakeCBlockIdToKNHoWoBlockClusterAdaptor(CGridDesc_K_N_Ho_Wo{}));
-
-    template <typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2>
-    __host__ __device__ static constexpr auto MakeBiasK0K1GridDescriptor(
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc)
-    {
-        const auto K0 = c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetLength(I0);
-        const auto K1 = c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetLength(I1);
-
-        return make_naive_tensor_descriptor_packed(make_tuple(K0, K1));
-    }
-
-    __host__ __device__ static constexpr auto MakeCK1NH2W2ThreadDescriptor()
-    {
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<KPerThread>{}, I1, Number<HoPerThread>{}, Number<WoPerThread>{}));
-        return c_k1_n_h2_w2_thread_gemm_desc;
-    }
-
-    // using CThreadDesc_K1_N_H2_W2 = decltype(MakeCK1NH2W2ThreadDescriptor());
-
-    __host__ __device__ static constexpr auto GetBlockWiseGemm()
-    {
-        constexpr auto max_lds_align = Number<ABlockTransferDstScalarPerVector_E2>{};
-
-        constexpr auto a_e1_k1_e2_block_gemm_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E1PerBlock>{}, Number<KPerBlock>{}, Number<E2>{}), max_lds_align);
-
-        constexpr auto b_e1_n_h_w_e2_block_gemm_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(Number<E1PerBlock>{},
-                                                           I1,
-                                                           Number<HoPerBlock>{},
-                                                           Number<WoPerBlock>{},
-                                                           Number<E2>{}));
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v3<BlockSize,
-                                                 FloatAB,
-                                                 FloatAB,
-                                                 FloatAcc,
-                                                 decltype(a_e1_k1_e2_block_gemm_desc),
-                                                 decltype(b_e1_n_h_w_e2_block_gemm_desc),
-                                                 decltype(c_k1_n_h2_w2_thread_gemm_desc),
-                                                 EPerThread,
-                                                 K2>{};
-
-        return blockwise_gemm;
-    }
-
-    __device__ static constexpr auto GetCThreadIndex()
-    {
-        auto blockwise_gemm = GetBlockWiseGemm();
-        auto c_thread_mtx_index =
-            blockwise_gemm.GetBeginOfCThreadDesc_K_N_Ho_Wo(get_thread_local_1d_id());
-
-        return c_thread_mtx_index;
-    };
-
-    __device__ static constexpr auto GetCBlockIndex(
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor)
-    {
-        const auto c_k_n_h_w_block_cluster_idx =
-            cblockid_to_k_n_h_w_block_cluster_adaptor.CalculateBottomIndex(
-                make_multi_index(get_block_1d_id()));
-        return c_k_n_h_w_block_cluster_idx;
-    }
-
-    template <typename BiasGlobalBuff,
-              typename CThreadBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename BiasGridDesc_K0_K1,
-              typename CThreadDesc_K1_N_H2_W2>
-    __device__ static void BiasOp(BiasGlobalBuff& bias_global_buf,
-                                  CThreadBuff& c_thread_buf,
-                                  const CBlockIndex& c_block_idx,
-                                  const CThreadIndex& c_thread_idx,
-                                  const BiasGridDesc_K0_K1& bias_k0_k1_grid_desc,
-                                  const CThreadDesc_K1_N_H2_W2&)
-
-    {
-        const index_t k_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-
-        const auto k_thread_id = c_thread_idx[I0];
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        constexpr auto bias_k0_k1_thread_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1, Number<KPerThread>{}));
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatC,
-                     bias_k0_k1_thread_desc.GetElementSpaceSize(),
-                     true>
-            bias_thread_buf;
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        auto bias_threadwise_transfer =
-            ThreadwiseTensorSliceTransfer_v2<FloatC,
-                                             FloatC,
-                                             decltype(bias_k0_k1_grid_desc),
-                                             decltype(bias_k0_k1_thread_desc),
-                                             Sequence<I1, Number<KPerThread>{}>,
-                                             Sequence<0, 1>,
-                                             1,
-                                             CThreadTransferDstScalarPerVector,
-                                             false,
-                                             true>(
-                bias_k0_k1_grid_desc, make_multi_index(k_block_work_id, k_thread_data_on_global));
-
-        constexpr auto bias_k0_k1_global_tensor_step_hacks = make_tuple(
-            make_tuple(Sequence<0>{}, Sequence<0>{}), make_tuple(Sequence<0>{}, Sequence<0>{}));
-
-        bias_threadwise_transfer.Run(bias_k0_k1_grid_desc,
-                                     bias_global_buf,
-                                     bias_k0_k1_thread_desc,
-                                     make_tuple(I0, I0),
-                                     bias_thread_buf,
-                                     bias_k0_k1_global_tensor_step_hacks);
-
-        static_for<0, KPerThread, 1>{}([&](auto ki) {
-            static_for<0, HoPerThread, 1>{}([&](auto hi) {
-                static_for<0, WoPerThread, 1>{}([&](auto wi) {
-                    constexpr index_t c_offset =
-                        c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(make_tuple(ki, 0, hi, wi));
-                    c_thread_buf(Number<c_offset>{}) =
-                        c_thread_buf[Number<c_offset>{}] + bias_thread_buf[ki];
-                });
-            });
-        });
-    }
-
-    template <typename CThreadBuff, typename CThreadDesc_K1_N_H2_W2, ActivTypeEnum activ_type_>
-    __device__ static void Activation(CThreadBuff& c_thread_buf,
-                                      const CThreadDesc_K1_N_H2_W2&,
-                                      integral_constant<ActivTypeEnum, activ_type_>)
-    {
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        static_for<0, c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(), 1>{}([&](auto i) {
-            if constexpr(activ_type_ == 1)
-            {
-                c_thread_buf(i) = c_thread_buf[i] >= 0 ? c_thread_buf[i] : alpha * c_thread_buf[i];
-            }
-            else if constexpr(activ_type_ == 2)
-            {
-                FloatAcc x = 1.0 + exp(-c_thread_buf[i]);
-
-                asm volatile("\n \
-                        v_rcp_f32 %0, %1 \n"
-                             : "=v"(x)
-                             : "0"(x));
-
-                c_thread_buf(i) = x;
-            }
-        });
-    }
-
-    template <typename CThreadBuff,
-              typename CGlobalBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2>
-    __device__ static void
-    WriteOut(const CThreadBuff& c_thread_buf,
-             CGlobalBuff& c_global_buf,
-             const CBlockIndex& c_block_idx,
-             const CThreadIndex& c_thread_idx,
-             const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc)
-    {
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        const auto k_thread_id  = c_thread_idx[I0];
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        // hack to control index calculation when iterating over c_k_n_h0_h1_h2_w0_w1_w2_global
-        // tensor
-        constexpr auto c_k_n_h0_h1_h2_w0_w1_w2_global_tensor_step_hacks = CGlobalStepHacks{};
-
-        constexpr auto c_k0_k1_n_h0_h1_h2_w0_w1_w2_thread_copy_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<KPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThread>{}));
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        ThreadwiseTensorSliceTransfer_v1r3<
-            FloatAcc,
-            FloatC,
-            decltype(c_k0_k1_n_h0_h1_h2_w0_w1_w2_thread_copy_desc),
-            decltype(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc),
-            Sequence<I1, KPerThread, I1, I1, I1, HoPerThread, I1, I1, WoPerThread>,
-            CThreadTransferSrcDstAccessOrder,
-            CThreadTransferSrcDstVectorDim,
-            CThreadTransferDstScalarPerVector,
-            CGlobalMemoryDataOperation,
-            1,
-            true>(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                  make_multi_index(k_block_work_id,
-                                   k_thread_data_on_global,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0))
-            .Run(c_k0_k1_n_h0_h1_h2_w0_w1_w2_thread_copy_desc,
-                 make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                 c_thread_buf,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-                 c_global_buf,
-                 c_k_n_h0_h1_h2_w0_w1_w2_global_tensor_step_hacks);
-    }
-
-    template <typename CThreadBuff,
-              typename DGlobalBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename CThreadDesc_K1_N_H2_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx>
-    __device__ static void
-    MaxPool(const CThreadBuff& c_thread_buf,
-            DGlobalBuff& d_global_buf,
-            const CBlockIndex& c_block_idx,
-            const CThreadIndex& c_thread_idx,
-            const CThreadDesc_K1_N_H2_W2&,
-            const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc)
-    {
-
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        const auto k_thread_id  = c_thread_idx[I0];
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        static_assert(HoPerThread % 2 == 0 && WoPerThread % 2 == 0, "");
-
-        constexpr auto HoPerThread_2 = HoPerThread / 2;
-        constexpr auto WoPerThread_2 = WoPerThread / 2;
-
-        constexpr auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<KPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThread_2>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThread_2>{}));
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatC,
-                     d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.GetElementSpaceSize(),
-                     true>
-            d_thread_buf;
-
-        static_for<0, KPerThread, 1>{}([&](auto ki) {
-            static_for<0, HoPerThread_2, 1>{}([&](auto hi) {
-                static_for<0, WoPerThread_2, 1>{}([&](auto wi) {
-                    constexpr index_t d_offset =
-                        d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.CalculateOffset(
-                            make_tuple(0, ki, 0, 0, 0, hi, 0, 0, wi));
-
-                    constexpr index_t c_offset_0 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2, wi * 2));
-                    constexpr index_t c_offset_1 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2, wi * 2 + 1));
-                    constexpr index_t c_offset_2 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2 + 1, wi * 2));
-                    constexpr index_t c_offset_3 = c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                        make_tuple(ki, 0, hi * 2 + 1, wi * 2 + 1));
-
-                    d_thread_buf(Number<d_offset>{}) = c_thread_buf[Number<c_offset_0>{}];
-                    d_thread_buf(Number<d_offset>{}) =
-                        fmaxf(c_thread_buf[Number<c_offset_1>{}], d_thread_buf(Number<d_offset>{}));
-                    d_thread_buf(Number<d_offset>{}) =
-                        fmaxf(c_thread_buf[Number<c_offset_2>{}], d_thread_buf(Number<d_offset>{}));
-                    d_thread_buf(Number<d_offset>{}) =
-                        fmax(c_thread_buf[Number<c_offset_3>{}], d_thread_buf(Number<d_offset>{}));
-                });
-            });
-        });
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        constexpr auto d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks = DGlobalStepHacks{};
-
-        ThreadwiseTensorSliceTransfer_v1r3<
-            FloatC,
-            FloatC,
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc),
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc),
-            Sequence<I1, KPerThread, I1, I1, I1, HoPerThread_2, I1, I1, WoPerThread_2>,
-            CThreadTransferSrcDstAccessOrder,
-            CThreadTransferSrcDstVectorDim,
-            CThreadTransferDstScalarPerVector,
-            InMemoryDataOperationEnum::Set,
-            1,
-            true>(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                  make_multi_index(k_block_work_id,
-                                   k_thread_data_on_global,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0))
-            .Run(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc,
-                 make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                 d_thread_buf,
-                 d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                 d_global_buf,
-                 d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks);
-    }
-
-    template <typename CThreadBuff,
-              typename DGlobalBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename CThreadDesc_K1_N_H2_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx>
-    __device__ static void
-    ResizeAdd(const CThreadBuff& c_thread_buf,
-              DGlobalBuff& d_global_buf,
-              const CBlockIndex& c_block_idx,
-              const CThreadIndex& c_thread_idx,
-              const CThreadDesc_K1_N_H2_W2&,
-              const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc)
-    {
-
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        const auto k_thread_id  = c_thread_idx[I0];
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        constexpr auto HoPerThreadx2 = HoPerThread * 2;
-        constexpr auto WoPerThreadx2 = WoPerThread * 2;
-
-        constexpr auto d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<KPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThreadx2>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThreadx2>{}));
-
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatC,
-                     d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.GetElementSpaceSize(),
-                     true>
-            d_thread_buf;
-
-        static_for<0, KPerThread, 1>{}([&](auto k_i) {
-            static_for<0, HoPerThreadx2, 1>{}([&](auto h_i) {
-                static_for<0, WoPerThreadx2, 1>{}([&](auto w_i) {
-                    d_thread_buf(Number<d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc.CalculateOffset(
-                                     make_tuple(0, k_i, 0, 0, 0, h_i, 0, 0, w_i))>{}) =
-                        c_thread_buf[Number<c_k1_n_h2_w2_thread_gemm_desc.CalculateOffset(
-                            make_tuple(k_i, 0, h_i / 2, w_i / 2))>{}];
-                });
-            });
-        });
-
-        // hack to control index calculation when iterating over d_k_n_ho_wo_global tensor
-        constexpr auto d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks = DGlobalStepHacks{};
-
-        const index_t k_thread_data_on_global = k_thread_id * KPerThread;
-
-        ThreadwiseTensorSliceTransfer_v1r3<
-            FloatC,
-            FloatC,
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc),
-            decltype(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc),
-            Sequence<I1, KPerThread, I1, I1, I1, HoPerThreadx2, I1, I1, WoPerThreadx2>,
-            CThreadTransferSrcDstAccessOrder,
-            CThreadTransferSrcDstVectorDim,
-            CThreadTransferDstScalarPerVector,
-            InMemoryDataOperationEnum::Add,
-            1,
-            true>(d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                  make_multi_index(k_block_work_id,
-                                   k_thread_data_on_global,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0))
-            .Run(d_k0_k1_n_h0_h1_hx_w0_w1_wx_thread_desc,
-                 make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                 d_thread_buf,
-                 d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-                 d_global_buf,
-                 d_k_n_h0_h1_hx_w0_w1_wx_global_tensor_step_hacks);
-    }
-
-    template <typename AGlobalBuff,
-              typename BGlobalBuff,
-              typename CThreadBuff,
-              typename CBlockIndex,
-              typename CThreadIndex,
-              typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CThreadDesc_K1_N_H2_W2,
-              bool HasMainE0BlockLoop>
-    __device__ static void
-    GemmOp(const AGlobalBuff& a_global_buf,
-           const BGlobalBuff& b_global_buf,
-           CThreadBuff& c_thread_buf,
-           FloatAB* __restrict__ p_shared_block,
-           const CBlockIndex& c_block_idx,
-           const CThreadIndex& c_thread_idx,
-           const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-           const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-           const CThreadDesc_K1_N_H2_W2&,
-           integral_constant<bool, HasMainE0BlockLoop>)
-    {
-        constexpr auto HasMainE1BlockLoop       = CalculateHasMainE1BlockLoop();
-        constexpr auto HasDoubleTailE1BlockLoop = CalculateHasDoubleTailE1BlockLoop();
-
-        // const auto c_k_n_h_w_block_cluster_idx =
-        // GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-        // cblockid_to_k_n_h_w_block_cluster_adaptor.CalculateBottomIndex(
-        // make_multi_index(get_block_1d_id()));
-
-        const index_t k_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I0]);
-        const index_t n_block_work_id  = __builtin_amdgcn_readfirstlane(c_block_idx[I1]);
-        const index_t ho_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I2]);
-        const index_t wo_block_work_id = __builtin_amdgcn_readfirstlane(c_block_idx[I3]);
-
-        constexpr auto max_lds_align = Number<ABlockTransferDstScalarPerVector_E2>{};
-
-        constexpr auto a_e1_k1_e2_block_gemm_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<E1PerBlock>{}, Number<KPerBlock>{}, Number<E2>{}), max_lds_align);
-
-        constexpr auto b_e1_n_h_w_e2_block_gemm_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(Number<E1PerBlock>{},
-                                                           I1,
-                                                           Number<HoPerBlock>{},
-                                                           Number<WoPerBlock>{},
-                                                           Number<E2>{}));
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = CThreadDesc_K1_N_H2_W2{};
-
-        auto blockwise_gemm =
-            BlockwiseGemmDlops_km_kn_m0m1n0n1_v3<BlockSize,
-                                                 FloatAB,
-                                                 FloatAB,
-                                                 FloatAcc,
-                                                 decltype(a_e1_k1_e2_block_gemm_desc),
-                                                 decltype(b_e1_n_h_w_e2_block_gemm_desc),
-                                                 decltype(c_k1_n_h2_w2_thread_gemm_desc),
-                                                 EPerThread,
-                                                 K2>{};
-        // blockwise_gemm.GetBeginOfCThreadDesc_K_N_Ho_Wo(get_thread_local_1d_id());
-
-        const auto ho_thread_id = c_thread_idx[I2];
-        const auto wo_thread_id = c_thread_idx[I3];
-
-        constexpr auto a_e0_e1_k0_k1_e2_block_copy_desc = make_naive_tensor_descriptor_aligned(
-            make_tuple(Number<I1>{}, Number<E1>{}, I1, Number<KPerBlock>{}, Number<E2>{}),
-            max_lds_align);
-
-        // A matrix blockwise copy
-        auto a_blockwise_copy =
-            BlockwiseTensorSliceTransfer_v4<BlockSize,
-                                            InMemoryDataOperationEnum::Set,
-                                            Sequence<I1, E1, I1, KPerBlock, E2>,
-                                            ABlockTransferThreadSliceLengths_E0_E1_K0_K1_E2,
-                                            ABlockTransferThreadClusterLengths_E0_E1_K0_K1_E2,
-                                            ABlockTransferThreadClusterArrangeOrder,
-                                            FloatAB,
-                                            FloatAB,
-                                            decltype(a_e0_e1_k0_k1_e2_grid_desc),
-                                            decltype(a_e0_e1_k0_k1_e2_block_copy_desc),
-                                            ABlockTransferSrcAccessOrder,
-                                            Sequence<0, 1, 2, 3, 4>,
-                                            ABlockTransferSrcVectorDim,
-                                            4,
-                                            ABlockTransferSrcScalarPerVector,
-                                            ABlockTransferDstScalarPerVector_E2,
-                                            1,
-                                            1,
-                                            AThreadTransferSrcResetCoordinateAfterRun,
-                                            false>(a_e0_e1_k0_k1_e2_grid_desc,
-                                                   make_multi_index(0, 0, k_block_work_id, 0, 0),
-                                                   a_e0_e1_k0_k1_e2_block_copy_desc,
-                                                   make_multi_index(0, 0, 0, 0, 0));
-
-        constexpr auto a_block_slice_copy_step = make_multi_index(I1, 0, 0, 0, 0);
-
-        constexpr auto b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(I1,
-                                                           Number<E1PerBlock>{},
-                                                           I1,
-                                                           I1,
-                                                           I1,
-                                                           Number<HoPerThread>{},
-                                                           I1,
-                                                           I1,
-                                                           Number<WoPerThread>{},
-                                                           Number<E2>{}));
-
-        auto b_threadwise_transfer = ThreadwiseTensorSliceTransfer_v2<
-            FloatAB,
-            FloatAB,
-            decltype(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc),
-            decltype(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc),
-            Sequence<I1, E1PerBlock, I1, I1, I1, HoPerThread, I1, I1, WoPerThread, E2>,
-            BBlockTransferSrcAccessOrder,
-            BBlockTransferSrcVectorDim,
-            BBlockTransferSrcScalarPerVector,
-            BThreadTransferSrcResetCoordinateAfterRun,
-            true>(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                  make_multi_index(0,
-                                   0,
-                                   n_block_work_id,
-                                   ho_block_work_id,
-                                   ho_thread_id,
-                                   0,
-                                   wo_block_work_id,
-                                   wo_thread_id,
-                                   0,
-                                   0));
-
-        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            p_shared_block, a_e0_e1_k0_k1_e2_block_copy_desc.GetElementSpaceSize());
-
-        //// register allocation for output
-        // StaticBuffer<AddressSpaceEnum::Vgpr,
-        // FloatAcc,
-        // c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-        // true>
-        // c_thread_buf;
-
-        // initialize output thread tensor
-        ThreadwiseTensorSliceSet_v1<FloatAcc,
-                                    decltype(c_k1_n_h2_w2_thread_gemm_desc),
-                                    Sequence<KPerThread, I1, HoPerThread, WoPerThread>>{}
-            .Run(c_k1_n_h2_w2_thread_gemm_desc,
-                 make_tuple(I0, I0, I0, I0),
-                 c_thread_buf,
-                 FloatAcc{0});
-
-        constexpr auto b_thread_slice_copy_step =
-            make_multi_index(0, E1PerBlock, 0, 0, 0, 0, 0, 0, 0, 0);
-
-        // hack to control index calculation when iterating over A and B matrix for threadwise copy
-        constexpr auto a_e0_e1_k_e2_global_step_hacks                   = AGlobalStepHacks{};
-        constexpr auto b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks = BGlobalStepHacks{};
-
-        // double regsiter buffer for b
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAB,
-                     b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc.GetElementSpaceSize(),
-                     true>
-            b_thread_even_buf, b_thread_odd_buf;
-
-        if constexpr(HasMainE0BlockLoop)
-        {
-            const auto E0 = b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetLength(I0);
-
-            index_t e0_block_data_begin = 0;
-
-            do
-            {
-                // LDS double buffer: preload data
-                {
-                    a_blockwise_copy.RunRead(
-                        a_e0_e1_k0_k1_e2_grid_desc, a_global_buf, a_e0_e1_k_e2_global_step_hacks);
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_even_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    a_blockwise_copy.RunWrite(a_e0_e1_k0_k1_e2_block_copy_desc, a_block_buf);
-                }
-
-                __syncthreads();
-
-                if constexpr(HasMainE1BlockLoop)
-                {
-                    index_t e1_block_data_begin = 0;
-
-                    // LDS double buffer: main body
-                    // use Do-While loop instead of For loop to simplify control flow
-                    do
-                    {
-                        // even iteration
-                        b_threadwise_transfer.MoveSrcSliceWindow(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_thread_slice_copy_step,
-                            BGlobalMoveSliceWindowStepHacks{});
-
-                        b_threadwise_transfer.Run(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_global_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                            make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                            b_thread_odd_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                        // LDS double buffer: GEMM on current data
-                        blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                        blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                        b_threadwise_transfer.MoveSrcSliceWindow(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_thread_slice_copy_step,
-                            BGlobalMoveSliceWindowStepHacks{});
-
-                        b_threadwise_transfer.Run(
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                            b_global_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                            make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                            b_thread_even_buf,
-                            b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                        // LDS double buffer: GEMM on current data
-                        blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-
-                        blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                        e1_block_data_begin += 2 * E1PerBlock;
-
-                    } while(e1_block_data_begin < E1 - 2 * E1PerBlock);
-                }
-
-                // LDS double buffer: tail
-                if constexpr(HasDoubleTailE1BlockLoop) // if has 2 iteration left
-                {
-                    b_threadwise_transfer.MoveSrcSliceWindow(
-                        b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                        b_thread_slice_copy_step,
-                        BGlobalMoveSliceWindowStepHacks{});
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_odd_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    // LDS double buffer: GEMM on 2nd-last data
-                    blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                    blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                    // LDS double buffer: GEMM on last data
-                    blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-                }
-                else // if has 1 iteration left
-                {
-                    // LDS double buffer: GEMM on last data
-                    blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-                }
-
-                a_blockwise_copy.MoveSrcSliceWindow(a_e0_e1_k0_k1_e2_grid_desc,
-                                                    a_block_slice_copy_step,
-                                                    AGlobalMoveSliceWindowStepHacks{});
-
-                blockwise_gemm.MoveABlockSliceWindow(make_tuple(-(E1 - E1PerBlock), 0, 0));
-
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                                         b_thread_slice_copy_step,
-                                                         BGlobalMoveSliceWindowStepHacks{});
-
-                e0_block_data_begin += 1;
-
-            } while(e0_block_data_begin < E0);
-        }
-        else
-        {
-            // LDS double buffer: preload data
-            {
-                a_blockwise_copy.RunRead(
-                    a_e0_e1_k0_k1_e2_grid_desc, a_global_buf, a_e0_e1_k_e2_global_step_hacks);
-
-                b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                          b_global_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                          make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                          b_thread_even_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                a_blockwise_copy.RunWrite(a_e0_e1_k0_k1_e2_block_copy_desc, a_block_buf);
-            }
-
-            __syncthreads();
-
-            if constexpr(HasMainE1BlockLoop)
-            {
-                index_t e1_block_data_begin = 0;
-
-                // LDS double buffer: main body
-                // use Do-While loop instead of For loop to simplify control flow
-                do
-                {
-                    // even iteration
-                    b_threadwise_transfer.MoveSrcSliceWindow(
-                        b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                        b_thread_slice_copy_step,
-                        BGlobalMoveSliceWindowStepHacks{});
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_odd_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    // LDS double buffer: GEMM on current data
-                    blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                    blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                    b_threadwise_transfer.MoveSrcSliceWindow(
-                        b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                        b_thread_slice_copy_step,
-                        BGlobalMoveSliceWindowStepHacks{});
-
-                    b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                              b_global_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                              make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                              b_thread_even_buf,
-                                              b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                    // LDS double buffer: GEMM on current data
-                    blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-
-                    blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                    e1_block_data_begin += 2 * E1PerBlock;
-
-                } while(e1_block_data_begin < E1 - 2 * E1PerBlock);
-            }
-
-            // LDS double buffer: tail
-            if constexpr(HasDoubleTailE1BlockLoop) // if has 2 iteration left
-            {
-                b_threadwise_transfer.MoveSrcSliceWindow(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                                         b_thread_slice_copy_step,
-                                                         BGlobalMoveSliceWindowStepHacks{});
-
-                b_threadwise_transfer.Run(b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-                                          b_global_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_thread_copy_desc,
-                                          make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                          b_thread_odd_buf,
-                                          b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_global_step_hacks);
-
-                // LDS double buffer: GEMM on 2nd-last data
-                blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-
-                blockwise_gemm.MoveABlockSliceWindow(make_tuple(E1PerBlock, 0, 0));
-
-                // LDS double buffer: GEMM on last data
-                blockwise_gemm.Run(a_block_buf, b_thread_odd_buf, c_thread_buf);
-            }
-            else // if has 1 iteration left
-            {
-                // LDS double buffer: GEMM on last data
-                blockwise_gemm.Run(a_block_buf, b_thread_even_buf, c_thread_buf);
-            }
-        }
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop>
-    __device__ static void
-    Conv(const FloatAB* __restrict__ p_a_global,
-         const FloatAB* __restrict__ p_b_global,
-         const FloatC* __restrict__ p_bias_global,
-         FloatC* __restrict__ p_c_global,
-         FloatC* __restrict__ p_d_global,
-         FloatAB* __restrict__ p_shared_block,
-         const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-         const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-         const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-         const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-         const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-         integral_constant<bool, HasMainE0BlockLoop>)
-    {
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetElementSpaceSize());
-        auto d_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_d_global, d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Output
-        WriteOut(c_thread_buf,
-                 c_global_buf,
-                 c_k_n_h_w_block_cluster_idx,
-                 c_thread_mtx_index,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop,
-              ActivTypeEnum ActivType>
-    __device__ static void ConvBiasActiv(
-        const FloatAB* __restrict__ p_a_global,
-        const FloatAB* __restrict__ p_b_global,
-        const FloatC* __restrict__ p_bias_global,
-        FloatC* __restrict__ p_c_global,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-        const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-        integral_constant<bool, HasMainE0BlockLoop>,
-        integral_constant<ActivTypeEnum, ActivType>)
-    {
-        static constexpr auto activ_type = integral_constant<ActivTypeEnum, ActivType>{};
-
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Bias
-        BiasOp(bias_global_buf,
-               c_thread_buf,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               bias_k0_k1_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc);
-
-        // Activ
-        Activation(c_thread_buf, c_k1_n_h2_w2_thread_gemm_desc, activ_type);
-
-        // Output
-        WriteOut(c_thread_buf,
-                 c_global_buf,
-                 c_k_n_h_w_block_cluster_idx,
-                 c_thread_mtx_index,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop,
-              ActivTypeEnum ActivType>
-    __device__ static void ConvBiasActivMaxpool(
-        const FloatAB* __restrict__ p_a_global,
-        const FloatAB* __restrict__ p_b_global,
-        const FloatC* __restrict__ p_bias_global,
-        FloatC* __restrict__ p_c_global,
-        FloatC* __restrict__ p_d_global,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-        const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-        const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-        integral_constant<bool, HasMainE0BlockLoop>,
-        integral_constant<ActivTypeEnum, ActivType>)
-    {
-        static constexpr auto activ_type = integral_constant<ActivTypeEnum, ActivType>{};
-
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto c_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_c_global, c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc.GetElementSpaceSize());
-        auto d_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_d_global, d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Bias
-        BiasOp(bias_global_buf,
-               c_thread_buf,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               bias_k0_k1_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc);
-
-        // Activ
-        Activation(c_thread_buf, c_k1_n_h2_w2_thread_gemm_desc, activ_type);
-
-        // Output
-        WriteOut(c_thread_buf,
-                 c_global_buf,
-                 c_k_n_h_w_block_cluster_idx,
-                 c_thread_mtx_index,
-                 c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        // MaxPool
-        MaxPool(c_thread_buf,
-                d_global_buf,
-                c_k_n_h_w_block_cluster_idx,
-                c_thread_mtx_index,
-                c_k1_n_h2_w2_thread_gemm_desc,
-                d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc);
-    }
-
-    template <typename AGridDesc_E0_E1_K0_K1_E2,
-              typename BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2,
-              typename CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2,
-              typename DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx,
-              typename CBlockIdToBlockClusterAdaptor_K_N_H_W,
-              bool HasMainE0BlockLoop,
-              ActivTypeEnum ActivType>
-    __device__ static void ConvBiasActivResizeAdd(
-        const FloatAB* __restrict__ p_a_global,
-        const FloatAB* __restrict__ p_b_global,
-        const FloatC* __restrict__ p_bias_global,
-        FloatC* __restrict__ p_d_global,
-        FloatAB* __restrict__ p_shared_block,
-        const AGridDesc_E0_E1_K0_K1_E2& a_e0_e1_k0_k1_e2_grid_desc,
-        const BGridDesc_E0_E1_N_H0_H1_H2_W0_W1_W2_E2& b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-        const CGridDesc_K0_K1_N_H0_H1_H2_W0_W1_W2& c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc,
-        const DGridDesc_K0_K1_N_H0_H1_Hx_W0_W1_Wx& d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc,
-        const CBlockIdToBlockClusterAdaptor_K_N_H_W& cblockid_to_k_n_h_w_block_cluster_adaptor,
-        integral_constant<bool, HasMainE0BlockLoop>,
-        integral_constant<ActivTypeEnum, ActivType>)
-    {
-        static constexpr auto activ_type = integral_constant<ActivTypeEnum, ActivType>{};
-
-        const auto bias_k0_k1_grid_desc =
-            MakeBiasK0K1GridDescriptor(c_k0_k1_n_h0_h1_h2_w0_w1_w2_grid_desc);
-
-        const auto a_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_a_global, a_e0_e1_k0_k1_e2_grid_desc.GetElementSpaceSize());
-        const auto b_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_b_global, b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc.GetElementSpaceSize());
-        auto d_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_d_global, d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc.GetElementSpaceSize());
-        auto bias_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_bias_global, bias_k0_k1_grid_desc.GetElementSpaceSize());
-
-        constexpr auto c_k1_n_h2_w2_thread_gemm_desc = MakeCK1NH2W2ThreadDescriptor();
-
-        // register allocation for output
-        StaticBuffer<AddressSpaceEnum::Vgpr,
-                     FloatAcc,
-                     c_k1_n_h2_w2_thread_gemm_desc.GetElementSpaceSize(),
-                     true>
-            c_thread_buf;
-
-        const auto c_k_n_h_w_block_cluster_idx =
-            GetCBlockIndex(cblockid_to_k_n_h_w_block_cluster_adaptor);
-
-        const auto c_thread_mtx_index = GetCThreadIndex();
-
-        // GemmOp
-        GemmOp(a_global_buf,
-               b_global_buf,
-               c_thread_buf,
-               p_shared_block,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               a_e0_e1_k0_k1_e2_grid_desc,
-               b_e0_e1_n_h0_h1_h2_w0_w1_w2_e2_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc,
-               integral_constant<bool, HasMainE0BlockLoop>{});
-
-        // Bias
-        BiasOp(bias_global_buf,
-               c_thread_buf,
-               c_k_n_h_w_block_cluster_idx,
-               c_thread_mtx_index,
-               bias_k0_k1_grid_desc,
-               c_k1_n_h2_w2_thread_gemm_desc);
-
-        // Activ
-        Activation(c_thread_buf, c_k1_n_h2_w2_thread_gemm_desc, activ_type);
-
-        // Resize_Add
-        ResizeAdd(c_thread_buf,
-                  d_global_buf,
-                  c_k_n_h_w_block_cluster_idx,
-                  c_thread_mtx_index,
-                  c_k1_n_h2_w2_thread_gemm_desc,
-                  d_k0_k1_n_h0_h1_hx_w0_w1_wx_grid_desc);
-    }
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp
deleted file mode 100644
index 6a73466efa..0000000000
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r3.hpp
+++ /dev/null
@@ -1,886 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_THREADWISE_TENSOR_SLICE_TRANSFER_V3R3_HPP
-#define CK_THREADWISE_TENSOR_SLICE_TRANSFER_V3R3_HPP
-
-#include "common_header.hpp"
-#include "tensor_descriptor.hpp"
-#include "tensor_descriptor_helper.hpp"
-#include "static_tensor.hpp"
-
-namespace ck {
-
-namespace detail {
-// TODO: How to fix this? It uses an struct instead of lambda because lambda
-// doesn't have constructor
-template <index_t SrcVectorDim,
-          index_t SrcScalarPerVector,
-          index_t DstVectorDim,
-          index_t DstScalarPerVector>
-struct lambda_scalar_per_access_for_src_and_dst
-{
-    __host__ __device__ constexpr auto operator()(index_t i) const
-    {
-        if(i == SrcVectorDim && i == DstVectorDim)
-        {
-            return math::lcm(SrcScalarPerVector, DstScalarPerVector);
-        }
-        else if(i == SrcVectorDim)
-        {
-            return SrcScalarPerVector;
-        }
-        else if(i == DstVectorDim)
-        {
-            return DstScalarPerVector;
-        }
-        else
-        {
-            return 1;
-        }
-    }
-};
-
-} // namespace detail
-
-// Assume:
-//   1. src_desc and dst_desc are not known at compile-time
-//   2. SrcBuffer and DstBuffer are DynamicBuffer
-//   3. src_slice_origin and dst_slice_origin are not known at compile-time,
-//   4. Use thread buffer
-template <typename SliceLengths,
-          typename SrcElementwiseOperation,
-          typename DstElementwiseOperation,
-          InMemoryDataOperationEnum DstInMemOp,
-          typename SrcData,
-          typename DstData,
-          typename SrcDesc,
-          typename DstDesc,
-          typename Dst0Desc,
-          typename Dst1Desc,
-          typename SrcDimAccessOrder,
-          typename DstDimAccessOrder,
-          index_t SrcVectorDim,
-          index_t DstVectorDim,
-          index_t SrcScalarPerVector,
-          index_t DstScalarPerVector,
-          index_t SrcScalarStrideInVector,
-          index_t DstScalarStrideInVector,
-          bool SrcResetCoordinateAfterRun, // control whether to move back src coordinate after each
-                                           // RunRead(),  will be fused with MoveSrcSliceWindow to
-                                           // save addr computation
-          bool DstResetCoordinateAfterRun> // control whether to move back dst coordinate after each
-                                           // RunWrite(),  will be fused with MoveDstSliceWindow to
-                                           // save addr computation
-struct ThreadwiseTensorSliceTransfer_v3r3
-{
-    static constexpr index_t nDim = SliceLengths::Size();
-    using Index                   = MultiIndex<nDim>;
-
-    using SrcCoord  = decltype(make_tensor_coordinate(SrcDesc{}, Index{}));
-    using DstCoord  = decltype(make_tensor_coordinate(DstDesc{}, Index{}));
-    using Dst0Coord = decltype(make_tensor_coordinate(Dst0Desc{}, Index{}));
-    using Dst1Coord = decltype(make_tensor_coordinate(Dst1Desc{}, Index{}));
-
-    using SrcCoordStep  = decltype(make_tensor_coordinate_step(SrcDesc{}, Index{}));
-    using DstCoordStep  = decltype(make_tensor_coordinate_step(DstDesc{}, Index{}));
-    using Dst0CoordStep = decltype(make_tensor_coordinate_step(Dst0Desc{}, Index{}));
-    using Dst1CoordStep = decltype(make_tensor_coordinate_step(Dst1Desc{}, Index{}));
-
-    __device__ constexpr ThreadwiseTensorSliceTransfer_v3r3(
-        const SrcDesc& src_desc,
-        const Index& src_slice_origin,
-        const SrcElementwiseOperation& src_element_op,
-        const DstDesc& dst_desc,
-        const Dst0Desc& dst0_desc,
-        const Dst1Desc& dst1_desc,
-        const Index& dst_slice_origin,
-        const DstElementwiseOperation& dst_element_op)
-        : src_coord_(make_tensor_coordinate(src_desc, src_slice_origin)),
-          dst_coord_(make_tensor_coordinate(dst_desc, dst_slice_origin)),
-          dst0_coord_(make_tensor_coordinate(dst0_desc, dst_slice_origin)),
-          dst1_coord_(make_tensor_coordinate(dst1_desc, dst_slice_origin)),
-          src_element_op_(src_element_op),
-          dst_element_op_(dst_element_op)
-    {
-    }
-
-    __device__ void SetSrcSliceOrigin(const SrcDesc& src_desc, const Index& src_slice_origin_idx)
-    {
-        src_coord_ = make_tensor_coordinate(src_desc, src_slice_origin_idx);
-    }
-
-    __device__ void SetDstSliceOrigin(const DstDesc& dst_desc,
-                                      const Dst0Desc& dst0_desc,
-                                      const Dst1Desc& dst1_desc,
-                                      const Index& dst_slice_origin_idx)
-    {
-        dst_coord_  = make_tensor_coordinate(dst_desc, dst_slice_origin_idx);
-        dst0_coord_ = make_tensor_coordinate(dst0_desc, dst_slice_origin_idx);
-        dst1_coord_ = make_tensor_coordinate(dst1_desc, dst_slice_origin_idx);
-    }
-
-    template <typename SrcBuffer>
-    __device__ void RunRead(const SrcDesc& src_desc, const SrcBuffer& src_buf)
-    {
-        static_assert(SrcBuffer::GetAddressSpace() == AddressSpaceEnum::Global or
-                          SrcBuffer::GetAddressSpace() == AddressSpaceEnum::Lds,
-                      "wrong!");
-
-        static_assert(
-            is_same<remove_cvref_t<typename SrcBuffer::type>, remove_cvref_t<SrcData>>::value,
-            "wrong! SrcBuffer and SrcData data type are inconsistent");
-
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-
-        // scalar per access on each dim
-        // TODO: don't use lambda_scalar_per_access
-        constexpr auto src_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
-
-        constexpr auto src_dim_access_order = SrcDimAccessOrder{};
-
-        constexpr auto ordered_src_access_lengths =
-            container_reorder_given_new2old(src_access_lengths, src_dim_access_order);
-
-        // make forward steps
-        const auto src_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? src_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(src_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps
-        const auto src_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -src_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(src_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // loop over tensor and copy
-        static_ford<decltype(ordered_src_access_lengths)>{}([&](auto ordered_src_access_idx) {
-            // judge move forward or move backward
-            constexpr auto forward_sweep = [&]() {
-                StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-                forward_sweep_(I0) = true;
-
-                static_for<1, nDim, 1>{}([&](auto i) {
-                    index_t tmp = ordered_src_access_idx[I0];
-
-                    static_for<1, i, 1>{}([&](auto j) {
-                        tmp = tmp * ordered_src_access_lengths[j] + ordered_src_access_idx[j];
-                    });
-
-                    forward_sweep_(i) = tmp % 2 == 0;
-                });
-
-                return forward_sweep_;
-            }();
-
-            // calculate src data index
-            constexpr auto src_data_idx = [&]() {
-                Index ordered_idx;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    ordered_idx(i) = forward_sweep[i] ? ordered_src_access_idx[i]
-                                                      : ordered_src_access_lengths[i] - 1 -
-                                                            ordered_src_access_idx[i];
-                });
-
-                return container_reorder_given_old2new(ordered_idx, src_dim_access_order) *
-                       src_scalar_per_access;
-            }();
-
-            constexpr auto src_data_idx_seq = generate_sequence_v2(
-                [&](auto i) { return Number<src_data_idx[i]>{}; }, Number<src_data_idx.Size()>{});
-
-            const bool is_src_valid =
-                coordinate_has_valid_offset_assuming_visible_index_is_valid(src_desc, src_coord_);
-
-            using src_vector_type = vector_type_maker_t<SrcData, SrcScalarPerVector>;
-            using src_vector_t    = typename src_vector_type::type;
-
-            // copy data from src_buf into src_vector_container
-            auto src_vector_container = src_vector_type{
-                src_buf.template Get<src_vector_t>(src_coord_.GetOffset(), is_src_valid)};
-
-            // apply SrcElementwiseOperation on src_vector_container
-            static_for<0, SrcScalarPerVector, 1>{}([&](auto i) {
-                src_vector_container.template AsType<SrcData>()(i) =
-                    src_element_op_(src_vector_container.template AsType<SrcData>()[i]);
-            });
-
-            // copy data from src_vector_container into src_thread_scratch_
-            src_thread_scratch_.template SetAsType<src_vector_t>(
-                src_data_idx_seq, src_vector_container.template AsType<src_vector_t>()[I0]);
-
-            constexpr auto move_on_dim = [&]() constexpr
-            {
-                StaticallyIndexedArray<bool, nDim> move_on_dim_;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    move_on_dim_(i) = ordered_src_access_idx[i] < ordered_src_access_lengths[i] - 1;
-
-                    static_for<i + 1, nDim, 1>{}([&](auto j) {
-                        move_on_dim_(i) &=
-                            ordered_src_access_idx[j] == ordered_src_access_lengths[j] - 1;
-                    });
-                });
-
-                return move_on_dim_;
-            }
-            ();
-
-            // move src coord
-            static_for<0, nDim, 1>{}([&](auto i) {
-                if constexpr(move_on_dim[i])
-                {
-                    if constexpr(forward_sweep[i])
-                    {
-                        move_tensor_coordinate(
-                            src_desc, src_coord_, src_forward_steps[src_dim_access_order[i]]);
-                    }
-                    else
-                    {
-                        move_tensor_coordinate(
-                            src_desc, src_coord_, src_backward_steps[src_dim_access_order[i]]);
-                    }
-                }
-            });
-        });
-
-        // move src coordinate back to slice origin (or not)
-        if constexpr(SrcResetCoordinateAfterRun)
-        {
-            const auto src_reset_step =
-                make_tensor_coordinate_step(src_desc, GetSrcCoordinateResetStep());
-
-            move_tensor_coordinate(src_desc, src_coord_, src_reset_step);
-        }
-    }
-
-    __device__ void TransferDataFromSrcThreadScratchToDstThreadScratch()
-    {
-#if !CK_EXPERIMENTAL_USE_IN_REGISTER_SUB_DWORD_TRANSPOSE
-        static_ford<SliceLengths>{}([&](auto idx) {
-            // convert from SrcData to DstData here
-            dst_thread_scratch_(idx) = type_convert<DstData>(src_thread_scratch_[idx]);
-        });
-#else
-        // sub-dword transpose between src_thread_scratch_ and dst_thread_scratch_
-        // TODO make this logic more generic for more sub-dword datatype
-        if constexpr(SrcVectorDim != DstVectorDim &&
-                     is_same<half_t, remove_cvref_t<SrcData>>::value &&
-                     is_same<half_t, remove_cvref_t<DstData>>::value &&
-                     SrcScalarPerVector % 2 == 0 && DstScalarPerVector % 2 == 0)
-        {
-            // each transpose does
-            // DstScalarPerVector # of src vectors in src_thread_scratch_
-            // SrcScalarPerVector # of dst vectors in dst_thread_scratch_
-            constexpr index_t num_src_vector = Number<DstScalarPerVector>{};
-            constexpr index_t num_dst_vector = Number<SrcScalarPerVector>{};
-
-            // Assume SrcVectorDim is not the same as DstVectorDim, so we do transpose
-            // TODO: make this logic generic for all scenario
-            static_assert(SrcVectorDim != DstVectorDim, "wrong");
-
-            constexpr auto src_scalar_step_in_vector = generate_sequence(
-                detail::lambda_scalar_step_in_vector<SrcVectorDim>{}, Number<nDim>{});
-
-            constexpr auto dst_scalar_step_in_vector = generate_sequence(
-                detail::lambda_scalar_step_in_vector<DstVectorDim>{}, Number<nDim>{});
-
-            constexpr auto scalar_per_access = generate_sequence(
-                detail::lambda_scalar_per_access_for_src_and_dst<SrcVectorDim,
-                                                                 SrcScalarPerVector,
-                                                                 DstVectorDim,
-                                                                 DstScalarPerVector>{},
-                Number<nDim>{});
-
-            constexpr auto access_lengths = SliceLengths{} / scalar_per_access;
-
-            static_ford<decltype(access_lengths)>{}([&](auto access_idx) {
-                constexpr auto data_idx = access_idx * scalar_per_access;
-
-                constexpr auto data_idx_seq = generate_sequence_v2(
-                    [&](auto i) { return Number<data_idx[i]>{}; }, Number<nDim>{});
-
-                // TODO type_convert is not used yet!!!!!
-                using src_vector_t = vector_type_maker_t<SrcData, SrcScalarPerVector>;
-                using dst_vector_t = vector_type_maker_t<DstData, DstScalarPerVector>;
-
-                // get DstScalarPerVector # of read-only references to src vectors from
-                // src_thread_scratch_
-                const auto src_vector_refs = generate_tie(
-                    [&](auto i) -> const src_vector_t& {
-                        // i increment corresponds to movement in DstVectorDim
-                        return src_thread_scratch_.GetVectorTypeReference(
-                            data_idx_seq + i * dst_scalar_step_in_vector);
-                    },
-                    Number<num_src_vector>{});
-
-                // get SrcScalarPerVector # of references to dst vectors from dst_thread_scratch_
-                auto dst_vector_refs = generate_tie(
-                    [&](auto i) -> dst_vector_t& {
-                        // i increment corresponds to movement in SrcVectorDim
-                        return dst_thread_scratch_.GetVectorTypeReference(
-                            data_idx_seq + i * src_scalar_step_in_vector);
-                    },
-                    Number<num_dst_vector>{});
-
-                // do data transpose
-                // TODO type_convert is not used yet!!!!!
-                transpose_vectors<SrcData, DstScalarPerVector, SrcScalarPerVector>{}(
-                    src_vector_refs, dst_vector_refs);
-            });
-        }
-        else
-        {
-            static_ford<SliceLengths>{}([&](auto idx) {
-                // convert from SrcData to DstData here
-                dst_thread_scratch_(idx) = type_convert<DstData>(src_thread_scratch_[idx]);
-            });
-        }
-#endif
-    }
-
-    template <typename DstBuffer, typename Dst0Buffer, typename Dst1Buffer>
-    __device__ void RunWrite(const DstDesc& dst_desc,
-                             DstBuffer& dst_buf,
-                             const Dst0Desc& dst0_desc,
-                             const Dst0Buffer& dst0_buf,
-                             const Dst1Desc& dst1_desc,
-                             const Dst1Buffer& dst1_buf)
-    {
-        // if there is transpose, it's done here
-        // TODO move this elsewhere
-        TransferDataFromSrcThreadScratchToDstThreadScratch();
-
-        static_assert(DstBuffer::GetAddressSpace() == AddressSpaceEnum::Global or
-                          DstBuffer::GetAddressSpace() == AddressSpaceEnum::Lds,
-                      "wrong!");
-
-        static_assert(
-            is_same<remove_cvref_t<typename DstBuffer::type>, remove_cvref_t<DstData>>::value,
-            "wrong! SrcBuffer or DstBuffer data type is wrong");
-
-        constexpr auto I0 = Number<0>{};
-        constexpr auto I1 = Number<1>{};
-
-        // src scalar per access on each dim
-        // TODO: don't use this
-        constexpr auto dst_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto dst_access_lengths = SliceLengths{} / dst_scalar_per_access;
-
-        constexpr auto dst_dim_access_order = DstDimAccessOrder{};
-
-        constexpr auto ordered_dst_access_lengths =
-            container_reorder_given_new2old(dst_access_lengths, dst_dim_access_order);
-
-        // make forward steps
-        const auto dst_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make forward steps: dst0
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst0_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst0_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make forward steps: dst1
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst1_forward_steps = generate_tuple(
-            [&](auto i) {
-                Index forward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    forward_step_idx(j) = (i.value == j.value) ? dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst1_desc, forward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps
-        const auto dst_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps: dst0
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst0_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst0_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // make backward steps: dst1
-        // WARNING!!!!!!: this logic is only correct if dst/dst0/dst1 can use the same
-        // DstScalarPerVector
-        // TODO: fix this
-        const auto dst1_backward_steps = generate_tuple(
-            [&](auto i) {
-                Index backward_step_idx;
-
-                static_for<0, nDim, 1>{}([&](auto j) {
-                    backward_step_idx(j) = (i.value == j.value) ? -dst_scalar_per_access[i] : 0;
-                });
-
-                return make_tensor_coordinate_step(dst1_desc, backward_step_idx);
-            },
-            Number<nDim>{});
-
-        // loop over tensor and copy
-        static_ford<decltype(ordered_dst_access_lengths)>{}([&](auto ordered_dst_access_idx) {
-            // judge move forward or move backward
-            constexpr auto forward_sweep = [&]() {
-                StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-                forward_sweep_(I0) = true;
-
-                static_for<1, nDim, 1>{}([&](auto i) {
-                    index_t tmp = ordered_dst_access_idx[I0];
-
-                    static_for<1, i, 1>{}([&](auto j) {
-                        tmp = tmp * ordered_dst_access_lengths[j] + ordered_dst_access_idx[j];
-                    });
-
-                    forward_sweep_(i) = tmp % 2 == 0;
-                });
-
-                return forward_sweep_;
-            }();
-
-            // calculate dst data index
-            constexpr auto dst_data_idx = [&]() {
-                Index ordered_idx;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    ordered_idx(i) = forward_sweep[i] ? ordered_dst_access_idx[i]
-                                                      : ordered_dst_access_lengths[i] - 1 -
-                                                            ordered_dst_access_idx[i];
-                });
-
-                return container_reorder_given_old2new(ordered_idx, dst_dim_access_order) *
-                       dst_scalar_per_access;
-            }();
-
-            constexpr auto dst_data_idx_seq = generate_sequence_v2(
-                [&](auto i) { return Number<dst_data_idx[i]>{}; }, Number<dst_data_idx.Size()>{});
-
-            const bool is_dst_valid =
-                coordinate_has_valid_offset_assuming_visible_index_is_valid(dst_desc, dst_coord_);
-
-            using dst_vector_type = vector_type_maker_t<DstData, DstScalarPerVector>;
-            using dst_vector_t    = typename dst_vector_type::type;
-
-            // copy data from dst_thread_scratch_ into dst_vector_container
-            auto dst_vector_container = dst_vector_type{
-                dst_thread_scratch_.template GetAsType<dst_vector_t>(dst_data_idx_seq)};
-
-            // apply DstElementwiseOperation on dst_vector_container
-            static_for<0, DstScalarPerVector, 1>{}([&](auto i) {
-                dst_vector_container.template AsType<DstData>()(i) =
-                    dst_element_op_(dst_vector_container.template AsType<DstData>()[i]);
-            });
-
-            // copy data from dst_vector_container to dst_buf
-            dst_buf.template Set<dst_vector_t>(
-                dst_coord_.GetOffset(),
-                is_dst_valid,
-                dst_vector_container.template AsType<dst_vector_t>()[I0]);
-
-            constexpr auto move_on_dim = [&]() constexpr
-            {
-                StaticallyIndexedArray<bool, nDim> move_on_dim_;
-
-                static_for<0, nDim, 1>{}([&](auto i) {
-                    move_on_dim_(i) = ordered_dst_access_idx[i] < ordered_dst_access_lengths[i] - 1;
-
-                    static_for<i + 1, nDim, 1>{}([&](auto j) {
-                        move_on_dim_(i) &=
-                            ordered_dst_access_idx[j] == ordered_dst_access_lengths[j] - 1;
-                    });
-                });
-
-                return move_on_dim_;
-            }
-            ();
-
-            // move dst coord
-            static_for<0, nDim, 1>{}([&](auto i) {
-                if constexpr(move_on_dim[i])
-                {
-                    if constexpr(forward_sweep[i])
-                    {
-                        move_tensor_coordinate(
-                            dst_desc, dst_coord_, dst_forward_steps[dst_dim_access_order[i]]);
-                    }
-                    else
-                    {
-                        move_tensor_coordinate(
-                            dst_desc, dst_coord_, dst_backward_steps[dst_dim_access_order[i]]);
-                    }
-                }
-            });
-        });
-
-        // move dst coordinate back to slice origin (or not)
-        if constexpr(DstResetCoordinateAfterRun)
-        {
-            const auto dst_reset_step =
-                make_tensor_coordinate_step(dst_desc, GetDstCoordinateResetStep());
-
-            move_tensor_coordinate(dst_desc, dst_coord_, dst_reset_step);
-        }
-    }
-
-    __device__ static constexpr auto GetSrcCoordinateResetStep()
-    {
-        constexpr auto I0 = Number<0>{};
-
-        // scalar per access on each dim
-        // TODO: don't use lambda_scalar_per_access
-        constexpr auto src_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
-
-        constexpr auto src_dim_access_order = SrcDimAccessOrder{};
-
-        constexpr auto ordered_src_access_lengths =
-            container_reorder_given_new2old(src_access_lengths, src_dim_access_order);
-
-        // judge move forward or move backward during the last iteration
-        constexpr auto forward_sweep = [&]() {
-            StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-            forward_sweep_(I0) = true;
-
-            // TODO: BUG: should start at 1
-            static_for<1, nDim, 1>{}([&](auto i) {
-                index_t tmp = ordered_src_access_lengths[I0] - 1;
-
-                static_for<1, i, 1>{}([&](auto j) {
-                    tmp = tmp * ordered_src_access_lengths[j] + ordered_src_access_lengths[j] - 1;
-                });
-
-                forward_sweep_(i) = tmp % 2 == 0;
-            });
-
-            return forward_sweep_;
-        }();
-
-        // calculate src data index after last iteration in RunRead(), if it has not being reset by
-        // RunRead()
-        constexpr auto src_data_idx = [&]() {
-            Index ordered_idx;
-
-            static_for<0, nDim, 1>{}([&](auto i) {
-                ordered_idx(i) = forward_sweep[i] ? ordered_src_access_lengths[i] - 1 : 0;
-            });
-
-            return container_reorder_given_old2new(ordered_idx, src_dim_access_order) *
-                   src_scalar_per_access;
-        }();
-
-        //
-        constexpr auto reset_src_data_step = [&]() {
-            Index reset_src_data_step_;
-
-            static_for<0, nDim, 1>{}([&](auto i) { reset_src_data_step_(i) = -src_data_idx[i]; });
-
-            return reset_src_data_step_;
-        }();
-
-        return reset_src_data_step;
-    }
-
-    __device__ static constexpr auto GetDstCoordinateResetStep()
-    {
-        constexpr auto I0 = Number<0>{};
-
-        // scalar per access on each dim
-        // TODO: don't use lambda_scalar_per_access
-        constexpr auto dst_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto dst_access_lengths = SliceLengths{} / dst_scalar_per_access;
-
-        constexpr auto dst_dim_access_order = DstDimAccessOrder{};
-
-        constexpr auto ordered_dst_access_lengths =
-            container_reorder_given_new2old(dst_access_lengths, dst_dim_access_order);
-
-        // judge move forward or move backward during the last iteration
-        constexpr auto forward_sweep = [&]() {
-            StaticallyIndexedArray<bool, nDim> forward_sweep_;
-
-            forward_sweep_(I0) = true;
-
-            static_for<1, nDim, 1>{}([&](auto i) {
-                index_t tmp = ordered_dst_access_lengths[I0] - 1;
-
-                static_for<1, i, 1>{}([&](auto j) {
-                    tmp = tmp * ordered_dst_access_lengths[j] + ordered_dst_access_lengths[j] - 1;
-                });
-
-                forward_sweep_(i) = tmp % 2 == 0;
-            });
-
-            return forward_sweep_;
-        }();
-
-        // calculate dst data index after last iteration in RunWrite(), if it has not being reset by
-        // RunWrite()
-        constexpr auto dst_data_idx = [&]() {
-            Index ordered_idx;
-
-            static_for<0, nDim, 1>{}([&](auto i) {
-                ordered_idx(i) = forward_sweep[i] ? ordered_dst_access_lengths[i] - 1 : 0;
-            });
-
-            return container_reorder_given_old2new(ordered_idx, dst_dim_access_order) *
-                   dst_scalar_per_access;
-        }();
-
-        //
-        constexpr auto reset_dst_data_step = [&]() {
-            Index reset_dst_data_step_;
-
-            static_for<0, nDim, 1>{}([&](auto i) { reset_dst_data_step_(i) = -dst_data_idx[i]; });
-
-            return reset_dst_data_step_;
-        }();
-
-        return reset_dst_data_step;
-    }
-
-    // src_slice_origin_step_idx need to be known at compile-time, for performance reason
-    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc,
-                                       const Index& src_slice_origin_step_idx)
-    {
-        // if src coord was not reset by RunRead(), then need to adjust the step here
-        const auto adjusted_step_idx =
-            SrcResetCoordinateAfterRun ? src_slice_origin_step_idx
-                                       : src_slice_origin_step_idx + GetSrcCoordinateResetStep();
-
-        // is it OK to construct a new step every time?
-        const auto adjusted_step = make_tensor_coordinate_step(src_desc, adjusted_step_idx);
-
-        move_tensor_coordinate(src_desc, src_coord_, adjusted_step);
-    }
-
-    // src_slice_origin_step_idx need to be known at compile-time, for performance reason
-    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc,
-                                       const Index& src_slice_origin_step_idx)
-    {
-        // if src coord was not reset by RunRead(), then need to adjust the step here
-        const auto adjusted_step_idx =
-            SrcResetCoordinateAfterRun ? src_slice_origin_step_idx
-                                       : src_slice_origin_step_idx + GetSrcCoordinateResetStep();
-
-        // is it OK to construct a new step every time?
-        const auto adjusted_step = make_tensor_coordinate_step(src_desc, adjusted_step_idx);
-
-        move_tensor_coordinate(src_desc, src_coord_, adjusted_step);
-    }
-
-    // dst_slice_origin_step_idx need to be known at compile-time, for performance reason
-    __device__ void MoveDstSliceWindow(const DstDesc& dst_desc,
-                                       const Dst0Desc dst0_desc,
-                                       const Dst1Desc dst1_desc,
-                                       const Index& dst_slice_origin_step_idx)
-    {
-        // if dst coord was not reset by RunWrite(), then need to adjust the step here
-        const auto adjusted_step_idx =
-            DstResetCoordinateAfterRun ? dst_slice_origin_step_idx
-                                       : dst_slice_origin_step_idx + GetDstCoordinateResetStep();
-
-        // is it OK to construct a new step every time?
-        const auto adjusted_step = make_tensor_coordinate_step(dst_desc, adjusted_step_idx);
-
-        move_tensor_coordinate(dst_desc, dst_coord_, adjusted_step);
-        move_tensor_coordinate(dst0_desc, dst0_coord_, adjusted_step);
-        move_tensor_coordinate(dst1_desc, dst1_coord_, adjusted_step);
-    }
-
-    __device__ static constexpr auto GetSrcThreadScratchDescriptor()
-    {
-        constexpr auto src_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
-
-        constexpr auto src_access_lengths_and_vector_length = container_push_back(
-            sequence_to_tuple_of_number(src_access_lengths), Number<SrcScalarPerVector>{});
-
-        // 1st stage of transforms
-        constexpr auto desc0 =
-            make_naive_tensor_descriptor_packed(src_access_lengths_and_vector_length);
-
-        // 2nd stage of transforms
-        constexpr auto transforms = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == SrcVectorDim)
-                {
-                    return make_merge_transform_v3_division_mod(
-                        make_tuple(src_access_lengths_and_vector_length[i],
-                                   src_access_lengths_and_vector_length[Number<nDim>{}]));
-                }
-                else
-                {
-                    return make_pass_through_transform(src_access_lengths_and_vector_length[i]);
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto low_dim_idss = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == SrcVectorDim)
-                {
-                    return Sequence<i.value, nDim>{};
-                }
-                else
-                {
-                    return Sequence<i.value>{};
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto up_dim_idss =
-            generate_tuple([&](auto i) { return Sequence<i.value>{}; }, Number<nDim>{});
-
-        return transform_tensor_descriptor(desc0, transforms, low_dim_idss, up_dim_idss);
-    }
-
-    __device__ static constexpr auto GetDstThreadScratchDescriptor()
-    {
-        // 1st stage of transforms
-        constexpr auto dst_scalar_per_access = generate_sequence(
-            detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
-
-        constexpr auto dst_access_lengths = SliceLengths{} / dst_scalar_per_access;
-
-        constexpr auto dst_access_lengths_and_vector_length = container_push_back(
-            sequence_to_tuple_of_number(dst_access_lengths), Number<DstScalarPerVector>{});
-
-        constexpr auto desc0 =
-            make_naive_tensor_descriptor_packed(dst_access_lengths_and_vector_length);
-
-        // 2nd stage of transforms
-        constexpr auto transforms = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == DstVectorDim)
-                {
-                    return make_merge_transform_v3_division_mod(
-                        make_tuple(dst_access_lengths_and_vector_length[i],
-                                   dst_access_lengths_and_vector_length[Number<nDim>{}]));
-                }
-                else
-                {
-                    return make_pass_through_transform(dst_access_lengths_and_vector_length[i]);
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto low_dim_idss = generate_tuple(
-            [&](auto i) {
-                if constexpr(i == DstVectorDim)
-                {
-                    return Sequence<i.value, nDim>{};
-                }
-                else
-                {
-                    return Sequence<i.value>{};
-                }
-            },
-            Number<nDim>{});
-
-        constexpr auto up_dim_idss =
-            generate_tuple([&](auto i) { return Sequence<i.value>{}; }, Number<nDim>{});
-
-        return transform_tensor_descriptor(desc0, transforms, low_dim_idss, up_dim_idss);
-    }
-
-    private:
-    static constexpr auto src_thread_scratch_desc_ = decltype(GetSrcThreadScratchDescriptor()){};
-    static constexpr auto dst_thread_scratch_desc_ = decltype(GetDstThreadScratchDescriptor()){};
-
-    StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
-                                    SrcData,
-                                    SrcScalarPerVector,
-                                    decltype(src_thread_scratch_desc_),
-                                    true>
-        src_thread_scratch_;
-
-    StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
-                                    DstData,
-                                    DstScalarPerVector,
-                                    decltype(dst_thread_scratch_desc_),
-                                    true>
-        dst_thread_scratch_;
-
-    SrcCoord src_coord_;
-    DstCoord dst_coord_;
-    const SrcElementwiseOperation src_element_op_;
-    const DstElementwiseOperation dst_element_op_;
-};
-
-} // namespace ck
-#endif
diff --git a/include/ck/utility/amd_llvm_intrinsic.hpp b/include/ck/utility/amd_llvm_intrinsic.hpp
deleted file mode 100644
index 01e77d7be8..0000000000
--- a/include/ck/utility/amd_llvm_intrinsic.hpp
+++ /dev/null
@@ -1,14 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_AMD_LLVM_INTRINSIC_HPP
-#define CK_AMD_LLVM_INTRINSIC_HPP
-
-#include "data_type.hpp"
-
-namespace ck {
-
-__device__ int32_t llvm_amdgcn_readfirstlane_i32(int32_t i) __asm("llvm.amdgcn.readfirstlane");
-
-} // namespace ck
-#endif
diff --git a/include/ck/utility/print.hpp b/include/ck/utility/print.hpp
deleted file mode 100644
index eed1ca42c7..0000000000
--- a/include/ck/utility/print.hpp
+++ /dev/null
@@ -1,25 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#ifndef CK_PRINT_HPP
-#define CK_PRINT_HPP
-
-#include "array.hpp"
-#include "statically_indexed_array.hpp"
-#include "container_helper.hpp"
-#include "sequence.hpp"
-
-namespace ck {
-
-template <typename T>
-__host__ __device__ void print_array(const char* s, T a)
-{
-    constexpr index_t nsize = a.Size();
-
-    printf("%s size %d, {", s, nsize);
-    static_for<0, nsize, 1>{}([&a](auto i) constexpr { printf("%d, ", int32_t{a[i]}); });
-    printf("}\n");
-}
-
-} // namespace ck
-#endif
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp
deleted file mode 100644
index c77d22f4cd..0000000000
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_2d.hpp
+++ /dev/null
@@ -1,136 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace host {
-
-template <typename ADataType,
-          typename BDataType,
-          typename C0DataType,
-          typename CDataType,
-          typename AccDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation>
-struct ReferenceGemmBias2D : public device::BaseOperator
-{
-    // Argument
-    struct Argument : public device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_m_k,
-                 const Tensor<BDataType>& b_k_n,
-                 const Tensor<C0DataType>& c0_m_n,
-                 Tensor<CDataType>& c_m_n,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : a_m_k_{a_m_k},
-              b_k_n_{b_k_n},
-              c0_m_n_{c0_m_n},
-              c_m_n_{c_m_n},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_m_k_;
-        const Tensor<BDataType>& b_k_n_;
-        const Tensor<CDataType>& c0_m_n_;
-        Tensor<CDataType>& c_m_n_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public device::BaseInvoker
-    {
-        using Argument = ReferenceGemmBias2D::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_mk_kn_mn = [&](auto m, auto n) {
-                const int K = arg.a_m_k_.mDesc.GetLengths()[1];
-
-                AccDataType a   = 0;
-                AccDataType b   = 0;
-                AccDataType acc = 0;
-
-                for(int k = 0; k < K; ++k)
-                {
-                    arg.a_element_op_(a, ck::type_convert<AccDataType>(arg.a_m_k_(m, k)));
-                    arg.b_element_op_(b, ck::type_convert<AccDataType>(arg.b_k_n_(k, n)));
-                    acc += a * b;
-                }
-
-                CDataType cast_acc = static_cast<CDataType>(acc);
-                arg.c_element_op_(arg.c_m_n_(m, n), cast_acc, arg.c0_m_n_(m, n));
-            };
-
-            make_ParallelTensorFunctor(
-                f_mk_kn_mn, arg.c_m_n_.mDesc.GetLengths()[0], arg.c_m_n_.mDesc.GetLengths()[1])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_m_k,
-                             const Tensor<BDataType>& b_k_n,
-                             const Tensor<C0DataType>& c0_m_n,
-                             Tensor<CDataType>& c_m_n,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
-    {
-        return Argument{a_m_k, b_k_n, c0_m_n, c_m_n, a_element_op, b_element_op, c_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceGemmBias2D"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace host
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp
deleted file mode 100644
index 7dfc3c1ed4..0000000000
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation.hpp
+++ /dev/null
@@ -1,140 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace host {
-
-template <typename ADataType,
-          typename BDataType,
-          typename CDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation>
-struct ReferenceGemmBiasActivation : public device::BaseOperator
-{
-    // Argument
-    struct Argument : public device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_m_k,
-                 const Tensor<BDataType>& b_k_n,
-                 Tensor<CDataType>& c_m_n,
-                 const Tensor<CDataType>& c0_n,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : a_m_k_{a_m_k},
-              b_k_n_{b_k_n},
-              c_m_n_{c_m_n},
-              c0_n_{c0_n},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_m_k_;
-        const Tensor<BDataType>& b_k_n_;
-        Tensor<CDataType>& c_m_n_;
-        const Tensor<CDataType>& c0_n_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public device::BaseInvoker
-    {
-        using Argument = ReferenceGemmBiasActivation::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_mk_kn_mn = [&](auto m, auto n) {
-                const int K = arg.a_m_k_.mDesc.GetLengths()[1];
-
-                float v_acc = 0;
-
-                for(int k = 0; k < K; ++k)
-                {
-                    float v_a;
-                    float v_b;
-
-                    arg.a_element_op_(v_a, static_cast<const float>(arg.a_m_k_(m, k)));
-                    arg.b_element_op_(v_b, static_cast<const float>(arg.b_k_n_(k, n)));
-
-                    v_acc += v_a * v_b;
-                }
-
-                float v_c;
-
-                arg.c_element_op_(v_c, v_acc, static_cast<float>(arg.c0_n_(n)));
-
-                arg.c_m_n_(m, n) = v_c;
-            };
-
-            make_ParallelTensorFunctor(
-                f_mk_kn_mn, arg.c_m_n_.mDesc.GetLengths()[0], arg.c_m_n_.mDesc.GetLengths()[1])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_m_k,
-                             const Tensor<BDataType>& b_k_n,
-                             Tensor<CDataType>& c_m_n,
-                             const Tensor<CDataType>& c0_n,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
-    {
-        return Argument{a_m_k, b_k_n, c_m_n, c0_n, a_element_op, b_element_op, c_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceGemmBiasActivation"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace host
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp
deleted file mode 100644
index 99102a40d4..0000000000
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_bias_activation_add.hpp
+++ /dev/null
@@ -1,148 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace host {
-
-template <typename ADataType,
-          typename BDataType,
-          typename CDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation>
-struct ReferenceGemmBiasActivationAdd : public device::BaseOperator
-{
-    // Argument
-    struct Argument : public device::BaseArgument
-    {
-        Argument(const Tensor<ADataType>& a_m_k,
-                 const Tensor<BDataType>& b_k_n,
-                 Tensor<CDataType>& c_m_n,
-                 const Tensor<CDataType>& c0_n,
-                 const Tensor<CDataType>& c1_m_n,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : a_m_k_{a_m_k},
-              b_k_n_{b_k_n},
-              c_m_n_{c_m_n},
-              c0_n_{c0_n},
-              c1_m_n_{c1_m_n},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-        }
-
-        const Tensor<ADataType>& a_m_k_;
-        const Tensor<BDataType>& b_k_n_;
-        Tensor<CDataType>& c_m_n_;
-        const Tensor<CDataType>& c0_n_;
-        const Tensor<CDataType>& c1_m_n_;
-
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-    };
-
-    // Invoker
-    struct Invoker : public device::BaseInvoker
-    {
-        using Argument = ReferenceGemmBiasActivationAdd::Argument;
-
-        float Run(const Argument& arg)
-        {
-            auto f_mk_kn_mn = [&](auto m, auto n) {
-                const int K = arg.a_m_k_.mDesc.GetLengths()[1];
-
-                float v_acc = 0;
-
-                for(int k = 0; k < K; ++k)
-                {
-                    float v_a;
-                    float v_b;
-
-                    arg.a_element_op_(v_a, static_cast<const float>(arg.a_m_k_(m, k)));
-                    arg.b_element_op_(v_b, static_cast<const float>(arg.b_k_n_(k, n)));
-
-                    v_acc += v_a * v_b;
-                }
-
-                float v_c;
-
-                arg.c_element_op_(v_c,
-                                  v_acc,
-                                  static_cast<float>(arg.c0_n_(n)),
-                                  static_cast<float>(arg.c1_m_n_(m, n)));
-
-                arg.c_m_n_(m, n) = v_c;
-            };
-
-            make_ParallelTensorFunctor(
-                f_mk_kn_mn, arg.c_m_n_.mDesc.GetLengths()[0], arg.c_m_n_.mDesc.GetLengths()[1])(
-                std::thread::hardware_concurrency());
-
-            return 0;
-        }
-
-        float Run(const device::BaseArgument* p_arg,
-                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg));
-        }
-    };
-
-    static constexpr bool IsValidCompilationParameter()
-    {
-        // TODO: properly implement this check
-        return true;
-    }
-
-    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
-
-    static auto MakeArgument(const Tensor<ADataType>& a_m_k,
-                             const Tensor<BDataType>& b_k_n,
-                             Tensor<CDataType>& c_m_n,
-                             const Tensor<CDataType>& c0_n,
-                             const Tensor<CDataType>& c1_m_n,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
-    {
-        return Argument{
-            a_m_k, b_k_n, c_m_n, c0_n, c1_m_n, a_element_op, b_element_op, c_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "ReferenceGemmBiasActivationAdd"
-            << std::endl;
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace host
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
index 0655fd92e4..bb5f971c77 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
index 495c5f884f..0b025b33c8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
index 0aa7a5aa3d..593ef7cb96 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
index a6dcfa30d3..e1a4391c4a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
index 89df1a7a0d..34c86dd440 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index c116d999da..6a551c7268 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -3,10 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index e3f07606c2..fc9ec8d615 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -3,10 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index ec5d18fc21..07d5524762 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
index 62f28c9b11..2c529e06fc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
index 381a015eb0..8af400cb7a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
@@ -3,8 +3,7 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
index 682f546759..3d0c34062c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
@@ -4,7 +4,7 @@
 #pragma once
 
 #include <cstdlib>
-
+#include <vector>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_reduce.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index e230507e7e..0a8f2215ba 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -3,10 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
 #include <memory>
 #include <vector>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
index 90b6e11b9b..2d578cca4f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
@@ -3,10 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index ef70504f29..8f4cd4d968 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -3,10 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
 #include <vector>
 #include <memory>
-
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index 8986a79344..e888446948 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_splitk.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index 175932e632..fc4beb0ae9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -4,7 +4,7 @@
 #pragma once
 
 #include <vector>
-
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
index e38dad1650..e97484a5ad 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
index 55c67b7623..199ed73b4c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
@@ -3,8 +3,8 @@
 
 #pragma once
 
-#include <cstdlib>
-
+#include <vector>
+#include <memory>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_normalization.hpp"
diff --git a/library/include/ck/library/utility/host_conv.hpp b/library/include/ck/library/utility/host_conv.hpp
deleted file mode 100644
index 8348a3089f..0000000000
--- a/library/include/ck/library/utility/host_conv.hpp
+++ /dev/null
@@ -1,152 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-#include "host_tensor.hpp"
-#include "conv_common.hpp"
-
-template <typename TIn,
-          typename TWei,
-          typename TOut,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-void host_conv_nchw_kcyx_nkhw(const Tensor<TIn>& in,
-                              const Tensor<TWei>& wei,
-                              Tensor<TOut>& out,
-                              const ConvStrides& conv_strides,
-                              const ConvDilations& conv_dilations,
-                              const InLeftPads& in_left_pads,
-                              const InRightPads&)
-{
-    constexpr auto I0 = ck::Number<0>{};
-    constexpr auto I1 = ck::Number<1>{};
-
-    auto f_nchw = [&](auto n, auto k, auto ho, auto wo) {
-        float v = 0;
-        for(int c = 0; c < wei.mDesc.GetLengths()[1]; ++c)
-        {
-            for(int y = 0; y < wei.mDesc.GetLengths()[2]; ++y)
-            {
-                int hi = ho * conv_strides[I0] + y * conv_dilations[I0] - in_left_pads[I0];
-                for(int x = 0; x < wei.mDesc.GetLengths()[3]; ++x)
-                {
-                    int wi = wo * conv_strides[I1] + x * conv_dilations[I1] - in_left_pads[I1];
-                    if(hi >= 0 && hi < in.mDesc.GetLengths()[2] && wi >= 0 &&
-                       wi < in.mDesc.GetLengths()[3])
-                    {
-                        v += ck::type_convert<float>(in(n, c, hi, wi)) *
-                             ck::type_convert<float>(wei(k, c, y, x));
-                    }
-                }
-            }
-        }
-        out(n, k, ho, wo) = ck::type_convert<TOut>(v);
-    };
-
-    make_ParallelTensorFunctor(f_nchw,
-                               out.mDesc.GetLengths()[0],
-                               out.mDesc.GetLengths()[1],
-                               out.mDesc.GetLengths()[2],
-                               out.mDesc.GetLengths()[3])(std::thread::hardware_concurrency());
-}
-
-template <typename TIn,
-          typename TWei,
-          typename TOut,
-          typename ConvStrides,
-          typename ConvDilations,
-          typename InLeftPads,
-          typename InRightPads>
-void host_conv3d_ndhwc_kzyxc_ndhwk(const Tensor<TIn>& in,
-                                   const Tensor<TWei>& wei,
-                                   Tensor<TOut>& out,
-                                   const ConvStrides& conv_strides,
-                                   const ConvDilations& conv_dilations,
-                                   const InLeftPads& in_left_pads,
-                                   const InRightPads&)
-{
-    using namespace ck;
-
-    constexpr auto I0 = Number<0>{};
-    constexpr auto I1 = Number<1>{};
-    constexpr auto I2 = Number<2>{};
-    const auto Di     = in.mDesc.GetLengths()[1];
-    const auto Hi     = in.mDesc.GetLengths()[2];
-    const auto Wi     = in.mDesc.GetLengths()[3];
-    const auto Z      = wei.mDesc.GetLengths()[1];
-    const auto Y      = wei.mDesc.GetLengths()[2];
-    const auto X      = wei.mDesc.GetLengths()[3];
-    const auto C      = wei.mDesc.GetLengths()[4];
-
-    auto f_ndhwc = [&](auto n, auto do_tmp, auto ho_tmp, auto wo_tmp, auto k) {
-        // do__ must be converted to signed integer, otherwise zmin might be wrong in cases
-        // negative values.
-        const int do_ = static_cast<int>(do_tmp);
-        const int ho  = static_cast<int>(ho_tmp);
-        const int wo  = static_cast<int>(wo_tmp);
-        const int zmin =
-            std::max(0,
-                     (in_left_pads[I0] - do_ * conv_strides[I0] + conv_dilations[I0] - 1) /
-                         conv_dilations[I0]);
-        const int ymin =
-            std::max(0,
-                     (in_left_pads[I1] - ho * conv_strides[I1] + conv_dilations[I1] - 1) /
-                         conv_dilations[I1]);
-        const int xmin =
-            std::max(0,
-                     (in_left_pads[I2] - wo * conv_strides[I2] + conv_dilations[I2] - 1) /
-                         conv_dilations[I2]);
-        const int zmax =
-            std::min(Z, (in_left_pads[I0] - do_ * conv_strides[I0] + Di) / conv_dilations[I0]);
-        const int ymax =
-            std::min(Y, (in_left_pads[I1] - ho * conv_strides[I1] + Hi) / conv_dilations[I1]);
-        const int xmax =
-            std::min(X, (in_left_pads[I2] - wo * conv_strides[I2] + Wi) / conv_dilations[I2]);
-        const int di_min = do_ * conv_strides[I0] + zmin * conv_dilations[I0] - in_left_pads[I0];
-        const int hi_min = ho * conv_strides[I1] + ymin * conv_dilations[I1] - in_left_pads[I1];
-        const int wi_min = wo * conv_strides[I2] + xmin * conv_dilations[I2] - in_left_pads[I2];
-
-        double v = 0;
-
-        const TIn* in_n   = in.mData.data() + n * Di * Hi * Wi * C;
-        const TWei* wei_k = wei.mData.data() + k * Z * Y * X * C;
-
-        int di = di_min;
-        for(int z = zmin; z < zmax; ++z, di += conv_dilations[I0])
-        {
-            const TIn* in_n_di  = in_n + di * Hi * Wi * C;
-            const TWei* wei_k_z = wei_k + z * Y * X * C;
-            int hi              = hi_min;
-
-            for(int y = ymin; y < ymax; ++y, hi += conv_dilations[I1])
-            {
-                const TIn* in_n_di_hi = in_n_di + hi * Wi * C;
-                const TWei* wei_k_z_y = wei_k_z + y * X * C;
-                int wi                = wi_min;
-
-                for(int x = xmin; x < xmax; ++x, wi += conv_dilations[I2])
-                {
-                    const TIn* in_n_di_hi_wi = in_n_di_hi + wi * C;
-                    const TWei* wei_k_z_y_x  = wei_k_z_y + x * C;
-
-                    for(int c = 0; c < C; ++c)
-                    {
-                        v += static_cast<const double>(in_n_di_hi_wi[c]) *
-                             static_cast<const double>(wei_k_z_y_x[c]);
-                    }
-                }
-            }
-        }
-
-        out(n, do_, ho, wo, k) = v;
-    };
-
-    make_ParallelTensorFunctor(f_ndhwc,
-                               out.mDesc.GetLengths()[0],
-                               out.mDesc.GetLengths()[1],
-                               out.mDesc.GetLengths()[2],
-                               out.mDesc.GetLengths()[3],
-                               out.mDesc.GetLengths()[4])(std::thread::hardware_concurrency() - 4);
-}
diff --git a/library/include/ck/library/utility/op_instance_engine.hpp b/library/include/ck/library/utility/op_instance_engine.hpp
deleted file mode 100644
index 78812e8c81..0000000000
--- a/library/include/ck/library/utility/op_instance_engine.hpp
+++ /dev/null
@@ -1,249 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <cstdlib>
-#include <iostream>
-#include <limits>
-#include <memory>
-#include <stdexcept>
-#include <tuple>
-#include <utility>
-#include <vector>
-
-#include "ck/utility/functional2.hpp"
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-
-namespace ck {
-namespace utils {
-
-struct ProfileBestConfig
-{
-    std::string best_op_name;
-    float best_avg_time   = std::numeric_limits<float>::max();
-    float best_tflops     = std::numeric_limits<float>::max();
-    float best_gb_per_sec = std::numeric_limits<float>::max();
-};
-
-/**
- * @brief      This class describes an operation instance(s).
- *
- *             Op instance defines a particular specializations of operator
- *             template. Thanks to this specific input/output data types, data
- *             layouts and modifying elementwise operations it is able to create
- *             it's input/output tensors, provide pointers to instances which
- *             can execute it and all operation specific parameters.
- */
-template <typename OutDataType, typename... InArgTypes>
-class OpInstance
-{
-    public:
-    template <typename T>
-    using TensorPtr      = std::unique_ptr<Tensor<T>>;
-    using InTensorsTuple = std::tuple<TensorPtr<InArgTypes>...>;
-    using DeviceMemPtr   = std::unique_ptr<DeviceMem>;
-    using DeviceBuffers  = std::vector<DeviceMemPtr>;
-
-    OpInstance()                  = default;
-    OpInstance(const OpInstance&) = default;
-    OpInstance& operator=(const OpInstance&) = default;
-    virtual ~OpInstance(){};
-
-    virtual InTensorsTuple GetInputTensors() const         = 0;
-    virtual TensorPtr<OutDataType> GetOutputTensor() const = 0;
-    virtual std::unique_ptr<tensor_operation::device::BaseInvoker>
-    MakeInvokerPointer(tensor_operation::device::BaseOperator*) const = 0;
-    virtual std::unique_ptr<tensor_operation::device::BaseArgument>
-    MakeArgumentPointer(tensor_operation::device::BaseOperator*,
-                        const DeviceBuffers&,
-                        const DeviceMemPtr&) const = 0;
-    virtual std::size_t GetFlops() const           = 0;
-    virtual std::size_t GetBtype() const           = 0;
-};
-
-/**
- * @brief      A generic operation instance run engine.
- */
-template <typename OutDataType, typename... InArgTypes>
-class OpInstanceRunEngine
-{
-    public:
-    using OpInstanceT = OpInstance<InArgTypes..., OutDataType>;
-    template <typename T>
-    using TensorPtr        = std::unique_ptr<Tensor<T>>;
-    using DeviceMemPtr     = std::unique_ptr<DeviceMem>;
-    using InTensorsTuple   = std::tuple<TensorPtr<InArgTypes>...>;
-    using DeviceBuffers    = std::vector<DeviceMemPtr>;
-    using InArgsTypesTuple = std::tuple<InArgTypes...>;
-
-    OpInstanceRunEngine() = delete;
-
-    template <typename ReferenceOp = std::function<void()>>
-    OpInstanceRunEngine(const OpInstanceT& op_instance,
-                        const ReferenceOp& reference_op = ReferenceOp{},
-                        bool do_verification            = true)
-        : op_instance_{op_instance}
-    {
-        in_tensors_ = op_instance_.GetInputTensors();
-        out_tensor_ = op_instance_.GetOutputTensor();
-
-        if constexpr(std::is_invocable_v<ReferenceOp,
-                                         const Tensor<InArgTypes>&...,
-                                         Tensor<OutDataType>&>)
-        {
-            if(do_verification)
-            {
-                ref_output_ = op_instance_.GetOutputTensor();
-                CallRefOpUnpackArgs(reference_op, std::make_index_sequence<kNInArgs_>{});
-            }
-        }
-        AllocateDeviceInputTensors(std::make_index_sequence<kNInArgs_>{});
-        out_device_buffer_ = std::make_unique<DeviceMem>(sizeof(OutDataType) *
-                                                         out_tensor_->mDesc.GetElementSpaceSize());
-        out_device_buffer_->SetZero();
-    }
-
-    virtual ~OpInstanceRunEngine(){};
-
-    template <typename OpInstancePtr>
-    bool Test(const std::vector<OpInstancePtr>& op_ptrs)
-    {
-        bool res{true};
-        for(auto& op_ptr : op_ptrs)
-        {
-            auto invoker  = op_instance_.MakeInvokerPointer(op_ptr.get());
-            auto argument = op_instance_.MakeArgumentPointer(
-                op_ptr.get(), in_device_buffers_, out_device_buffer_);
-            if(op_ptr->IsSupportedArgument(argument.get()))
-            {
-                std::cout << "Testing instance: " << op_ptr->GetTypeString() << std::endl;
-                invoker->Run(argument.get());
-                out_device_buffer_->FromDevice(out_tensor_->mData.data());
-                if(!ref_output_)
-                {
-                    throw std::runtime_error(
-                        "OpInstanceRunEngine::Test: Reference value not availabe."
-                        " You have to provide reference function.");
-                }
-                // TODO: enable flexible use of custom check_error functions
-                bool inst_res = CheckErr(out_tensor_->mData, ref_output_->mData);
-                std::cout << (inst_res ? "SUCCESS" : "FAILURE") << std::endl;
-                res = res && inst_res;
-                out_device_buffer_->SetZero();
-            }
-            else
-            {
-                std::cout << "Given conv problem is not supported by instance: \n\t>>>>"
-                          << op_ptr->GetTypeString() << std::endl;
-            }
-        }
-        return res;
-    }
-
-    template <typename OpInstancePtr>
-    ProfileBestConfig Profile(const std::vector<OpInstancePtr>& op_ptrs,
-                              bool time_kernel     = false,
-                              bool do_verification = false,
-                              bool do_log          = false)
-    {
-        ProfileBestConfig best_config;
-
-        for(auto& op_ptr : op_ptrs)
-        {
-            auto invoker  = op_instance_.MakeInvokerPointer(op_ptr.get());
-            auto argument = op_instance_.MakeArgumentPointer(
-                op_ptr.get(), in_device_buffers_, out_device_buffer_);
-            if(op_ptr->IsSupportedArgument(argument.get()))
-            {
-                std::string op_name = op_ptr->GetTypeString();
-                float avg_time = invoker->Run(argument.get(), StreamConfig{nullptr, time_kernel});
-
-                std::size_t flops     = op_instance_.GetFlops();
-                std::size_t num_btype = op_instance_.GetBtype();
-                float tflops          = static_cast<float>(flops) / 1.E9 / avg_time;
-                float gb_per_sec      = num_btype / 1.E6 / avg_time;
-
-                std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                          << " GB/s, " << op_name << std::endl;
-
-                if(avg_time < best_config.best_avg_time)
-                {
-                    best_config.best_op_name    = op_name;
-                    best_config.best_tflops     = tflops;
-                    best_config.best_gb_per_sec = gb_per_sec;
-                    best_config.best_avg_time   = avg_time;
-                }
-
-                if(do_verification)
-                {
-                    out_device_buffer_->FromDevice(out_tensor_->mData.data());
-                    if(!ref_output_)
-                    {
-                        throw std::runtime_error(
-                            "OpInstanceRunEngine::Profile: Reference value not availabe."
-                            " You have to provide reference function.");
-                    }
-                    // TODO: enable flexible use of custom check_error functions
-                    CheckErr(out_tensor_->mData, ref_output_->mData);
-
-                    if(do_log) {}
-                }
-                out_device_buffer_->SetZero();
-            }
-        }
-        return best_config;
-    }
-
-    void SetAtol(double a) { atol_ = a; }
-    void SetRtol(double r) { rtol_ = r; }
-
-    private:
-    template <typename F, std::size_t... Is>
-    void CallRefOpUnpackArgs(const F& f, std::index_sequence<Is...>) const
-    {
-        f(*std::get<Is>(in_tensors_)..., *ref_output_);
-    }
-
-    template <std::size_t... Is>
-    void AllocateDeviceInputTensors(std::index_sequence<Is...>)
-    {
-        (AllocateDeviceInputTensorsImpl<Is>(), ...);
-    }
-
-    template <std::size_t Index>
-    void AllocateDeviceInputTensorsImpl()
-    {
-        const auto& ts = std::get<Index>(in_tensors_);
-        in_device_buffers_
-            .emplace_back(
-                std::make_unique<DeviceMem>(sizeof(std::tuple_element_t<Index, InArgsTypesTuple>) *
-                                            ts->mDesc.GetElementSpaceSize()))
-            ->ToDevice(ts->mData.data());
-    }
-
-    static constexpr std::size_t kNInArgs_ = std::tuple_size_v<InTensorsTuple>;
-    const OpInstanceT& op_instance_;
-    double rtol_{1e-5};
-    double atol_{1e-8};
-
-    InTensorsTuple in_tensors_;
-    TensorPtr<OutDataType> out_tensor_;
-    TensorPtr<OutDataType> ref_output_;
-
-    DeviceBuffers in_device_buffers_;
-    DeviceMemPtr out_device_buffer_;
-
-    template <typename T>
-    bool CheckErr(const std::vector<T>& dev_out, const std::vector<T>& ref_out) const
-    {
-        return ck::utils::check_err(dev_out, ref_out, "Error: incorrect results!", rtol_, atol_);
-    }
-};
-
-} // namespace utils
-} // namespace ck
diff --git a/profiler/include/profiler/data_type_enum_helper.hpp b/profiler/include/profiler/data_type_enum_helper.hpp
deleted file mode 100644
index d9bd5e1a40..0000000000
--- a/profiler/include/profiler/data_type_enum_helper.hpp
+++ /dev/null
@@ -1,77 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma
-
-#include "ck/utility/data_type.hpp"
-#include "profiler/data_type_enum.hpp"
-
-namespace ck {
-
-template <DataTypeEnum DataTypeEnum>
-struct get_datatype_from_enum;
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Int8>
-{
-    using type = int8_t;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Int32>
-{
-    using type = int32_t;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Half>
-{
-    using type = half_t;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Float>
-{
-    using type = float;
-};
-
-template <>
-struct get_datatype_from_enum<DataTypeEnum::Double>
-{
-    using type = double;
-};
-
-template <typename T>
-struct get_datatype_enum_from_type;
-
-template <>
-struct get_datatype_enum_from_type<int8_t>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Int8;
-};
-
-template <>
-struct get_datatype_enum_from_type<int32_t>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Int32;
-};
-
-template <>
-struct get_datatype_enum_from_type<half_t>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Half;
-};
-
-template <>
-struct get_datatype_enum_from_type<float>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Float;
-};
-
-template <>
-struct get_datatype_enum_from_type<double>
-{
-    static constexpr DataTypeEnum value = DataTypeEnum::Double;
-};
-
-} // namespace ck
diff --git a/profiler/include/profiler/profile_convnd_bwd_data_impl.hpp b/profiler/include/profiler/profile_convnd_bwd_data_impl.hpp
deleted file mode 100644
index 1e69ebc8bd..0000000000
--- a/profiler/include/profiler/profile_convnd_bwd_data_impl.hpp
+++ /dev/null
@@ -1,486 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/conv_util.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
-
-using F16  = ck::half_t;
-using F32  = float;
-using BF16 = ck::bhalf_t;
-using INT8 = int8_t;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using DeviceConvBwdDataNoOpPtr =
-    DeviceConvBwdDataPtr<ck::tensor_operation::element_wise::PassThrough,
-                         ck::tensor_operation::element_wise::PassThrough,
-                         ck::tensor_operation::element_wise::PassThrough>;
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
-    std::vector<DeviceConvBwdDataNoOpPtr>&);
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-namespace ck {
-namespace profiler {
-using DeviceConvBwdDataNoOpPtr = ck::tensor_operation::device::instance::DeviceConvBwdDataNoOpPtr;
-
-template <typename InLayout>
-HostTensorDescriptor get_input_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-template <typename WeiLayout>
-HostTensorDescriptor get_filters_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                        int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-template <typename OutLayout>
-HostTensorDescriptor get_output_host_ensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-template <typename InDataType, typename WeiDataType, typename OutDataType>
-void get_device_conv_bwd_data_op_ptr(
-    InDataType, WeiDataType, OutDataType, std::vector<DeviceConvBwdDataNoOpPtr>&, int)
-{
-    std::cout << "can not find device conv bwd data" << std::endl;
-    exit(1);
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    F32, F32, F32, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    F16, F16, F16, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    BF16, BF16, BF16, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-template <>
-void get_device_conv_bwd_data_op_ptr(
-    INT8, INT8, INT8, std::vector<DeviceConvBwdDataNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <typename T>
-static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
-{
-    float max_diff = 1e-6;
-
-    for(std::size_t i = 0; i < ref.mData.size(); ++i)
-    {
-        float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
-        if(max_diff < diff)
-        {
-            return false;
-        }
-    }
-    return true;
-}
-template <typename DataType>
-void show_data_nhwc_layout(Tensor<DataType>& nhwc)
-{
-    std::cout << "[";
-    for(int n = 0; n < ck::type_convert<int>(nhwc.mDesc.GetLengths()[0]); n++)
-    {
-        std::cout << "[";
-        for(int hi = 0; hi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[2]); hi++)
-        {
-            std::cout << "[";
-            for(int wi = 0; wi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[3]); wi++)
-            {
-                std::cout << "[";
-                for(int c = 0; c < ck::type_convert<int>(nhwc.mDesc.GetLengths()[1]); c++)
-                {
-                    std::cout << static_cast<float>(nhwc(n, c, hi, wi)) << "  ";
-                }
-                std::cout << "]";
-            }
-            std::cout << "]";
-        }
-        std::cout << "]";
-    }
-    std::cout << "]";
-}
-
-template <int NDimSpatial,
-          typename InDataType,
-          typename WeiDataType,
-          typename OutDataType,
-          typename AccDataType,
-          typename InLayout,
-          typename WeiLayout,
-          typename OutLayout>
-bool profile_convnd_bwd_data_impl(int do_verification,
-                                  int init_method,
-                                  bool do_log,
-                                  bool time_kernel,
-                                  ck::index_t N,
-                                  ck::index_t K,
-                                  ck::index_t C,
-                                  const std::vector<ck::index_t>& input_spatial_lengths,
-                                  const std::vector<ck::index_t>& filter_spatial_lengths,
-                                  const std::vector<ck::index_t>& output_spatial_lengths,
-                                  const std::vector<ck::index_t>& conv_filter_strides,
-                                  const std::vector<ck::index_t>& conv_filter_dilations,
-                                  const std::vector<ck::index_t>& input_left_pads,
-                                  const std::vector<ck::index_t>& input_right_pads)
-{
-    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-    const auto in_element_op  = InElementOp{};
-    const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{};
-
-    std::vector<std::size_t> input_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(C)};
-    input_dims.insert(
-        std::end(input_dims), std::begin(input_spatial_lengths), std::end(input_spatial_lengths));
-
-    std::vector<std::size_t> filter_dims{static_cast<std::size_t>(K), static_cast<std::size_t>(C)};
-    filter_dims.insert(std::end(filter_dims),
-                       std::begin(filter_spatial_lengths),
-                       std::end(filter_spatial_lengths));
-
-    std::vector<std::size_t> output_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(K)};
-    output_dims.insert(std::end(output_dims),
-                       std::begin(output_spatial_lengths),
-                       std::end(output_spatial_lengths));
-
-    Tensor<InDataType> input_host_result(
-        get_input_host_tensor_descriptor<InLayout>(input_dims, NDimSpatial));
-    Tensor<InDataType> input_device_result(
-        get_input_host_tensor_descriptor<InLayout>(input_dims, NDimSpatial));
-    Tensor<WeiDataType> weights(
-        get_filters_host_tensor_descriptor<WeiLayout>(filter_dims, NDimSpatial));
-    Tensor<OutDataType> output(
-        get_output_host_ensor_descriptor<OutLayout>(output_dims, NDimSpatial));
-
-    std::cout << "input: " << input_host_result.mDesc << std::endl;
-    std::cout << "weights: " << weights.mDesc << std::endl;
-    std::cout << "output: " << output.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        output.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
-        weights.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
-        break;
-    default:
-        output.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
-        weights.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
-    }
-
-    DeviceMem in_device_buf(sizeof(InDataType) * input_device_result.mDesc.GetElementSpace());
-    DeviceMem wei_device_buf(sizeof(WeiDataType) * weights.mDesc.GetElementSpace());
-    DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
-
-    out_device_buf.ToDevice(output.mData.data());
-    wei_device_buf.ToDevice(weights.mData.data());
-
-    // reset input to zero
-    in_device_buf.SetZero();
-
-    if(do_verification)
-    {
-        auto RunReference = [&](auto& ref_conv) {
-            auto ref_invoker = ref_conv.MakeInvoker();
-
-            auto ref_argument = ref_conv.MakeArgument(input_host_result,
-                                                      weights,
-                                                      output,
-                                                      conv_filter_strides,
-                                                      conv_filter_dilations,
-                                                      input_left_pads,
-                                                      input_right_pads,
-                                                      InElementOp{},
-                                                      WeiElementOp{},
-                                                      OutElementOp{});
-            ref_invoker.Run(ref_argument);
-        };
-
-        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdData<InDataType,
-                                                                         WeiDataType,
-                                                                         OutDataType,
-                                                                         AccDataType,
-                                                                         InElementOp,
-                                                                         WeiElementOp,
-                                                                         OutElementOp,
-                                                                         NDimSpatial>();
-        RunReference(ref_conv);
-    }
-
-    // add device Conv instances
-    std::vector<DeviceConvBwdDataNoOpPtr> conv_ptrs;
-    get_device_conv_bwd_data_op_ptr(
-        InDataType{}, WeiDataType{}, OutDataType{}, conv_ptrs, NDimSpatial);
-
-    if(conv_ptrs.size() <= 0)
-    {
-        throw std::runtime_error("wrong! no device Conv instance found");
-    }
-
-    std::string best_conv_name;
-    float best_ave_time   = 0;
-    float best_tflops     = 0;
-    float best_gb_per_sec = 0;
-
-    // profile device Conv instances
-    bool success = true;
-    for(auto& conv_ptr : conv_ptrs)
-    {
-        auto argument_ptr = conv_ptr->MakeArgumentPointer(
-            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-            static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
-            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-            N,
-            K,
-            C,
-            input_spatial_lengths,
-            filter_spatial_lengths,
-            output_spatial_lengths,
-            conv_filter_strides,
-            conv_filter_dilations,
-            input_left_pads,
-            input_right_pads,
-            in_element_op,
-            wei_element_op,
-            out_element_op);
-
-        auto invoker_ptr = conv_ptr->MakeInvokerPointer();
-
-        if(conv_ptr->IsSupportedArgument(argument_ptr.get()))
-        {
-            std::string conv_name = conv_ptr->GetTypeString();
-
-            float ave_time =
-                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-
-            std::size_t flop =
-                ck::utils::conv::get_flops(N, C, K, filter_spatial_lengths, output_spatial_lengths);
-            std::size_t num_btype =
-                ck::utils::conv::get_btype<InDataType, WeiDataType, OutDataType>(
-                    N, C, K, input_spatial_lengths, filter_spatial_lengths, output_spatial_lengths);
-
-            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
-            float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-            std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                      << " GB/s" << std::endl;
-
-            if(tflops > best_tflops)
-            {
-                best_conv_name  = conv_name;
-                best_tflops     = tflops;
-                best_ave_time   = ave_time;
-                best_gb_per_sec = gb_per_sec;
-            }
-
-            if(do_verification)
-            {
-                in_device_buf.FromDevice(input_device_result.mData.data());
-
-                if(!check_out(input_host_result, input_device_result))
-                {
-                    std::cout << "Fail Info: " << conv_ptr->GetTypeString() << std::endl;
-
-                    success = false;
-                }
-                else
-                {
-                    std::cout << "Pass Info: " << conv_ptr->GetTypeString() << std::endl;
-                }
-
-                success = ck::utils::check_err(input_host_result, input_device_result);
-
-                if(do_log)
-                {
-                    std::cout << "in : ";
-                    show_data_nhwc_layout(output);
-                    std::cout << std::endl;
-
-                    std::cout << "wei: ";
-                    show_data_nhwc_layout(weights);
-                    std::cout << std::endl;
-
-                    std::cout << "out_host  : ";
-                    show_data_nhwc_layout(input_host_result);
-                    std::cout << std::endl;
-
-                    std::cout << "out_device: ";
-                    show_data_nhwc_layout(input_device_result);
-                    std::cout << std::endl;
-                }
-            }
-        }
-    }
-
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-              << best_gb_per_sec << " GB/s, " << best_conv_name << std::endl;
-    return success;
-}
-
-} // namespace profiler
-} // namespace ck
diff --git a/profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp b/profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp
deleted file mode 100644
index e37c887a96..0000000000
--- a/profiler/include/profiler/profile_convnd_bwd_weight_impl.hpp
+++ /dev/null
@@ -1,474 +0,0 @@
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_conv_backward_weight.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/conv_util.hpp"
-#include "ck/library/host_tensor/device_memory.hpp"
-#include "ck/library/host_tensor/host_tensor.hpp"
-#include "ck/library/host_tensor/host_tensor_generator.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_conv_backward_weight.hpp"
-
-using F16  = ck::half_t;
-using F32  = float;
-using BF16 = ck::bhalf_t;
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using DeviceConvndBwdWeightNoOpPtr =
-    DeviceConvBwdWeightPtr<ck::tensor_operation::element_wise::PassThrough,
-                           ck::tensor_operation::element_wise::PassThrough,
-                           ck::tensor_operation::element_wise::PassThrough>;
-
-void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-
-void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-
-void add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-void add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
-    std::vector<DeviceConvndBwdWeightNoOpPtr>&);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
-
-namespace ck {
-namespace profiler {
-
-using DeviceConvndBwdWeightNoOpPtr =
-    ck::tensor_operation::device::instance::DeviceConvndBwdWeightNoOpPtr;
-
-template <typename InLayout>
-HostTensorDescriptor get_input_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, InLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-
-template <typename WeiLayout>
-HostTensorDescriptor get_filters_host_tensor_descriptor(const std::vector<std::size_t>& dims,
-                                                        int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, WeiLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-
-template <typename OutLayout>
-HostTensorDescriptor get_output_host_ensor_descriptor(const std::vector<std::size_t>& dims,
-                                                      int num_dim_spatial = 2)
-{
-    namespace tl = ck::tensor_layout::convolution;
-
-    switch(num_dim_spatial)
-    {
-    case 3: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 2: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    case 1: {
-        return ck::utils::conv::get_host_tensor_descriptor(dims, OutLayout{});
-    }
-    default: {
-        throw std::runtime_error("Unsupported number of spatial dimensions provided!");
-    }
-    }
-}
-
-template <typename InDataType, typename WeiDataType, typename OutDataType>
-void get_device_conv_bwd_weight_op_ptr(
-    InDataType, WeiDataType, OutDataType, std::vector<DeviceConvndBwdWeightNoOpPtr>&, int)
-{
-    std::cout << "can not find device conv bwd weight" << std::endl;
-    exit(1);
-}
-
-template <>
-void get_device_conv_bwd_weight_op_ptr(
-    F32, F32, F32, std::vector<DeviceConvndBwdWeightNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f32_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f32_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f32_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <>
-void get_device_conv_bwd_weight_op_ptr(
-    F16, F16, F16, std::vector<DeviceConvndBwdWeightNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_f16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_convnd_bwd_weight_xdl_nhwc_kyxc_nhwk_f16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_f16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <>
-void get_device_conv_bwd_weight_op_ptr(
-    BF16, BF16, BF16, std::vector<DeviceConvndBwdWeightNoOpPtr>& conv_ptrs, int num_dim_spatial)
-{
-    switch(num_dim_spatial)
-    {
-    case 1:
-        ck::tensor_operation::device::instance::
-            add_device_conv1d_bwd_weight_xdl_nwc_kxc_nwk_bf16_instances(conv_ptrs);
-        break;
-    case 2:
-        ck::tensor_operation::device::instance::
-            add_device_conv2d_bwd_weight_xdl_nhwc_kyxc_nhwk_bf16_instances(conv_ptrs);
-        break;
-    case 3:
-        ck::tensor_operation::device::instance::
-            add_device_conv3d_bwd_weight_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(conv_ptrs);
-        break;
-    default: break;
-    }
-}
-
-template <typename DataType>
-void show_data_nhwc_layout(Tensor<DataType>& nhwc)
-{
-    std::cout << "[";
-    for(int n = 0; n < ck::type_convert<int>(nhwc.mDesc.GetLengths()[0]); n++)
-    {
-        std::cout << "[";
-        for(int hi = 0; hi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[2]); hi++)
-        {
-            std::cout << "[";
-            for(int wi = 0; wi < ck::type_convert<int>(nhwc.mDesc.GetLengths()[3]); wi++)
-            {
-                std::cout << "[";
-                for(int c = 0; c < ck::type_convert<int>(nhwc.mDesc.GetLengths()[1]); c++)
-                {
-                    std::cout << static_cast<float>(nhwc(n, c, hi, wi)) << "  ";
-                }
-                std::cout << "]";
-            }
-            std::cout << "]";
-        }
-        std::cout << "]";
-    }
-    std::cout << "]";
-}
-
-template <int NDimSpatial,
-          typename InDataType,
-          typename WeiDataType,
-          typename OutDataType,
-          typename InLayout,
-          typename WeiLayout,
-          typename OutLayout>
-bool profile_convnd_bwd_weight_impl(int do_verification,
-                                    int init_method,
-                                    bool do_log,
-                                    bool time_kernel,
-                                    ck::index_t N,
-                                    ck::index_t K,
-                                    ck::index_t C,
-                                    std::vector<ck::index_t> input_spatial_lengths,
-                                    std::vector<ck::index_t> filter_spatial_lengths,
-                                    std::vector<ck::index_t> output_spatial_lengths,
-                                    std::vector<ck::index_t> conv_filter_strides,
-                                    std::vector<ck::index_t> conv_filter_dilations,
-                                    std::vector<ck::index_t> input_left_pads,
-                                    std::vector<ck::index_t> input_right_pads,
-                                    ck::index_t split_k)
-{
-    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
-    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
-
-    const auto in_element_op  = InElementOp{};
-    const auto wei_element_op = WeiElementOp{};
-    const auto out_element_op = OutElementOp{};
-
-    std::vector<std::size_t> input_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(C)};
-    input_dims.insert(
-        std::end(input_dims), std::begin(input_spatial_lengths), std::end(input_spatial_lengths));
-
-    std::vector<std::size_t> filter_dims{static_cast<std::size_t>(K), static_cast<std::size_t>(C)};
-    filter_dims.insert(std::end(filter_dims),
-                       std::begin(filter_spatial_lengths),
-                       std::end(filter_spatial_lengths));
-
-    std::vector<std::size_t> output_dims{static_cast<std::size_t>(N), static_cast<std::size_t>(K)};
-    output_dims.insert(std::end(output_dims),
-                       std::begin(output_spatial_lengths),
-                       std::end(output_spatial_lengths));
-
-    Tensor<InDataType> input(get_input_host_tensor_descriptor<InLayout>(input_dims, NDimSpatial));
-    Tensor<WeiDataType> weights_host_result(
-        get_filters_host_tensor_descriptor<WeiLayout>(filter_dims, NDimSpatial));
-    Tensor<WeiDataType> weights_device_result(
-        get_filters_host_tensor_descriptor<WeiLayout>(filter_dims, NDimSpatial));
-    Tensor<OutDataType> output(
-        get_output_host_ensor_descriptor<OutLayout>(output_dims, NDimSpatial));
-
-    std::cout << "input: " << input.mDesc << std::endl;
-    std::cout << "weights: " << weights_host_result.mDesc << std::endl;
-    std::cout << "output: " << output.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        input.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-2, 2});
-        output.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
-        break;
-    default:
-        input.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
-        output.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
-    }
-
-    DeviceMem in_device_buf(sizeof(InDataType) * input.mDesc.GetElementSpace());
-    DeviceMem wei_device_buf(sizeof(WeiDataType) * weights_device_result.mDesc.GetElementSpace());
-    DeviceMem out_device_buf(sizeof(OutDataType) * output.mDesc.GetElementSpace());
-
-    in_device_buf.ToDevice(input.mData.data());
-    out_device_buf.ToDevice(output.mData.data());
-
-    // reset input to zero
-    wei_device_buf.SetZero();
-
-    if(do_verification)
-    {
-        auto RunReference = [&](auto& ref_conv) {
-            auto ref_invoker = ref_conv.MakeInvoker();
-
-            auto ref_argument = ref_conv.MakeArgument(input,
-                                                      weights_host_result,
-                                                      output,
-                                                      conv_filter_strides,
-                                                      conv_filter_dilations,
-                                                      input_left_pads,
-                                                      input_right_pads,
-                                                      InElementOp{},
-                                                      WeiElementOp{},
-                                                      OutElementOp{});
-            ref_invoker.Run(ref_argument);
-        };
-
-        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdWeight<InDataType,
-                                                                           WeiDataType,
-                                                                           OutDataType,
-                                                                           InElementOp,
-                                                                           WeiElementOp,
-                                                                           OutElementOp,
-                                                                           NDimSpatial>();
-        RunReference(ref_conv);
-    }
-
-    // add device Conv instances
-    std::vector<DeviceConvndBwdWeightNoOpPtr> conv_ptrs;
-    get_device_conv_bwd_weight_op_ptr(
-        InDataType{}, WeiDataType{}, OutDataType{}, conv_ptrs, NDimSpatial);
-
-    if(conv_ptrs.size() <= 0)
-    {
-        throw std::runtime_error("wrong! no device Conv instance found");
-    }
-
-    std::string best_conv_name;
-    float best_ave_time   = 0;
-    float best_tflops     = 0;
-    float best_gb_per_sec = 0;
-
-    // profile device Conv instances
-    bool success = true;
-    for(auto& conv_ptr : conv_ptrs)
-    {
-        // using atomic, so need to reset input, setzero is done in invoker
-        // if(split_k > 1)
-        //{
-        //    wei_device_buf.SetZero();
-        //}
-
-        auto argument_ptr = conv_ptr->MakeArgumentPointer(
-            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-            static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
-            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-            N,
-            K,
-            C,
-            input_spatial_lengths,
-            filter_spatial_lengths,
-            output_spatial_lengths,
-            conv_filter_strides,
-            conv_filter_dilations,
-            input_left_pads,
-            input_right_pads,
-            in_element_op,
-            wei_element_op,
-            out_element_op,
-            split_k);
-
-        if(!conv_ptr->IsSupportedArgument(argument_ptr.get()))
-        {
-            std::cout << "wrong! device_conv with the specified compilation parameters does "
-                         "not support this Conv problem"
-                      << std::endl;
-            continue;
-        }
-
-        auto invoker_ptr      = conv_ptr->MakeInvokerPointer();
-        std::string conv_name = conv_ptr->GetTypeString();
-        float ave_time        = 0;
-
-        if(std::is_same<InDataType, ck::bhalf_t>::value && split_k > 1)
-        {
-            // alloc work space
-            size_t bwd_weight_workspace_size = conv_ptr->GetWorkSpaceSize(argument_ptr.get());
-            if(bwd_weight_workspace_size <= 0)
-            {
-                printf("wrong work space size\n");
-                exit(1);
-            }
-            DeviceMem wei_work_space_device_buf(bwd_weight_workspace_size);
-            wei_work_space_device_buf.SetZero();
-            conv_ptr->SetWorkSpacePointer(argument_ptr.get(),
-                                          wei_work_space_device_buf.GetDeviceBuffer());
-            ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-        }
-        else
-        {
-            ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-        }
-
-        std::size_t flop =
-            ck::utils::conv::get_flops(N, C, K, filter_spatial_lengths, output_spatial_lengths);
-        std::size_t num_btype = ck::utils::conv::get_btype<InDataType, WeiDataType, OutDataType>(
-            N, C, K, input_spatial_lengths, filter_spatial_lengths, output_spatial_lengths);
-
-        float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
-        float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                  << " GB/s" << std::endl;
-
-        if(tflops > best_tflops)
-        {
-            best_conv_name  = conv_name;
-            best_tflops     = tflops;
-            best_ave_time   = ave_time;
-            best_gb_per_sec = gb_per_sec;
-        }
-
-        if(do_verification)
-        {
-            wei_device_buf.FromDevice(weights_device_result.mData.data());
-
-            success = ck::utils::check_err(weights_host_result, weights_device_result);
-
-            if(success == false)
-            {
-                std::cout << "Fail Info: " << conv_ptr->GetTypeString() << std::endl;
-            }
-            else
-            {
-                std::cout << "Pass Info: " << conv_ptr->GetTypeString() << std::endl;
-            }
-
-            if(do_log)
-            {
-                std::cout << "in : ";
-                show_data_nhwc_layout(output);
-                std::cout << std::endl;
-
-                std::cout << "wei: ";
-                show_data_nhwc_layout(weights_host_result);
-                std::cout << std::endl;
-
-                std::cout << "out  : ";
-                show_data_nhwc_layout(input);
-                std::cout << std::endl;
-
-                std::cout << "wei_device: ";
-                show_data_nhwc_layout(weights_device_result);
-                std::cout << std::endl;
-            }
-        }
-    }
-
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-              << best_gb_per_sec << " GB/s, " << best_conv_name << std::endl;
-    return success;
-}
-
-} // namespace profiler
-} // namespace ck

From c2d7a29dec6a9a6f05195fba561d8bcb4b305c3d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 30 May 2023 14:07:17 +0200
Subject: [PATCH 019/194] Add instances for fp16/int8 Gemm kernels (Navi21)
 (#717)

* Add instances for fp16/int8 Gemm kernels (Navi21)

* Extend instances with smaller tiles

* Fix SrcVectorTensor for km_kn_mn int8
---
 .../tensor_operation_instance/gpu/gemm.hpp    | 48 +++++++++++
 .../gpu/gemm/CMakeLists.txt                   |  8 ++
 ..._gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp | 34 +++++++-
 ...16_f16_f16_km_kn_mn_irregular_instance.cpp | 73 +++++++++++++++++
 ..._gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp | 34 +++++++-
 ...16_f16_f16_km_nk_mn_irregular_instance.cpp | 73 +++++++++++++++++
 ..._gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp | 34 +++++++-
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp | 73 +++++++++++++++++
 ..._gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp | 34 +++++++-
 ...16_f16_f16_mk_nk_mn_irregular_instance.cpp | 74 +++++++++++++++++
 ...ice_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp | 35 +++++++-
 ...l_i8_i8_i8_km_kn_mn_irregular_instance.cpp | 80 +++++++++++++++++++
 ...ice_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp | 35 +++++++-
 ...l_i8_i8_i8_km_nk_mn_irregular_instance.cpp | 80 +++++++++++++++++++
 ...ice_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp | 35 +++++++-
 ...l_i8_i8_i8_mk_kn_mn_irregular_instance.cpp | 80 +++++++++++++++++++
 ...ice_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp | 35 +++++++-
 ...l_i8_i8_i8_mk_nk_mn_irregular_instance.cpp | 80 +++++++++++++++++++
 18 files changed, 937 insertions(+), 8 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp

diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 0a8f2215ba..732d980699 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -22,21 +22,41 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -63,21 +83,41 @@ void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -295,6 +335,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
@@ -302,6 +343,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
             }
@@ -310,6 +352,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
@@ -317,6 +360,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
@@ -352,24 +396,28 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
             }
         }
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index e20d592c84..d66010af73 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -33,11 +33,19 @@ add_instance_library(device_gemm_instance
    device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
    device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
    device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
index 5d2f18e14e..35df85b7af 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..d444e29aa3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
index 01e3b3793a..7d0863c953 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..f8f8a0bd33
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64       
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
index 804e86a066..c1b11b19df 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -34,7 +34,39 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..840a4fabe1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
index 159fa90f74..9a889e8d55 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -35,7 +35,39 @@ using device_gemm_dl_f16_f16_f16_mk_nk_mn_instances =
         //  ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         //  ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         //  ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..82b1b5dc29
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances =
+    std::tuple<
+        // clang-format off
+        //  ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        //  ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        //  ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        //  ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        // clang-format on
+        >;
+
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
index a4208245e5..9f5cebcab0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_km_kn_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..0439201511
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 4, 4>,      S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,      S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
index 06fab7f68a..41afb519f1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_km_nk_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..350834f7e5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
index b6d72fa221..d1173095fc 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_mk_kn_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..27397527bc
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
index 67d2e3ce4b..efda345a8d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -31,7 +31,40 @@ using device_gemm_dl_i8_i8_i8_mk_nk_mn_instances = std::tuple<
         // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,       S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,               5,                  4>
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..b99f3f2b63
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances = std::tuple<
+    // clang-format off
+        // #########|  AData|   BData|   CData|    AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+        // #########|   Type|    Type|    Type|       Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+        // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+        // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 70e4eb567fcad81c57598ab9ee6f81b4136ecca5 Mon Sep 17 00:00:00 2001
From: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Date: Tue, 30 May 2023 14:09:06 +0200
Subject: [PATCH 020/194] Multiple fixes to GroupedGemm+SplitK (#707)

* Add license header.

* Reduce number of logged output. Add constant initialization.

* Add functional tests for grouped_gemm with different kbatch value.

* Add debug log informations + remove unused code.

* Don't pass kbatch to CalculateKPadded.

* Turn on logging in grouped gemm and gemm splitk profiler

* Debug: limit number of test cases to run;

* Log more information and initialize with constant value.

* Turn on DEBUG_LOG

* Add more debug log informations.

* Limit the number of instances to compile.

* Use GridwiseGemmPipeline

* Use KBatch to calculate K0

* Multiple DebugLog messages.

* Unit tests for multiple KBatch values.

* Refactoring

* Disable logging
* extract out of if statement KBatch update.

* Uncomment instances.

* Disable DebugLog.

* Use Kbatch when calculate KPadded.

* Fix CGridDesc padding.

* Use available helper functions.

* Uncomment code commented for debuggin.

* Remove unnecessary debug log messages.

* Uncomment previously commented code for debug purposes.

* Add KBatch info to profiler output summary log.

* Add gtests for gemm splitk using ckProfiler API.

* Add more test-cases for different data layout.

* Add more test cases for gemm splitk

* Remove old test.

* Unit tests for MKNK ggemm interface.

* Fix and add more unit-tests.

* Constepxr everything!

* Increase error threshold for fp16 and splitk.

Since we're using fp16 atomic add for splitk there's a
known precision loss.

---------

Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../gpu/device/device_grouped_gemm.hpp        |   4 +
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |  14 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp |  28 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  | 180 +++++++-----
 ...gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp |   1 +
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |   6 +-
 ...16_f16_f16_mk_nk_mn_irregular_instance.cpp |   4 +-
 .../profiler/profile_gemm_splitk_impl.hpp     |   6 +-
 .../profiler/profile_grouped_gemm_impl.hpp    | 120 +++++---
 test/gemm_split_k/CMakeLists.txt              |   5 +-
 test/gemm_split_k/gemm_split_k.cpp            | 261 ------------------
 test/gemm_split_k/test_gemm_splitk.cpp        |  66 +++++
 .../test_gemm_splitk_ut_cases.inc             | 217 +++++++++++++++
 test/gemm_split_k/test_gemm_splitk_util.hpp   |  78 ++++++
 test/grouped_gemm/CMakeLists.txt              |  10 +-
 test/grouped_gemm/grouped_gemm_fp16.cpp       |  69 -----
 .../test_grouped_gemm_interface.cpp           | 202 ++++++++++++++
 .../grouped_gemm/test_grouped_gemm_splitk.cpp |  34 +++
 .../test_grouped_gemm_ut_cases.inc            | 180 ++++++++++++
 test/grouped_gemm/test_grouped_gemm_util.hpp  | 249 +++++++++++++++++
 20 files changed, 1263 insertions(+), 471 deletions(-)
 delete mode 100644 test/gemm_split_k/gemm_split_k.cpp
 create mode 100644 test/gemm_split_k/test_gemm_splitk.cpp
 create mode 100644 test/gemm_split_k/test_gemm_splitk_ut_cases.inc
 create mode 100644 test/gemm_split_k/test_gemm_splitk_util.hpp
 delete mode 100644 test/grouped_gemm/grouped_gemm_fp16.cpp
 create mode 100644 test/grouped_gemm/test_grouped_gemm_interface.cpp
 create mode 100644 test/grouped_gemm/test_grouped_gemm_splitk.cpp
 create mode 100644 test/grouped_gemm/test_grouped_gemm_ut_cases.inc
 create mode 100644 test/grouped_gemm/test_grouped_gemm_util.hpp

diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
index 4b1106c122..1e03405536 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm.hpp
@@ -1,4 +1,8 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
 #pragma once
+
 #include <iostream>
 #include <vector>
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 776f96e8e6..89bfc180a5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -73,6 +73,11 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     static constexpr auto I2 = Number<2>{};
     static constexpr auto I3 = Number<3>{};
 
+    // TODO: should be exposed as Tparams.
+    static constexpr index_t NumGemmKPrefetchStage = 1;
+    static constexpr LoopScheduler LoopSched       = make_default_loop_scheduler();
+    static constexpr PipelineVersion PipelineVer   = PipelineVersion::v2;
+
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
         ADataType, // TODO: distinguish A/B datatype
@@ -85,6 +90,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         BElementwiseOperation,
         CElementwiseOperation,
         GemmSpec,
+        NumGemmKPrefetchStage,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -112,7 +118,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         CShuffleMRepeatPerShuffle,
         CShuffleNRepeatPerShuffle,
         CBlockTransferScalarPerVector_NWaveNPerXDL,
-        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        LoopSched,
+        PipelineVer>;
 
     using Argument              = typename GridwiseGemm::Argument;
     using DefaultBlock2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
@@ -257,7 +265,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         StrideC,
                         GridwiseGemm::CalculateMPadded(M),
                         GridwiseGemm::CalculateNPadded(N),
-                        GridwiseGemm::CalculateKPadded(K),
+                        GridwiseGemm::CalculateKPadded(K, KBatch),
                         GridwiseGemm::CalculateK0(K, KBatch),
                         KBatch};
     }
@@ -290,7 +298,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                           StrideC,
                                           GridwiseGemm::CalculateMPadded(M),
                                           GridwiseGemm::CalculateNPadded(N),
-                                          GridwiseGemm::CalculateKPadded(K),
+                                          GridwiseGemm::CalculateKPadded(K, KBatch),
                                           GridwiseGemm::CalculateK0(K, KBatch),
                                           KBatch);
     }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 467a8429ab..54ad9eb065 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -85,7 +85,7 @@ template <typename ALayout,
           typename BElementwiseOperation,
           typename CDEElementwiseOperation,
           GemmSpecialization GemmSpec,
-          ck::index_t NumPrefetch,
+          ck::index_t NumGemmKPrefetchStage,
           ck::index_t BlockSize,
           ck::index_t MPerBlock,
           ck::index_t NPerBlock,
@@ -152,6 +152,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         BElementwiseOperation,
         CDEElementwiseOperation,
         GemmSpec,
+        NumGemmKPrefetchStage,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -179,7 +180,9 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         CShuffleMXdlPerWavePerShuffle,
         CShuffleNXdlPerWavePerShuffle,
         CDEBlockTransferScalarPerVector_NPerBlock,
-        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        LoopSched,
+        PipelineVersion::v2>;
 
     using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
     using Block2ETileMapKSplit =
@@ -265,8 +268,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
                 const index_t k0       = GridwiseGemm::CalculateK0(K, K_BATCH);
 
-                const auto c_grid_desc_m_n =
-                    GridwiseGemm::MakeCGridDescriptor_M_N(M, N, m_padded, n_padded, stride_c);
+                const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
 
                 const auto local_b2c_tile_map =
                     Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
@@ -319,8 +321,8 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
                 const index_t k0       = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
 
-                const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(
-                    karg.M, karg.N, karg.MPadded, karg.NPadded, karg.StrideC);
+                const auto c_grid_desc_m_n =
+                    GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
 
                 const auto local_b2c_tile_map =
                     Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
@@ -501,6 +503,11 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
             arg.skipped_group_count_) != arg.group_count_)
         {
+#if DEBUG_LOG
+            std::cout << "The group count is not equal to sum of skipped groups "
+                         "and kernel args size!"
+                      << std::endl;
+#endif // DEBUG_LOG
             return false;
         }
 
@@ -509,14 +516,15 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         {
             const auto& a        = arg.gemm_kernel_args_[i].karg_;
             bool group_arg_valid = GridwiseGemm::CheckValidity(a);
-#if DEBUG_LOG
             if(not group_arg_valid)
             {
-                std::cout << "[" << __func__ << "] group id: " << i << " is not supported!\n";
+#if DEBUG_LOG
+                std::cout << "[" << __func__ << "] group id: " << i
+                          << " has invalid GridwiseGemm settings!" << std::endl;
                 a.Print();
-            }
 #endif // DEBUG_LOG
-            supported &= group_arg_valid;
+            }
+            supported = supported && group_arg_valid;
         }
         return supported;
     }
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index b393c4897e..d56d1986e2 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -8,14 +8,14 @@
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
 #include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
 #include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 
 namespace ck {
 
@@ -55,6 +55,7 @@ template <index_t BlockSize,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
           tensor_operation::device::GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
           index_t MPerBlock,
           index_t NPerBlock,
           index_t K0PerBlock,
@@ -82,7 +83,9 @@ template <index_t BlockSize,
           index_t CShuffleMRepeatPerShuffle,
           index_t CShuffleNRepeatPerShuffle,
           index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
-          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>
+          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
 struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 {
     static constexpr auto I0 = Number<0>{};
@@ -99,8 +102,15 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     static constexpr auto M01 = 1;
     static constexpr auto N01 = 1;
 
+    static constexpr auto gemm_padder =
+        tensor_operation::device::GemmPadder<GemmSpec, index_t, index_t, index_t>{
+            MPerBlock, NPerBlock, K1* K0PerBlock};
+
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    using GridwiseGemmPipe = remove_cvref_t<decltype(
+        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
     struct Argument : public ck::tensor_operation::device::BaseArgument
     {
         const FloatAB* p_a_grid;
@@ -176,12 +186,12 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     // prefer this to be called on host
     __host__ __device__ static auto CalculateMPadded(index_t M)
     {
-        return (M + MPerBlock - 1) / MPerBlock * MPerBlock;
+        return math::integer_least_multiple(M, MPerBlock);
     }
 
     __host__ __device__ static auto CalculateNPadded(index_t N)
     {
-        return (N + NPerBlock - 1) / NPerBlock * NPerBlock;
+        return math::integer_least_multiple(N, NPerBlock);
     }
 
     __host__ __device__ static auto CalculateK0(index_t K, index_t K_Batch = 1)
@@ -295,8 +305,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
     }
 
-    __host__ __device__ static auto
-    MakeCGridDescriptor_M_N(index_t M, index_t N, index_t MPad, index_t NPad, index_t StrideC)
+    __host__ __device__ static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
     {
         const auto c_grid_desc_m_n = [&]() {
             if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
@@ -309,22 +318,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             }
         }();
 
-        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
-        {
-            return transform_tensor_descriptor(c_grid_desc_m_n,
-                                               make_tuple(make_right_pad_transform(M, MPad - M),
-                                                          make_right_pad_transform(N, NPad - N)),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
+        return gemm_padder.PadCDescriptor_M_N(c_grid_desc_m_n);
     }
 
     __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
@@ -383,7 +377,15 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
         {
             if(!(karg.M % MPerBlock == 0))
+            {
+#if DEBUG_LOG
+                std::cout << "Arg M value is not a multiple of MPerBlock! M: " << karg.M << " "
+                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
+                          << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
                        GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
@@ -391,40 +393,116 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                        GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
         {
             if(!(karg.N % NPerBlock == 0))
+            {
+#if DEBUG_LOG
+                std::cout << "Arg N value is not a multiple of NPerBlock! N: " << karg.N << " "
+                          << __FILE__ << ":" << __LINE__ << ", in function: " << __func__
+                          << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
 
         if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
         {
             if(karg.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg K (" << karg.K
+                          << ") value is not a multiple of ABlockTransferSrcScalarPerVector ("
+                          << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         else
         {
             if(karg.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg M (" << karg.M
+                          << ") value is not a multiple of ABlockTransferSrcScalarPerVector ("
+                          << ABlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
 
         if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
         {
             if(karg.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg N (" << karg.N
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         else
         {
             if(karg.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+#if DEBUG_LOG
+                std::cout << "Arg K (" << karg.K
+                          << ") value is not a multiple of BBlockTransferSrcScalarPerVector ("
+                          << BBlockTransferSrcScalarPerVector << " )! " << __FILE__ << ":"
+                          << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
 
         if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
         {
             if(karg.N % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+            {
+#if DEBUG_LOG
+                std::cout
+                    << "Arg N (" << karg.N
+                    << ") value is not a multiple of CBlockTransferScalarPerVector_NWaveNPerXDL ("
+                    << CBlockTransferScalarPerVector_NWaveNPerXDL << " )! " << __FILE__ << ":"
+                    << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
         }
         else
         {
             if(karg.M % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+            {
+#if DEBUG_LOG
+                std::cout
+                    << "Arg M (" << karg.M
+                    << ") value is not a multiple of CBlockTransferScalarPerVector_NWaveNPerXDL ("
+                    << CBlockTransferScalarPerVector_NWaveNPerXDL << " )! " << __FILE__ << ":"
+                    << __LINE__ << ", in function: " << __func__ << std::endl;
+
+#endif // DEBUG_LOG
                 return false;
+            }
+        }
+
+        const auto num_k_loop = karg.K0 / K0PerBlock;
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+#if DEBUG_LOG
+            std::cout << "The number of k loops (" << num_k_loop
+                      << ") value is not supported by GridwiseGemm Pipeline."
+                      << " K0: " << karg.K0 << ", K0PerBlock: " << K0PerBlock << " " << __FILE__
+                      << ":" << __LINE__ << ", in function: " << __func__ << std::endl;
+#endif // DEBUG_LOG
+            return false;
         }
 
         return true;
@@ -439,9 +517,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
     __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
     {
-        const bool has_main_k0_block_loop = K0 > K0PerBlock;
-
-        return has_main_k0_block_loop;
+        const index_t num_loop = K0 / K0PerBlock;
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
     template <typename CGridDesc>
@@ -490,7 +567,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         return BlockToCTileMap_3DGrid_KSplit<MPerBlock, NPerBlock>();
     }
 
-    using CGridDesc_M_N         = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1, 1, 1))>;
+    using CGridDesc_M_N         = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1))>;
     using DefaultBlock2CTileMap = remove_cvref_t<decltype(MakeDefaultBlock2CTileMap())>;
 
     template <bool HasMainKBlockLoop,
@@ -507,8 +584,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
         const auto b_b_k0_n_k1_grid_desc = MakeBGridDescriptor_KBatch_K0_N_K1(
             karg.K, karg.NPadded, karg.N, karg.StrideB, karg.k_batch, karg.K0, karg.KPadded);
-        const auto c_grid_desc_m_n =
-            MakeCGridDescriptor_M_N(karg.M, karg.N, karg.MPadded, karg.NPadded, karg.StrideC);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
 
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
@@ -680,20 +756,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-#if 1
-        auto blockwise_gemm =
-            BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatAB,
-                                                                FloatAcc,
-                                                                decltype(a_k0_m_k1_block_desc),
-                                                                decltype(b_k0_n_k1_block_desc),
-                                                                MPerXDL,
-                                                                NPerXDL,
-                                                                MRepeat,
-                                                                NRepeat,
-                                                                K1>{};
-#else
-        auto blockwise_gemm = BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatAB,
             FloatAcc,
@@ -703,9 +767,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             NPerXDL,
             MRepeat,
             NRepeat,
-            K1>{};
-
-#endif
+            K1,
+            LoopSched>();
 
         auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
 
@@ -761,7 +824,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                 b_blockwise_copy.RunWrite(b_b_k0_n_k1_block_desc, b_block_buf);
 
                 k0_block_data_begin += K0PerBlock;
-            } while(k0_block_data_begin < (K0 - K0PerBlock));
+            } while(k0_block_data_begin < (karg.K0 - K0PerBlock));
         }
 
         // tail
@@ -772,13 +835,12 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
 #else
         // gridwise GEMM pipeline
-        const auto gridwise_gemm_pipeline =
-            GridwiseGemmPipeline_Selector<PipelineVersion::v2, 1, LoopScheduler::Default>();
-
         const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
             (a_b_k0_m_k1_grid_desc.GetLength(I1) * a_b_k0_m_k1_grid_desc.GetLength(I3)) /
             (K0PerBlock * K1));
 
+        const auto gridwise_gemm_pipeline = GridwiseGemmPipe{};
+
         gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_b_k0_m_k1_grid_desc,
                                                                a_b_k0_m_k1_block_desc,
                                                                a_blockwise_copy,
@@ -993,24 +1055,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
     }
 
-    template <typename Layout>
-    struct LStr
-    {
-        static std::string Get() { return ""; }
-    };
-
-    template <>
-    struct LStr<ck::tensor_layout::gemm::RowMajor>
-    {
-        static std::string Get() { return "R"; }
-    };
-
-    template <>
-    struct LStr<ck::tensor_layout::gemm::ColumnMajor>
-    {
-        static std::string Get() { return "C"; }
-    };
-
     static std::string GetTypeString()
     {
         auto str = std::stringstream();
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index a93cb7fc84..5f5d6c9b5a 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -64,6 +64,7 @@ using device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = st
         //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index 0385b0fc0c..a3d73440eb 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -44,14 +44,14 @@ using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instanc
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
index 5933ff61ec..dddfa2aa44 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -37,7 +37,7 @@ using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instanc
         //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
@@ -45,7 +45,7 @@ using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instanc
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-//      DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index 4cc62509d7..ab1bce258a 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -246,9 +246,9 @@ bool profile_gemm_splitk_impl(int do_verification,
     }
 
     std::cout << " M = " << M << " N = " << N << " K = " << K << " StrideA = " << StrideA
-              << " StrideB = " << StrideB << " StrideC = " << StrideC << " : " << best_ave_time
-              << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec << " GB/s, "
-              << best_op_name << std::endl;
+              << " StrideB = " << StrideB << " StrideC = " << StrideC << " KBatch = " << KBatch
+              << " : " << best_ave_time << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec
+              << " GB/s, " << best_op_name << std::endl;
 
     return pass;
 }
diff --git a/profiler/include/profiler/profile_grouped_gemm_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
index 23dca244dc..9abb5e7a53 100644
--- a/profiler/include/profiler/profile_grouped_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
@@ -19,6 +19,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/utility/literals.hpp"
+#include "ck/library/utility/fill.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
 
 namespace ck {
@@ -43,7 +44,6 @@ bool profile_grouped_gemm_impl(int do_verification,
                                const std::vector<int>& StrideCs,
                                int kbatch = 1)
 {
-
     bool pass = true;
 
     auto f_host_tensor_descriptor =
@@ -81,11 +81,11 @@ bool profile_grouped_gemm_impl(int do_verification,
 
         c_m_n_device_results.push_back(
             Tensor<CDataType>(f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{})));
-
+#if DEBUG_LOG
         std::cout << "group: " << i << " a_m_k[" << i << "]:" << a_m_k[i].mDesc << ", b_k_n[" << i
                   << "]:" << b_k_n[i].mDesc << ", c_m_n_device_results[" << i
                   << "]:" << c_m_n_device_results[i].mDesc << std::endl;
-
+#endif // DEBUG_LOG
         std::size_t num_thread = 1;
         switch(init_method)
         {
@@ -191,65 +191,71 @@ bool profile_grouped_gemm_impl(int do_verification,
         DeviceMem gemm_desc_workspace(gemm_ptr->GetWorkSpaceSize(argument_ptr.get()));
 
         gemm_ptr->SetWorkSpacePointer(argument_ptr.get(), gemm_desc_workspace.GetDeviceBuffer());
+        std::string gemm_name = gemm_ptr->GetTypeString();
+
+        if(kbatch > 1)
+        {
+            using DeviceOpSplitK =
+                ck::tensor_operation::device::DeviceGroupedGemmSplitK<ALayout,
+                                                                      BLayout,
+                                                                      ck::Tuple<>,
+                                                                      CLayout,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      ck::Tuple<>,
+                                                                      CDataType,
+                                                                      AElementOp,
+                                                                      BElementOp,
+                                                                      CElementOp>;
+
+            if(dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get()) != nullptr)
+            {
+                dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get())
+                    ->SetKBatchSize(argument_ptr.get(), kbatch);
+            }
+        }
 
         if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
         {
-            std::string gemm_name = gemm_ptr->GetTypeString();
-
-            if(kbatch > 1)
-            {
-                using DeviceOpSplitK =
-                    ck::tensor_operation::device::DeviceGroupedGemmSplitK<ALayout,
-                                                                          BLayout,
-                                                                          ck::Tuple<>,
-                                                                          CLayout,
-                                                                          ADataType,
-                                                                          BDataType,
-                                                                          ck::Tuple<>,
-                                                                          CDataType,
-                                                                          AElementOp,
-                                                                          BElementOp,
-                                                                          CElementOp>;
-
-                if(dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get()) != nullptr)
-                {
-                    dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get())
-                        ->SetKBatchSize(argument_ptr.get(), kbatch);
-                }
-            }
 
             float ave_time =
                 invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
-            std::size_t flop = 0, num_btype = 0;
-            for(std::size_t i = 0; i < gemm_descs.size(); i++)
+            if(time_kernel)
             {
-                flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
+                std::size_t flop = 0, num_btype = 0;
+                for(std::size_t i = 0; i < gemm_descs.size(); i++)
+                {
+                    flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
 
-                num_btype += sizeof(ADataType) * Ms[i] * Ks[i] + sizeof(BDataType) * Ks[i] * Ns[i] +
-                             sizeof(CDataType) * Ms[i] * Ns[i];
-            }
+                    num_btype += sizeof(ADataType) * Ms[i] * Ks[i] +
+                                 sizeof(BDataType) * Ks[i] * Ns[i] +
+                                 sizeof(CDataType) * Ms[i] * Ns[i];
+                }
 
-            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+                float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
 
-            float gb_per_sec = num_btype / 1.E6 / ave_time;
-            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
-                      << gb_per_sec << " GB/s, " << gemm_name << std::endl;
+                float gb_per_sec = num_btype / 1.E6 / ave_time;
+                std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
+                          << " TFlops, " << gb_per_sec << " GB/s, " << gemm_name << std::endl;
 
-            if(tflops > best_tflops)
-            {
-                best_gemm_name  = gemm_name;
-                best_tflops     = tflops;
-                best_ave_time   = ave_time;
-                best_gb_per_sec = gb_per_sec;
+                if(tflops > best_tflops)
+                {
+                    best_gemm_name  = gemm_name;
+                    best_tflops     = tflops;
+                    best_ave_time   = ave_time;
+                    best_gb_per_sec = gb_per_sec;
+                }
             }
 
             if(do_verification)
             {
+                bool instance_pass = true;
                 for(std::size_t i = 0; i < gemm_descs.size(); i++)
                 {
 
                     c_device_buf[i]->FromDevice(c_m_n_device_results[i].mData.data());
+                    c_device_buf[i]->SetZero();
 
                     Tensor<CDataType> c_m_n_host_result(
                         f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{}));
@@ -274,7 +280,20 @@ bool profile_grouped_gemm_impl(int do_verification,
                                                               c_element_op);
 
                     ref_invoker.Run(ref_argument);
-                    pass = pass && ck::utils::check_err(c_m_n_device_results[i], c_m_n_host_result);
+                    if(std::is_same_v<CDataType, ck::half_t> && kbatch > 1)
+                    {
+                        instance_pass =
+                            instance_pass && ck::utils::check_err(c_m_n_device_results[i],
+                                                                  c_m_n_host_result,
+                                                                  "Error: Incorrect results!",
+                                                                  0.06);
+                    }
+                    else
+                    {
+                        instance_pass =
+                            instance_pass &&
+                            ck::utils::check_err(c_m_n_device_results[i], c_m_n_host_result);
+                    }
 
                     if(do_log)
                     {
@@ -289,16 +308,25 @@ bool profile_grouped_gemm_impl(int do_verification,
                             << std::endl;
                     }
                 }
+
+                std::cout << "Instance: " << gemm_name << " verification "
+                          << (instance_pass ? "SUCCEED" : "FAILED") << std::endl;
+
+                pass = pass && instance_pass;
             }
         }
         else
         {
-            std::cout << "does not support this GEMM problem" << std::endl;
+            std::cout << "Instance: " << gemm_name << ", does not support this GEMM problem"
+                      << std::endl;
         }
     }
 
-    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-              << best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
+    if(time_kernel)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+                  << best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
+    }
 
     return pass;
 }
diff --git a/test/gemm_split_k/CMakeLists.txt b/test/gemm_split_k/CMakeLists.txt
index 09bbf79389..2274854f88 100644
--- a/test/gemm_split_k/CMakeLists.txt
+++ b/test/gemm_split_k/CMakeLists.txt
@@ -1,5 +1,4 @@
 if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
-   add_test_executable(test_gemm_split_k gemm_split_k.cpp)
-   target_link_libraries(test_gemm_split_k PRIVATE utility)
-   target_link_libraries(test_gemm_split_k PRIVATE device_gemm_splitk_instance)
+   add_gtest_executable(test_gemm_splitk test_gemm_splitk.cpp)
+   target_link_libraries(test_gemm_splitk PRIVATE utility device_gemm_splitk_instance)
 endif()
diff --git a/test/gemm_split_k/gemm_split_k.cpp b/test/gemm_split_k/gemm_split_k.cpp
deleted file mode 100644
index 1edb5769c6..0000000000
--- a/test/gemm_split_k/gemm_split_k.cpp
+++ /dev/null
@@ -1,261 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <initializer_list>
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp"
-
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-
-#include "ck/library/utility/host_gemm.hpp"
-
-enum struct GemmMatrixLayout
-{
-    MK_KN_MN, // 0
-    MK_NK_MN, // 1
-    KM_KN_MN, // 2
-    KM_NK_MN, // 3
-};
-
-template <typename T>
-static bool check_out(const Tensor<T>& ref, const Tensor<T>& result)
-{
-    float max_diff = 1e-6;
-
-    for(std::size_t i = 0; i < ref.mData.size(); ++i)
-    {
-        float diff = std::abs(double(ref.mData[i]) - double(result.mData[i]));
-        if(max_diff < diff)
-        {
-            return false;
-        }
-    }
-
-    return true;
-}
-
-struct gemmArgs
-{
-    GemmMatrixLayout layout;
-    int M;
-    int N;
-    int K;
-    int StrideA;
-    int StrideB;
-    int StrideC;
-    int KBatch;
-};
-
-int test_gemm(const gemmArgs& args)
-{
-    using Row = ck::tensor_layout::gemm::RowMajor;
-    using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-    bool a_row_major, b_row_major, c_row_major;
-
-    switch(args.layout)
-    {
-    case GemmMatrixLayout::MK_KN_MN:
-        a_row_major = true;
-        b_row_major = true;
-        c_row_major = true;
-        break;
-    case GemmMatrixLayout::MK_NK_MN:
-        a_row_major = true;
-        b_row_major = false;
-        c_row_major = true;
-        break;
-    case GemmMatrixLayout::KM_KN_MN:
-        a_row_major = false;
-        b_row_major = true;
-        c_row_major = true;
-        break;
-    case GemmMatrixLayout::KM_NK_MN:
-        a_row_major = false;
-        b_row_major = false;
-        c_row_major = true;
-        break;
-    default: printf("not supported layout"); return 1;
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, bool row_major) {
-            using namespace ck::literals;
-
-            if(row_major)
-            {
-                return HostTensorDescriptor({row, col}, {stride, 1_uz});
-            }
-            else
-            {
-                return HostTensorDescriptor({row, col}, {1_uz, stride});
-            }
-        };
-
-    Tensor<float> a_m_k(f_host_tensor_descriptor(args.M, args.K, args.StrideA, a_row_major));
-    Tensor<float> b_k_n(f_host_tensor_descriptor(args.K, args.N, args.StrideB, b_row_major));
-    Tensor<float> c_m_n_host_result(
-        f_host_tensor_descriptor(args.M, args.N, args.StrideC, c_row_major));
-    Tensor<float> c_m_n_device_result(
-        f_host_tensor_descriptor(args.M, args.N, args.StrideC, c_row_major));
-
-    // init data
-    std::size_t num_thread = 1;
-    a_m_k.GenerateTensorValue(GeneratorTensor_2<float>{-5, 5}, num_thread);
-    b_k_n.GenerateTensorValue(GeneratorTensor_2<float>{-5, 5}, num_thread);
-    // set zero to c_device_buf
-    c_m_n_device_result.GenerateTensorValue(GeneratorTensor_0<float>{}, num_thread);
-
-    host_gemm_mk_kn_mn(a_m_k,
-                       b_k_n,
-                       c_m_n_host_result,
-                       ck::tensor_operation::element_wise::PassThrough{},
-                       ck::tensor_operation::element_wise::PassThrough{},
-                       ck::tensor_operation::element_wise::PassThrough{});
-
-    DeviceMem a_device_buf(sizeof(float) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(float) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem c_device_buf(sizeof(float) * c_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_m_k.mData.data());
-    b_device_buf.ToDevice(b_k_n.mData.data());
-    c_device_buf.ToDevice(c_m_n_device_result.mData.data());
-
-    auto test = [&](auto a_layout, auto b_layout, auto c_layout) {
-        bool success = false;
-
-        using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<decltype(a_layout),
-                                                                        decltype(b_layout),
-                                                                        decltype(c_layout),
-                                                                        float,
-                                                                        float,
-                                                                        float,
-                                                                        PassThrough,
-                                                                        PassThrough,
-                                                                        PassThrough>;
-
-        const auto gemm_ptrs =
-            ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
-                DeviceOp>::GetInstances();
-
-        for(auto& gemm_ptr : gemm_ptrs)
-        {
-            auto argument_ptr =
-                gemm_ptr->MakeArgumentPointer(static_cast<float*>(a_device_buf.GetDeviceBuffer()),
-                                              static_cast<float*>(b_device_buf.GetDeviceBuffer()),
-                                              static_cast<float*>(c_device_buf.GetDeviceBuffer()),
-                                              args.M,
-                                              args.N,
-                                              args.K,
-                                              args.StrideA,
-                                              args.StrideB,
-                                              args.StrideC,
-                                              ck::tensor_operation::element_wise::PassThrough{},
-                                              ck::tensor_operation::element_wise::PassThrough{},
-                                              ck::tensor_operation::element_wise::PassThrough{},
-                                              args.KBatch);
-
-            auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
-
-            if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
-            {
-                invoker_ptr->Run(argument_ptr.get());
-
-                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-                if(!check_out(c_m_n_host_result, c_m_n_device_result))
-                {
-                    success = false;
-                    break;
-                }
-                success = true;
-            }
-        }
-
-        return success;
-    };
-
-    bool success = false;
-
-    if(args.layout == GemmMatrixLayout::MK_KN_MN)
-    {
-        success = test(Row{}, Row{}, Row{});
-    }
-    else if(args.layout == GemmMatrixLayout::MK_NK_MN)
-    {
-        success = test(Row{}, Col{}, Row{});
-    }
-    else if(args.layout == GemmMatrixLayout::KM_KN_MN)
-    {
-        success = test(Col{}, Row{}, Row{});
-    }
-    else
-    {
-        success = test(Col{}, Col{}, Row{});
-    }
-
-    auto error_code = 0;
-    if(success)
-    {
-        std::cout << "test split k : Pass" << std::endl;
-    }
-    else
-    {
-        std::cout << "test split k: Fail " << std::endl;
-        error_code = -1; // test needs to report failure
-    }
-    return error_code;
-}
-
-int main(int argc, char* argv[])
-{
-    std::vector<gemmArgs> test_cases;
-    if(argc == 1)
-    {
-        test_cases = {{GemmMatrixLayout::MK_KN_MN, 1024, 1024, 1024, 1024, 1024, 1024, 2},
-                      {GemmMatrixLayout::MK_KN_MN, 1024, 1024, 1024, 1024, 1024, 1024, 8}};
-    }
-    else if(argc == 9)
-    {
-        const auto layout = static_cast<GemmMatrixLayout>(std::stoi(argv[1]));
-
-        const int M = std::stoi(argv[2]);
-        const int N = std::stoi(argv[3]);
-        const int K = std::stoi(argv[4]);
-
-        const int StrideA = std::stoi(argv[5]);
-        const int StrideB = std::stoi(argv[6]);
-        const int StrideC = std::stoi(argv[7]);
-        const int KBatch  = std::stoi(argv[8]);
-        test_cases        = {{layout, M, N, K, StrideA, StrideB, StrideC, KBatch}};
-    }
-    else
-    {
-        printf("arg1: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
-        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
-        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
-        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
-        printf("arg2 to 7: M, N, K, StrideA, StrideB, StrideC KBatch\n");
-        return -1;
-    }
-    bool error = false;
-    for(const auto& kinder : test_cases)
-    {
-        error |= test_gemm(kinder);
-    }
-    return error ? 1 : 0;
-}
diff --git a/test/gemm_split_k/test_gemm_splitk.cpp b/test/gemm_split_k/test_gemm_splitk.cpp
new file mode 100644
index 0000000000..9eba5bba37
--- /dev/null
+++ b/test/gemm_split_k/test_gemm_splitk.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+
+#include "gtest/gtest.h"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "test_gemm_splitk_util.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+namespace {
+
+template <typename X, typename Y>
+struct tuple_concat;
+
+template <typename... Xs, typename... Ys>
+struct tuple_concat<std::tuple<Xs...>, std::tuple<Ys...>>
+{
+    using type = std::tuple<Xs..., Ys...>;
+};
+
+} // namespace
+
+template <typename Tuple>
+class TestGemmSplitK_MK_KN
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Row, Row>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmSplitK_MK_NK
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Row, Col>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmSplitK_KM_KN
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Col, Row>, Tuple>::type>
+{
+};
+
+template <typename Tuple>
+class TestGemmSplitK_KM_NK
+    : public ck::test::TestGemmSplitK<typename tuple_concat<std::tuple<Col, Col>, Tuple>::type>
+{
+};
+
+// clang-format off
+using KernelTypes = ::testing::Types<
+    //         ADataType, BDataType, CDataType
+    std::tuple<      F16,       F16,       F16>,
+    std::tuple<      F32,       F32,       F32>
+    >;
+// clang-format on
+
+TYPED_TEST_SUITE(TestGemmSplitK_MK_KN, KernelTypes);
+TYPED_TEST_SUITE(TestGemmSplitK_MK_NK, KernelTypes);
+TYPED_TEST_SUITE(TestGemmSplitK_KM_KN, KernelTypes);
+TYPED_TEST_SUITE(TestGemmSplitK_KM_NK, KernelTypes);
+
+#include "test_gemm_splitk_ut_cases.inc"
diff --git a/test/gemm_split_k/test_gemm_splitk_ut_cases.inc b/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
new file mode 100644
index 0000000000..54b9c6c9e3
--- /dev/null
+++ b/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
@@ -0,0 +1,217 @@
+#pragma once
+
+TYPED_TEST(TestGemmSplitK_MK_KN, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, SmallM)
+{
+    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_KN, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, MidLargeM)
+{
+    std::vector<int> Ms{127, 255, 312, 799, 1573};
+    constexpr int N = 512;
+    constexpr int K = 320;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_KN, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, PaddK)
+{
+    std::vector<int> Ms{127};
+    constexpr int N = 512;
+    constexpr int K = 437;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_KN, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_MK_NK, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideA = K;
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, StrideA, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_KN, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideB = N;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
+
+TYPED_TEST(TestGemmSplitK_KM_NK, Regular)
+{
+    std::vector<int> Ms{512};
+    constexpr int N = 512;
+    constexpr int K = 512;
+
+    constexpr int StrideB = K;
+    constexpr int StrideC = N;
+
+    for(int M : Ms)
+        this->Run(M, N, K, M, StrideB, StrideC);
+}
diff --git a/test/gemm_split_k/test_gemm_splitk_util.hpp b/test/gemm_split_k/test_gemm_splitk_util.hpp
new file mode 100644
index 0000000000..8243747a69
--- /dev/null
+++ b/test/gemm_split_k/test_gemm_splitk_util.hpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <string>
+#include <sstream>
+#include <tuple>
+#include <vector>
+#include <gtest/gtest.h>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "include/ck/utility/data_type.hpp"
+#include "profiler/profile_gemm_splitk_impl.hpp"
+
+namespace ck {
+namespace test {
+
+template <typename Tuple>
+class TestGemmSplitK : public testing::Test
+{
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using F32 = float;
+
+    protected:
+    using ALayout   = std::tuple_element_t<0, Tuple>;
+    using BLayout   = std::tuple_element_t<1, Tuple>;
+    using CLayout   = Row;
+    using ADataType = std::tuple_element_t<2, Tuple>;
+    using BDataType = std::tuple_element_t<3, Tuple>;
+    using CDataType = std::tuple_element_t<4, Tuple>;
+
+    public:
+    static constexpr bool verify_     = true;
+    static constexpr int init_method_ = 1; // decimal value initialization
+    static constexpr bool log_        = false;
+    static constexpr bool bench_      = false; // measure kernel performance
+    std::vector<int> k_batches_;
+
+    void SetUp() override { k_batches_ = {1, 2, 3, 5, 8}; }
+
+    void Run(const int M,
+             const int N,
+             const int K,
+             const int StrideA,
+             const int StrideB,
+             const int StrideC)
+    {
+        for(auto kb : k_batches_)
+        {
+            RunSingle(M, N, K, StrideA, StrideB, StrideC, kb);
+        }
+    }
+
+    void RunSingle(const int M,
+                   const int N,
+                   const int K,
+                   const int StrideA,
+                   const int StrideB,
+                   const int StrideC,
+                   int kbatch = 1)
+    {
+        bool pass = ck::profiler::profile_gemm_splitk_impl<ADataType,
+                                                           BDataType,
+                                                           F32,
+                                                           CDataType,
+                                                           ALayout,
+                                                           BLayout,
+                                                           CLayout>(
+            verify_, init_method_, log_, bench_, M, N, K, StrideA, StrideB, StrideC, kbatch);
+        EXPECT_TRUE(pass);
+    }
+};
+
+} // namespace test
+} // namespace ck
diff --git a/test/grouped_gemm/CMakeLists.txt b/test/grouped_gemm/CMakeLists.txt
index a7619eac6e..40f634d8b3 100644
--- a/test/grouped_gemm/CMakeLists.txt
+++ b/test/grouped_gemm/CMakeLists.txt
@@ -1,5 +1,9 @@
 if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
-   add_test_executable(test_grouped_gemm_fp16 grouped_gemm_fp16.cpp)
-   target_link_libraries(test_grouped_gemm_fp16 PRIVATE utility)
-   target_link_libraries(test_grouped_gemm_fp16 PRIVATE device_grouped_gemm_instance)
+   add_custom_target(test_grouped_gemm)
+   add_gtest_executable(test_grouped_gemm_splitk test_grouped_gemm_splitk.cpp)
+   add_gtest_executable(test_grouped_gemm_interface test_grouped_gemm_interface.cpp)
+   target_link_libraries(test_grouped_gemm_splitk PRIVATE utility device_grouped_gemm_instance)
+   target_link_libraries(test_grouped_gemm_interface PRIVATE utility device_grouped_gemm_instance)
+   
+   add_dependencies(test_grouped_gemm test_grouped_gemm_splitk test_grouped_gemm_interface)
 endif()
diff --git a/test/grouped_gemm/grouped_gemm_fp16.cpp b/test/grouped_gemm/grouped_gemm_fp16.cpp
deleted file mode 100644
index f20d750d36..0000000000
--- a/test/grouped_gemm/grouped_gemm_fp16.cpp
+++ /dev/null
@@ -1,69 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <random>
-
-#include "profiler/profile_grouped_gemm_impl.hpp"
-
-namespace {
-
-using ADataType   = ck::half_t;
-using BDataType   = ck::half_t;
-using CDataType   = ck::half_t;
-using AccDataType = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <typename ALayout, typename BLayout, typename CLayout>
-bool TestGroupedGemm()
-{
-
-    std::mt19937 gen(19391);
-    std::uniform_int_distribution<> distrib(1, 10);
-    int group_count = distrib(gen);
-
-    // GEMM shape
-    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
-    std::vector<const void*> p_a, p_b;
-    std::vector<void*> p_c;
-
-    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideCs;
-
-    for(int i = 0; i < group_count; i++)
-    {
-        Ms.push_back(256 + 256 * distrib(gen));
-        Ns.push_back(256 + 256 * distrib(gen));
-        Ks.push_back(128 + 128 * distrib(gen));
-
-        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
-        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
-        StrideCs.push_back(std::is_same<Row, CLayout>::value ? Ns[i] : Ms[i]);
-    }
-
-    return ck::profiler::profile_grouped_gemm_impl<ADataType,
-                                                   BDataType,
-                                                   CDataType,
-                                                   AccDataType,
-                                                   ALayout,
-                                                   BLayout,
-                                                   CLayout>(
-        true, 1, false, 1, Ms, Ns, Ks, StrideAs, StrideBs, StrideCs);
-}
-
-} // anonymous namespace
-
-int main()
-{
-    bool res = true;
-
-    res = res && TestGroupedGemm<Row, Row, Row>();
-    res = res && TestGroupedGemm<Row, Col, Row>();
-    res = res && TestGroupedGemm<Col, Row, Row>();
-    res = res && TestGroupedGemm<Col, Col, Row>();
-
-    std::cout << "TestGroupedGemm ..... " << (res ? "SUCCESS" : "FAILURE") << std::endl;
-
-    return res ? 0 : 1;
-}
diff --git a/test/grouped_gemm/test_grouped_gemm_interface.cpp b/test/grouped_gemm/test_grouped_gemm_interface.cpp
new file mode 100644
index 0000000000..ffa8840fc7
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_interface.cpp
@@ -0,0 +1,202 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <stdexcept>
+#include <vector>
+#include "gtest/gtest.h"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "test_grouped_gemm_util.hpp"
+
+class TestGGemmSplitKInterface_MKNKMN : public ::testing::Test
+{
+    protected:
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using ALayout = Row;
+    using BLayout = Col;
+    using ELayout = Row;
+
+    static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+    template <ck::tensor_operation::device::GemmSpecialization GemmSpec,
+              ck::index_t KPerBlock,
+              ck::index_t K1,
+              ck::index_t ABlockTransferSrcScalarPerVector,
+              ck::index_t BBlockTransferSrcScalarPerVector,
+              ck::index_t CDEBlockTransferScalarPerVector_NPerBlock>
+    using GGemmInstance =
+        ck::test::DeviceGroupedGemmSplitkInstanceWrapper<ALayout,
+                                                         BLayout,
+                                                         ELayout,
+                                                         GemmSpec,
+                                                         KPerBlock,
+                                                         K1,
+                                                         ABlockTransferSrcScalarPerVector,
+                                                         BBlockTransferSrcScalarPerVector,
+                                                         CDEBlockTransferScalarPerVector_NPerBlock>;
+
+    using DefaultGGemmInstance = GGemmInstance<GemmDefault, 32, 8, 4, 8, 8>;
+};
+
+TEST_F(TestGGemmSplitKInterface_MKNKMN, TileSize)
+{
+    std::vector<int> Ms{128, 256, 188, 512};
+    constexpr int N = 256;
+    constexpr int K = 128;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // M % MPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ms = std::vector<int>{256, 128, 128, 512};
+    Ns = std::vector<int>{256, 177, 128, 512};
+    // N % NPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
+
+TEST_F(TestGGemmSplitKInterface_MKNKMN, VectorLoadWidth)
+{
+    static constexpr auto GemmMNKPadding =
+        ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+    using PaddedGGemmInstance = GGemmInstance<GemmMNKPadding, 32, 8, 4, 8, 8>;
+
+    std::vector<int> Ms{128, 256, 256, 512};
+    constexpr int N = 256;
+    constexpr int K = 512;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // K % ABlockTransferSrcScalarPerVector
+    Ks = std::vector<int>{256, 177, 128, 512};
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ks = std::vector<int>{256, 164, 128, 512};
+    // K % BBlockTransferSrcScalarPerVector
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ks = std::vector<int>(4, 128);
+    Ns = std::vector<int>{256, 127, 128, 512};
+    // N % CBlockTransferScalarPerVector_NWaveNPerXDL
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
+
+TEST_F(TestGGemmSplitKInterface_MKNKMN, KLoops)
+{
+    std::vector<int> Ms{128, 256, 256, 512};
+    constexpr int N      = 256;
+    constexpr int K      = 128;
+    constexpr int kbatch = 4;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // kloops % 2
+    Ks = std::vector<int>{256, 512, 320, 768};
+    EXPECT_FALSE(
+        DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));
+
+    // Not all gemms have same value for main_k0_block_loop!
+    Ks = std::vector<int>{256, 512, 512, 512};
+    EXPECT_THROW(DefaultGGemmInstance{}.Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch),
+                 std::runtime_error);
+}
+
+class TestGGemmSplitKInterface_KMKNNM : public ::testing::Test
+{
+    protected:
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using ALayout = Col;
+    using BLayout = Row;
+    using ELayout = Col;
+
+    static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+    template <ck::tensor_operation::device::GemmSpecialization GemmSpec,
+              ck::index_t KPerBlock,
+              ck::index_t K1,
+              ck::index_t ABlockTransferSrcScalarPerVector,
+              ck::index_t BBlockTransferSrcScalarPerVector,
+              ck::index_t CDEBlockTransferScalarPerVector_NPerBlock>
+    using GGemmInstance =
+        ck::test::DeviceGroupedGemmSplitkInstanceWrapper<ALayout,
+                                                         BLayout,
+                                                         ELayout,
+                                                         GemmSpec,
+                                                         KPerBlock,
+                                                         K1,
+                                                         ABlockTransferSrcScalarPerVector,
+                                                         BBlockTransferSrcScalarPerVector,
+                                                         CDEBlockTransferScalarPerVector_NPerBlock>;
+
+    using DefaultGGemmInstance = GGemmInstance<GemmDefault, 32, 8, 4, 8, 4>;
+};
+
+TEST_F(TestGGemmSplitKInterface_KMKNNM, TileSize)
+{
+    std::vector<int> Ms{128, 256, 188, 512};
+    constexpr int N = 256;
+    constexpr int K = 128;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // M % MPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ms = std::vector<int>{128, 256, 256, 512};
+    Ns = std::vector<int>{256, 177, 128, 512};
+    // N % NPerBlock
+    EXPECT_FALSE(DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
+
+TEST_F(TestGGemmSplitKInterface_KMKNNM, VectorLoadWidth)
+{
+    static constexpr auto GemmMNKPadding =
+        ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+    using PaddedGGemmInstance = GGemmInstance<GemmMNKPadding, 32, 8, 2, 8, 4>;
+
+    std::vector<int> Ms{128, 256, 256, 512};
+    constexpr int N = 256;
+    constexpr int K = 512;
+
+    std::vector<int> Ns(Ms.size(), N);
+    std::vector<int> Ks(Ms.size(), K);
+    std::vector<int> StrideAs(Ms.size(), K);
+    std::vector<int> StrideBs(Ms.size(), K);
+    std::vector<int> StrideCs(Ms.size(), N);
+
+    // M % ABlockTransferSrcScalarPerVector
+    Ms = std::vector<int>{256, 177, 128, 512};
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ms = std::vector<int>{128, 256, 256, 512};
+    Ns = std::vector<int>{256, 164, 128, 512};
+    // N % BBlockTransferSrcScalarPerVector
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+
+    Ns = std::vector<int>{128, 256, 256, 512};
+    Ms = std::vector<int>{256, 130, 128, 512};
+    // M % CBlockTransferScalarPerVector_NWaveNPerXDL
+    EXPECT_FALSE(PaddedGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs));
+}
diff --git a/test/grouped_gemm/test_grouped_gemm_splitk.cpp b/test/grouped_gemm/test_grouped_gemm_splitk.cpp
new file mode 100644
index 0000000000..d9282fa924
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_splitk.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <tuple>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "gtest/gtest.h"
+#include "test_grouped_gemm_util.hpp"
+
+using F16 = ck::half_t;
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using RRR_F16_F16_F16 = ck::test::TestGroupedGemm<std::tuple<Row, Row, Row, F16, F16, F16>>;
+using RCR_F16_F16_F16 = ck::test::TestGroupedGemm<std::tuple<Row, Col, Row, F16, F16, F16>>;
+
+using RRR_F16_F16_F16_LargeK = ck::test::TestGroupedGemm<std::tuple<Row, Row, Row, F16, F16, F16>>;
+using RCR_F16_F16_F16_LargeK = ck::test::TestGroupedGemm<std::tuple<Row, Col, Row, F16, F16, F16>>;
+
+const std::vector<int> KBATCH{1, 2, 3, 5, 8};
+
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_MK_KN, RRR_F16_F16_F16, testing::ValuesIn(KBATCH));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_MK_NK, RCR_F16_F16_F16, testing::ValuesIn(KBATCH));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_LargeK_MK_KN,
+                         RRR_F16_F16_F16_LargeK,
+                         testing::Values(32, 64));
+INSTANTIATE_TEST_SUITE_P(TestGroupedGemm_splitk_LargeK_MK_NK,
+                         RCR_F16_F16_F16_LargeK,
+                         testing::Values(32, 64));
+
+#include "test_grouped_gemm_ut_cases.inc"
diff --git a/test/grouped_gemm/test_grouped_gemm_ut_cases.inc b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
new file mode 100644
index 0000000000..d94d140d97
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
@@ -0,0 +1,180 @@
+#pragma once
+
+TEST_P(RRR_F16_F16_F16, TinyCases)
+{
+    const std::vector<int> Ms{0, 1};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, SmallCases)
+{
+    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, MidCases)
+{
+    const std::vector<int> Ms{167, 183, 177, 153, 139, 204};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, Regular)
+{
+    const std::vector<int> Ms{64, 128, 256};
+    constexpr int N = 768;
+    constexpr int K = 320;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16, MNKPadded)
+{
+    const std::vector<int> Ms{127, 150, 188, 210};
+    constexpr int N = 136;
+    constexpr int K = 280;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, TinyCases)
+{
+    const std::vector<int> Ms{0, 1};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, SmallCases)
+{
+    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, MidCases)
+{
+    const std::vector<int> Ms{167, 183, 177, 153, 139, 204};
+    constexpr int N = 768;
+    constexpr int K = 544;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, Regular)
+{
+    const std::vector<int> Ms{32, 64, 128, 256};
+    constexpr int N = 768;
+    constexpr int K = 320;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16, MNKPadded)
+{
+    const std::vector<int> Ms{127, 150, 188, 210};
+    constexpr int N = 136;
+    constexpr int K = 280;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RRR_F16_F16_F16_LargeK, TestLargeKBatch)
+{
+    const std::vector<int> Ms{188, 210};
+    constexpr int N = 768;
+    constexpr int K = 4096;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), N);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
+
+TEST_P(RCR_F16_F16_F16_LargeK, TestLargeKBatch)
+{
+    const std::vector<int> Ms{188, 210};
+    constexpr int N = 768;
+    constexpr int K = 4096;
+
+    const std::vector<int> Ns(Ms.size(), N);
+    const std::vector<int> Ks(Ms.size(), K);
+    const std::vector<int> StrideAs(Ms.size(), K);
+    const std::vector<int> StrideBs(Ms.size(), K);
+    const std::vector<int> StrideCs(Ms.size(), N);
+
+    this->Run(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, this->GetParam());
+}
diff --git a/test/grouped_gemm/test_grouped_gemm_util.hpp b/test/grouped_gemm/test_grouped_gemm_util.hpp
new file mode 100644
index 0000000000..b61118b512
--- /dev/null
+++ b/test/grouped_gemm/test_grouped_gemm_util.hpp
@@ -0,0 +1,249 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+#include <string>
+#include <sstream>
+#include <tuple>
+#include <vector>
+#include <gtest/gtest.h>
+
+#include "ck/ck.hpp"
+#include "ck/stream_config.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/sequence.hpp"
+#include "ck/utility/tuple.hpp"
+#include "ck/utility/number.hpp"
+#include "profiler/profile_grouped_gemm_impl.hpp"
+
+namespace ck {
+namespace test {
+
+template <typename Range>
+std::string serialize_range(const Range& range)
+{
+    std::stringstream ss;
+    for(auto& r : range)
+    {
+        ss << r << ", ";
+    }
+    std::string str = ss.str();
+    return std::string(str.begin(), str.end() - 2);
+}
+
+template <typename Tuple>
+class TestGroupedGemm : public testing::TestWithParam<int>
+{
+    protected:
+    using ALayout   = std::tuple_element_t<0, Tuple>;
+    using BLayout   = std::tuple_element_t<1, Tuple>;
+    using ELayout   = std::tuple_element_t<2, Tuple>;
+    using ADataType = std::tuple_element_t<3, Tuple>;
+    using BDataType = std::tuple_element_t<4, Tuple>;
+    using EDataType = std::tuple_element_t<5, Tuple>;
+
+    public:
+    static constexpr bool verify_     = true;
+    static constexpr int init_method_ = 1; // decimal value initialization
+    static constexpr bool log_        = false;
+    static constexpr bool bench_      = false; // measure kernel performance
+
+    void SetUp() override {}
+
+    void Run(const std::vector<int>& Ms,
+             const std::vector<int>& Ns,
+             const std::vector<int>& Ks,
+             const std::vector<int>& StrideAs,
+             const std::vector<int>& StrideBs,
+             const std::vector<int>& StrideCs,
+             int kbatch = 1)
+    {
+        bool pass = ck::profiler::profile_grouped_gemm_impl<ADataType,
+                                                            BDataType,
+                                                            EDataType,
+                                                            float,
+                                                            ALayout,
+                                                            BLayout,
+                                                            ELayout>(
+            verify_, init_method_, log_, bench_, Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch);
+        EXPECT_TRUE(pass);
+    }
+};
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          tensor_operation::device::GemmSpecialization GemmSpec,
+          ck::index_t KPerBlock,
+          ck::index_t K1,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock>
+struct DeviceGroupedGemmSplitkInstanceWrapper
+{
+    using F16         = half_t;
+    using F32         = float;
+    using Row         = ck::tensor_layout::gemm::RowMajor;
+    using Col         = ck::tensor_layout::gemm::ColumnMajor;
+    using PassThrough = tensor_operation::element_wise::PassThrough;
+
+    using EmptyTuple = ck::Tuple<>;
+
+    template <ck::index_t... Is>
+    using S = ck::Sequence<Is...>;
+
+    template <ck::index_t N>
+    using I = ck::Number<N>;
+
+    using ABlockTransferThreadClusterArrageOrder =
+        std::conditional_t<std::is_same_v<ALayout, Row>, S<0, 2, 1, 3>, S<0, 1, 3, 2>>;
+    using ABlockTransferSrcAccessOrder =
+        std::conditional_t<std::is_same_v<ALayout, Row>, S<0, 2, 1, 3>, S<0, 1, 3, 2>>;
+    using ABlockTransferSrcVectorDim = std::conditional_t<std::is_same_v<ALayout, Row>, I<3>, I<2>>;
+    using ABlockTransferDstScalarPerVector_K1 =
+        std::conditional_t<std::is_same_v<ALayout, Row>, I<8>, I<2>>;
+    using ABlockLdsAddExtraM = std::conditional_t<std::is_same_v<ALayout, Row>, I<1>, I<0>>;
+
+    using BBlockTransferThreadClusterArrageOrder =
+        std::conditional_t<std::is_same_v<BLayout, Row>, S<0, 1, 3, 2>, S<0, 2, 1, 3>>;
+    using BBlockTransferSrcAccessOrder =
+        std::conditional_t<std::is_same_v<BLayout, Row>, S<0, 1, 3, 2>, S<0, 2, 1, 3>>;
+    using BBlockTransferSrcVectorDim = std::conditional_t<std::is_same_v<BLayout, Row>, I<2>, I<3>>;
+    using BBlockTransferDstScalarPerVector_K1 =
+        std::conditional_t<std::is_same_v<ALayout, Row>, I<2>, I<8>>;
+    using BBlockLdsAddExtraM = std::conditional_t<std::is_same_v<ALayout, Row>, I<0>, I<1>>;
+
+    using DeviceGroupedGemmSplitKInstance =
+        tensor_operation::device::DeviceGroupedGemmXdlSplitKCShuffle<
+            ALayout,
+            BLayout,
+            EmptyTuple,
+            ELayout,
+            F16,
+            F16,
+            F32,
+            F16,
+            EmptyTuple,
+            F16,
+            PassThrough,
+            PassThrough,
+            PassThrough,
+            GemmSpec,
+            1,
+            128,
+            128,
+            128,
+            KPerBlock,
+            K1,
+            K1,
+            32,
+            32,
+            4,
+            2,
+            S<1, 4, 32, 1>,
+            ABlockTransferThreadClusterArrageOrder,
+            ABlockTransferSrcAccessOrder,
+            ABlockTransferSrcVectorDim::value,
+            ABlockTransferSrcScalarPerVector,
+            ABlockTransferDstScalarPerVector_K1::value,
+            ABlockLdsAddExtraM::value,
+            S<1, 4, 32, 1>,
+            BBlockTransferThreadClusterArrageOrder,
+            BBlockTransferSrcAccessOrder,
+            BBlockTransferSrcVectorDim::value,
+            BBlockTransferSrcScalarPerVector,
+            BBlockTransferDstScalarPerVector_K1::value,
+            BBlockLdsAddExtraM::value,
+            1,
+            1,
+            S<1, 16, 1, 8>,
+            CDEBlockTransferScalarPerVector_NPerBlock>;
+
+    bool IsSupported(const std::vector<int>& Ms,
+                     const std::vector<int>& Ns,
+                     const std::vector<int>& Ks,
+                     const std::vector<int>& StrideAs,
+                     const std::vector<int>& StrideBs,
+                     const std::vector<int>& StrideCs,
+                     int kbatch = 1) const
+    {
+        std::size_t n_groups = Ms.size();
+        EXPECT_TRUE(Ns.size() == n_groups && Ks.size() == n_groups && StrideAs.size() == n_groups &&
+                    StrideBs.size() == n_groups && StrideCs.size() == n_groups)
+            << "The number of groups is not consistent!";
+
+        std::vector<tensor_operation::device::GemmDesc> gemm_descs;
+
+        for(std::size_t i = 0; i < n_groups; ++i)
+        {
+            gemm_descs.push_back(tensor_operation::device::GemmDesc{
+                Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
+        }
+
+        std::vector<const void*> p_As(n_groups, nullptr);
+        std::vector<const void*> p_Bs(n_groups, nullptr);
+        std::vector<void*> p_Cs(n_groups, nullptr);
+        auto p_Ds = std::vector<std::array<const void*, 0>>{};
+
+        auto ggemm_instance = DeviceGroupedGemmSplitKInstance{};
+        auto argument       = ggemm_instance.MakeArgument(
+            p_As, p_Bs, p_Ds, p_Cs, gemm_descs, PassThrough{}, PassThrough{}, PassThrough{});
+        if(kbatch > 1)
+        {
+            ggemm_instance.SetKBatchSize(argument, kbatch);
+        }
+
+        return ggemm_instance.IsSupportedArgument(argument);
+    }
+
+    float Run(const std::vector<int>& Ms,
+              const std::vector<int>& Ns,
+              const std::vector<int>& Ks,
+              const std::vector<int>& StrideAs,
+              const std::vector<int>& StrideBs,
+              const std::vector<int>& StrideCs,
+              int kbatch = 1) const
+    {
+        std::size_t n_groups = Ms.size();
+        EXPECT_TRUE(Ns.size() == n_groups && Ks.size() == n_groups && StrideAs.size() == n_groups &&
+                    StrideBs.size() == n_groups && StrideCs.size() == n_groups)
+            << "The number of groups is not consistent!";
+
+        std::vector<tensor_operation::device::GemmDesc> gemm_descs;
+
+        for(std::size_t i = 0; i < n_groups; ++i)
+        {
+            gemm_descs.push_back(tensor_operation::device::GemmDesc{
+                Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
+        }
+
+        std::vector<const void*> p_As(n_groups, nullptr);
+        std::vector<const void*> p_Bs(n_groups, nullptr);
+        std::vector<void*> p_Cs(n_groups, nullptr);
+        auto p_Ds = std::vector<std::array<const void*, 0>>{};
+
+        auto ggemm_instance = DeviceGroupedGemmSplitKInstance{};
+        auto argument       = ggemm_instance.MakeArgument(
+            p_As, p_Bs, p_Ds, p_Cs, gemm_descs, PassThrough{}, PassThrough{}, PassThrough{});
+        if(kbatch > 1)
+        {
+            ggemm_instance.SetKBatchSize(argument, kbatch);
+        }
+
+        EXPECT_TRUE(ggemm_instance.IsSupportedArgument(argument));
+        auto invoker = ggemm_instance.MakeInvoker();
+        DeviceMem gemm_desc_workspace(ggemm_instance.GetWorkSpaceSize(&argument));
+        ggemm_instance.SetWorkSpacePointer(&argument, gemm_desc_workspace.GetDeviceBuffer());
+        return invoker.Run(argument, StreamConfig{nullptr, false});
+    }
+};
+
+} // namespace test
+} // namespace ck

From 1344a0f25b5da1b74267cbe3adf97b084c05708f Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Tue, 30 May 2023 20:09:55 +0800
Subject: [PATCH 021/194] Simplify kernel argument of device operator
 DeviceGemm_Xdl_CShuffle<> (#696)

* Remove M/N/KPad local variables

* Use M/N/KPad to name padded lengths

* Replace duplicated local variable by parameters

* Rename variables M/N/KRaw to M/N/K

* Move AK0/BK0 compute logic into GridwiseGemm

* Use macro to shorten code

* Move CalculateGridSize() logic into GridwiseGemm

* Add comment to credit the implementation source

* Reuse the existing implementation

* Remove no-longer used data members

* Remove elementwise-op objects from interfaces

* Reserve kernel arg as whole object in interfaces

* Remove redundant data member

* Make 3rd type parameter optional

* Remove unnesscary type parameters

* Remove no-longer used descriptor-creation methods

* Move kernel arg type definition into GridwiseGemm

* Add macro to switch between code sections

* Move argument field computing logic into device op side

* Make utility method 'static'

* Declare special methods

* Unify MakeArgument() usage

* Adapt the new GridwiseGemm interface

* Push-down class 'GridwiseGemm::Argument' fields

* Remove no-longer used methods

* Add unused parameters

* Force copying parameters in 'Embed' ctor

* Remove no-longer used descriptors

* Fallback change on BaseArgument

* Remove macro 'INTEGER_DIVIDE_CEIL'

* Make variable naming more consistent

* Make sure methods are only invoked on right place

* Remove tailing underscore in public attribute name

* Remove necessary methods

* Hide computing logic of derived attributes

* Make new 'Embed' ctor only available for device code

* Make sure 'Embed' type args are not references

* Move check for karg.K into CheckValidity()

* Remove more integer division logic form device code

* Undo changes on Embed

* Separate 'Problem' concept out from 'Argument'

* Share same name for kernel interfaces

* Reject unsupported argument

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../impl/device_cgemm_4gemm_xdl_cshuffle.hpp  | 751 +++++-------------
 .../device/impl/device_gemm_xdl_cshuffle.hpp  | 486 +-----------
 .../gpu/grid/block_to_ctile_map.hpp           |  68 +-
 .../grid/gridwise_gemm_xdl_cshuffle_v1.hpp    | 636 ++++++++++++---
 4 files changed, 802 insertions(+), 1139 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
index 29978458bb..0c9cce97fb 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
@@ -118,277 +118,11 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         return PadDescriptor_M_1d(desc_m, gridSize, blockSize);
     }
 
-    static auto MakeAGridDescriptor_AK0_M_AK1(index_t MRaw, index_t KRaw, index_t StrideA)
-    {
-        const auto a_grid_desc_mraw_kraw = [&]() {
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both M and K
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(M)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad M, but not K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_right_pad_transform(MRaw, MPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad K, but not M
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k = transform_tensor_descriptor(
-                a_grid_desc_mraw_kraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else
-        {
-            // not pad M or K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-    }
-
-    static auto MakeBGridDescriptor_BK0_N_BK1(index_t KRaw, index_t NRaw, index_t StrideB)
-    {
-        const auto b_grid_desc_nraw_kraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(I1, StrideB));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(StrideB, I1));
-            }
-        }();
-
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto NPad = N - NRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both N and K
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_right_pad_transform(NRaw, NPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(N)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad N, but not K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_right_pad_transform(NRaw, NPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad K, but not N
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k = transform_tensor_descriptor(
-                b_grid_desc_nraw_kraw,
-                make_tuple(make_pass_through_transform(NRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else
-        {
-            // not pad N or K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideC)
-    {
-        const auto c_grid_desc_mraw_nraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(I1, StrideC));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto NPad = N - NRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad M and N
-            return transform_tensor_descriptor(c_grid_desc_mraw_nraw,
-                                               make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                          make_right_pad_transform(NRaw, NPad)),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad M, but not N
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_right_pad_transform(MRaw, MPad), make_pass_through_transform(NRaw)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad N, but not M
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(NRaw, NPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-            // not pad M or N
-            return c_grid_desc_mraw_nraw;
-        }
-    }
-
-    using AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1(1, 1, 1));
-    using BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1(1, 1, 1));
-    using CGridDesc_M_N       = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-    using CGridDesc_M         = decltype(MakeDescriptor_M({1, 1}, {1, 1}, 1, 1));
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1<
+        ALayout,
+        BLayout,
+        CLayout,
         ADataType, // TODO: distinguish A/B datatype
         GemmAccDataType,
         CShuffleDataType,
@@ -396,10 +130,8 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_AK0_M_AK1,
-        BGridDesc_BK0_N_BK1,
-        CGridDesc_M_N,
         NumGemmKPrefetchStage,
         BlockSize,
         MPerBlock,
@@ -433,108 +165,82 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched>;
 
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid_real,
-                 const ADataType* p_a_grid_imag,
-                 const BDataType* p_b_grid_real,
-                 const BDataType* p_b_grid_imag,
-                 CDataType* p_c_grid_real,
-                 CDataType* p_c_grid_imag,
-                 CDataType* p_workspace,
-                 index_t MRaw,
-                 index_t NRaw,
-                 index_t KRaw,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_real_{p_a_grid_real},
-              p_a_grid_imag_{p_a_grid_imag},
-              p_b_grid_real_{p_b_grid_real},
-              p_b_grid_imag_{p_b_grid_imag},
-              p_c_grid_real_{p_c_grid_real},
-              p_c_grid_imag_{p_c_grid_imag},
-              p_aux_grid_{p_workspace},
-              a_grid_desc_ak0_m_ak1_{DeviceOp::MakeAGridDescriptor_AK0_M_AK1(MRaw, KRaw, StrideA)},
-              b_grid_desc_bk0_n_bk1_{DeviceOp::MakeBGridDescriptor_BK0_N_BK1(KRaw, NRaw, StrideB)},
-              c_grid_desc_m_n_{DeviceOp::MakeCGridDescriptor_M_N(MRaw, NRaw, StrideC)},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_)},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op}
-        {
-            if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
-                                           b_grid_desc_bk0_n_bk1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        c_grid_desc_m_n_);
-            }
+    using CGridDesc_M = decltype(MakeDescriptor_M({1, 1}, {1, 1}, 1, 1));
 
-            const index_t grid_size = block_2_ctile_map_.CalculateGridSize(c_grid_desc_m_n_);
+    // Argument
+    struct Argument : public tensor_operation::device::BaseArgument, public GridwiseGemm::Problem
+    {
+        using Problem = typename GridwiseGemm::Problem;
+
+        Argument(const ADataType* p_a_grid_real_,
+                 const ADataType* p_a_grid_imag_,
+                 const BDataType* p_b_grid_real_,
+                 const BDataType* p_b_grid_imag_,
+                 CDataType* p_c_grid_real_,
+                 CDataType* p_c_grid_imag_,
+                 CDataType* p_workspace,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid_real{p_a_grid_real_},
+              p_a_grid_imag{p_a_grid_imag_},
+              p_b_grid_real{p_b_grid_real_},
+              p_b_grid_imag{p_b_grid_imag_},
+              p_c_grid_real{p_c_grid_real_},
+              p_c_grid_imag{p_c_grid_imag_},
+              p_aux_grid{p_workspace}
+        {
+            const index_t grid_size = std::get<1>(GridwiseGemm::CalculateGridSize(M_, N_));
 
             if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
             {
-                c_grid_desc_m_ =
-                    DeviceOp::MakeDescriptor_M({MRaw, NRaw}, {StrideC, I1}, grid_size, BlockSize);
+                c_grid_desc_m =
+                    DeviceOp::MakeDescriptor_M({M_, N_}, {StrideC_, I1}, grid_size, BlockSize);
             }
             else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
             {
-                c_grid_desc_m_ =
-                    DeviceOp::MakeDescriptor_M({MRaw, NRaw}, {I1, StrideC}, grid_size, BlockSize);
+                c_grid_desc_m =
+                    DeviceOp::MakeDescriptor_M({M_, N_}, {I1, StrideC_}, grid_size, BlockSize);
             }
 
-            p_aux_2_grid_ = p_workspace + c_grid_desc_m_n_.GetElementSpaceSize();
+            p_aux_2_grid = p_workspace + GetCElementSpaceSize(M_, N_, StrideC_);
         }
 
         //  private:
-        const ADataType* p_a_grid_real_;
-        const ADataType* p_a_grid_imag_;
-        const BDataType* p_b_grid_real_;
-        const BDataType* p_b_grid_imag_;
-        CDataType* p_c_grid_real_;
-        CDataType* p_c_grid_imag_;
-        CDataType* p_aux_grid_;
-        CDataType* p_aux_2_grid_;
-        AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
-        BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_M c_grid_desc_m_;
-        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            c_grid_desc_mblock_mperblock_nblock_nperblock_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
+        const ADataType* p_a_grid_real;
+        const ADataType* p_a_grid_imag;
+        const BDataType* p_b_grid_real;
+        const BDataType* p_b_grid_imag;
+        CDataType* p_c_grid_real;
+        CDataType* p_c_grid_imag;
+        CDataType* p_aux_grid;
+        CDataType* p_aux_2_grid;
+        CGridDesc_M c_grid_desc_m;
     };
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceOp::Argument;
-
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                            arg.b_grid_desc_bk0_n_bk1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(stream_config.log_level_ > 0)
+            {
+                arg.Print();
+            }
+
+            if(!GridwiseGemm::CheckValidity(arg))
             {
                 throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
+            index_t gdx, gdy, gdz;
+            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(arg.M, arg.N);
 
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+            const auto K = GridwiseGemm::CalculateAK0(arg.K) * AK1;
 
             float ave_time = 0;
 
@@ -578,224 +284,148 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    true>;
+                const auto kernel =
+                    kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, ADataType, CDataType, true>;
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_grid,
+                                                   arg);
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_real = aux - aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     subtract_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_real_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_real),
                     Subtract{});
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_grid,
+                                                   arg);
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_imag = aux + aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     add_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_imag_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_imag),
                     Add{});
             }
             else
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    false>;
+                const auto kernel =
+                    kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, ADataType, CDataType, false>;
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_grid,
+                                                   arg);
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_real = aux - aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     subtract_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_real_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_real),
                     Subtract{});
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_real_,
-                                           arg.p_b_grid_imag_,
-                                           arg.p_aux_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_real,
+                                                   arg.p_b_grid_imag,
+                                                   arg.p_aux_grid,
+                                                   arg);
 
-                ave_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_imag_,
-                                           arg.p_b_grid_real_,
-                                           arg.p_aux_2_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(gdx, gdy, gdz),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.p_a_grid_imag,
+                                                   arg.p_b_grid_real,
+                                                   arg.p_aux_2_grid,
+                                                   arg);
 
                 // c_imag = aux + aux_2
                 ave_time += launch_and_time_kernel(
                     stream_config,
                     add_kernel,
-                    dim3(grid_size),
+                    dim3(gdx, gdy, gdz),
                     dim3(BlockSize),
                     0,
-                    make_tuple(arg.c_grid_desc_m_, arg.c_grid_desc_m_),
-                    make_tuple(arg.c_grid_desc_m_),
-                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid_),
-                               const_cast<const CDataType*>(arg.p_aux_2_grid_)),
-                    make_tuple(arg.p_c_grid_imag_),
+                    make_tuple(arg.c_grid_desc_m, arg.c_grid_desc_m),
+                    make_tuple(arg.c_grid_desc_m),
+                    make_tuple(const_cast<const CDataType*>(arg.p_aux_grid),
+                               const_cast<const CDataType*>(arg.p_aux_2_grid)),
+                    make_tuple(arg.p_c_grid_imag),
                     Add{});
             }
 
@@ -818,10 +448,7 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(arg);
     }
 
     // polymorphic
@@ -837,15 +464,15 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                              CDataType* p_c_real,
                              CDataType* p_c_imag,
                              CDataType* p_workspace,
-                             index_t MRaw,
-                             index_t NRaw,
-                             index_t KRaw,
+                             index_t M,
+                             index_t N,
+                             index_t K,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
     {
         return Argument{p_a_real,
                         p_a_imag,
@@ -854,15 +481,12 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                         p_c_real,
                         p_c_imag,
                         p_workspace,
-                        MRaw,
-                        NRaw,
-                        KRaw,
+                        M,
+                        N,
+                        K,
                         StrideA,
                         StrideB,
-                        StrideC,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+                        StrideC};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -875,15 +499,15 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                                                       void* p_c_real,
                                                       void* p_c_imag,
                                                       void* p_workspace,
-                                                      index_t MRaw,
-                                                      index_t NRaw,
-                                                      index_t KRaw,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op,
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation,
                                                       index_t /* KBatch */ = 1) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a_real),
@@ -893,15 +517,12 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
                                           static_cast<CDataType*>(p_c_real),
                                           static_cast<CDataType*>(p_c_imag),
                                           static_cast<CDataType*>(p_workspace),
-                                          MRaw,
-                                          NRaw,
-                                          KRaw,
+                                          M,
+                                          N,
+                                          K,
                                           StrideA,
                                           StrideB,
-                                          StrideC,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          StrideC);
     }
 
     // polymorphic
@@ -930,16 +551,22 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         return str.str();
     }
 
-    std::size_t GetWorkspaceSize(index_t MRaw,
-                                 index_t NRaw,
-                                 [[maybe_unused]] index_t KRaw,
+    static std::size_t GetCElementSpaceSize(index_t M, index_t N, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(
+            M, GridwiseGemm::CalculateMPadded(M), N, GridwiseGemm::CalculateNPadded(N), StrideC);
+
+        return c_grid_desc_m_n.GetElementSpaceSize();
+    }
+
+    std::size_t GetWorkspaceSize(index_t M,
+                                 index_t N,
+                                 [[maybe_unused]] index_t K,
                                  [[maybe_unused]] index_t StrideA,
                                  [[maybe_unused]] index_t StrideB,
                                  index_t StrideC) override
     {
-        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(MRaw, NRaw, StrideC);
-
-        return 2 * sizeof(CDataType) * c_grid_desc_m_n.GetElementSpaceSize();
+        return 2 * sizeof(CDataType) * GetCElementSpaceSize(M, N, StrideC);
     }
 };
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index 7cd0ff72e8..13a30911a1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -82,276 +82,11 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
 
-    static auto MakeAGridDescriptor_AK0_M_AK1(index_t MRaw, index_t KRaw, index_t StrideA)
-    {
-        const auto a_grid_desc_mraw_kraw = [&]() {
-            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
-                                                    make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both M and K
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(M)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad M, but not K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_right_pad_transform(MRaw, MPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad K, but not M
-            assert(K % AK1 == 0);
-
-            const auto AK0 = K / AK1;
-
-            const auto a_grid_desc_m_k = transform_tensor_descriptor(
-                a_grid_desc_mraw_kraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_m_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-        else
-        {
-            // not pad M or K
-            assert(KRaw % AK1 == 0);
-
-            const auto AK0 = KRaw / AK1;
-
-            const auto a_grid_desc_ak0_m_ak1 =
-                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
-                                                       make_pass_through_transform(MRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return a_grid_desc_ak0_m_ak1;
-        }
-    }
-
-    static auto MakeBGridDescriptor_BK0_N_BK1(index_t KRaw, index_t NRaw, index_t StrideB)
-    {
-        const auto b_grid_desc_nraw_kraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(I1, StrideB));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
-                                                    make_tuple(StrideB, I1));
-            }
-        }();
-
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-        const auto K = math::integer_divide_ceil(KRaw, KPerBlock) * KPerBlock;
-
-        const auto NPad = N - NRaw;
-        const auto KPad = K - KRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad both N and K
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_right_pad_transform(NRaw, NPad),
-                                                       make_right_pad_transform(KRaw, KPad)),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(N)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::MNPadding)
-        {
-            // pad N, but not K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_right_pad_transform(NRaw, NPad)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad K, but not N
-            assert(K % BK1 == 0);
-
-            const auto BK0 = K / BK1;
-
-            const auto b_grid_desc_n_k = transform_tensor_descriptor(
-                b_grid_desc_nraw_kraw,
-                make_tuple(make_pass_through_transform(NRaw), make_right_pad_transform(KRaw, KPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_n_k,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-        else
-        {
-            // not pad N or K
-            assert(KRaw % BK1 == 0);
-
-            const auto BK0 = KRaw / BK1;
-
-            const auto b_grid_desc_bk0_n_bk1 =
-                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_pass_through_transform(NRaw)),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-            return b_grid_desc_bk0_n_bk1;
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideC)
-    {
-        const auto c_grid_desc_mraw_nraw = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
-                                                    make_tuple(I1, StrideC));
-            }
-        }();
-
-        const auto M = math::integer_divide_ceil(MRaw, MPerBlock) * MPerBlock;
-        const auto N = math::integer_divide_ceil(NRaw, NPerBlock) * NPerBlock;
-
-        const auto MPad = M - MRaw;
-        const auto NPad = N - NRaw;
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
-                     GemmSpec == GemmSpecialization::MNKPadding)
-        {
-            // pad M and N
-            return transform_tensor_descriptor(c_grid_desc_mraw_nraw,
-                                               make_tuple(make_right_pad_transform(MRaw, MPad),
-                                                          make_right_pad_transform(NRaw, NPad)),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
-                          GemmSpec == GemmSpecialization::MKPadding)
-        {
-            // pad M, but not N
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_right_pad_transform(MRaw, MPad), make_pass_through_transform(NRaw)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
-                          GemmSpec == GemmSpecialization::NKPadding)
-        {
-            // pad N, but not M
-            return transform_tensor_descriptor(
-                c_grid_desc_mraw_nraw,
-                make_tuple(make_pass_through_transform(MRaw), make_right_pad_transform(NRaw, NPad)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-            // not pad M or N
-            return c_grid_desc_mraw_nraw;
-        }
-    }
-
-    using AGridDesc_AK0_M_AK1 = decltype(MakeAGridDescriptor_AK0_M_AK1(1, 1, 1));
-    using BGridDesc_BK0_N_BK1 = decltype(MakeBGridDescriptor_BK0_N_BK1(1, 1, 1));
-    using CGridDesc_M_N       = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1<
+        ALayout,
+        BLayout,
+        CLayout,
         ADataType, // TODO: distinguish A/B datatype
         GemmAccDataType,
         CShuffleDataType,
@@ -359,10 +94,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_AK0_M_AK1,
-        BGridDesc_BK0_N_BK1,
-        CGridDesc_M_N,
         NumGemmKPrefetchStage,
         BlockSize,
         MPerBlock,
@@ -397,162 +130,43 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         LoopSched,
         PipelineVer>;
 
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t MRaw,
-                 index_t NRaw,
-                 index_t KRaw,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              a_grid_desc_ak0_m_ak1_{DeviceOp::MakeAGridDescriptor_AK0_M_AK1(MRaw, KRaw, StrideA)},
-              b_grid_desc_bk0_n_bk1_{DeviceOp::MakeBGridDescriptor_BK0_N_BK1(KRaw, NRaw, StrideB)},
-              c_grid_desc_m_n_{DeviceOp::MakeCGridDescriptor_M_N(MRaw, NRaw, StrideC)},
-              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_ctile_map_{GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_)},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              kraw_{KRaw}
-        {
-            if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
-                                           b_grid_desc_bk0_n_bk1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        c_grid_desc_m_n_);
-            }
-        }
-
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1_;
-        BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-            c_grid_desc_mblock_mperblock_nblock_nperblock_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t kraw_;
-    };
+    using Argument = typename GridwiseGemm::Argument;
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceOp::Argument;
-
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-#if DEBUG_LOG
+            if(stream_config.log_level_ > 0)
             {
-                std::cout << "arg.a_grid_desc_ak0_m_ak1_{"
-                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) << ", "
-                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_ak0_m_ak1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_bk0_n_bk1_{"
-                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I0) << ", "
-                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_bk0_n_bk1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+                arg.Print();
             }
-#endif
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                            arg.b_grid_desc_bk0_n_bk1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(arg))
             {
                 throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
-            const auto K =
-                arg.a_grid_desc_ak0_m_ak1_.GetLength(I0) * arg.a_grid_desc_ak0_m_ak1_.GetLength(I2);
+            index_t gdx, gdy, gdz;
+            std::tie(gdx, gdy, gdz) = GridwiseGemm::CalculateGridSize(arg.M, arg.N);
+
+            const auto K = GridwiseGemm::CalculateAK0(arg.K) * AK1;
 
             float ave_time = 0;
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    true>;
+                const auto kernel = kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, true>;
 
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_,
-                                           arg.p_b_grid_,
-                                           arg.p_c_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, arg);
             }
             else
             {
-                const auto kernel = kernel_gemm_xdl_cshuffle_v1<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    DeviceOp::AGridDesc_AK0_M_AK1,
-                    DeviceOp::BGridDesc_BK0_N_BK1,
-                    typename GridwiseGemm::CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-                    typename GridwiseGemm::DefaultBlock2CTileMap,
-                    false>;
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(grid_size),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.p_a_grid_,
-                                           arg.p_b_grid_,
-                                           arg.p_c_grid_,
-                                           arg.a_element_op_,
-                                           arg.b_element_op_,
-                                           arg.c_element_op_,
-                                           arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                           arg.block_2_ctile_map_);
+                const auto kernel = kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, false>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, arg);
             }
 
             return ave_time;
@@ -580,19 +194,15 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
             return false;
         }
 
-        if((arg.kraw_ % AK1 != 0 || arg.kraw_ % BK1 != 0) &&
-           !(GemmSpec == GemmSpecialization::MKPadding ||
-             GemmSpec == GemmSpecialization::NKPadding ||
-             GemmSpec == GemmSpecialization::MNKPadding ||
-             GemmSpec == GemmSpecialization::KPadding))
+        if((arg.K % AK1 != 0 || arg.K % BK1 != 0) && !(GemmSpec == GemmSpecialization::MKPadding ||
+                                                       GemmSpec == GemmSpecialization::NKPadding ||
+                                                       GemmSpec == GemmSpecialization::MNKPadding ||
+                                                       GemmSpec == GemmSpecialization::KPadding))
         {
             return false;
         }
 
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
-                                           arg.b_grid_desc_bk0_n_bk1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(arg);
     }
 
     // polymorphic
@@ -604,28 +214,17 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     static auto MakeArgument(const ADataType* p_a,
                              const BDataType* p_b,
                              CDataType* p_c,
-                             index_t MRaw,
-                             index_t NRaw,
-                             index_t KRaw,
+                             index_t M,
+                             index_t N,
+                             index_t K,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
     {
-        return Argument{p_a,
-                        p_b,
-                        p_c,
-                        MRaw,
-                        NRaw,
-                        KRaw,
-                        StrideA,
-                        StrideB,
-                        StrideC,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+        return Argument{p_a, p_b, p_c, M, N, K, StrideA, StrideB, StrideC};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -634,28 +233,25 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
                                                       const void* p_b,
                                                       void* p_c,
-                                                      index_t MRaw,
-                                                      index_t NRaw,
-                                                      index_t KRaw,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op) override
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
                                           static_cast<CDataType*>(p_c),
-                                          MRaw,
-                                          NRaw,
-                                          KRaw,
+                                          M,
+                                          N,
+                                          K,
                                           StrideA,
                                           StrideB,
-                                          StrideC,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          StrideC);
     }
 
     // polymorphic
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index 9bd860f397..ad91c3c687 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -109,30 +109,37 @@ struct BlockToCTileMap_M00_N0_M01
 
 // Rows of column-vectors
 // This C-tile map dynamically adjusts M01 when C-tile index is out of range
-template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
-struct BlockToCTileMap_M00_N0_M01Adapt
+template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N = void>
+struct BlockToCTileMap_M00_N0_M01Adapt;
+
+template <index_t MPerBlock, index_t NPerBlock>
+struct BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
 
     __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt() = default;
 
-    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
-                                                        index_t M01 = 8)
-        : M01_(M01), c_grid_desc_m_n_(c_grid_desc_m_n)
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const BlockToCTileMap_M00_N0_M01Adapt&) =
+        default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(BlockToCTileMap_M00_N0_M01Adapt&&) =
+        default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt&
+    operator=(const BlockToCTileMap_M00_N0_M01Adapt&) = default;
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt&
+    operator=(BlockToCTileMap_M00_N0_M01Adapt&&) = default;
+
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(index_t M, index_t N, index_t M01 = 8)
+        : M_(M), N_(N), M01_(M01)
     {
     }
 
-    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+    __host__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
     {
-        const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I0), MPerBlock);
-        const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I1), NPerBlock);
+        const auto M0 = math::integer_divide_ceil(M, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N, NPerBlock);
 
-        const index_t grid_size = M0 * N0;
-
-        return grid_size;
+        return M0 * N0;
     }
 
     template <typename TopIdx>
@@ -140,8 +147,8 @@ struct BlockToCTileMap_M00_N0_M01Adapt
     {
         auto block_1d_id = idx_top[I0];
 
-        const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n_.GetLength(I0), MPerBlock);
-        const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n_.GetLength(I1), NPerBlock);
+        const auto M0 = math::integer_divide_ceil(M_, MPerBlock);
+        const auto N0 = math::integer_divide_ceil(N_, NPerBlock);
 
         block_1d_id = block_1d_id % (M0 * N0); // swallow batch index
 
@@ -209,11 +216,36 @@ struct BlockToCTileMap_M00_N0_M01Adapt
         return true; // always valid provided that user gets grid size from CalculateGridSize()
     }
 
-    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const { return true; }
-
     private:
+    index_t M_;
+    index_t N_;
     index_t M01_;
-    CGridDesc_M_N c_grid_desc_m_n_;
+};
+
+template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
+struct BlockToCTileMap_M00_N0_M01Adapt : BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
+{
+    using Parent = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>;
+
+    using Parent::I0;
+    using Parent::I1;
+
+    using Parent::Parent;
+    using Parent::operator=;
+
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
+                                                        index_t M01 = 8)
+        : Parent(c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1), M01)
+    {
+    }
+
+    __host__ static constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        return Parent::CalculateGridSize(c_grid_desc_m_n.GetLength(I0),
+                                         c_grid_desc_m_n.GetLength(I1));
+    }
+
+    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const { return true; }
 };
 
 // 2D slices of column-vectors in 3D space
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index 1213cdc263..e6303d76c1 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -17,17 +17,25 @@
 
 namespace ck {
 
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename AGridDesc_AK0_M_AK1,
-          typename BGridDesc_BK0_N_BK1,
-          typename CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename Block2CTileMap,
-          bool HasMainKBlockLoop>
+template <typename GridwiseGemm, bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_xdl_cshuffle_v1(typename GridwiseGemm::Argument karg)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(
+        karg.p_a_grid, karg.p_b_grid, karg.p_c_grid, p_shared, karg);
+#else
+    ignore = karg;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <typename GridwiseGemm, typename FloatAB, typename FloatC, bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
@@ -35,55 +43,33 @@ __global__ void
         kernel_gemm_xdl_cshuffle_v1(const FloatAB* __restrict__ p_a_grid,
                                     const FloatAB* __restrict__ p_b_grid,
                                     FloatC* __restrict__ p_c_grid,
-                                    const AElementwiseOperation a_element_op,
-                                    const BElementwiseOperation b_element_op,
-                                    const CElementwiseOperation c_element_op,
-                                    const AGridDesc_AK0_M_AK1 a_grid_desc_ak0_m_ak1,
-                                    const BGridDesc_BK0_N_BK1 b_grid_desc_bk0_n_bk1,
-                                    const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                                        c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                    const Block2CTileMap block_2_ctile_map)
+                                    typename GridwiseGemm::Problem problem)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
-                                                  p_b_grid,
-                                                  p_c_grid,
-                                                  p_shared,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  a_grid_desc_ak0_m_ak1,
-                                                  b_grid_desc_bk0_n_bk1,
-                                                  c_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  block_2_ctile_map);
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid, p_b_grid, p_c_grid, p_shared, problem);
 #else
     ignore = p_a_grid;
     ignore = p_b_grid;
     ignore = p_c_grid;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = a_grid_desc_ak0_m_ak1;
-    ignore = b_grid_desc_bk0_n_bk1;
-    ignore = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = block_2_ctile_map;
+    ignore = problem;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
-template <typename FloatAB,
+template <typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename FloatAB,
           typename FloatGemmAcc,
           typename FloatCShuffle,
           typename FloatC,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_AK0_M_AK1,
-          typename BGridDesc_BK0_N_BK1,
-          typename CGridDesc_M_N,
           index_t NumGemmKPrefetchStage,
           index_t BlockSize,
           index_t MPerBlock,
@@ -129,35 +115,396 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     static constexpr auto I7 = Number<7>{};
 
     // K1 should be Number<...>
-    static constexpr auto AK0 = Number<KPerBlock / AK1Value>{};
-    static constexpr auto BK0 = Number<KPerBlock / BK1Value>{};
-    static constexpr auto AK1 = Number<AK1Value>{};
-    static constexpr auto BK1 = Number<BK1Value>{};
+    static constexpr auto AK0Number = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0Number = Number<KPerBlock / BK1Value>{};
+    static constexpr auto AK1Number = Number<AK1Value>{};
+    static constexpr auto BK1Number = Number<BK1Value>{};
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    __host__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_divide_ceil(M, MPerBlock) * MPerBlock;
+    }
+
+    __host__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_divide_ceil(N, NPerBlock) * NPerBlock;
+    }
+
+    __host__ static auto CalculateKPadded(index_t K)
+    {
+        return math::integer_divide_ceil(K, KPerBlock) * KPerBlock;
+    }
+
+    __host__ static auto CalculateAK0(index_t K)
+    {
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding ||
+                     GemmSpec == GemmSpecialization::KPadding ||
+                     GemmSpec == GemmSpecialization::NKPadding)
+        {
+            return CalculateKPadded(K) / AK1Value;
+        }
+        else
+        {
+            return K / AK1Value;
+        }
+    }
+
+    __host__ static auto CalculateBK0(index_t K)
+    {
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding ||
+                     GemmSpec == GemmSpecialization::KPadding ||
+                     GemmSpec == GemmSpecialization::MKPadding)
+        {
+            return CalculateKPadded(K) / BK1Value;
+        }
+        else
+        {
+            return K / BK1Value;
+        }
+    }
+
+    __host__ static auto CalculateMBlock(index_t M)
+    {
+        return math::integer_divide_floor(M, MPerBlock);
+    }
+
+    __host__ static auto CalculateNBlock(index_t N)
+    {
+        return math::integer_divide_floor(N, NPerBlock);
+    }
+
+    __device__ static auto MakeAGridDescriptor_AK0_M_AK1(
+        index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA, index_t AK0)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::MKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding)
+        {
+            // pad both M and K
+            const auto a_grid_desc_m_k =
+                transform_tensor_descriptor(a_grid_desc_mraw_kraw,
+                                            make_tuple(make_right_pad_transform(M, MPad - M),
+                                                       make_right_pad_transform(K, KPad - K)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_pass_through_transform(MPad)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
+                          GemmSpec == GemmSpecialization::MNPadding)
+        {
+            // pad M, but not K
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_mraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
+                          GemmSpec == GemmSpecialization::NKPadding)
+        {
+            // pad K, but not M
+            const auto a_grid_desc_m_k = transform_tensor_descriptor(
+                a_grid_desc_mraw_kraw,
+                make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+        else
+        {
+            // not pad M or K
+            const auto a_grid_desc_ak0_m_ak1 = transform_tensor_descriptor(
+                a_grid_desc_mraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return a_grid_desc_ak0_m_ak1;
+        }
+    }
+
+    __device__ static auto MakeBGridDescriptor_BK0_N_BK1(
+        index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB, index_t BK0)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N, K), make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N, K), make_tuple(StrideB, I1));
+            }
+        }();
+
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::NKPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding)
+        {
+            // pad both N and K
+            const auto b_grid_desc_n_k =
+                transform_tensor_descriptor(b_grid_desc_nraw_kraw,
+                                            make_tuple(make_right_pad_transform(N, NPad - N),
+                                                       make_right_pad_transform(K, KPad - K)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_n_k,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_pass_through_transform(NPad)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
+                          GemmSpec == GemmSpecialization::MNPadding)
+        {
+            // pad N, but not K
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_nraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::KPadding ||
+                          GemmSpec == GemmSpecialization::MKPadding)
+        {
+            // pad K, but not N
+            const auto b_grid_desc_n_k = transform_tensor_descriptor(
+                b_grid_desc_nraw_kraw,
+                make_tuple(make_pass_through_transform(N), make_right_pad_transform(K, KPad - K)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_n_k,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+        else
+        {
+            // not pad N or K
+            const auto b_grid_desc_bk0_n_bk1 = transform_tensor_descriptor(
+                b_grid_desc_nraw_kraw,
+                make_tuple(make_unmerge_transform(make_tuple(BK0, BK1Value)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return b_grid_desc_bk0_n_bk1;
+        }
+    }
+
+    __host__ __device__ static auto
+    MakeCGridDescriptor_M_N(index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC)
+    {
+        const auto c_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        using GemmSpecialization = tensor_operation::device::GemmSpecialization;
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding ||
+                     GemmSpec == GemmSpecialization::MNKPadding)
+        {
+            // pad M and N
+            return transform_tensor_descriptor(c_grid_desc_mraw_nraw,
+                                               make_tuple(make_right_pad_transform(M, MPad - M),
+                                                          make_right_pad_transform(N, NPad - N)),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::MPadding ||
+                          GemmSpec == GemmSpecialization::MKPadding)
+        {
+            // pad M, but not N
+            return transform_tensor_descriptor(
+                c_grid_desc_mraw_nraw,
+                make_tuple(make_right_pad_transform(M, MPad - M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else if constexpr(GemmSpec == GemmSpecialization::NPadding ||
+                          GemmSpec == GemmSpecialization::NKPadding)
+        {
+            // pad N, but not M
+            return transform_tensor_descriptor(
+                c_grid_desc_mraw_nraw,
+                make_tuple(make_pass_through_transform(M), make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+            // not pad M or N
+            return c_grid_desc_mraw_nraw;
+        }
+    }
+
+    struct Problem
+    {
+        __host__ Problem(index_t M_,
+                         index_t N_,
+                         index_t K_,
+                         index_t StrideA_,
+                         index_t StrideB_,
+                         index_t StrideC_)
+            : M{M_},
+              N{N_},
+              K{K_},
+              StrideA{StrideA_},
+              StrideB{StrideB_},
+              StrideC{StrideC_},
+              MPadded{CalculateMPadded(M_)},
+              NPadded{CalculateNPadded(N_)},
+              KPadded{CalculateKPadded(K_)},
+              AK0{CalculateAK0(K_)},
+              BK0{CalculateBK0(K_)},
+              MBlock{CalculateMBlock(M_)},
+              NBlock{CalculateNBlock(N_)}
+        {
+        }
+
+        __host__ void Print() const
+        {
+            std::cout << "problem {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "KP:" << KPadded << ", "
+                      << "AK0:" << AK0 << ", "
+                      << "BK0:" << BK0 << ", "
+                      << "MBlock: " << MBlock << ", "
+                      << "NBlock: " << NBlock << "}" << std::endl;
+        }
+
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t KPadded;
+        index_t AK0;
+        index_t BK0;
+        index_t MBlock;
+        index_t NBlock;
+    };
+
+    // Argument
+    struct Argument : public tensor_operation::device::BaseArgument, public Problem
+    {
+        __host__ Argument(const FloatAB* p_a_grid_,
+                          const FloatAB* p_b_grid_,
+                          FloatC* p_c_grid_,
+                          index_t M_,
+                          index_t N_,
+                          index_t K_,
+                          index_t StrideA_,
+                          index_t StrideB_,
+                          index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_}
+        {
+        }
+
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+    };
+
     // FIXME: pass GridwiseGemmPipe as a template arguement into GridwiseGemm
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
-    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
         // A matrix in LDS memory, dst of blockwise copy
         return make_naive_tensor_descriptor(
-            make_tuple(AK0, Number<MPerBlock>{}, AK1),
-            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+            make_tuple(AK0Number, Number<MPerBlock>{}, AK1Number),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1Number, AK1Number, I1));
     }
 
-    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
     {
         // B matrix in LDS memory, dst of blockwise copy
         return make_naive_tensor_descriptor(
-            make_tuple(BK0, Number<NPerBlock>{}, BK1),
-            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+            make_tuple(BK0Number, Number<NPerBlock>{}, BK1Number),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1Number, BK1Number, I1));
     }
 
-    __host__ __device__ static constexpr auto
-    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    __device__ static constexpr auto GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
     {
         constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
         constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
@@ -172,14 +519,14 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
     }
 
-    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
     {
         // LDS allocation for A and B: be careful of alignment
         constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
         constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
 
         // lds max alignment
-        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+        constexpr auto max_lds_align = math::lcm(AK1Number, BK1Number);
 
         constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
@@ -200,36 +547,102 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     }
 
     // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    template <typename Block2CTileMap>
-    __host__ __device__ static constexpr bool
-    CheckValidity(const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
-                  const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
-                  const CGridDesc_M_N& c_grid_desc_m_n,
-                  const Block2CTileMap& block_2_ctile_map)
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
     {
         static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
                           (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
                       "Invalid tuning param!");
 
-        const auto M = a_grid_desc_ak0_m_ak1.GetLength(I1);
-        const auto N = b_grid_desc_bk0_n_bk1.GetLength(I1);
-        const auto K = a_grid_desc_ak0_m_ak1.GetLength(I0) * a_grid_desc_ak0_m_ak1.GetLength(I2);
-
-        if(!(M == c_grid_desc_m_n.GetLength(I0) && N == c_grid_desc_m_n.GetLength(I1)))
-            return false;
-
-        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
-            return false;
-
-        // check gridwise gemm pipeline
-        const auto num_k_loop = K / KPerBlock;
-
-        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
         {
-            return false;
+            if(!(problem.M % MPerBlock == 0))
+            {
+                return false;
+            }
         }
 
-        if(!block_2_ctile_map.CheckValidity(c_grid_desc_m_n))
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.N % NPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::KPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding)
+        {
+            if(!(CalculateKPadded(problem.K) % AK1Value == 0) ||
+               !(CalculateKPadded(problem.K) % BK1Value == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(!(problem.K % AK1Value == 0) || !(problem.K % BK1Value == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(problem.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(problem.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+        {
+            if(problem.N % CShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % CShuffleBlockTransferScalarPerVector_NPerBlock != 0)
+            {
+                return false;
+            }
+        }
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = (CalculateAK0(problem.K) * AK1Value) / KPerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {
             return false;
         }
@@ -238,22 +651,17 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         return true;
     }
 
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    __host__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
     {
         const index_t num_loop = K / KPerBlock;
 
         return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
-    __host__ __device__ static constexpr auto
-    MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc_M_N& c_grid_desc_m_n)
+    template <typename CGridDesc>
+    __device__ static constexpr auto MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+        const CGridDesc& c_grid_desc_m_n, index_t MBlock, index_t NBlock)
     {
-        const auto M = c_grid_desc_m_n.GetLength(I0);
-        const auto N = c_grid_desc_m_n.GetLength(I1);
-
-        const auto MBlock = M / MPerBlock;
-        const auto NBlock = N / NPerBlock;
-
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
             c_grid_desc_m_n,
             make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
@@ -265,33 +673,26 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     }
 
     // return block_id to C matrix tile idx (m0, n0) mapping
-    __host__ __device__ static constexpr auto
-    MakeDefaultBlock2CTileMap(const CGridDesc_M_N& c_grid_desc_m_n)
-    {
-        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>(
-            c_grid_desc_m_n);
-    }
+    using Block2CTileMap = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>;
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
-
-    using DefaultBlock2CTileMap =
-        remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}))>;
-
-    template <bool HasMainKBlockLoop, typename Block2CTileMap>
+    template <bool HasMainKBlockLoop>
     __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
                                const FloatAB* __restrict__ p_b_grid,
                                FloatC* __restrict__ p_c_grid,
                                void* __restrict__ p_shared,
-                               const AElementwiseOperation& a_element_op,
-                               const BElementwiseOperation& b_element_op,
-                               const CElementwiseOperation& c_element_op,
-                               const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
-                               const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
-                               const CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock&
-                                   c_grid_desc_mblock_mperblock_nblock_nperblock,
-                               const Block2CTileMap& block_2_ctile_map)
+                               const Problem& problem)
     {
+        const auto a_grid_desc_ak0_m_ak1 = MakeAGridDescriptor_AK0_M_AK1(
+            problem.M, problem.MPadded, problem.K, problem.KPadded, problem.StrideA, problem.AK0);
+        const auto b_grid_desc_bk0_n_bk1 = MakeBGridDescriptor_BK0_N_BK1(
+            problem.K, problem.KPadded, problem.N, problem.NPadded, problem.StrideB, problem.BK0);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(
+            problem.M, problem.MPadded, problem.N, problem.NPadded, problem.StrideC);
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                c_grid_desc_m_n, problem.MBlock, problem.NBlock);
+
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_grid_desc_ak0_m_ak1.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
@@ -299,7 +700,13 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
 
+        const AElementwiseOperation a_element_op{};
+        const BElementwiseOperation b_element_op{};
+        const CElementwiseOperation c_element_op{};
+
         // divide block work by [M, N]
+        const auto block_2_ctile_map = Block2CTileMap{problem.M, problem.N};
+
         const auto block_work_idx =
             block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
 
@@ -319,7 +726,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
 
         // lds max alignment
-        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+        constexpr auto max_lds_align = math::lcm(AK1Number, BK1Number);
 
         // A matrix in LDS memory, dst of blockwise copy
         constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
@@ -333,7 +740,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 AElementwiseOperation,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 InMemoryDataOperationEnum::Set,
-                                                Sequence<AK0, MPerBlock, AK1>,
+                                                Sequence<AK0Number, MPerBlock, AK1Number>,
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
@@ -364,7 +771,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 BElementwiseOperation,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 InMemoryDataOperationEnum::Set,
-                                                Sequence<BK0, NPerBlock, BK1>,
+                                                Sequence<BK0Number, NPerBlock, BK1Number>,
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 FloatAB,
@@ -396,8 +803,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-        constexpr index_t KPack = math::max(
-            math::lcm(AK1, BK1), MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+        constexpr index_t KPack =
+            math::max(math::lcm(AK1Number, BK1Number),
+                      MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
@@ -425,8 +833,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             static_cast<FloatAB*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
-        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
-        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1, 0, 0);
+        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1Number, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1Number, 0, 0);
 
         // gridwise GEMM pipeline
         static_assert(std::is_default_constructible_v<GridwiseGemmPipe>);

From 6eef0755c923c8dbf7860e31fb8c2e1b8859bc6e Mon Sep 17 00:00:00 2001
From: Haocong WANG <haocwang@amd.com>
Date: Tue, 30 May 2023 20:18:53 +0800
Subject: [PATCH 022/194] fix wmma gemm int8; add grouped conv int8 example
 (#716)

---
 .../CMakeLists.txt                            |  1 +
 ...ouped_conv_fwd_bias_relu_add_wmma_int8.cpp | 26 +++++++++++++++++++
 .../tensor_operation/gpu/warp/wmma_gemm.hpp   |  8 +++---
 3 files changed, 31 insertions(+), 4 deletions(-)
 create mode 100644 example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp

diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index 0472996858..9780a64cc2 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -20,4 +20,5 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
 endif()
 if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+  add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
 endif()
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp
new file mode 100644
index 0000000000..793324970e
--- /dev/null
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_int8.cpp
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common_wmma.hpp"
+
+// kernel data types
+using InKernelDataType       = I8;
+using WeiKernelDataType      = I8;
+using AccDataType            = I32;
+using CShuffleDataType       = I8;
+using BiasKernelDataType     = I8;
+using ResidualKernelDataType = I8;
+using OutKernelDataType      = I8;
+
+// tensor data types
+using InUserDataType  = InKernelDataType;
+using WeiUserDataType = WeiKernelDataType;
+using OutUserDataType = OutKernelDataType;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = ck::tensor_operation::element_wise::AddReluAdd;
+
+#include "run_grouped_conv_fwd_bias_relu_add_wmma_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_conv_fwd_bias_relu_add_example(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp b/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
index 0672bf8e5b..24efeb2dea 100644
--- a/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
@@ -262,12 +262,12 @@ struct wmma_type<WmmaInstr::wmma_i32_16x16x16_iu8,
 
     template <index_t MPerWmma,
               index_t NPerWmma,
-              bool neg_a,
-              bool neg_b,
-              bool clamp,
               class FloatA,
               class FloatB,
-              class FloatC>
+              class FloatC,
+              bool neg_a = false,
+              bool neg_b = false,
+              bool clamp = false>
     __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
     {
         if constexpr(wave_size == 32)

From 582e31e88d185f7bb8b24086a0d8d3804d17501f Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Wed, 31 May 2023 23:25:25 +0800
Subject: [PATCH 023/194] Add class type support for
 __builtin_amdgcn_readfirstlane() (#711)

* Add overloaded version of __builtin_amdgcn_readfirstlane()

* Remove 'static' specifiers

* Remove more 'static' specifier

* Replace unsigne char by std::byte

* Add 'const' specifier to never changing variable

* Add 'inline' specifier to funcion definition

* Fix wrong boundar calculation logic

* Rename type trait

* Remove std:: qualifier from standard types

* Replace 'size_t' by 'unsigned'

* Use type alias to hint usage

* Replace static_for<> by ordinary 'for' loop

* Rename readfirstlane() to amd_wave_read_first_lane()

* Rename file readfirstlance.hpp as amd_wave_read_first_lane.hpp

* Reorder statements
---
 .../ck/utility/amd_wave_read_first_lane.hpp   | 83 +++++++++++++++++++
 include/ck/utility/common_header.hpp          |  1 +
 2 files changed, 84 insertions(+)
 create mode 100644 include/ck/utility/amd_wave_read_first_lane.hpp

diff --git a/include/ck/utility/amd_wave_read_first_lane.hpp b/include/ck/utility/amd_wave_read_first_lane.hpp
new file mode 100644
index 0000000000..4652ce7a74
--- /dev/null
+++ b/include/ck/utility/amd_wave_read_first_lane.hpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/utility/functional2.hpp"
+#include "ck/utility/math.hpp"
+
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+namespace ck {
+namespace detail {
+
+template <unsigned Size>
+struct get_unsigned_int;
+
+template <>
+struct get_unsigned_int<1>
+{
+    using type = uint8_t;
+};
+
+template <>
+struct get_unsigned_int<2>
+{
+    using type = uint16_t;
+};
+
+template <>
+struct get_unsigned_int<4>
+{
+    using type = uint32_t;
+};
+
+template <unsigned Size>
+using get_unsigned_int_t = typename get_unsigned_int<Size>::type;
+
+} // namespace detail
+
+__device__ inline int32_t amd_wave_read_first_lane(int32_t value)
+{
+    return __builtin_amdgcn_readfirstlane(value);
+}
+
+template <
+    typename Object,
+    typename = std::enable_if_t<std::is_class_v<Object> && std::is_trivially_copyable_v<Object>>>
+__device__ auto amd_wave_read_first_lane(const Object& obj)
+{
+    using Size                = unsigned;
+    constexpr Size SgprSize   = 4;
+    constexpr Size ObjectSize = sizeof(Object);
+
+    auto* const from_obj = reinterpret_cast<const std::byte*>(&obj);
+    alignas(Object) std::byte to_obj[ObjectSize];
+
+    constexpr Size RemainedSize             = ObjectSize % SgprSize;
+    constexpr Size CompleteSgprCopyBoundary = ObjectSize - RemainedSize;
+    for(Size offset = 0; offset < CompleteSgprCopyBoundary; offset += SgprSize)
+    {
+        using Sgpr = detail::get_unsigned_int_t<SgprSize>;
+
+        *reinterpret_cast<Sgpr*>(to_obj + offset) =
+            amd_wave_read_first_lane(*reinterpret_cast<const Sgpr*>(from_obj + offset));
+    }
+
+    if constexpr(0 < RemainedSize)
+    {
+        using Carrier = detail::get_unsigned_int_t<RemainedSize>;
+
+        *reinterpret_cast<Carrier>(to_obj + CompleteSgprCopyBoundary) = amd_wave_read_first_lane(
+            *reinterpret_cast<const Carrier*>(from_obj + CompleteSgprCopyBoundary));
+    }
+
+    /// NOTE: Implicitly start object lifetime. It's better to use std::start_lifetime_at() in this
+    /// scenario
+    return *reinterpret_cast<Object*>(to_obj);
+}
+
+} // namespace ck
diff --git a/include/ck/utility/common_header.hpp b/include/ck/utility/common_header.hpp
index 1378bbe448..8da87c8766 100644
--- a/include/ck/utility/common_header.hpp
+++ b/include/ck/utility/common_header.hpp
@@ -33,6 +33,7 @@
 #include "ck/utility/debug.hpp"
 
 #include "ck/utility/amd_buffer_addressing.hpp"
+#include "ck/utility/amd_wave_read_first_lane.hpp"
 #include "ck/utility/generic_memory_space_atomic.hpp"
 #include "ck/utility/get_id.hpp"
 #include "ck/utility/thread_group.hpp"

From b94fd0b2279c6476c6e109e99dc5d0e6d8ce313c Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 31 May 2023 16:46:57 -0700
Subject: [PATCH 024/194] update copyright headers (#726)

---
 CONTRIBUTORS.md                                                 | 2 +-
 LICENSE                                                         | 2 +-
 client_example/01_gemm/gemm.cpp                                 | 2 +-
 .../02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp          | 2 +-
 client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp   | 2 +-
 client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp       | 2 +-
 .../03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp          | 2 +-
 .../03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp   | 2 +-
 client_example/04_contraction/contraction_bilinear_fp32.cpp     | 2 +-
 client_example/04_contraction/contraction_bilinear_fp64.cpp     | 2 +-
 .../04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp        | 2 +-
 client_example/04_contraction/contraction_scale_fp32.cpp        | 2 +-
 client_example/04_contraction/contraction_scale_fp64.cpp        | 2 +-
 client_example/05_layernorm/layernorm2d.cpp                     | 2 +-
 client_example/06_softmax/softmax4d.cpp                         | 2 +-
 client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp     | 2 +-
 client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp     | 2 +-
 client_example/08_fused_attention/fused_attention.cpp           | 2 +-
 client_example/08_fused_attention/fused_attention_bias.cpp      | 2 +-
 .../conv2d_fwd_bias_relu_perchannel_quantization.cpp            | 2 +-
 .../conv2d_fwd_bias_relu_perlayer_quantization.cpp              | 2 +-
 .../conv2d_fwd_bias_tanh_perchannel_quantization.cpp            | 2 +-
 .../conv2d_fwd_bias_tanh_perlayer_quantization.cpp              | 2 +-
 .../09_quantization/conv2d_fwd_perchannel_quantization.cpp      | 2 +-
 .../09_quantization/conv2d_fwd_perlayer_quantization.cpp        | 2 +-
 client_example/09_quantization/gemm_quantization.cpp            | 2 +-
 .../10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp      | 2 +-
 client_example/11_grouped_conv_bwd_weight/common.hpp            | 2 +-
 .../12_elementwise_normalization/elementwise_layernorm2d.cpp    | 2 +-
 client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp              | 2 +-
 client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp              | 2 +-
 client_example/13_batchnorm/batchnorm_infer_nhwc.cpp            | 2 +-
 client_example/14_instance_id/batchnorm_fwd_instance_id.cpp     | 2 +-
 client_example/15_gemm_add_multiply/gemm_add_multiply.cpp       | 2 +-
 client_example/15_reduce/reduce_nhwc_c.cpp                      | 2 +-
 .../17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp          | 2 +-
 client_example/18_groupnorm/groupnorm_swish.cpp                 | 2 +-
 client_example/19_pool_fwd/avg_pool3d_fwd.cpp                   | 2 +-
 client_example/19_pool_fwd/max_pool2d_fwd.cpp                   | 2 +-
 example/01_gemm/common.hpp                                      | 2 +-
 example/01_gemm/gemm_dl_fp16.cpp                                | 2 +-
 example/01_gemm/gemm_dl_fp32.cpp                                | 2 +-
 example/01_gemm/gemm_dl_int4.cpp                                | 2 +-
 example/01_gemm/gemm_dl_int8.cpp                                | 2 +-
 example/01_gemm/gemm_wmma_fp16.cpp                              | 2 +-
 example/01_gemm/gemm_xdl_bf16.cpp                               | 2 +-
 example/01_gemm/gemm_xdl_fp16.cpp                               | 2 +-
 example/01_gemm/gemm_xdl_fp64.cpp                               | 2 +-
 example/01_gemm/gemm_xdl_int4.cpp                               | 2 +-
 example/01_gemm/gemm_xdl_int8.cpp                               | 2 +-
 example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp                    | 2 +-
 example/01_gemm/gemm_xdl_wavelet_fp16.cpp                       | 2 +-
 example/01_gemm/run_gemm_example.inc                            | 2 +-
 example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp            | 2 +-
 example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp             | 2 +-
 example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp           | 2 +-
 example/04_gemm_add_add_fastgelu/common.hpp                     | 2 +-
 .../04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp | 2 +-
 .../04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp | 2 +-
 .../04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp | 2 +-
 .../04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp | 2 +-
 .../04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp | 2 +-
 example/09_convnd_fwd/convnd_fwd_common.hpp                     | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_common.hpp                  | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp                    | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp                    | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_int8.cpp                    | 2 +-
 example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp                   | 2 +-
 example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp                   | 2 +-
 example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp                   | 2 +-
 example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp                   | 2 +-
 example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp                   | 2 +-
 example/09_convnd_fwd/run_convnd_fwd_dl_example.inc             | 2 +-
 example/09_convnd_fwd/run_convnd_fwd_example.inc                | 2 +-
 example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp     | 2 +-
 .../convnd_fwd_max_xdl_bf16.cpp                                 | 2 +-
 .../convnd_fwd_max_xdl_fp16.cpp                                 | 2 +-
 .../convnd_fwd_max_xdl_fp32.cpp                                 | 2 +-
 .../convnd_fwd_max_xdl_int4.cpp                                 | 2 +-
 .../convnd_fwd_max_xdl_int8.cpp                                 | 2 +-
 .../run_convnd_fwd_max_example.inc                              | 2 +-
 example/12_reduce/reduce_blockwise.cpp                          | 2 +-
 example/12_reduce/reduce_blockwise_impl.hpp                     | 2 +-
 example/12_reduce/reduce_blockwise_two_call.cpp                 | 2 +-
 example/12_reduce/reduce_example_common.hpp                     | 2 +-
 example/12_reduce/reduce_multiblock_atomic_add.cpp              | 2 +-
 example/12_reduce/reduce_multiblock_atomic_add_impl.hpp         | 2 +-
 example/13_pool2d_fwd/pool2d_fwd_common.hpp                     | 2 +-
 example/13_pool2d_fwd/pool2d_fwd_fp16.cpp                       | 2 +-
 example/13_pool2d_fwd/pool2d_fwd_fp32.cpp                       | 2 +-
 example/14_gemm_quantization/gemm_dl_quantization_int8.cpp      | 2 +-
 .../gemm_xdl_bias_relu_quantization_int8.cpp                    | 2 +-
 example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp     | 2 +-
 example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp     | 2 +-
 example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp              | 2 +-
 example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp               | 2 +-
 example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp               | 2 +-
 example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp               | 2 +-
 example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp               | 2 +-
 example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp        | 2 +-
 .../gemm_add_add_mean_meansquare_xdl_fp16.cpp                   | 2 +-
 .../gemm_add_addsquare_xdl_int8.cpp                             | 2 +-
 example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp     | 2 +-
 example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp     | 2 +-
 example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp     | 2 +-
 example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp     | 2 +-
 example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp     | 2 +-
 .../gemm_mean_meansquare_xdl_bf16.cpp                           | 2 +-
 .../gemm_mean_meansquare_xdl_fp16.cpp                           | 2 +-
 .../gemm_mean_meansquare_xdl_fp32.cpp                           | 2 +-
 .../16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp    | 2 +-
 example/17_convnd_bwd_data/convnd_bwd_data_common.hpp           | 2 +-
 example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp          | 2 +-
 example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp         | 2 +-
 example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp | 2 +-
 example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp       | 2 +-
 example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp      | 2 +-
 example/19_binary_elementwise/elementwise_add_1d.cpp            | 2 +-
 example/19_binary_elementwise/elementwise_add_4d.cpp            | 2 +-
 example/20_grouped_conv_bwd_weight/common.hpp                   | 2 +-
 .../grouped_conv_bwd_weight_xdl_bf16.cpp                        | 2 +-
 .../grouped_conv_bwd_weight_xdl_fp16.cpp                        | 2 +-
 .../run_grouped_conv_bwd_weight_example.inc                     | 2 +-
 .../gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp             | 2 +-
 .../gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp           | 2 +-
 example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp     | 2 +-
 .../gemm_xdl_layernorm_naive_single_kernel_fp16.cpp             | 2 +-
 example/22_cgemm/cgemm_xdl_bf16.cpp                             | 2 +-
 example/22_cgemm/cgemm_xdl_common.hpp                           | 2 +-
 example/22_cgemm/cgemm_xdl_fp16.cpp                             | 2 +-
 example/22_cgemm/cgemm_xdl_fp32.cpp                             | 2 +-
 example/22_cgemm/cgemm_xdl_int4.cpp                             | 2 +-
 example/22_cgemm/cgemm_xdl_int8.cpp                             | 2 +-
 example/23_softmax/softmax_blockwise.cpp                        | 2 +-
 .../gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp                   | 2 +-
 .../gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp                   | 2 +-
 example/26_contraction/contraction_bilinear_xdl_fp32.cpp        | 2 +-
 example/26_contraction/contraction_bilinear_xdl_fp64.cpp        | 2 +-
 example/26_contraction/contraction_scale_xdl_fp32.cpp           | 2 +-
 example/26_contraction/contraction_scale_xdl_fp64.cpp           | 2 +-
 example/27_layernorm/common.hpp                                 | 2 +-
 example/27_layernorm/layernorm_fp16.cpp                         | 2 +-
 example/27_layernorm/layernorm_splitk_fp16.cpp                  | 2 +-
 example/27_layernorm/run_layernorm_example.inc                  | 2 +-
 .../grouped_gemm_bias_e_permute_xdl_fp16.cpp                    | 2 +-
 .../batched_gemm_bias_e_permute_wmma_fp16.cpp                   | 2 +-
 .../batched_gemm_bias_e_permute_xdl_fp16.cpp                    | 2 +-
 example/30_grouped_conv_fwd_multiple_d/common.hpp               | 2 +-
 example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp          | 2 +-
 .../grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp                | 2 +-
 .../grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp                 | 2 +-
 .../grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp                 | 2 +-
 .../grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp                 | 2 +-
 .../grouped_conv_fwd_bias_relu_add_xdl_int4.cpp                 | 2 +-
 .../grouped_conv_fwd_bias_relu_add_xdl_int8.cpp                 | 2 +-
 .../run_grouped_conv_fwd_bias_relu_add_example.inc              | 2 +-
 .../run_grouped_conv_fwd_bias_relu_add_wmma_example.inc         | 2 +-
 .../run_grouped_conv_fwd_example.inc                            | 2 +-
 example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp     | 2 +-
 example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp     | 2 +-
 example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp     | 2 +-
 example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp     | 2 +-
 example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp     | 2 +-
 example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc  | 2 +-
 ..._gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp | 2 +-
 .../batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp        | 2 +-
 .../batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp        | 2 +-
 .../batched_gemm_scale_softmax_gemm_xdl_bf16.cpp                | 2 +-
 .../batched_gemm_scale_softmax_gemm_xdl_fp16.cpp                | 2 +-
 ..._gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp | 2 +-
 .../grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp        | 2 +-
 .../run_batched_gemm_scale_softmax_gemm.inc                     | 2 +-
 .../run_batched_gemm_scale_softmax_gemm_permute.inc             | 2 +-
 .../run_grouped_gemm_scale_softmax_gemm_permute.inc             | 2 +-
 example/33_multiple_reduce/dual_reduce_common.hpp               | 2 +-
 example/33_multiple_reduce/dual_reduce_multiblock.cpp           | 2 +-
 example/33_multiple_reduce/dual_reduce_threadwise.cpp           | 2 +-
 example/34_batchnorm/batchnorm_backward_nhwc.cpp                | 2 +-
 example/34_batchnorm/batchnorm_common.hpp                       | 2 +-
 example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp       | 2 +-
 example/34_batchnorm/batchnorm_forward_training_nhwc.cpp        | 2 +-
 example/34_batchnorm/batchnorm_infer_impl.hpp                   | 2 +-
 example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp                | 2 +-
 example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp                 | 2 +-
 example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp                 | 2 +-
 example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp                 | 2 +-
 example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp                 | 2 +-
 .../36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp | 2 +-
 .../batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp             | 2 +-
 example/38_grouped_conv_bwd_data_multiple_d/common.hpp          | 2 +-
 .../grouped_conv_bwd_data_bias_relu_fp16.cpp                    | 2 +-
 .../grouped_conv_bwd_data_fp16.cpp                              | 2 +-
 .../run_grouped_conv_bwd_data_bias_relu_example.inc             | 2 +-
 .../run_grouped_conv_bwd_data_example.inc                       | 2 +-
 example/39_permute/common.hpp                                   | 2 +-
 example/39_permute/permute_1xHxW_fp16.cpp                       | 2 +-
 example/39_permute/permute_HxWx4_fp16.cpp                       | 2 +-
 example/39_permute/permute_NxHxW_fp16.cpp                       | 2 +-
 example/39_permute/run_permute_bundle_example.inc               | 2 +-
 example/39_permute/run_permute_element_example.inc              | 2 +-
 example/40_conv2d_fwd_quantization/common.hpp                   | 2 +-
 .../conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp    | 2 +-
 .../conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp      | 2 +-
 .../conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp    | 2 +-
 .../conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp      | 2 +-
 .../conv2d_fwd_dl_perchannel_quantization_int8.cpp              | 2 +-
 .../conv2d_fwd_dl_perlayer_quantization_int8.cpp                | 2 +-
 .../conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp   | 2 +-
 .../conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp     | 2 +-
 .../conv2d_fwd_xdl_perchannel_quantization_int8.cpp             | 2 +-
 .../conv2d_fwd_xdl_perlayer_quantization_int8.cpp               | 2 +-
 .../run_conv2d_fwd_bias_perchannel_quantization_example.inc     | 2 +-
 .../run_conv2d_fwd_bias_perlayer_quantization_example.inc       | 2 +-
 .../run_conv2d_fwd_perchannel_quantization_example.inc          | 2 +-
 .../run_conv2d_fwd_perlayer_quantization_example.inc            | 2 +-
 .../41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp | 2 +-
 .../41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp | 2 +-
 .../41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp | 2 +-
 .../41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp | 2 +-
 .../41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp | 2 +-
 .../run_grouped_conv_conv_fwd_example.inc                       | 2 +-
 example/42_groupnorm/common.hpp                                 | 2 +-
 example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp             | 2 +-
 example/42_groupnorm/groupnorm_splitk_fp16.cpp                  | 2 +-
 example/42_groupnorm/groupnorm_swish_fp16.cpp                   | 2 +-
 example/42_groupnorm/run_groupnorm_example.inc                  | 2 +-
 .../splitk_gemm_bias_e_permute_xdl_fp16.cpp                     | 2 +-
 .../splitk_gemm_bias_e_permute_xdl_fp32.cpp                     | 2 +-
 .../elementwise_layernorm_blockwise.cpp                         | 2 +-
 example/46_gemm_add_multiply/common.hpp                         | 2 +-
 example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp      | 2 +-
 example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp     | 2 +-
 .../gemm_bias_softmax_gemm_permute.cpp                          | 2 +-
 example/48_pool3d_fwd/pool3d_fwd_common.hpp                     | 2 +-
 example/48_pool3d_fwd/pool3d_fwd_fp16.cpp                       | 2 +-
 include/ck/ck.hpp                                               | 2 +-
 include/ck/host_utility/device_prop.hpp                         | 2 +-
 include/ck/host_utility/hip_check_error.hpp                     | 2 +-
 include/ck/host_utility/io.hpp                                  | 2 +-
 include/ck/host_utility/kernel_launch.hpp                       | 2 +-
 ...forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp | 2 +-
 include/ck/stream_config.hpp                                    | 2 +-
 include/ck/tensor/static_tensor.hpp                             | 2 +-
 include/ck/tensor_description/cluster_descriptor.hpp            | 2 +-
 include/ck/tensor_description/multi_index_transform.hpp         | 2 +-
 include/ck/tensor_description/multi_index_transform_helper.hpp  | 2 +-
 include/ck/tensor_description/tensor_adaptor.hpp                | 2 +-
 include/ck/tensor_description/tensor_descriptor.hpp             | 2 +-
 include/ck/tensor_description/tensor_descriptor_helper.hpp      | 2 +-
 include/ck/tensor_description/tensor_space_filling_curve.hpp    | 2 +-
 .../ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp    | 2 +-
 .../ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp | 2 +-
 .../ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp   | 2 +-
 include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp   | 2 +-
 include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp | 2 +-
 .../gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp              | 2 +-
 include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp     | 2 +-
 .../gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp          | 2 +-
 include/ck/tensor_operation/gpu/block/blockwise_welford.hpp     | 2 +-
 .../gpu/block/reduction_functions_blockwise.hpp                 | 2 +-
 .../gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp       | 2 +-
 .../gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp       | 2 +-
 .../gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp       | 2 +-
 .../gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp       | 2 +-
 .../gpu/block/thread_group_tensor_slice_transfer_v7.hpp         | 2 +-
 .../gpu/device/convolution_backward_data_specialization.hpp     | 2 +-
 .../gpu/device/convolution_backward_weight_specialization.hpp   | 2 +-
 .../gpu/device/convolution_forward_specialization.hpp           | 2 +-
 include/ck/tensor_operation/gpu/device/device_base.hpp          | 2 +-
 .../gpu/device/device_batched_contraction_multiple_d.hpp        | 2 +-
 include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp  | 2 +-
 .../ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp | 2 +-
 .../tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp | 2 +-
 .../device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp   | 2 +-
 .../gpu/device/device_batched_gemm_softmax_gemm.hpp             | 2 +-
 .../gpu/device/device_batched_gemm_softmax_gemm_permute.hpp     | 2 +-
 .../tensor_operation/gpu/device/device_batchnorm_backward.hpp   | 2 +-
 .../ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp | 2 +-
 .../ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp   | 2 +-
 include/ck/tensor_operation/gpu/device/device_cgemm.hpp         | 2 +-
 .../gpu/device/device_contraction_multiple_d.hpp                | 2 +-
 include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp | 2 +-
 include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp      | 2 +-
 .../gpu/device/device_conv_fwd_bias_activation.hpp              | 2 +-
 .../gpu/device/device_conv_fwd_bias_activation_add.hpp          | 2 +-
 include/ck/tensor_operation/gpu/device/device_elementwise.hpp   | 2 +-
 .../gpu/device/device_elementwise_normalization.hpp             | 2 +-
 include/ck/tensor_operation/gpu/device/device_gemm.hpp          | 2 +-
 .../tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp  | 2 +-
 .../ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp   | 2 +-
 .../gpu/device/device_gemm_multiple_d_layernorm.hpp             | 2 +-
 .../gpu/device/device_gemm_multiple_d_multiple_r.hpp            | 2 +-
 include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp   | 2 +-
 include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp   | 2 +-
 .../gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp        | 2 +-
 .../gpu/device/device_grouped_contraction_multiple_d.hpp        | 2 +-
 .../gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp      | 2 +-
 .../gpu/device/device_grouped_conv_bwd_weight.hpp               | 2 +-
 .../ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp  | 2 +-
 .../device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp    | 2 +-
 .../gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp    | 2 +-
 .../gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp     | 2 +-
 .../device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp   | 2 +-
 .../ck/tensor_operation/gpu/device/device_multiple_reduce.hpp   | 2 +-
 include/ck/tensor_operation/gpu/device/device_normalization.hpp | 2 +-
 include/ck/tensor_operation/gpu/device/device_permute.hpp       | 2 +-
 include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp      | 2 +-
 include/ck/tensor_operation/gpu/device/device_reduce.hpp        | 2 +-
 include/ck/tensor_operation/gpu/device/device_softmax.hpp       | 2 +-
 .../gpu/device/device_splitk_contraction_multiple_d.hpp         | 2 +-
 .../device_splitk_contraction_multiple_d_xdl_cshuffle.hpp       | 2 +-
 include/ck/tensor_operation/gpu/device/gemm_specialization.hpp  | 2 +-
 .../device_batched_contraction_multiple_d_wmma_cshuffle.hpp     | 2 +-
 .../impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp | 2 +-
 .../gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp   | 2 +-
 .../gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp         | 2 +-
 ...ice_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp | 2 +-
 .../gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp | 2 +-
 .../device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp   | 2 +-
 .../impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp      | 2 +-
 .../gpu/device/impl/device_batched_gemm_xdl.hpp                 | 2 +-
 .../gpu/device/impl/device_batchnorm_backward_impl.hpp          | 2 +-
 .../gpu/device/impl/device_batchnorm_forward_impl.hpp           | 2 +-
 .../gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp         | 2 +-
 .../device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp  | 2 +-
 ...vice_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp | 2 +-
 .../device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp   | 2 +-
 ...v2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp | 2 +-
 ..._conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp | 2 +-
 .../impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp     | 2 +-
 .../gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp    | 2 +-
 .../device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp   | 2 +-
 .../gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp | 2 +-
 .../gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp   | 2 +-
 .../gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp  | 2 +-
 .../gpu/device/impl/device_elementwise_2d_impl.hpp              | 2 +-
 .../gpu/device/impl/device_elementwise_impl.hpp                 | 2 +-
 .../gpu/device/impl/device_elementwise_normalization_impl.hpp   | 2 +-
 .../device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp    | 2 +-
 include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp  | 2 +-
 .../gpu/device/impl/device_gemm_multiple_d_dl.hpp               | 2 +-
 .../impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp      | 2 +-
 .../impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp     | 2 +-
 .../gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp    | 2 +-
 .../gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp     | 2 +-
 .../gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp         | 2 +-
 .../ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp    | 2 +-
 include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp | 2 +-
 .../gpu/device/impl/device_gemm_xdl_cshuffle.hpp                | 2 +-
 .../gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp      | 2 +-
 .../gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp              | 2 +-
 .../gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp        | 2 +-
 .../impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp | 2 +-
 .../device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp | 2 +-
 ...vice_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp | 2 +-
 .../impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp      | 2 +-
 ...vice_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp | 2 +-
 .../impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp   | 2 +-
 .../gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp       | 2 +-
 .../gpu/device/impl/device_grouped_gemm_xdl.hpp                 | 2 +-
 .../gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp | 2 +-
 .../gpu/device/impl/device_multiple_reduce_multiblock.hpp       | 2 +-
 .../gpu/device/impl/device_multiple_reduce_threadwise.hpp       | 2 +-
 .../gpu/device/impl/device_normalization_impl.hpp               | 2 +-
 .../gpu/device/impl/device_normalization_splitk_impl.hpp        | 2 +-
 .../ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp | 2 +-
 .../gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp             | 2 +-
 .../gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp           | 2 +-
 .../tensor_operation/gpu/device/impl/device_reduce_common.hpp   | 2 +-
 .../gpu/device/impl/device_reduce_multiblock.hpp                | 2 +-
 .../gpu/device/impl/device_reduce_threadwise.hpp                | 2 +-
 .../ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp | 2 +-
 .../device/impl/device_sparse_embeddings_forward_layernorm.hpp  | 2 +-
 .../ck/tensor_operation/gpu/device/masking_specialization.hpp   | 2 +-
 include/ck/tensor_operation/gpu/device/matrix_padder.hpp        | 2 +-
 .../tensor_operation/gpu/device/reduction_operator_mapping.hpp  | 2 +-
 include/ck/tensor_operation/gpu/device/tensor_layout.hpp        | 2 +-
 .../ck/tensor_operation/gpu/device/tensor_specialization.hpp    | 2 +-
 include/ck/tensor_operation/gpu/device/welford_helper.hpp       | 2 +-
 .../gpu/element/binary_element_wise_operation.hpp               | 2 +-
 .../ck/tensor_operation/gpu/element/element_wise_operation.hpp  | 2 +-
 .../gpu/element/unary_element_wise_operation.hpp                | 2 +-
 ...e_multiblock_reduce_second_half_batchnorm_backward_final.hpp | 2 +-
 .../gridwise_multiblock_welford_first_half.hpp                  | 2 +-
 ...e_multiblock_welford_second_half_batchnorm_forward_final.hpp | 2 +-
 ...tiblock_welford_second_half_multiblock_reduce_first_half.hpp | 2 +-
 include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp     | 2 +-
 ...gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp | 2 +-
 .../gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp | 2 +-
 .../gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp      | 2 +-
 .../gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp      | 2 +-
 .../gpu/grid/gridwise_2d_reduction_multiblock.hpp               | 2 +-
 .../gpu/grid/gridwise_2d_reduction_threadwise.hpp               | 2 +-
 .../gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp     | 2 +-
 ..._batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp | 2 +-
 ...ise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp | 2 +-
 .../grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp | 2 +-
 .../gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp  | 2 +-
 .../gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp   | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp    | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp    | 2 +-
 .../grid/gridwise_elementwise_layernorm_welford_variance.hpp    | 2 +-
 .../gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp  | 2 +-
 .../tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp   | 2 +-
 .../grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp   | 2 +-
 .../gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp         | 2 +-
 .../gpu/grid/gridwise_gemm_pipeline_selector.hpp                | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp  | 2 +-
 .../gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp           | 2 +-
 .../gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp  | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp | 2 +-
 include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp     | 2 +-
 .../tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp | 2 +-
 .../gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp        | 2 +-
 .../gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp        | 2 +-
 .../gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp             | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp  | 2 +-
 .../tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp   | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp  | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp  | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp  | 2 +-
 include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp       | 2 +-
 .../ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp  | 2 +-
 .../gpu/grid/gridwise_set_multiple_buffer_value.hpp             | 2 +-
 include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp       | 2 +-
 .../gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp   | 2 +-
 .../normalization/gridwise_normalization_naive_variance.hpp     | 2 +-
 .../gpu/grid/normalization/gridwise_normalization_selector.hpp  | 2 +-
 .../grid/normalization/gridwise_normalization_splitk_1st.hpp    | 2 +-
 .../grid/normalization/gridwise_normalization_splitk_2nd.hpp    | 2 +-
 .../normalization/gridwise_normalization_welford_variance.hpp   | 2 +-
 .../gpu/thread/reduction_functions_threadwise.hpp               | 2 +-
 .../tensor_operation/gpu/thread/threadwise_contraction_dl.hpp   | 2 +-
 .../ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp | 2 +-
 .../tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer.hpp             | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp        | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp        | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp        | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp        | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp        | 2 +-
 .../gpu/thread/threadwise_tensor_slice_transfer_v7.hpp          | 2 +-
 include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp   | 2 +-
 include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp              | 2 +-
 include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp            | 2 +-
 .../operator_transform/transform_contraction_to_gemm.hpp        | 2 +-
 .../operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp   | 2 +-
 .../operator_transform/transform_conv_fwd_to_gemm.hpp           | 2 +-
 include/ck/utility/amd_address_space.hpp                        | 2 +-
 include/ck/utility/amd_buffer_addressing.hpp                    | 2 +-
 include/ck/utility/amd_inline_asm.hpp                           | 2 +-
 include/ck/utility/amd_wmma.hpp                                 | 2 +-
 include/ck/utility/amd_xdlops.hpp                               | 2 +-
 include/ck/utility/array.hpp                                    | 2 +-
 include/ck/utility/array_multi_index.hpp                        | 2 +-
 include/ck/utility/c_style_pointer_cast.hpp                     | 2 +-
 include/ck/utility/common_header.hpp                            | 2 +-
 include/ck/utility/container_element_picker.hpp                 | 2 +-
 include/ck/utility/container_helper.hpp                         | 2 +-
 include/ck/utility/debug.hpp                                    | 2 +-
 include/ck/utility/dynamic_buffer.hpp                           | 2 +-
 include/ck/utility/enable_if.hpp                                | 2 +-
 include/ck/utility/functional.hpp                               | 2 +-
 include/ck/utility/functional2.hpp                              | 2 +-
 include/ck/utility/functional3.hpp                              | 2 +-
 include/ck/utility/functional4.hpp                              | 2 +-
 include/ck/utility/generic_memory_space_atomic.hpp              | 2 +-
 include/ck/utility/get_id.hpp                                   | 2 +-
 include/ck/utility/ignore.hpp                                   | 2 +-
 include/ck/utility/inner_product.hpp                            | 2 +-
 include/ck/utility/integral_constant.hpp                        | 2 +-
 include/ck/utility/is_known_at_compile_time.hpp                 | 2 +-
 include/ck/utility/magic_division.hpp                           | 2 +-
 include/ck/utility/math.hpp                                     | 2 +-
 include/ck/utility/math_v2.hpp                                  | 2 +-
 include/ck/utility/multi_index.hpp                              | 2 +-
 include/ck/utility/number.hpp                                   | 2 +-
 include/ck/utility/reduction_common.hpp                         | 2 +-
 include/ck/utility/reduction_enums.hpp                          | 2 +-
 include/ck/utility/reduction_functions_accumulate.hpp           | 2 +-
 include/ck/utility/reduction_operator.hpp                       | 2 +-
 include/ck/utility/sequence.hpp                                 | 2 +-
 include/ck/utility/sequence_helper.hpp                          | 2 +-
 include/ck/utility/span.hpp                                     | 2 +-
 include/ck/utility/static_buffer.hpp                            | 2 +-
 include/ck/utility/statically_indexed_array.hpp                 | 2 +-
 include/ck/utility/statically_indexed_array_multi_index.hpp     | 2 +-
 include/ck/utility/synchronization.hpp                          | 2 +-
 include/ck/utility/thread_group.hpp                             | 2 +-
 include/ck/utility/transpose_vectors.hpp                        | 2 +-
 include/ck/utility/tuple.hpp                                    | 2 +-
 include/ck/utility/tuple_helper.hpp                             | 2 +-
 include/ck/utility/type.hpp                                     | 2 +-
 .../reference_tensor_operation/cpu/reference_batched_gemm.hpp   | 2 +-
 .../cpu/reference_batchnorm_backward.hpp                        | 2 +-
 .../cpu/reference_batchnorm_forward.hpp                         | 2 +-
 .../cpu/reference_batchnorm_infer.hpp                           | 2 +-
 .../library/reference_tensor_operation/cpu/reference_cgemm.hpp  | 2 +-
 .../reference_tensor_operation/cpu/reference_conv_bwd_data.hpp  | 2 +-
 .../cpu/reference_conv_bwd_weight.hpp                           | 2 +-
 .../reference_tensor_operation/cpu/reference_conv_fwd.hpp       | 2 +-
 .../cpu/reference_conv_fwd_bias_activation.hpp                  | 2 +-
 .../cpu/reference_conv_fwd_bias_activation_add.hpp              | 2 +-
 .../library/reference_tensor_operation/cpu/reference_gemm.hpp   | 2 +-
 .../reference_tensor_operation/cpu/reference_gemm_layernorm.hpp | 2 +-
 .../reference_tensor_operation/cpu/reference_groupnorm.hpp      | 2 +-
 .../reference_tensor_operation/cpu/reference_layernorm.hpp      | 2 +-
 .../reference_tensor_operation/cpu/reference_pool_fwd.hpp       | 2 +-
 .../library/reference_tensor_operation/cpu/reference_reduce.hpp | 2 +-
 .../reference_tensor_operation/cpu/reference_softmax.hpp        | 2 +-
 .../cpu/reference_sparse_embedding3_forward_layernorm.hpp       | 2 +-
 .../library/reference_tensor_operation/gpu/naive_conv_fwd.hpp   | 2 +-
 .../tensor_operation_instance/add_device_operation_instance.hpp | 2 +-
 .../device_operation_instance_factory.hpp                       | 2 +-
 .../ck/library/tensor_operation_instance/gpu/batched_gemm.hpp   | 2 +-
 .../gpu/batched_gemm_add_relu_gemm_add.hpp                      | 2 +-
 .../tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp | 2 +-
 .../gpu/batched_gemm_bias_softmax_gemm_permute.hpp              | 2 +-
 .../library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp | 2 +-
 .../tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp | 2 +-
 .../gpu/batched_gemm_softmax_gemm_permute.hpp                   | 2 +-
 .../tensor_operation_instance/gpu/batchnorm_backward.hpp        | 2 +-
 .../library/tensor_operation_instance/gpu/batchnorm_forward.hpp | 2 +-
 .../library/tensor_operation_instance/gpu/batchnorm_infer.hpp   | 2 +-
 .../tensor_operation_instance/gpu/contraction_bilinear.hpp      | 2 +-
 .../library/tensor_operation_instance/gpu/contraction_scale.hpp | 2 +-
 .../tensor_operation_instance/gpu/convolution_backward_data.hpp | 2 +-
 .../tensor_operation_instance/gpu/convolution_forward.hpp       | 2 +-
 .../gpu/device_elementwise_instance.hpp                         | 2 +-
 .../gpu/device_gemm_mean_squaremean_instance.hpp                | 2 +-
 .../tensor_operation_instance/gpu/elementwise_normalization.hpp | 2 +-
 .../include/ck/library/tensor_operation_instance/gpu/gemm.hpp   | 2 +-
 .../tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp     | 2 +-
 .../library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp | 2 +-
 .../library/tensor_operation_instance/gpu/gemm_add_multiply.hpp | 2 +-
 .../gpu/gemm_add_relu_add_layernorm.hpp                         | 2 +-
 .../ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp  | 2 +-
 .../ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp  | 2 +-
 .../ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp    | 2 +-
 .../gpu/grouped_convolution_backward_data.hpp                   | 2 +-
 .../gpu/grouped_convolution_backward_weight.hpp                 | 2 +-
 .../gpu/grouped_convolution_forward.hpp                         | 2 +-
 .../ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp   | 2 +-
 .../ck/library/tensor_operation_instance/gpu/normalization.hpp  | 2 +-
 .../tensor_operation_instance/gpu/normalization_swish.hpp       | 2 +-
 .../ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp     | 2 +-
 .../ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp     | 2 +-
 .../gpu/quantization/gemm_quantization.hpp                      | 2 +-
 ...grouped_convolution_bias_forward_perchannel_quantization.hpp | 2 +-
 .../grouped_convolution_bias_forward_perlayer_quantization.hpp  | 2 +-
 .../grouped_convolution_forward_perchannel_quantization.hpp     | 2 +-
 .../grouped_convolution_forward_perlayer_quantization.hpp       | 2 +-
 .../gpu/reduce/device_reduce_instance.hpp                       | 2 +-
 .../gpu/reduce/device_reduce_instance_blockwise.hpp             | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp | 2 +-
 .../device_reduce_instance_blockwise_b16_f32_b16_amax.hpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp | 2 +-
 .../device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp      | 2 +-
 .../device_reduce_instance_blockwise_f16_f16_f16_amax.hpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp | 2 +-
 .../device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp | 2 +-
 .../device_reduce_instance_blockwise_f32_f32_f32_amax.hpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp | 2 +-
 .../device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp | 2 +-
 .../device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp | 2 +-
 .../device_reduce_instance_blockwise_f64_f64_f64_amax.hpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp | 2 +-
 .../device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp   | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp   | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp   | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp    | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp    | 2 +-
 .../gpu/reduce/device_reduce_instance_impl_common.hpp           | 2 +-
 .../gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp | 2 +-
 .../gpu/reduce/device_reduce_instance_threadwise.hpp            | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_add.hpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_amax.hpp      | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_avg.hpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_max.hpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_min.hpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp     | 2 +-
 .../device_reduce_instance_threadwise_f16_f16_f16_amax.hpp      | 2 +-
 .../device_reduce_instance_threadwise_f16_f16_f16_max.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f16_f16_min.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f32_f16_add.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f32_f16_avg.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp     | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_add.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_amax.hpp      | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_avg.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_max.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_min.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp     | 2 +-
 .../device_reduce_instance_threadwise_f32_f64_f32_add.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f64_f32_avg.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp     | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_add.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_amax.hpp      | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_avg.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_max.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_min.hpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp     | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp  | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp  | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp  | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp   | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp   | 2 +-
 .../ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp  | 2 +-
 .../ck/library/tensor_operation_instance/gpu/softmax.hpp        | 2 +-
 .../gpu/softmax/device_softmax_f16_f16_instance.hpp             | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp   | 2 +-
 .../gpu/softmax/device_softmax_f16_f16_instance_type.hpp        | 2 +-
 .../gpu/softmax/device_softmax_f32_f32_instance.hpp             | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp   | 2 +-
 .../gpu/softmax/device_softmax_f32_f32_instance_type.hpp        | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance.hpp               | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_type.hpp          | 2 +-
 .../gpu/softmax/device_softmax_instance.hpp                     | 2 +-
 library/include/ck/library/utility/algorithm.hpp                | 2 +-
 library/include/ck/library/utility/check_err.hpp                | 2 +-
 library/include/ck/library/utility/conv_common.hpp              | 2 +-
 .../utility/convolution_host_tensor_descriptor_helper.hpp       | 2 +-
 library/include/ck/library/utility/convolution_parameter.hpp    | 2 +-
 library/include/ck/library/utility/device_memory.hpp            | 2 +-
 library/include/ck/library/utility/fill.hpp                     | 2 +-
 library/include/ck/library/utility/host_common_util.hpp         | 2 +-
 library/include/ck/library/utility/host_gemm.hpp                | 2 +-
 library/include/ck/library/utility/host_tensor.hpp              | 2 +-
 library/include/ck/library/utility/host_tensor_generator.hpp    | 2 +-
 library/include/ck/library/utility/iterator.hpp                 | 2 +-
 library/include/ck/library/utility/literals.hpp                 | 2 +-
 library/include/ck/library/utility/numeric.hpp                  | 2 +-
 library/include/ck/library/utility/ranges.hpp                   | 2 +-
 ...ice_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp | 2 +-
 ...device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp | 2 +-
 ...ice_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp | 2 +-
 ...dd_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 ...dd_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp | 2 +-
 ..._permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp | 2 +-
 ...mm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 ...mm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp | 2 +-
 ...ce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp | 2 +-
 ...ce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp | 2 +-
 ...ce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp | 2 +-
 ...ce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp | 2 +-
 ...mm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 ...dl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 ...te_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 ...dl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 ...te_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp | 2 +-
 .../gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp   | 2 +-
 .../gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp    | 2 +-
 .../gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp    | 2 +-
 .../gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp    | 2 +-
 .../gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp    | 2 +-
 .../gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp     | 2 +-
 .../gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp     | 2 +-
 .../gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp     | 2 +-
 .../gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp      | 2 +-
 .../gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp       | 2 +-
 .../gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp       | 2 +-
 .../gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp       | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp | 2 +-
 ...ear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp | 2 +-
 ...on_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp | 2 +-
 .../device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp    | 2 +-
 .../device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp     | 2 +-
 .../device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp     | 2 +-
 .../device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp    | 2 +-
 .../device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp   | 2 +-
 .../device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp   | 2 +-
 .../device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp  | 2 +-
 .../device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp | 2 +-
 .../device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp  | 2 +-
 .../device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp  | 2 +-
 .../device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp | 2 +-
 ...ice_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp | 2 +-
 .../device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp      | 2 +-
 .../device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp       | 2 +-
 .../device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp       | 2 +-
 .../device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp      | 2 +-
 ..._fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp | 2 +-
 ..._xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp | 2 +-
 ...vice_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp | 2 +-
 ...evice_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp | 2 +-
 ...evice_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp | 2 +-
 ...vice_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp | 2 +-
 .../gpu/elementwise/device_normalize_instance.cpp               | 2 +-
 .../device_elementwise_normalization_f16_instance.cpp           | 2 +-
 .../gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp   | 2 +-
 .../gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp      | 2 +-
 .../gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp      | 2 +-
 .../gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp      | 2 +-
 .../gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp      | 2 +-
 ...gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp | 2 +-
 ...vice_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp | 2 +-
 ...vice_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp | 2 +-
 ...vice_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp | 2 +-
 ...vice_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp    | 2 +-
 .../gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp  | 2 +-
 .../gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp  | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp | 2 +-
 ...tgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp | 2 +-
 ...tgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp | 2 +-
 ...tgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp | 2 +-
 ...tgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp | 2 +-
 ...dl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp | 2 +-
 ..._add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp | 2 +-
 ..._add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp | 2 +-
 ..._add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp | 2 +-
 ..._add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp | 2 +-
 ...emean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp | 2 +-
 ...emean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp | 2 +-
 ...emean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp | 2 +-
 ...emean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp | 2 +-
 ...inear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp | 2 +-
 ...inear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp | 2 +-
 ...inear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp | 2 +-
 ...inear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp | 2 +-
 ...emm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp | 2 +-
 ...emm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp | 2 +-
 ...emm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp | 2 +-
 ...emm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp | 2 +-
 ...educe_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp | 2 +-
 ...educe_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp | 2 +-
 ...educe_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp | 2 +-
 ...educe_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp | 2 +-
 .../device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp    | 2 +-
 .../device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp    | 2 +-
 ...ouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp | 2 +-
 ...rouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp | 2 +-
 ...rouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp | 2 +-
 ...vice_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp | 2 +-
 ...evice_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp | 2 +-
 ...evice_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp | 2 +-
 ...vice_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp | 2 +-
 ...ouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp | 2 +-
 ...ed_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp | 2 +-
 ...ped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp | 2 +-
 ...ped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp | 2 +-
 .../gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp | 2 +-
 ...ice_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp | 2 +-
 ...ice_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp | 2 +-
 .../device_grouped_conv2d_fwd_dl_instance.hpp                   | 2 +-
 ...e_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp | 2 +-
 ...ce_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp | 2 +-
 ...ce_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp | 2 +-
 .../device_grouped_conv2d_fwd_xdl_instance.hpp                  | 2 +-
 ...e_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp | 2 +-
 ...ce_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp | 2 +-
 ...ce_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp | 2 +-
 ...conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp | 2 +-
 ..._conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp | 2 +-
 ..._conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp | 2 +-
 ...rouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp | 2 +-
 ...grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp | 2 +-
 ...grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp | 2 +-
 ...rouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp | 2 +-
 ...rouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp | 2 +-
 ...grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp | 2 +-
 ...grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp | 2 +-
 ...rouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp | 2 +-
 .../device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp   | 2 +-
 .../device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp   | 2 +-
 .../device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp   | 2 +-
 .../device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp   | 2 +-
 ...ce_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp | 2 +-
 ..._gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp | 2 +-
 ...ce_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp | 2 +-
 ..._gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp | 2 +-
 ..._grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp | 2 +-
 ..._grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp | 2 +-
 ..._grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp | 2 +-
 ..._grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp | 2 +-
 .../gpu/normalization/device_groupnorm_f16_instance.cpp         | 2 +-
 .../gpu/normalization/device_groupnorm_f32_instance.cpp         | 2 +-
 .../device_groupnorm_swish_f16_f32_f32_f16_instance.cpp         | 2 +-
 .../gpu/normalization/device_groupnorm_swish_f16_instance.cpp   | 2 +-
 .../gpu/normalization/device_groupnorm_swish_f32_instance.cpp   | 2 +-
 .../gpu/normalization/device_layernorm2d_f16_instance.cpp       | 2 +-
 .../gpu/normalization/device_layernorm2d_f32_instance.cpp       | 2 +-
 .../gpu/normalization/device_layernorm4d_f16_instance.cpp       | 2 +-
 .../gpu/normalization/device_layernorm4d_f32_instance.cpp       | 2 +-
 .../gpu/normalization/normalization_instance_common.hpp         | 2 +-
 .../gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp    | 2 +-
 .../gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp    | 2 +-
 .../gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp   | 2 +-
 .../gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp   | 2 +-
 .../gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp    | 2 +-
 .../gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp    | 2 +-
 .../gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp   | 2 +-
 .../gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp   | 2 +-
 .../gpu/pool_fwd/pool_fwd_instance_common.hpp                   | 2 +-
 .../gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp  | 2 +-
 ...ice_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp | 2 +-
 ...evice_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp | 2 +-
 .../quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp  | 2 +-
 .../device_conv2d_dl_perchannel_quantization_int8_instance.cpp  | 2 +-
 .../device_conv2d_dl_perlayer_quantization_int8_instance.cpp    | 2 +-
 ...ce_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp | 2 +-
 ...vice_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp | 2 +-
 .../quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp | 2 +-
 .../device_conv2d_xdl_perchannel_quantization_int8_instance.cpp | 2 +-
 .../device_conv2d_xdl_perlayer_quantization_int8_instance.cpp   | 2 +-
 .../device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp | 2 +-
 ...emm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp | 2 +-
 ...emm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp | 2 +-
 ...emm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp | 2 +-
 ...emm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp | 2 +-
 ...device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp | 2 +-
 ...mm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp | 2 +-
 ...mm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp | 2 +-
 ...mm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp | 2 +-
 ...mm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp | 2 +-
 .../gpu/quantization/gemm/gemm_quantization_common.hpp          | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp | 2 +-
 .../device_reduce_instance_blockwise_b16_f32_b16_amax.cpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp | 2 +-
 .../device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp      | 2 +-
 .../device_reduce_instance_blockwise_f16_f16_f16_amax.cpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp | 2 +-
 .../device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp | 2 +-
 .../device_reduce_instance_blockwise_f32_f32_f32_amax.cpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp | 2 +-
 .../device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp | 2 +-
 .../device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp | 2 +-
 .../device_reduce_instance_blockwise_f64_f64_f64_amax.cpp       | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp | 2 +-
 .../reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp | 2 +-
 .../device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp      | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp   | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp   | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp   | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp    | 2 +-
 .../reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp    | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp | 2 +-
 ...ce_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_add.cpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_amax.cpp      | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_avg.cpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_max.cpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_min.cpp       | 2 +-
 .../device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp     | 2 +-
 .../device_reduce_instance_threadwise_f16_f16_f16_amax.cpp      | 2 +-
 .../device_reduce_instance_threadwise_f16_f16_f16_max.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f16_f16_min.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f32_f16_add.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f32_f16_avg.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp     | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_add.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_amax.cpp      | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_avg.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_max.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_min.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp     | 2 +-
 .../device_reduce_instance_threadwise_f32_f64_f32_add.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f64_f32_avg.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp     | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_add.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_amax.cpp      | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_avg.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_max.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_min.cpp       | 2 +-
 .../device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp     | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp  | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp  | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp  | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp   | 2 +-
 .../reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp   | 2 +-
 .../gpu/softmax/device_softmax_f16_f16_instance.cpp             | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp   | 2 +-
 .../softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp   | 2 +-
 .../gpu/softmax/device_softmax_f32_f32_instance.cpp             | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp   | 2 +-
 .../softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp   | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance.cpp               | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp | 2 +-
 .../gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp | 2 +-
 library/src/utility/convolution_parameter.cpp                   | 2 +-
 library/src/utility/device_memory.cpp                           | 2 +-
 library/src/utility/host_tensor.cpp                             | 2 +-
 profiler/include/profiler/data_type_enum.hpp                    | 2 +-
 .../profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp    | 2 +-
 .../profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp     | 2 +-
 profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp    | 2 +-
 profiler/include/profiler/profile_batched_gemm_impl.hpp         | 2 +-
 profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp  | 2 +-
 .../include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp | 2 +-
 .../profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp | 2 +-
 profiler/include/profiler/profile_batchnorm_backward_impl.hpp   | 2 +-
 profiler/include/profiler/profile_batchnorm_forward_impl.hpp    | 2 +-
 profiler/include/profiler/profile_batchnorm_infer_impl.hpp      | 2 +-
 profiler/include/profiler/profile_conv_bwd_data_impl.hpp        | 2 +-
 .../include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp    | 2 +-
 profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp   | 2 +-
 profiler/include/profiler/profile_conv_fwd_impl.hpp             | 2 +-
 .../include/profiler/profile_elementwise_layernorm_impl.hpp     | 2 +-
 .../include/profiler/profile_gemm_add_add_fastgelu_impl.hpp     | 2 +-
 profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp    | 2 +-
 profiler/include/profiler/profile_gemm_add_multiply_impl.hpp    | 2 +-
 .../profiler/profile_gemm_add_relu_add_layernorm_impl.hpp       | 2 +-
 profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp | 2 +-
 profiler/include/profiler/profile_gemm_bilinear_impl.hpp        | 2 +-
 profiler/include/profiler/profile_gemm_fastgelu_impl.hpp        | 2 +-
 profiler/include/profiler/profile_gemm_impl.hpp                 | 2 +-
 profiler/include/profiler/profile_gemm_reduce_impl.hpp          | 2 +-
 profiler/include/profiler/profile_gemm_splitk_impl.hpp          | 2 +-
 .../include/profiler/profile_grouped_conv_bwd_weight_impl.hpp   | 2 +-
 profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp     | 2 +-
 .../include/profiler/profile_grouped_gemm_fastgelu_impl.hpp     | 2 +-
 profiler/include/profiler/profile_grouped_gemm_impl.hpp         | 2 +-
 profiler/include/profiler/profile_groupnorm_impl.hpp            | 2 +-
 profiler/include/profiler/profile_layernorm_impl.hpp            | 2 +-
 profiler/include/profiler/profile_pool2d_fwd_impl.hpp           | 2 +-
 profiler/include/profiler/profile_pool3d_fwd_impl.hpp           | 2 +-
 profiler/include/profiler/profile_reduce_impl.hpp               | 2 +-
 profiler/include/profiler/profile_softmax_impl.hpp              | 2 +-
 profiler/src/profile_avg_pool2d_fwd.cpp                         | 2 +-
 profiler/src/profile_batched_gemm.cpp                           | 2 +-
 profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp         | 2 +-
 profiler/src/profile_batched_gemm_gemm.cpp                      | 2 +-
 profiler/src/profile_batched_gemm_reduce.cpp                    | 2 +-
 profiler/src/profile_batchnorm_bwd.cpp                          | 2 +-
 profiler/src/profile_batchnorm_fwd.cpp                          | 2 +-
 profiler/src/profile_batchnorm_infer.cpp                        | 2 +-
 profiler/src/profile_conv_bwd_data.cpp                          | 2 +-
 profiler/src/profile_conv_fwd.cpp                               | 2 +-
 profiler/src/profile_conv_fwd_bias_relu.cpp                     | 2 +-
 profiler/src/profile_conv_fwd_bias_relu_add.cpp                 | 2 +-
 profiler/src/profile_gemm.cpp                                   | 2 +-
 profiler/src/profile_gemm_add_add_fastgelu.cpp                  | 2 +-
 profiler/src/profile_gemm_add_fastgelu.cpp                      | 2 +-
 profiler/src/profile_gemm_add_multiply.cpp                      | 2 +-
 profiler/src/profile_gemm_add_relu_add_layernorm.cpp            | 2 +-
 profiler/src/profile_gemm_bias_add_reduce.cpp                   | 2 +-
 profiler/src/profile_gemm_bilinear.cpp                          | 2 +-
 profiler/src/profile_gemm_fastgelu.cpp                          | 2 +-
 profiler/src/profile_gemm_reduce.cpp                            | 2 +-
 profiler/src/profile_gemm_splitk.cpp                            | 2 +-
 profiler/src/profile_grouped_conv_bwd_weight.cpp                | 2 +-
 profiler/src/profile_grouped_conv_fwd.cpp                       | 2 +-
 profiler/src/profile_grouped_gemm.cpp                           | 2 +-
 profiler/src/profile_grouped_gemm_fastgelu.cpp                  | 2 +-
 profiler/src/profile_groupnorm.cpp                              | 2 +-
 profiler/src/profile_layernorm.cpp                              | 2 +-
 profiler/src/profile_max_pool3d_fwd.cpp                         | 2 +-
 profiler/src/profile_reduce.cpp                                 | 2 +-
 profiler/src/profile_softmax.cpp                                | 2 +-
 profiler/src/profiler.cpp                                       | 2 +-
 profiler/src/profiler_operation_registry.hpp                    | 2 +-
 test/batched_gemm/batched_gemm_bf16.cpp                         | 2 +-
 test/batched_gemm/batched_gemm_fp16.cpp                         | 2 +-
 test/batched_gemm/batched_gemm_fp32.cpp                         | 2 +-
 test/batched_gemm/batched_gemm_int8.cpp                         | 2 +-
 test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp          | 2 +-
 test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp          | 2 +-
 test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp           | 2 +-
 .../test_batched_gemm_softmax_gemm_fp16.cpp                     | 2 +-
 .../test_batched_gemm_softmax_gemm_util.hpp                     | 2 +-
 .../test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp        | 2 +-
 .../test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp        | 2 +-
 .../test_batched_gemm_bias_softmax_gemm_permute_util.hpp        | 2 +-
 .../test_batched_gemm_softmax_gemm_permute_bf16.cpp             | 2 +-
 .../test_batched_gemm_softmax_gemm_permute_fp16.cpp             | 2 +-
 .../test_batched_gemm_softmax_gemm_permute_util.hpp             | 2 +-
 test/batchnorm/batchnorm_bwd_rank_4.cpp                         | 2 +-
 test/batchnorm/batchnorm_fwd_rank_4.cpp                         | 2 +-
 test/batchnorm/batchnorm_infer_rank_4.cpp                       | 2 +-
 test/block_to_ctile_map/test_block_to_ctile_map.cpp             | 2 +-
 test/conv_util/conv_util.cpp                                    | 2 +-
 test/convnd_bwd_data/convnd_bwd_data.cpp                        | 2 +-
 test/convnd_fwd/convnd_fwd.cpp                                  | 2 +-
 test/data_type/int4.cpp                                         | 2 +-
 .../test_elementwise_layernorm_fp16.cpp                         | 2 +-
 test/gemm/gemm_bf16.cpp                                         | 2 +-
 test/gemm/gemm_fp16.cpp                                         | 2 +-
 test/gemm/gemm_fp32.cpp                                         | 2 +-
 test/gemm/gemm_fp64.cpp                                         | 2 +-
 test/gemm/gemm_int8.cpp                                         | 2 +-
 test/gemm/gemm_standalone_xdl_fp16.cpp                          | 2 +-
 test/gemm/gemm_util.hpp                                         | 2 +-
 test/gemm/instance/gemm_f16_nn_instance.cpp                     | 2 +-
 test/gemm/instance/gemm_f16_nn_instance.hpp                     | 2 +-
 test/gemm/instance/gemm_f16_nt_instance.cpp                     | 2 +-
 test/gemm/instance/gemm_f16_nt_instance.hpp                     | 2 +-
 test/gemm/instance/gemm_f16_tn_instance.cpp                     | 2 +-
 test/gemm/instance/gemm_f16_tn_instance.hpp                     | 2 +-
 test/gemm/instance/gemm_f16_tt_instance.cpp                     | 2 +-
 test/gemm/instance/gemm_f16_tt_instance.hpp                     | 2 +-
 test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp             | 2 +-
 test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp             | 2 +-
 test/gemm/run_gemm_test.inc                                     | 2 +-
 test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp   | 2 +-
 test/gemm_reduce/gemm_reduce_fp16.cpp                           | 2 +-
 test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp    | 2 +-
 test/grouped_convnd_fwd/grouped_convnd_fwd.cpp                  | 2 +-
 test/magic_number_division/magic_number_division.cpp            | 2 +-
 test/normalization/test_groupnorm_fp16.cpp                      | 2 +-
 test/normalization/test_groupnorm_fp32.cpp                      | 2 +-
 test/normalization/test_layernorm2d_fp16.cpp                    | 2 +-
 test/normalization/test_layernorm2d_fp32.cpp                    | 2 +-
 test/pool_fwd/test_avg_pool2d_fwd.cpp                           | 2 +-
 test/pool_fwd/test_avg_pool3d_fwd.cpp                           | 2 +-
 test/pool_fwd/test_max_pool2d_fwd.cpp                           | 2 +-
 test/pool_fwd/test_max_pool3d_fwd.cpp                           | 2 +-
 test/pool_fwd/test_pool_fwd_common.hpp                          | 2 +-
 test/reduce/reduce_no_index.cpp                                 | 2 +-
 test/reduce/reduce_with_index.cpp                               | 2 +-
 test/reference_conv_fwd/reference_conv_fwd.cpp                  | 2 +-
 test/softmax/test_softmax_interface.cpp                         | 2 +-
 test/softmax/test_softmax_rank3.cpp                             | 2 +-
 test/softmax/test_softmax_rank4.cpp                             | 2 +-
 test/softmax/test_softmax_util.hpp                              | 2 +-
 test/space_filling_curve/space_filling_curve.cpp                | 2 +-
 test/wmma_op/wmma_op.cpp                                        | 2 +-
 test/wmma_op/wmma_op_util.hpp                                   | 2 +-
 1161 files changed, 1161 insertions(+), 1161 deletions(-)

diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index 8ccfe99c3c..07d8368817 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -4,7 +4,7 @@ This is the list of developers and contributors to Composable Kernel library
 
 
 ## Developers
-[Chao Liu](https://github.com/asroy), [Jing Zhang](https://github.com/zjing14), 2018-2022
+[Chao Liu](https://github.com/asroy), [Jing Zhang](https://github.com/zjing14), 2018-2023
 
 [Letao Qin](https://github.com/ltqin), [Qianfeng Zhang](https://github.com/qianfengz), [Liang Huang](https://github.com/carlushuang), [Shaojie Wang](https://github.com/shaojiewang), 2019-2022
 
diff --git a/LICENSE b/LICENSE
index 2fe9a8455e..e03fddaf78 100644
--- a/LICENSE
+++ b/LICENSE
@@ -7,7 +7,7 @@ Copyright (c) 2020     , Advanced Micro Devices, Inc. (Xiaoyan Zhou)
 Copyright (c) 2021-2022, Advanced Micro Devices, Inc. (Jianfeng Yan)
 
 SPDX-License-Identifier: MIT
-Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal
diff --git a/client_example/01_gemm/gemm.cpp b/client_example/01_gemm/gemm.cpp
index ba7118ba39..c37f208db1 100644
--- a/client_example/01_gemm/gemm.cpp
+++ b/client_example/01_gemm/gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
index 08f297f58a..756889562e 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
index 658c1e9e8f..5965e9d1d6 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
index ea269545a5..319fdb0b03 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp b/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
index caa6573788..1129dfa6b4 100644
--- a/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
+++ b/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp b/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp
index d4f0c2048b..3d5fb60048 100644
--- a/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp
+++ b/client_example/03_gemm_layernorm/gemm_add_relu_add_layernorm_welford.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/04_contraction/contraction_bilinear_fp32.cpp b/client_example/04_contraction/contraction_bilinear_fp32.cpp
index 91dead41a4..89f834b982 100644
--- a/client_example/04_contraction/contraction_bilinear_fp32.cpp
+++ b/client_example/04_contraction/contraction_bilinear_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/04_contraction/contraction_bilinear_fp64.cpp b/client_example/04_contraction/contraction_bilinear_fp64.cpp
index 9238e4cd80..1aa3ba7de5 100644
--- a/client_example/04_contraction/contraction_bilinear_fp64.cpp
+++ b/client_example/04_contraction/contraction_bilinear_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp b/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp
index 62be3377a2..f8ea2258c2 100644
--- a/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp
+++ b/client_example/04_contraction/contraction_g1m2n3k1_add_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/04_contraction/contraction_scale_fp32.cpp b/client_example/04_contraction/contraction_scale_fp32.cpp
index 4e08ee19cd..ba7b0633c3 100644
--- a/client_example/04_contraction/contraction_scale_fp32.cpp
+++ b/client_example/04_contraction/contraction_scale_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/04_contraction/contraction_scale_fp64.cpp b/client_example/04_contraction/contraction_scale_fp64.cpp
index 3c36aa21eb..24e52eb5aa 100644
--- a/client_example/04_contraction/contraction_scale_fp64.cpp
+++ b/client_example/04_contraction/contraction_scale_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <numeric>
diff --git a/client_example/05_layernorm/layernorm2d.cpp b/client_example/05_layernorm/layernorm2d.cpp
index 856a4cc219..4af4d7abe8 100644
--- a/client_example/05_layernorm/layernorm2d.cpp
+++ b/client_example/05_layernorm/layernorm2d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/06_softmax/softmax4d.cpp b/client_example/06_softmax/softmax4d.cpp
index e939ce8dfe..987ac95690 100644
--- a/client_example/06_softmax/softmax4d.cpp
+++ b/client_example/06_softmax/softmax4d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp b/client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp
index 9fbdb83b1c..70be0101c6 100644
--- a/client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp
+++ b/client_example/07_grouped_convnd_fwd/grouped_conv1d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
diff --git a/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
index 0a798be270..57a210fa1f 100644
--- a/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
+++ b/client_example/07_grouped_convnd_fwd/grouped_conv2d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
diff --git a/client_example/08_fused_attention/fused_attention.cpp b/client_example/08_fused_attention/fused_attention.cpp
index fe927da124..df6bc11a70 100644
--- a/client_example/08_fused_attention/fused_attention.cpp
+++ b/client_example/08_fused_attention/fused_attention.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/client_example/08_fused_attention/fused_attention_bias.cpp b/client_example/08_fused_attention/fused_attention_bias.cpp
index 3113b78560..6c9f3bc8f6 100644
--- a/client_example/08_fused_attention/fused_attention_bias.cpp
+++ b/client_example/08_fused_attention/fused_attention_bias.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
index 43a4779f5f..cd504e942e 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_relu_perchannel_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
index 2ff91fe966..f4aa3666b1 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_relu_perlayer_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
index 6ea5dd2231..ebdbbf52c0 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_tanh_perchannel_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
index 33407c9a1c..9d60baee06 100644
--- a/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_bias_tanh_perlayer_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp b/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
index 6f51124707..dd81d9ee6b 100644
--- a/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_perchannel_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp b/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
index 6a11f9fc26..9c088a21d3 100644
--- a/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
+++ b/client_example/09_quantization/conv2d_fwd_perlayer_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/09_quantization/gemm_quantization.cpp b/client_example/09_quantization/gemm_quantization.cpp
index 242504b44f..b14e68fa08 100644
--- a/client_example/09_quantization/gemm_quantization.cpp
+++ b/client_example/09_quantization/gemm_quantization.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp b/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
index 55c7898042..1b2e8abc20 100644
--- a/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
+++ b/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
diff --git a/client_example/11_grouped_conv_bwd_weight/common.hpp b/client_example/11_grouped_conv_bwd_weight/common.hpp
index a906263333..62eb7bcf55 100644
--- a/client_example/11_grouped_conv_bwd_weight/common.hpp
+++ b/client_example/11_grouped_conv_bwd_weight/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iomanip>
diff --git a/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp b/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp
index de68f46d39..bc4a6fe0bf 100644
--- a/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp
+++ b/client_example/12_elementwise_normalization/elementwise_layernorm2d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp b/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
index 8ef21986a4..c0140f71c1 100644
--- a/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp b/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
index 322667a46b..3653733436 100644
--- a/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp b/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp
index 3117d162db..5e6627ce14 100644
--- a/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_infer_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp b/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
index 9cfeee1cfe..d45782d8e0 100644
--- a/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
+++ b/client_example/14_instance_id/batchnorm_fwd_instance_id.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp b/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp
index 28524a9eee..c74d7c6bd8 100644
--- a/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp
+++ b/client_example/15_gemm_add_multiply/gemm_add_multiply.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/15_reduce/reduce_nhwc_c.cpp b/client_example/15_reduce/reduce_nhwc_c.cpp
index 2275158bcb..b45b72f0de 100644
--- a/client_example/15_reduce/reduce_nhwc_c.cpp
+++ b/client_example/15_reduce/reduce_nhwc_c.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <numeric>
diff --git a/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp b/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp
index 223ed29be9..7ba3224fc3 100644
--- a/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp
+++ b/client_example/17_grouped_gemm_fastgelu/grouped_gemm_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <iostream>
diff --git a/client_example/18_groupnorm/groupnorm_swish.cpp b/client_example/18_groupnorm/groupnorm_swish.cpp
index 84f62ceac4..308061a324 100644
--- a/client_example/18_groupnorm/groupnorm_swish.cpp
+++ b/client_example/18_groupnorm/groupnorm_swish.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
index 47bd7738f6..2edaf474b5 100644
--- a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
+++ b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/client_example/19_pool_fwd/max_pool2d_fwd.cpp b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
index 12ee619201..c776dc12da 100644
--- a/client_example/19_pool_fwd/max_pool2d_fwd.cpp
+++ b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iomanip>
 #include <vector>
diff --git a/example/01_gemm/common.hpp b/example/01_gemm/common.hpp
index 495a815962..144c9aaccd 100644
--- a/example/01_gemm/common.hpp
+++ b/example/01_gemm/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/01_gemm/gemm_dl_fp16.cpp b/example/01_gemm/gemm_dl_fp16.cpp
index cf585a8c51..b5fecb9752 100644
--- a/example/01_gemm/gemm_dl_fp16.cpp
+++ b/example/01_gemm/gemm_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_dl_fp32.cpp b/example/01_gemm/gemm_dl_fp32.cpp
index 93f085cdee..212b72f2a6 100644
--- a/example/01_gemm/gemm_dl_fp32.cpp
+++ b/example/01_gemm/gemm_dl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_dl_int4.cpp b/example/01_gemm/gemm_dl_int4.cpp
index e392c490f2..e55ae14013 100644
--- a/example/01_gemm/gemm_dl_int4.cpp
+++ b/example/01_gemm/gemm_dl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/01_gemm/gemm_dl_int8.cpp b/example/01_gemm/gemm_dl_int8.cpp
index be9e387718..1840390aa9 100644
--- a/example/01_gemm/gemm_dl_int8.cpp
+++ b/example/01_gemm/gemm_dl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_wmma_fp16.cpp b/example/01_gemm/gemm_wmma_fp16.cpp
index 58f965be88..b11fe76ab2 100644
--- a/example/01_gemm/gemm_wmma_fp16.cpp
+++ b/example/01_gemm/gemm_wmma_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_bf16.cpp b/example/01_gemm/gemm_xdl_bf16.cpp
index 9aaae6ade9..3cac55ef47 100644
--- a/example/01_gemm/gemm_xdl_bf16.cpp
+++ b/example/01_gemm/gemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_fp16.cpp b/example/01_gemm/gemm_xdl_fp16.cpp
index 50d35fd9ac..54fbd9cdd4 100644
--- a/example/01_gemm/gemm_xdl_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_fp64.cpp b/example/01_gemm/gemm_xdl_fp64.cpp
index 99253b743d..8361576299 100644
--- a/example/01_gemm/gemm_xdl_fp64.cpp
+++ b/example/01_gemm/gemm_xdl_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_int4.cpp b/example/01_gemm/gemm_xdl_int4.cpp
index 7f1283a47b..f6238c7aa5 100644
--- a/example/01_gemm/gemm_xdl_int4.cpp
+++ b/example/01_gemm/gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/01_gemm/gemm_xdl_int8.cpp b/example/01_gemm/gemm_xdl_int8.cpp
index e67594c5bc..cc03200b9d 100644
--- a/example/01_gemm/gemm_xdl_int8.cpp
+++ b/example/01_gemm/gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp b/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
index 12a6992597..3afd0ebdb9 100644
--- a/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
index 3a0ddd90b7..d7176f75da 100644
--- a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/01_gemm/run_gemm_example.inc b/example/01_gemm/run_gemm_example.inc
index 4e2cedb52a..38c72afc60 100644
--- a/example/01_gemm/run_gemm_example.inc
+++ b/example/01_gemm/run_gemm_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp b/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp
index ff99bf4641..877792d740 100644
--- a/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp
+++ b/example/02_gemm_bilinear/gemm_bilinear_wmma_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp b/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
index 917b6b1c31..c3e6ef7d5d 100644
--- a/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
+++ b/example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp b/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
index aee51d05de..dffeff2337 100644
--- a/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
+++ b/example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/04_gemm_add_add_fastgelu/common.hpp b/example/04_gemm_add_add_fastgelu/common.hpp
index 839587c148..91d17df95f 100644
--- a/example/04_gemm_add_add_fastgelu/common.hpp
+++ b/example/04_gemm_add_add_fastgelu/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp
index ba0476b9b9..e630f67837 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
index b940bfd897..71f6677bae 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp
index fa651a34ea..4665c3932f 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_fp32.cpp
@@ -1,4 +1,4 @@
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp
index 9f9c423de2..f206bbeb41 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp
index fadc4ef5ee..e46483ab38 100644
--- a/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp
+++ b/example/04_gemm_add_add_fastgelu/gemm_add_add_fastgelu_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_common.hpp b/example/09_convnd_fwd/convnd_fwd_common.hpp
index 4c594ccdf8..109b8f9ee3 100644
--- a/example/09_convnd_fwd/convnd_fwd_common.hpp
+++ b/example/09_convnd_fwd/convnd_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_common.hpp b/example/09_convnd_fwd/convnd_fwd_dl_common.hpp
index 855710b9d9..aeddd4fc59 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_common.hpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp b/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
index db5a7f0bc3..7b6f18f46a 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_dl_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp b/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
index 964d784c85..551655b177 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_dl_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp b/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
index b0cd88f214..27a3f2e2a9 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_dl_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp
index d55d315491..74cf91d160 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
index d84afba642..f6d69bafd4 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
index f5acc540cf..6c3171f615 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp
index 8d697976ab..9977a496d2 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp b/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
index 99f7f2565c..bf084b3cc0 100644
--- a/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_fwd_common.hpp"
 
diff --git a/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc b/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc
index 697ada14ba..6474df1c35 100644
--- a/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc
+++ b/example/09_convnd_fwd/run_convnd_fwd_dl_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/09_convnd_fwd/run_convnd_fwd_example.inc b/example/09_convnd_fwd/run_convnd_fwd_example.inc
index 36a68056f1..49852ff667 100644
--- a/example/09_convnd_fwd/run_convnd_fwd_example.inc
+++ b/example/09_convnd_fwd/run_convnd_fwd_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp
index 00e370f296..137b0d1ff0 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cassert>
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp
index 6ff29b4b0f..4ccacb0bce 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp
index 02c19c2b63..bf495725e8 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp
index 679bb5c0c4..5848785673 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp
index abdbdaf74d..bf7127502f 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp
index cf86afa8e9..3e1694cbe8 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/convnd_fwd_max_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc b/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc
index b3a3891781..cebfeb51d6 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/run_convnd_fwd_max_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/12_reduce/reduce_blockwise.cpp b/example/12_reduce/reduce_blockwise.cpp
index a7ee9990c1..9a736d4cfa 100644
--- a/example/12_reduce/reduce_blockwise.cpp
+++ b/example/12_reduce/reduce_blockwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <initializer_list>
diff --git a/example/12_reduce/reduce_blockwise_impl.hpp b/example/12_reduce/reduce_blockwise_impl.hpp
index e6e3cc8d52..7f8394a730 100644
--- a/example/12_reduce/reduce_blockwise_impl.hpp
+++ b/example/12_reduce/reduce_blockwise_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/12_reduce/reduce_blockwise_two_call.cpp b/example/12_reduce/reduce_blockwise_two_call.cpp
index dbb18a0d83..eb8b5c76d3 100644
--- a/example/12_reduce/reduce_blockwise_two_call.cpp
+++ b/example/12_reduce/reduce_blockwise_two_call.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/12_reduce/reduce_example_common.hpp b/example/12_reduce/reduce_example_common.hpp
index 05f0a0edb2..5f9a48804a 100644
--- a/example/12_reduce/reduce_example_common.hpp
+++ b/example/12_reduce/reduce_example_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/12_reduce/reduce_multiblock_atomic_add.cpp b/example/12_reduce/reduce_multiblock_atomic_add.cpp
index c4d63a3add..120e3f0595 100644
--- a/example/12_reduce/reduce_multiblock_atomic_add.cpp
+++ b/example/12_reduce/reduce_multiblock_atomic_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <initializer_list>
diff --git a/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp b/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp
index 905242fb6b..fed6218644 100644
--- a/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp
+++ b/example/12_reduce/reduce_multiblock_atomic_add_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_common.hpp b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
index 9abc986713..1157ccd387 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_common.hpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
index 20c3e47017..daf8540d49 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
index 34ff6f435e..323e3f61f8 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp b/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp
index 044f3c166a..2585072dfe 100644
--- a/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp
+++ b/example/14_gemm_quantization/gemm_dl_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp b/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp
index d5f4e6f62c..aa3e011695 100644
--- a/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp
+++ b/example/14_gemm_quantization/gemm_xdl_bias_relu_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp b/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp
index 2371737382..4b207df5c6 100644
--- a/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp
+++ b/example/14_gemm_quantization/gemm_xdl_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp
index a5c51ceb0c..3e1f7f0893 100644
--- a/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_multiple_d_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstddef>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
index 05d572a1f5..680cee1f81 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
index 3f78dafa89..90a12bc1dd 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
index fd93bb5f87..28b0fcd0ce 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
index faf41bbf0b..60c4a71a35 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
index 7cb09778c5..0c96ef56d3 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
index a89937b2e2..743ab96be6 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
index eb3832a668..2f6533d448 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_add_add_mean_meansquare_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
index e1248002f7..b28e7f85d3 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_add_addsquare_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp
index c2feffeb89..b30ce2c48a 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp
index 363390add3..31e2efd6f6 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp
index de6b7eb480..d3c7c1d99c 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp
index 9666fc6622..9a4a6bc6e1 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp
index 00e0b767a4..1a8457a8bf 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_max_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp
index 652c0e6ea6..5c2706c79a 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp
index 7eee24fed8..c119e24370 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp
index c250b99692..0f5e588383 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_mean_meansquare_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_reduce_xdl_common.hpp"
 
diff --git a/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp b/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
index 62992de597..1bea1bcf3e 100644
--- a/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
+++ b/example/16_gemm_multi_d_multi_reduces/gemm_reduce_xdl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <numeric>
 #include <initializer_list>
diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp b/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
index 26fa9e9821..b4b544aabe 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp b/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
index f0896e9771..6b84eaba47 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_bwd_data_common.hpp"
 
diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp b/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp
index c4f2c1f02b..c9989c60ac 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "convnd_bwd_data_common.hpp"
 
diff --git a/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp b/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
index c2e3602a7b..e363dc5c12 100644
--- a/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
+++ b/example/18_batched_gemm_reduce/batched_gemm_reduce_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp b/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
index bee5dea546..24c8d82f67 100644
--- a/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
+++ b/example/19_binary_elementwise/broadcast_add_2d_amn_bn.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp b/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
index 6fc63b899e..3c04c56140 100644
--- a/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
+++ b/example/19_binary_elementwise/broadcast_add_3d_am_bmnk.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/19_binary_elementwise/elementwise_add_1d.cpp b/example/19_binary_elementwise/elementwise_add_1d.cpp
index a5a6bc0a8b..1ac09641a1 100644
--- a/example/19_binary_elementwise/elementwise_add_1d.cpp
+++ b/example/19_binary_elementwise/elementwise_add_1d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/19_binary_elementwise/elementwise_add_4d.cpp b/example/19_binary_elementwise/elementwise_add_4d.cpp
index cc209b12e3..e571aa8468 100644
--- a/example/19_binary_elementwise/elementwise_add_4d.cpp
+++ b/example/19_binary_elementwise/elementwise_add_4d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/20_grouped_conv_bwd_weight/common.hpp b/example/20_grouped_conv_bwd_weight/common.hpp
index 3f4818d2e3..15727495f0 100644
--- a/example/20_grouped_conv_bwd_weight/common.hpp
+++ b/example/20_grouped_conv_bwd_weight/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
index aed6d22b02..3cd70d0f36 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
index 4a2a6195d9..966b586767 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index d46b374766..39b9100bf8 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <ck::index_t NDimSpatial>
 using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
diff --git a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
index 192fe87b62..96d04dcb37 100644
--- a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
index 3f01e69477..fc58ca19f8 100644
--- a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp b/example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
index 4da6da65f7..bd1d6932ac 100644
--- a/example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_layernorm_xdl_naive_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp b/example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp
index e7d857c4a0..90d80f9f03 100644
--- a/example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_xdl_layernorm_naive_single_kernel_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/22_cgemm/cgemm_xdl_bf16.cpp b/example/22_cgemm/cgemm_xdl_bf16.cpp
index 92ed90ce4a..fa4482a984 100644
--- a/example/22_cgemm/cgemm_xdl_bf16.cpp
+++ b/example/22_cgemm/cgemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_common.hpp b/example/22_cgemm/cgemm_xdl_common.hpp
index 6aa06b7c32..26137a7c2e 100644
--- a/example/22_cgemm/cgemm_xdl_common.hpp
+++ b/example/22_cgemm/cgemm_xdl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <numeric>
 #include <initializer_list>
diff --git a/example/22_cgemm/cgemm_xdl_fp16.cpp b/example/22_cgemm/cgemm_xdl_fp16.cpp
index 11373736ee..89a581e865 100644
--- a/example/22_cgemm/cgemm_xdl_fp16.cpp
+++ b/example/22_cgemm/cgemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_fp32.cpp b/example/22_cgemm/cgemm_xdl_fp32.cpp
index 0f45c18c48..cf96599599 100644
--- a/example/22_cgemm/cgemm_xdl_fp32.cpp
+++ b/example/22_cgemm/cgemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_int4.cpp b/example/22_cgemm/cgemm_xdl_int4.cpp
index c26a83baaf..f69cc2b3cc 100644
--- a/example/22_cgemm/cgemm_xdl_int4.cpp
+++ b/example/22_cgemm/cgemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/22_cgemm/cgemm_xdl_int8.cpp b/example/22_cgemm/cgemm_xdl_int8.cpp
index 2f24189861..c4835b853e 100644
--- a/example/22_cgemm/cgemm_xdl_int8.cpp
+++ b/example/22_cgemm/cgemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/example/23_softmax/softmax_blockwise.cpp b/example/23_softmax/softmax_blockwise.cpp
index 41afd72f5a..d09e434bcf 100644
--- a/example/23_softmax/softmax_blockwise.cpp
+++ b/example/23_softmax/softmax_blockwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
index c934d35019..420a7cf74f 100644
--- a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
+++ b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
index 98835f98fa..9d606db205 100644
--- a/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
+++ b/example/25_gemm_bias_e_permute/gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
index 6004db6e0e..78522160c8 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
index 9576ce3f2c..6cceed5bc1 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/26_contraction/contraction_scale_xdl_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp32.cpp
index 3aa2a7ba93..1574f5d18f 100644
--- a/example/26_contraction/contraction_scale_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/26_contraction/contraction_scale_xdl_fp64.cpp b/example/26_contraction/contraction_scale_xdl_fp64.cpp
index cccf6505c6..3dacc70887 100644
--- a/example/26_contraction/contraction_scale_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/27_layernorm/common.hpp b/example/27_layernorm/common.hpp
index 8d833a3ae9..62a71713df 100644
--- a/example/27_layernorm/common.hpp
+++ b/example/27_layernorm/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/27_layernorm/layernorm_fp16.cpp b/example/27_layernorm/layernorm_fp16.cpp
index c15ffabf50..bb8b954f0a 100644
--- a/example/27_layernorm/layernorm_fp16.cpp
+++ b/example/27_layernorm/layernorm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/27_layernorm/layernorm_splitk_fp16.cpp b/example/27_layernorm/layernorm_splitk_fp16.cpp
index 01ee7161eb..e0378d028b 100644
--- a/example/27_layernorm/layernorm_splitk_fp16.cpp
+++ b/example/27_layernorm/layernorm_splitk_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/27_layernorm/run_layernorm_example.inc b/example/27_layernorm/run_layernorm_example.inc
index 678d8df281..95200b540a 100644
--- a/example/27_layernorm/run_layernorm_example.inc
+++ b/example/27_layernorm/run_layernorm_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp b/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp
index f8e6501ead..24e9b1d9b7 100644
--- a/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/28_grouped_gemm_bias_e_permute/grouped_gemm_bias_e_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp
index 30ad38a566..62233e5351 100644
--- a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp
+++ b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_wmma_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp
index 25d815b9cd..08158bfc25 100644
--- a/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/29_batched_gemm_bias_e_permute/batched_gemm_bias_e_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/30_grouped_conv_fwd_multiple_d/common.hpp b/example/30_grouped_conv_fwd_multiple_d/common.hpp
index e7c6ed9b93..e60ebee6e4 100644
--- a/example/30_grouped_conv_fwd_multiple_d/common.hpp
+++ b/example/30_grouped_conv_fwd_multiple_d/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp b/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp
index eb6975a6d8..ae769ff1d3 100644
--- a/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp
+++ b/example/30_grouped_conv_fwd_multiple_d/common_wmma.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp
index 9d1d257a28..039d250299 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common_wmma.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
index ee300d073a..43c0d57dc2 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
index 5a9df0b1e8..40b4132b35 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
index c2906cc9dd..e05d384f26 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
index 3d5a243e6b..5494563fdd 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
index eaf680fa43..6bf2e8d963 100644
--- a/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
+++ b/example/30_grouped_conv_fwd_multiple_d/grouped_conv_fwd_bias_relu_add_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
index 4561156e0b..eb242203ea 100644
--- a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
+++ b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <typename BiasLay, typename ResidualLay>
 struct LayoutSetting
diff --git a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc
index a6888649c0..360b2c8947 100644
--- a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc
+++ b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_bias_relu_add_wmma_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <typename BiasLay, typename ResidualLay>
 struct LayoutSetting
diff --git a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
index d087c31af5..58ed69182e 100644
--- a/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
+++ b/example/30_grouped_conv_fwd_multiple_d/run_grouped_conv_fwd_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 template <ck::index_t NDimSpatial>
 using DeviceConvFwdInstance =
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp
index 74e0e07e62..7605d9c4f8 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp
index d5fadb8081..33ed04fb30 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp
index 0dd4e0914f..e0eb193ad0 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
index 1fd93622a1..d166214c33 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp
index 15d98abab7..40f87d1f55 100644
--- a/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp
+++ b/example/31_batched_gemm_gemm/batched_gemm_gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Gemm fused operation. Computes C_m_o = A_m_k * B0_k_n * B1_n_o
diff --git a/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc b/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc
index 7e5f1614bc..f329146728 100644
--- a/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc
+++ b/example/31_batched_gemm_gemm/run_batched_gemm_gemm_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
index 0eb1565330..1d1566d575 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
index 8f1db577c6..bae88d4b8e 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
index 2ce91a8c60..a098ce6675 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
index 1fd2bf6930..ce8caf7588 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
index f4a8589052..138db14963 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/batched_gemm_scale_softmax_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
index e4a71b0431..d0eb8fcc34 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
index 38b5badc6e..1d97474d20 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g_k_n) * B1_g_n_o
diff --git a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
index 4e43dbdd8f..27602e2313 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
+++ b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
diff --git a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
index 0b876af952..fa76faea84 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
+++ b/example/32_batched_gemm_scale_softmax_gemm/run_batched_gemm_scale_softmax_gemm_permute.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
diff --git a/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc b/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
index ef2acf61f5..ea1e2734a6 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
+++ b/example/32_batched_gemm_scale_softmax_gemm/run_grouped_gemm_scale_softmax_gemm_permute.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run(int argc, char* argv[])
 {
diff --git a/example/33_multiple_reduce/dual_reduce_common.hpp b/example/33_multiple_reduce/dual_reduce_common.hpp
index 326606752b..cd21790be6 100644
--- a/example/33_multiple_reduce/dual_reduce_common.hpp
+++ b/example/33_multiple_reduce/dual_reduce_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/33_multiple_reduce/dual_reduce_multiblock.cpp b/example/33_multiple_reduce/dual_reduce_multiblock.cpp
index 9360599ed9..198931749b 100644
--- a/example/33_multiple_reduce/dual_reduce_multiblock.cpp
+++ b/example/33_multiple_reduce/dual_reduce_multiblock.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/33_multiple_reduce/dual_reduce_threadwise.cpp b/example/33_multiple_reduce/dual_reduce_threadwise.cpp
index 56255839e5..7609edad35 100644
--- a/example/33_multiple_reduce/dual_reduce_threadwise.cpp
+++ b/example/33_multiple_reduce/dual_reduce_threadwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <cstdlib>
diff --git a/example/34_batchnorm/batchnorm_backward_nhwc.cpp b/example/34_batchnorm/batchnorm_backward_nhwc.cpp
index a6ca9d150b..3756310fd7 100644
--- a/example/34_batchnorm/batchnorm_backward_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_backward_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>
diff --git a/example/34_batchnorm/batchnorm_common.hpp b/example/34_batchnorm/batchnorm_common.hpp
index bdbc8ea8b8..a1b8d253bf 100644
--- a/example/34_batchnorm/batchnorm_common.hpp
+++ b/example/34_batchnorm/batchnorm_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp b/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
index dc2984851a..6a8002025a 100644
--- a/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_forward_inferring_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>
diff --git a/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp b/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
index da36d65a29..d680818157 100644
--- a/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <limits>
 #include <iostream>
diff --git a/example/34_batchnorm/batchnorm_infer_impl.hpp b/example/34_batchnorm/batchnorm_infer_impl.hpp
index 15170586b6..d0b545b2a3 100644
--- a/example/34_batchnorm/batchnorm_infer_impl.hpp
+++ b/example/34_batchnorm/batchnorm_infer_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
index 7191ecf50a..1dc21a6c23 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
index efdb315b4e..74fb16e15b 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
index bc2e3d1d52..7506f69420 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
index 4eb2782462..7ebf914408 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
index eefdbca6b1..0fc7a5cc29 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp b/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
index f0a0cdf6f1..d2337dcda5 100644
--- a/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
+++ b/example/36_sparse_embedding/sparse_embedding3_forward_layernorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
index 071e8a7431..b3d0ab6bf7 100644
--- a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
+++ b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 /*
 Computes C_m_o = Relu(A0[m, k] * B0[n, k] + D00[m, n] + D01[mn]) * B1[n, o] + D1[m, o]
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/common.hpp b/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
index d07ee7bdc1..ca824b1075 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
index 55ea8c3a31..a3533bb4cc 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
index ddf82ec512..fb688b6f3f 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc
index 0afd8bd70d..0f0b120cbc 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc
+++ b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_bias_relu_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 bool run_conv_bwd_data_bias_relu(const ExecutionConfig& config,
                                  const ck::utils::conv::ConvParam& conv_params,
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc
index e50c98bbe8..25678491ce 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc
+++ b/example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 bool run_conv_bwd_data(const ExecutionConfig& config,
                        const ck::utils::conv::ConvParam& conv_params,
diff --git a/example/39_permute/common.hpp b/example/39_permute/common.hpp
index ab612cea17..54f3a78809 100644
--- a/example/39_permute/common.hpp
+++ b/example/39_permute/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/39_permute/permute_1xHxW_fp16.cpp b/example/39_permute/permute_1xHxW_fp16.cpp
index d7f9b80544..7336c3b631 100644
--- a/example/39_permute/permute_1xHxW_fp16.cpp
+++ b/example/39_permute/permute_1xHxW_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/39_permute/permute_HxWx4_fp16.cpp b/example/39_permute/permute_HxWx4_fp16.cpp
index 342aa134ec..6c24919ded 100644
--- a/example/39_permute/permute_HxWx4_fp16.cpp
+++ b/example/39_permute/permute_HxWx4_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/39_permute/permute_NxHxW_fp16.cpp b/example/39_permute/permute_NxHxW_fp16.cpp
index b53975eb2c..3551d2a7c8 100644
--- a/example/39_permute/permute_NxHxW_fp16.cpp
+++ b/example/39_permute/permute_NxHxW_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/39_permute/run_permute_bundle_example.inc b/example/39_permute/run_permute_bundle_example.inc
index 70406d63f9..2c19872922 100644
--- a/example/39_permute/run_permute_bundle_example.inc
+++ b/example/39_permute/run_permute_bundle_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/39_permute/run_permute_element_example.inc b/example/39_permute/run_permute_element_example.inc
index bc62353030..3587134456 100644
--- a/example/39_permute/run_permute_element_example.inc
+++ b/example/39_permute/run_permute_element_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/40_conv2d_fwd_quantization/common.hpp b/example/40_conv2d_fwd_quantization/common.hpp
index 6ee14d750e..266b09145c 100644
--- a/example/40_conv2d_fwd_quantization/common.hpp
+++ b/example/40_conv2d_fwd_quantization/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
index 5c445d9c50..40b33852b5 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
index 0ff85f008f..fc081ddc54 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
index f8f996d17e..c390f016a9 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
index 3b25fec0c4..10b131a527 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
index a98a1e240b..e59d0d0759 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
index 262594d58b..aee5fe9e60 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
index 6b22055053..06c839e4e2 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
index 1ac8679743..7a9b42d39f 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp
index f28abe5ebc..3495636297 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perchannel_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
index f468e8adcd..2611337254 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_xdl_perlayer_quantization_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
index 5675db77fa..e5b924ad51 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perchannel_quantization_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 template <ck::index_t NDimSpatial,
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
index 9fd19c1c4e..9f3a769dcf 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_bias_perlayer_quantization_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
index cacedfdadd..9b08fc690d 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perchannel_quantization_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
index 77332cb6d9..267c737e00 100644
--- a/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
+++ b/example/40_conv2d_fwd_quantization/run_conv2d_fwd_perlayer_quantization_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp
index 2aea08c400..e37d413695 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp
index b7f80e76d6..496e676a40 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp
index 15e460948e..35d50721dc 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
index 2cc4c07c0d..80f6e9ae05 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 #error Should compile this file with ck::int4_t support
diff --git a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp
index 40ff0f69cc..3ade6c811a 100644
--- a/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp
+++ b/example/41_grouped_conv_conv_fwd/grouped_conv_conv_fwd_xdl_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc b/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc
index a2c97f4d42..0722d497d8 100644
--- a/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc
+++ b/example/41_grouped_conv_conv_fwd/run_grouped_conv_conv_fwd_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/42_groupnorm/common.hpp b/example/42_groupnorm/common.hpp
index 780154b26c..c8f91eb53b 100644
--- a/example/42_groupnorm/common.hpp
+++ b/example/42_groupnorm/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
index b07a26c4c9..cc107b63dc 100644
--- a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/42_groupnorm/groupnorm_splitk_fp16.cpp b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
index fd4bfe3807..057b240a63 100644
--- a/example/42_groupnorm/groupnorm_splitk_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/42_groupnorm/groupnorm_swish_fp16.cpp b/example/42_groupnorm/groupnorm_swish_fp16.cpp
index c52243bfb0..363f22ed4c 100644
--- a/example/42_groupnorm/groupnorm_swish_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_swish_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 
diff --git a/example/42_groupnorm/run_groupnorm_example.inc b/example/42_groupnorm/run_groupnorm_example.inc
index d1016a3b12..16065c8d46 100644
--- a/example/42_groupnorm/run_groupnorm_example.inc
+++ b/example/42_groupnorm/run_groupnorm_example.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
index 7ac4b68272..b6d9b29a5b 100644
--- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
index 764e55ef55..60a0e01fe3 100644
--- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
+++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp b/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
index 7d6ff12eea..76361f87a5 100644
--- a/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
+++ b/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/example/46_gemm_add_multiply/common.hpp b/example/46_gemm_add_multiply/common.hpp
index 3ba78dfe47..2c656cf441 100644
--- a/example/46_gemm_add_multiply/common.hpp
+++ b/example/46_gemm_add_multiply/common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp b/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp
index 28c3939fa6..58a399f226 100644
--- a/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp
+++ b/example/46_gemm_add_multiply/gemm_add_multiply_dl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp"
diff --git a/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp b/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp
index d5aa41f1b6..56417b101d 100644
--- a/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp
+++ b/example/46_gemm_add_multiply/gemm_add_multiply_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
diff --git a/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp b/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
index 30c98e534a..cfb42c6e17 100644
--- a/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
+++ b/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/example/48_pool3d_fwd/pool3d_fwd_common.hpp b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
index 5706deb6d1..565bb94e47 100644
--- a/example/48_pool3d_fwd/pool3d_fwd_common.hpp
+++ b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 #pragma once
 
 #include <iostream>
diff --git a/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
index 4d3686bcb0..9afb51201d 100644
--- a/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
+++ b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 1626597ed2..314e6a8134 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/device_prop.hpp b/include/ck/host_utility/device_prop.hpp
index e2cbdb7332..bd02d5d88a 100644
--- a/include/ck/host_utility/device_prop.hpp
+++ b/include/ck/host_utility/device_prop.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/hip_check_error.hpp b/include/ck/host_utility/hip_check_error.hpp
index d3dc8eaf1e..af7bebd9d6 100644
--- a/include/ck/host_utility/hip_check_error.hpp
+++ b/include/ck/host_utility/hip_check_error.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/io.hpp b/include/ck/host_utility/io.hpp
index ac8719592d..55734bab2e 100644
--- a/include/ck/host_utility/io.hpp
+++ b/include/ck/host_utility/io.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/host_utility/kernel_launch.hpp b/include/ck/host_utility/kernel_launch.hpp
index 24f2121674..58740b4351 100644
--- a/include/ck/host_utility/kernel_launch.hpp
+++ b/include/ck/host_utility/kernel_launch.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp b/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp
index 4126753655..6b118e972e 100644
--- a/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/problem_transform/transform_forward_convolution3d_into_gemm_v4r4r4_ndhwc_kzyxc_ndhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
 #define CK_TRANSFORM_FORWARD_CONVOLUTION3D_INTO_GEMM_V4R4R4_NHWC_KYXC_NHWK_HPP
diff --git a/include/ck/stream_config.hpp b/include/ck/stream_config.hpp
index 70ca34555a..505a602b24 100644
--- a/include/ck/stream_config.hpp
+++ b/include/ck/stream_config.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor/static_tensor.hpp b/include/ck/tensor/static_tensor.hpp
index fee679f910..d719ef9760 100644
--- a/include/ck/tensor/static_tensor.hpp
+++ b/include/ck/tensor/static_tensor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_STATIC_TENSOR_HPP
 #define CK_STATIC_TENSOR_HPP
diff --git a/include/ck/tensor_description/cluster_descriptor.hpp b/include/ck/tensor_description/cluster_descriptor.hpp
index 0c9ea2ff2a..2dfcad8e04 100644
--- a/include/ck/tensor_description/cluster_descriptor.hpp
+++ b/include/ck/tensor_description/cluster_descriptor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/multi_index_transform.hpp b/include/ck/tensor_description/multi_index_transform.hpp
index 4e4d7593e9..6854226dd4 100644
--- a/include/ck/tensor_description/multi_index_transform.hpp
+++ b/include/ck/tensor_description/multi_index_transform.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/multi_index_transform_helper.hpp b/include/ck/tensor_description/multi_index_transform_helper.hpp
index 044a903700..af0a8a34d0 100644
--- a/include/ck/tensor_description/multi_index_transform_helper.hpp
+++ b/include/ck/tensor_description/multi_index_transform_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_adaptor.hpp b/include/ck/tensor_description/tensor_adaptor.hpp
index d42e0a6ff0..3ffac32469 100644
--- a/include/ck/tensor_description/tensor_adaptor.hpp
+++ b/include/ck/tensor_description/tensor_adaptor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_descriptor.hpp b/include/ck/tensor_description/tensor_descriptor.hpp
index f07d5b1733..f1df2eedd4 100644
--- a/include/ck/tensor_description/tensor_descriptor.hpp
+++ b/include/ck/tensor_description/tensor_descriptor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_descriptor_helper.hpp b/include/ck/tensor_description/tensor_descriptor_helper.hpp
index 461aae72cf..f3ac041bf9 100644
--- a/include/ck/tensor_description/tensor_descriptor_helper.hpp
+++ b/include/ck/tensor_description/tensor_descriptor_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_description/tensor_space_filling_curve.hpp b/include/ck/tensor_description/tensor_space_filling_curve.hpp
index 17c9100b9f..9a326092d2 100644
--- a/include/ck/tensor_description/tensor_space_filling_curve.hpp
+++ b/include/ck/tensor_description/tensor_space_filling_curve.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
index 8b1b7be11e..b3caa3214a 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp
index 33120bd86f..b0143366c1 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v2r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_BLOCKWISE_GEMM_DLOPS_V2R2_HPP
 #define CK_BLOCKWISE_GEMM_DLOPS_V2R2_HPP
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp
index f45655721f..0d092da516 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dlops_v3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_BLOCKWISE_GEMM_DLOPS_V3_HPP
 #define CK_BLOCKWISE_GEMM_DLOPS_V3_HPP
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
index d75f37d7b3..5ec964bd3f 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
index 5328dfde9b..d5a64d7aa6 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
index aa814ab009..8ae1ba3f34 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp b/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
index 7e62a822a8..82bcff6947 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp b/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp
index 03e4d42d3a..d8da134a34 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp b/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
index 316508651e..a3813ea248 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp b/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
index 2163ad3238..6c13513cfb 100644
--- a/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
+++ b/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
index 04ad75bd7d..c8690e5f68 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
index 5c47a49b38..905a59f56e 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp
index aa33fc083f..17110c8358 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp
index eb5f589a4a..9a5317dd12 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp
index 3bd7806389..993d90e356 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
index a4a29f5d5e..f3263c7216 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp
index 20b2a152b9..01bb806789 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp
index 953ff1e06e..adfa1689c6 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_base.hpp b/include/ck/tensor_operation/gpu/device/device_base.hpp
index 5946daf21e..1981690111 100644
--- a/include/ck/tensor_operation/gpu/device/device_base.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_base.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp
index 9fcd893c7a..ee7af0117d 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp
index e755913280..6cc2c7bb2f 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp
index af681127f3..91b4b6b91b 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp
index 116e62c009..f18dc32906 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp
index eacc5976d3..8234e29486 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_multiple_d_gemm_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp
index c1f85e575c..09259224e7 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp
index bde71806da..be8105c967 100644
--- a/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batched_gemm_softmax_gemm_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp b/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp
index d39f3b7cbc..2c0da69257 100644
--- a/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batchnorm_backward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp b/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp
index aa93dd9c19..e3962e177e 100644
--- a/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp b/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp
index 8a00fd9db3..69103b6f44 100644
--- a/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_batchnorm_infer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_cgemm.hpp b/include/ck/tensor_operation/gpu/device/device_cgemm.hpp
index aedae53800..8484212118 100644
--- a/include/ck/tensor_operation/gpu/device/device_cgemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_cgemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "device_base.hpp"
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
index dbc525c099..118ade8978 100644
--- a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp b/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp
index 82054a3c94..eb1b85ec82 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_bwd_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp
index 4b9881088d..4dc11dbefd 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp
index 5a627deeb2..7d3845666c 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp
index cc139303c9..3a49ac632e 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_fwd_bias_activation_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_elementwise.hpp b/include/ck/tensor_operation/gpu/device/device_elementwise.hpp
index f9f913a7c1..db0e4bd83f 100644
--- a/include/ck/tensor_operation/gpu/device/device_elementwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_elementwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp b/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp
index 9491a92247..c56a947ec9 100644
--- a/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_elementwise_normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm.hpp b/include/ck/tensor_operation/gpu/device/device_gemm.hpp
index c0af6f80fa..adf909821d 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp
index 4c2161eaed..a7f42c3b35 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_bias_e_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp
index 9113bb7b74..a44356dc24 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp
index a67a09b874..0258858fe5 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp
index f4881e32f6..539e83f7cb 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_d_multiple_r.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp
index fcc088ca43..eaa7671c64 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
index c701bff57f..6407aa7e09 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
index af38f14254..d00e19b440 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp
index 173c613a32..ba81948440 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
index 3350aec8d3..7e4bca2bd6 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
index 1258aed71c..de54f9be20 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
index 644c7ee9a9..025c43e75c 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 6180333350..f26974ccb9 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
index 80c864c836..9080936589 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
index b066a44585..fae6509740 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 0b1db28465..9c8b5f462e 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp b/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
index ee4b53e2fc..f68022ca04 100644
--- a/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_multiple_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_normalization.hpp b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
index 03601ce831..1f178f9fcb 100644
--- a/include/ck/tensor_operation/gpu/device/device_normalization.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_permute.hpp b/include/ck/tensor_operation/gpu/device/device_permute.hpp
index 9daa2be373..c994cf02c6 100644
--- a/include/ck/tensor_operation/gpu/device/device_permute.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
index e801e98a25..8b227fdfbe 100644
--- a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_reduce.hpp b/include/ck/tensor_operation/gpu/device/device_reduce.hpp
index c9209f2d7d..c2721b1845 100644
--- a/include/ck/tensor_operation/gpu/device/device_reduce.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_softmax.hpp b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
index 94f788e517..a96ba89e24 100644
--- a/include/ck/tensor_operation/gpu/device/device_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
index f59e6093e2..eeccd977cc 100644
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index 70990e795c..95517b1074 100644
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp b/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
index fc913e9ba0..0bb45b18c3 100644
--- a/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/gemm_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
index 493822aeb2..4d599e8017 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_wmma_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index 9bf8f5ccd9..bc86e78b62 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
index 20e9920d93..09220813bf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index 0df3460941..2d91c620c6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index 196dc86da1..e39d8f0696 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
index ef9b90ba7c..eff503f650 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 0c6c0ef7ad..e54c013cf9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
index 84edde63ec..9310d0752a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index d35f194171..b100967065 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp
index ab16a757f6..f46237e005 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_backward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
index 5a16ff765b..ce0320b2d7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
index 0c9cce97fb..9ee6c6f46b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index 1eaffe705a..b6f38698cd 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index b65afce8df..2ab09ba5c8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
index ea3020663a..8cef0eaf95 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
index e7e2bf3354..bb8f531613 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
index 6c4957b9b1..1bd1e553cd 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index 027f1a1954..de6bf27fec 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
index 6278220c2f..710ea91760 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
index f69d8f18ae..cd89f3232c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_naive_ndhwc_kzyxc_ndhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef DEVICE_CONV3D_FWD_NAIVE_HPP
 #define DEVICE_CONV3D_FWD_NAIVE_HPP
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
index d52879cd90..fd8c88da9a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef DEVICE_CONV3D_FWD_XDL_HPP
 #define DEVICE_CONV3D_FWD_XDL_HPP
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
index aff25aa7c6..3178f73f4b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
index 1fd4b76cec..822e1da4ec 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
index 83ed6198bd..c5f90e40fe 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
index a11b5d0398..5618fba517 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp
index 1fa69288a4..c3416758d1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
index a9a58c8ac4..63f7fa7068 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
index 36366a7631..13e9f96910 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
index c91328ff71..22f66931f0 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
index 580087e00c..1ab8362472 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index 4c1c3ab7bf..be174e599e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
index 750df31a3f..44b3518e2c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index 2488101484..bd5be99f8b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
index f358bd176f..e60fae370f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
index 03ffcf8e59..f64450b75f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
index a5051455bf..528d9bf42f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index 13a30911a1..a3dda82aa6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
index 8ee138f827..14ac5420ae 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
index 36b01f677f..ef5e267819 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 89bfc180a5..3801993412 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 76dd5a3666..a275ee1023 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 92c20a308f..3f6238d21f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
index c921c9f1b4..3bcc7bd64b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp
index 03185d5b1d..face627e1f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
index de40d71293..71e4e28bf2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
index 9d4b68c0b6..526108b874 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
index e88bf8ed75..d1f1b7fcc9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -1,6 +1,6 @@
 #pragma once
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index e3795060be..77a8e6ecbc 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -1,6 +1,6 @@
 #pragma once
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 54ad9eb065..cd39cc983b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp
index b49e109682..aec5a65ccf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp
index 17a96e9f6f..6d1d5c8e23 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_multiple_reduce_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
index 6a8037a324..ea0d805043 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
index 0026a87593..8b2b3c41bf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp
index 7b96373c0f..17dab08332 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_permute_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
index 6933db68dc..3f27c629dd 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
index d330fda8ce..0ab6c24758 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp
index 5dc051be3c..2481c5c769 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp
index c7868537fe..bf3deeb57a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp b/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
index aa255da645..6c5895b010 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_reduce_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
index ed96b7340c..4aa02dfd32 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp b/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
index 2f29224a75..7a62ec0465 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_sparse_embeddings_forward_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/masking_specialization.hpp b/include/ck/tensor_operation/gpu/device/masking_specialization.hpp
index ea0f5897a7..d6d6f74abd 100644
--- a/include/ck/tensor_operation/gpu/device/masking_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/masking_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/matrix_padder.hpp b/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
index 70e61bc772..c66d2fc516 100644
--- a/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
+++ b/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp b/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp
index d35318357a..5351d4ef24 100644
--- a/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp
+++ b/include/ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/tensor_layout.hpp b/include/ck/tensor_operation/gpu/device/tensor_layout.hpp
index b44427411f..b2d141fd61 100644
--- a/include/ck/tensor_operation/gpu/device/tensor_layout.hpp
+++ b/include/ck/tensor_operation/gpu/device/tensor_layout.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp b/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp
index 0ec0df2c9b..713fc93ebb 100644
--- a/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/tensor_specialization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/welford_helper.hpp b/include/ck/tensor_operation/gpu/device/welford_helper.hpp
index 6c909b767d..d7772d8764 100644
--- a/include/ck/tensor_operation/gpu/device/welford_helper.hpp
+++ b/include/ck/tensor_operation/gpu/device/welford_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
index 136017c6d1..1dd96809d5 100644
--- a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
index ceb2b665b9..3fdb391a02 100644
--- a/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index ef250b8bfd..c3e7706ef3 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
index a72a4ee068..e73e7e6817 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
index 08cb0dd191..fc263138f9 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
index 548d7fd40a..1f8990e6df 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
index 42b7e172b2..6fe78edb34 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index ad91c3c687..91ed1112f6 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
index aa34cfbf84..523e7f7c53 100644
--- a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp
index fbe89e7e5e..69468c25be 100644
--- a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_welford_second_half_layernorm2d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp
index bdebe3816f..bd1e0585fc 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp
index 1313ec9435..fc4f27e33b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_multiple_reduction_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp
index 6836a66047..203be3c42d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_multiblock.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
index 5986641c6f..910c926c7e 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
index fccb127d0f..a8a1f803fc 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
index b9f4a3080a..59d6bad5d9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
index 6a6f19d71e..74171ea9d9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
index d6d2051113..9eb2bf8aa8 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp
index ede6a96dc9..ed1ffdd857 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_backward_blockwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp
index 33c45a0f03..b6c83af13a 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp
index 8b82b65540..d686c14b35 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp
index 05257d1627..bf0e8c186c 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_2d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// // Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 //
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp
index b09a735902..3ea72b8534 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_elementwise_layernorm_welford_variance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
index bebcdceb43..c5c09e909e 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp
index 9c68b4f5c3..27f48a84ba 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index 578665ea85..7289a20da0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
index d3f81566e9..119c1ea596 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
index 98331d8544..9090c16dcd 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
index 3281b910d3..f54345b04f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
index a3f5324713..bb3e6a80bf 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
index aa89bff9ee..e7577bdcbd 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp
index 2d3a36fca0..de5a424198 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_waveletmodel.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
index 397ae1c1b9..740dde5c65 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index e6303d76c1..d5552656e3 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
index 2d4ebe7076..d805c9fa2b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
index acece0fbba..31c59d14ed 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
index 8d86f3c1d7..e9881d645b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
index 55f465a037..33a4f2b2c5 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index d56d1986e2..a4b320ddfa 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
index 8259927fec..df543c063f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
index 5d5fdae170..ec98fc9c93 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
index dc83f8e984..3a752dd749 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp
index de1ae91592..61d0f9e0d5 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
index 901e7aee98..41352fabeb 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_set_buffer_value.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp
index 88c7b6acfe..0ad36b418a 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_set_multiple_buffer_value.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp
index 0344e68305..5f56ac6fc4 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
index ff2511fa6e..ee68660a06 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
index 792ffabcb9..c3f122106d 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
index 632690e1ef..e50fb98133 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
index 129b4e1165..fc42e97629 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
index d796d1afc0..136ac94e7f 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
index 3a7ae459e5..ff9712276c 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp b/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp
index 188c62d93b..c6eecc067d 100644
--- a/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp
+++ b/include/ck/tensor_operation/gpu/thread/reduction_functions_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp
index 94cdfe0108..44730d551c 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp
index e045e3b545..e97aa433a6 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_gemm_dlops_v3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_THREADWISE_GEMM_DLOPS_V3_HPP
 #define CK_THREADWISE_GEMM_DLOPS_V3_HPP
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp
index 0a1197a163..6774a35bcb 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
index 16484ddcc5..605f2569c6 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp
index 6e8a23930b..6a6c1f2ac5 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp
index f13da341f9..bd01108b03 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v5r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
index 9c91cd9ca8..6ec9abc417 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp
index 68bc2726f4..cf2c7a2aee 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp
index 0f5fb88b04..b5847e51b4 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp
index 2eb1b0ee90..db7dee2199 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp
index 12ba2c5381..eb6715e8eb 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_welford.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp b/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
index 24efeb2dea..979f3567e9 100644
--- a/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/wmma_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index 319487bc05..faaa2c5a95 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp b/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp
index 5fc11d9158..ea27a40ce3 100644
--- a/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_contraction_to_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
index 13d0a28cfe..505ed33d50 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
index 1b5e64b66c..cee3d2825b 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
@@ -1,6 +1,6 @@
 
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/amd_address_space.hpp b/include/ck/utility/amd_address_space.hpp
index 9f1525914c..d54f70e750 100644
--- a/include/ck/utility/amd_address_space.hpp
+++ b/include/ck/utility/amd_address_space.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index bdfb4f2758..ef3d2032c5 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "data_type.hpp"
diff --git a/include/ck/utility/amd_inline_asm.hpp b/include/ck/utility/amd_inline_asm.hpp
index 1f7df70bcd..43baa817d3 100644
--- a/include/ck/utility/amd_inline_asm.hpp
+++ b/include/ck/utility/amd_inline_asm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_AMD_INLINE_ASM_HPP
 #define CK_AMD_INLINE_ASM_HPP
diff --git a/include/ck/utility/amd_wmma.hpp b/include/ck/utility/amd_wmma.hpp
index bf09142548..dd7f0b770a 100644
--- a/include/ck/utility/amd_wmma.hpp
+++ b/include/ck/utility/amd_wmma.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_AMD_WMMA_HPP
 #define CK_AMD_WMMA_HPP
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index a742496fc1..f4a04e2814 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_AMD_XDLOPS_HPP
 #define CK_AMD_XDLOPS_HPP
diff --git a/include/ck/utility/array.hpp b/include/ck/utility/array.hpp
index 370a457fe9..f63ce5e5a0 100644
--- a/include/ck/utility/array.hpp
+++ b/include/ck/utility/array.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_ARRAY_HPP
 #define CK_ARRAY_HPP
diff --git a/include/ck/utility/array_multi_index.hpp b/include/ck/utility/array_multi_index.hpp
index 9b8d5b95e9..c0c1ea65fc 100644
--- a/include/ck/utility/array_multi_index.hpp
+++ b/include/ck/utility/array_multi_index.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_ARRAY_MULTI_INDEX_HPP
 #define CK_ARRAY_MULTI_INDEX_HPP
diff --git a/include/ck/utility/c_style_pointer_cast.hpp b/include/ck/utility/c_style_pointer_cast.hpp
index 6e8b008158..610e393a77 100644
--- a/include/ck/utility/c_style_pointer_cast.hpp
+++ b/include/ck/utility/c_style_pointer_cast.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_C_STYLE_POINTER_CAST_HPP
 #define CK_C_STYLE_POINTER_CAST_HPP
diff --git a/include/ck/utility/common_header.hpp b/include/ck/utility/common_header.hpp
index 8da87c8766..41a9d0b585 100644
--- a/include/ck/utility/common_header.hpp
+++ b/include/ck/utility/common_header.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/container_element_picker.hpp b/include/ck/utility/container_element_picker.hpp
index abc5185e04..838147e420 100644
--- a/include/ck/utility/container_element_picker.hpp
+++ b/include/ck/utility/container_element_picker.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_CONTAINER_ELEMENT_PICKER_HPP
 #define CK_CONTAINER_ELEMENT_PICKER_HPP
diff --git a/include/ck/utility/container_helper.hpp b/include/ck/utility/container_helper.hpp
index c8b02bc5ac..9c7b954565 100644
--- a/include/ck/utility/container_helper.hpp
+++ b/include/ck/utility/container_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_CONTAINER_HELPER_HPP
 #define CK_CONTAINER_HELPER_HPP
diff --git a/include/ck/utility/debug.hpp b/include/ck/utility/debug.hpp
index 593bbb7116..80346f0d9f 100644
--- a/include/ck/utility/debug.hpp
+++ b/include/ck/utility/debug.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef UTILITY_DEBUG_HPP
 #define UTILITY_DEBUG_HPP
diff --git a/include/ck/utility/dynamic_buffer.hpp b/include/ck/utility/dynamic_buffer.hpp
index c6f0d299ef..9ea0d6c001 100644
--- a/include/ck/utility/dynamic_buffer.hpp
+++ b/include/ck/utility/dynamic_buffer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/enable_if.hpp b/include/ck/utility/enable_if.hpp
index 297434b0dd..c0a3c99f1f 100644
--- a/include/ck/utility/enable_if.hpp
+++ b/include/ck/utility/enable_if.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional.hpp b/include/ck/utility/functional.hpp
index 08e730782f..91797d2409 100644
--- a/include/ck/utility/functional.hpp
+++ b/include/ck/utility/functional.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional2.hpp b/include/ck/utility/functional2.hpp
index 6f125ca4c9..99c65f4eb8 100644
--- a/include/ck/utility/functional2.hpp
+++ b/include/ck/utility/functional2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional3.hpp b/include/ck/utility/functional3.hpp
index 06b67ef7e3..97605a7ade 100644
--- a/include/ck/utility/functional3.hpp
+++ b/include/ck/utility/functional3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/functional4.hpp b/include/ck/utility/functional4.hpp
index 6eeaf15c9b..b5f3df8d7c 100644
--- a/include/ck/utility/functional4.hpp
+++ b/include/ck/utility/functional4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_FUNCTIONAL4_HPP
 #define CK_FUNCTIONAL4_HPP
diff --git a/include/ck/utility/generic_memory_space_atomic.hpp b/include/ck/utility/generic_memory_space_atomic.hpp
index 6a1ca96652..98f40a4363 100644
--- a/include/ck/utility/generic_memory_space_atomic.hpp
+++ b/include/ck/utility/generic_memory_space_atomic.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "data_type.hpp"
diff --git a/include/ck/utility/get_id.hpp b/include/ck/utility/get_id.hpp
index 44ff438155..77564c6130 100644
--- a/include/ck/utility/get_id.hpp
+++ b/include/ck/utility/get_id.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/ignore.hpp b/include/ck/utility/ignore.hpp
index ac33cbf9a5..f70a182fd4 100644
--- a/include/ck/utility/ignore.hpp
+++ b/include/ck/utility/ignore.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index b65640bfff..7828d21d7f 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 #include "data_type.hpp"
diff --git a/include/ck/utility/integral_constant.hpp b/include/ck/utility/integral_constant.hpp
index 9aab4e2421..376070eb3d 100644
--- a/include/ck/utility/integral_constant.hpp
+++ b/include/ck/utility/integral_constant.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/is_known_at_compile_time.hpp b/include/ck/utility/is_known_at_compile_time.hpp
index 8198154422..2cafc3e6f2 100644
--- a/include/ck/utility/is_known_at_compile_time.hpp
+++ b/include/ck/utility/is_known_at_compile_time.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/magic_division.hpp b/include/ck/utility/magic_division.hpp
index a5e8e92165..f19030d4e9 100644
--- a/include/ck/utility/magic_division.hpp
+++ b/include/ck/utility/magic_division.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/math.hpp b/include/ck/utility/math.hpp
index 72071992f6..326b0e61ef 100644
--- a/include/ck/utility/math.hpp
+++ b/include/ck/utility/math.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/math_v2.hpp b/include/ck/utility/math_v2.hpp
index a3732b2fe0..1cac2cc0c7 100644
--- a/include/ck/utility/math_v2.hpp
+++ b/include/ck/utility/math_v2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/multi_index.hpp b/include/ck/utility/multi_index.hpp
index 1d544c0906..9f7ba8bff6 100644
--- a/include/ck/utility/multi_index.hpp
+++ b/include/ck/utility/multi_index.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/number.hpp b/include/ck/utility/number.hpp
index f3ca6b61dc..d29afd31a7 100644
--- a/include/ck/utility/number.hpp
+++ b/include/ck/utility/number.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_NUMBER_HPP
 #define CK_NUMBER_HPP
diff --git a/include/ck/utility/reduction_common.hpp b/include/ck/utility/reduction_common.hpp
index aceef7b296..3777d297c8 100644
--- a/include/ck/utility/reduction_common.hpp
+++ b/include/ck/utility/reduction_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/reduction_enums.hpp b/include/ck/utility/reduction_enums.hpp
index 6785633105..23b7149f8e 100644
--- a/include/ck/utility/reduction_enums.hpp
+++ b/include/ck/utility/reduction_enums.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/reduction_functions_accumulate.hpp b/include/ck/utility/reduction_functions_accumulate.hpp
index 724e5599d6..b9765ff0d2 100644
--- a/include/ck/utility/reduction_functions_accumulate.hpp
+++ b/include/ck/utility/reduction_functions_accumulate.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/reduction_operator.hpp b/include/ck/utility/reduction_operator.hpp
index b4e770a64e..0f5b73cb03 100644
--- a/include/ck/utility/reduction_operator.hpp
+++ b/include/ck/utility/reduction_operator.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/sequence.hpp b/include/ck/utility/sequence.hpp
index 97b597221c..d6bfb2eba1 100644
--- a/include/ck/utility/sequence.hpp
+++ b/include/ck/utility/sequence.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/sequence_helper.hpp b/include/ck/utility/sequence_helper.hpp
index db25c27e70..8c493a2822 100644
--- a/include/ck/utility/sequence_helper.hpp
+++ b/include/ck/utility/sequence_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/span.hpp b/include/ck/utility/span.hpp
index 1e50121454..5e7567a847 100644
--- a/include/ck/utility/span.hpp
+++ b/include/ck/utility/span.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/static_buffer.hpp b/include/ck/utility/static_buffer.hpp
index dd25c96203..835f565730 100644
--- a/include/ck/utility/static_buffer.hpp
+++ b/include/ck/utility/static_buffer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/statically_indexed_array.hpp b/include/ck/utility/statically_indexed_array.hpp
index 3438776f41..a2d70045a4 100644
--- a/include/ck/utility/statically_indexed_array.hpp
+++ b/include/ck/utility/statically_indexed_array.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_STATICALLY_INDEXED_ARRAY_HPP
 #define CK_STATICALLY_INDEXED_ARRAY_HPP
diff --git a/include/ck/utility/statically_indexed_array_multi_index.hpp b/include/ck/utility/statically_indexed_array_multi_index.hpp
index 21b2941b21..4a8b96ae8a 100644
--- a/include/ck/utility/statically_indexed_array_multi_index.hpp
+++ b/include/ck/utility/statically_indexed_array_multi_index.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef CK_STATICALLY_INDEXED_ARRAY_MULTI_INDEX_HPP
 #define CK_STATICALLY_INDEXED_ARRAY_MULTI_INDEX_HPP
diff --git a/include/ck/utility/synchronization.hpp b/include/ck/utility/synchronization.hpp
index 0e247ed0f8..775e7ac3a3 100644
--- a/include/ck/utility/synchronization.hpp
+++ b/include/ck/utility/synchronization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/thread_group.hpp b/include/ck/utility/thread_group.hpp
index d469dec899..1cd6b2f3ce 100644
--- a/include/ck/utility/thread_group.hpp
+++ b/include/ck/utility/thread_group.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/transpose_vectors.hpp b/include/ck/utility/transpose_vectors.hpp
index 2b0075d600..6faf5c133b 100644
--- a/include/ck/utility/transpose_vectors.hpp
+++ b/include/ck/utility/transpose_vectors.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/tuple.hpp b/include/ck/utility/tuple.hpp
index d8664be550..b616b3123f 100644
--- a/include/ck/utility/tuple.hpp
+++ b/include/ck/utility/tuple.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/tuple_helper.hpp b/include/ck/utility/tuple_helper.hpp
index 6f5b142a5e..e39ae1c23d 100644
--- a/include/ck/utility/tuple_helper.hpp
+++ b/include/ck/utility/tuple_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/utility/type.hpp b/include/ck/utility/type.hpp
index 90b9df2950..9609afba43 100644
--- a/include/ck/utility/type.hpp
+++ b/include/ck/utility/type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp
index 46a1fa559a..a1b1e0d91b 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batched_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp
index 0b621e88a0..a2eabdf5c1 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_backward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp
index dd0db31680..20c1fcd736 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp
index 463c655ac1..7d652fe4c4 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_batchnorm_infer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp
index b0149d88fd..24f754e598 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_cgemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
index 225f7b7e36..449734f434 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
index 7d62158f00..ec5df238ab 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
index b8d47d218b..8f4182a231 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp
index be22003fd9..71c84a1f5c 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp
index f949f27fde..0b90b4b50e 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd_bias_activation_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
index be69f297b2..9b797be925 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp
index 28132aa1eb..ce2a83da61 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
index fedd4dce62..6a48528c54 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
index 2bac5bc5c8..9994a2f9f7 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
index 3fc35a83c7..b4b7a5a032 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp
index c04baca574..944f34007e 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp
index a4fd46c932..9916a03b9c 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp
index b6a9b0fb5e..f949260ca4 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_sparse_embedding3_forward_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp
index df4fca6562..0b7887efb6 100644
--- a/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/gpu/naive_conv_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #ifndef NAIVE_CONV_FWD_HPP
 #define NAIVE_CONV_FWD_HPP
diff --git a/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp b/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp
index 20df1b3616..f57fed9c07 100644
--- a/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/add_device_operation_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index 1886439528..851d9e497f 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
index bb5f971c77..c3c8c0e5a7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
index 0b025b33c8..73f6004257 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
index 59d50e1bd2..70bff27899 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
index 593ef7cb96..33653a3080 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
index e1a4391c4a..28ccf61a3a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
index 8a0b1b1fa7..0a30c210dd 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
index 34c86dd440..2ff64675c9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
index c84ffcff8c..0e1f6f04e8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
index 8e40d60c17..8fd1c7665d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
index 342ade69cd..f6f4df7e2e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index 6a551c7268..2ed8255f6d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index fc9ec8d615..5d9567731d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index 07d5524762..1efe073669 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
index 2c529e06fc..dd8c1987cf 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
index 8af400cb7a..7e6267c87b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
index 3d0c34062c..b151395109 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_gemm_mean_squaremean_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
index c87ae159be..a1c006cf62 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 732d980699..adac7d0dcf 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
index 2d578cca4f..99b2ad1315 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
index 554437f490..fd3550c2f0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
index c07ca3134b..481915d00b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_multiply.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
index 7beae83cdc..de21f325a6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index 8f4cd4d968..f80efd5511 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
index fbc5df98a4..09b1c2190f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index e888446948..5348751511 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index 81b2b4fcf3..fadfd19951 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index ef6920e52a..377ce083c7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index fc4beb0ae9..a82ec543c6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
index e97484a5ad..b482e97ee0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
index 199ed73b4c..778f625d84 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
index 367180dea4..2391775299 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
index 44d89cf361..ccb5cb5a9e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
index 88523c7039..3a006b00a5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
index 2fd7ce22f7..2ed4b0d5f8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
index daec48050c..8a96f6707e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
index b7d81021e2..e17bea5fdf 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
index 2d54879ea6..9236b5c79e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
index f278cfa224..7c1eb4e429 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
index 550a7b0345..9930b1a6fa 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
index 2cdbfbb0c2..c9c1475f18 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp
index 4e3fa81f75..4dd5569cee 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp
index 7ca8bc258a..d52310a3f3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp
index 37398146b8..025500764e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp
index 5eacd358c8..2314e94980 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp
index 94ae02bf3d..5c2bff16cc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp
index e41e8de6a5..a1279eeccb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp
index 99762aa64b..aff6a2542b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp
index 1fc557a95d..be8da2243f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp
index ca3ba4eb0b..652984ae22 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp
index 28a85782d1..be60d1b32d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp
index ba74400793..27e3aa53e3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp
index f5c813de78..f7f4870a97 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp
index e25b6e8493..790b5a92ad 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
index a264d11262..ec3bc852e8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp
index 8b1d8c95ba..8c0c065670 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp
index 49a60d88c3..8631495df1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp
index 04a7c2d238..59849c2d45 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp
index d0feefb50d..33cbea85ef 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp
index 35f35f202c..3861829575 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp
index 63eb7221b5..f7c05b20a7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp
index 1bca3c1f43..2dfecc6b07 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp
index 1791a186f5..a687938965 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp
index 3f56c057ef..cf9c268a64 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp
index a3b8bcf9a0..aeb578f948 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp
index 18e0e084d7..73480262c6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp
index 4a106463a3..74293553e1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp
index 23e1c49fe9..8a91f76b34 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp
index 62e2d24f02..0ff2c30f34 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp
index 18a54d8686..932299008e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp
index 9f408906a7..c902e80a2c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp
index c40052562f..e7a1369277 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp
index 532bfb417e..2d1ab69fc7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp
index 8c08e5ef2f..9f782c11b6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_impl_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
index 0d08377a22..bb45295ed8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp
index 4cdd45e85b..5814b84bb7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp
index a36cafb27a..f1a609a6f4 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp
index 13b0780497..489bef1f56 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp
index 75e1f10242..2507d82308 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp
index 00ec17fada..df4b7b7b49 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp
index 7b762bc932..2748b9b9d0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp
index 2a2b284b22..26230d5456 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp
index 444d8ddc86..d5e1499aef 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp
index f3c0701761..cbc8befb3d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp
index c57edd0846..574cc3dd3e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
index 325ed1e6d4..04bab9d283 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp
index 8960ba7c5b..6c50d222ec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp
index 95d9c07265..5e91f53b0f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp
index dd6734061e..1718031560 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp
index 85f75110df..03bff5912d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp
index 7f62f4e010..676af19336 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp
index eee771b133..c67c0a41f8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp
index 64f1e9c22b..c9058a9b1d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp
index 078561e153..1fbd3c6e7d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp
index 5a9144186b..89496a4ed1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp
index dc4740aa3d..fb79ce8adf 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp
index 9ecc96797f..7915bd2630 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp
index ccce78e2f1..d769fe8154 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp
index 6d3749d868..49938d4434 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp
index 7594dde74d..810249994f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp
index 3272e7f9af..6640ffec70 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp
index 519ec8271d..441bea275c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp
index 77b2fb9306..05912e1e96 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp
index 5abb5c5eec..b0e16cfeaa 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp
index 23bd988b8a..0ee0f7d0fe 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp
index 7ce5577d7f..feb9d99eba 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp
index 7e4c5b77f0..a9f0d77cb8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp
index 5eca5fea7f..bd1f72250a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp
index b0e98411bf..60189fd304 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp
index 84609a995d..b5aa9ce613 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp
index 2f816bb11d..0e20a17ff9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp
index 9cecd4a5b4..c8c1cd7e80 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp
index 42e9b7fc79..3e71b44673 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp
index 494f1c3d71..c8286e56d1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp
index a80abb9247..36fbc52f11 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp
index 53fd286383..0900cffa35 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp
index df5a4db484..726b56066b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp
index ed78acd926..0ae9838607 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp
index 0038fc26da..a3a39b7ccc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/reduce.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
index 36eb092f0f..c5c2d2cdd7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
index 83f52fc3ee..7c6f189cb9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
index 046ff57805..33d5cc6839 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
index 8e6a226f6a..7668248c3f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
index 518fa5f986..20eb7bbc94 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
index 10016cdd70..e8356a9292 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
index cdd5a3cd7b..b3f7d4890b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
index a8be272e02..4190f50a35 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
index ec8296ff22..b7f3344905 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
index b3877c4bb3..53c142f612 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <tuple>
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
index a6d9a359f4..41c67af7ad 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
index 6621a2c867..2d791ff979 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
index 3dfac98ed8..eb9cc1ee2b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
index 6d2a0c9325..68af443a54 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
index 97dd3dcb18..3bf8704b45 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
index 58f8760acc..43e54aaca7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
index df8d31f0da..32c4cd74b5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
index 1bd773227e..f8f5caddb8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
index 16f129d2d0..a034e41a07 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <tuple>
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
index f80f712ff5..3cd3742093 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
index 6f9952e7d5..f7d4dd045a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
index 2cbd13a1ba..c49dd4d858 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
index 7b12522a85..4074ee3b1f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
index 54d477f80c..479fcc92fe 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
index 4ffc44e3a9..0dd644fab7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
index 08cbb81272..50f39396ab 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
index 187d034b95..defa2dbda6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
index 7fc9ed6919..6ff07de236 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <tuple>
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
index 03be6e2bc7..206980cf10 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/algorithm.hpp b/library/include/ck/library/utility/algorithm.hpp
index 86f04dd362..57136f8a2a 100644
--- a/library/include/ck/library/utility/algorithm.hpp
+++ b/library/include/ck/library/utility/algorithm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/check_err.hpp b/library/include/ck/library/utility/check_err.hpp
index a89d03d324..7f63a81a02 100644
--- a/library/include/ck/library/utility/check_err.hpp
+++ b/library/include/ck/library/utility/check_err.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/conv_common.hpp b/library/include/ck/library/utility/conv_common.hpp
index 6fad9f7d77..085454f42d 100644
--- a/library/include/ck/library/utility/conv_common.hpp
+++ b/library/include/ck/library/utility/conv_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp b/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp
index 2b4f63b28b..ff697fb71c 100644
--- a/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp
+++ b/library/include/ck/library/utility/convolution_host_tensor_descriptor_helper.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/convolution_parameter.hpp b/library/include/ck/library/utility/convolution_parameter.hpp
index f4a2b56f75..df6efca108 100644
--- a/library/include/ck/library/utility/convolution_parameter.hpp
+++ b/library/include/ck/library/utility/convolution_parameter.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/device_memory.hpp b/library/include/ck/library/utility/device_memory.hpp
index 87940e1671..1c16ff5916 100644
--- a/library/include/ck/library/utility/device_memory.hpp
+++ b/library/include/ck/library/utility/device_memory.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/fill.hpp b/library/include/ck/library/utility/fill.hpp
index c0bc372764..c01e139ea0 100644
--- a/library/include/ck/library/utility/fill.hpp
+++ b/library/include/ck/library/utility/fill.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/host_common_util.hpp b/library/include/ck/library/utility/host_common_util.hpp
index 6f4466e8da..20a8f234db 100644
--- a/library/include/ck/library/utility/host_common_util.hpp
+++ b/library/include/ck/library/utility/host_common_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/host_gemm.hpp b/library/include/ck/library/utility/host_gemm.hpp
index 44036d0234..5eb7e3b8c9 100644
--- a/library/include/ck/library/utility/host_gemm.hpp
+++ b/library/include/ck/library/utility/host_gemm.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/host_tensor.hpp b/library/include/ck/library/utility/host_tensor.hpp
index 844c29ed14..91293d29f7 100644
--- a/library/include/ck/library/utility/host_tensor.hpp
+++ b/library/include/ck/library/utility/host_tensor.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/host_tensor_generator.hpp b/library/include/ck/library/utility/host_tensor_generator.hpp
index 4259862e65..31ff13aec2 100644
--- a/library/include/ck/library/utility/host_tensor_generator.hpp
+++ b/library/include/ck/library/utility/host_tensor_generator.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/iterator.hpp b/library/include/ck/library/utility/iterator.hpp
index 9fdc88ea76..b44e2d8e3c 100644
--- a/library/include/ck/library/utility/iterator.hpp
+++ b/library/include/ck/library/utility/iterator.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/literals.hpp b/library/include/ck/library/utility/literals.hpp
index a73a2ea054..a8bd6303f1 100644
--- a/library/include/ck/library/utility/literals.hpp
+++ b/library/include/ck/library/utility/literals.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/numeric.hpp b/library/include/ck/library/utility/numeric.hpp
index 70a7e87ab1..9ee118d475 100644
--- a/library/include/ck/library/utility/numeric.hpp
+++ b/library/include/ck/library/utility/numeric.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/utility/ranges.hpp b/library/include/ck/library/utility/ranges.hpp
index 55c322f1ac..f11e4204a0 100644
--- a/library/include/ck/library/utility/ranges.hpp
+++ b/library/include/ck/library/utility/ranges.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
index cc8787458d..e730e9f589 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
index 04200cfb52..f6696ffa9f 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
index 7b86f3cc72..32d6f258b5 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
index 2afb1afbc3..ee246ba56e 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
index 68d768949c..5a9483b309 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
index 737e5bfca3..0fa0719237 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
index e09d017360..2f42f62b0b 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
index 984d66e288..10b4cea7d7 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
index 12cada9c44..c687eb20f0 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
index 13f198862e..b19374ca6d 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
index 2ca1adc2f6..bbd318ba92 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
index fe5de52796..187ccb5ff6 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
index 5b55c8e15e..ec2b2646ff 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
index 9517e4577e..d76cd350c8 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
index 43b9124406..ef65106c2c 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
index 326500fcbf..078b241f9d 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index e1bfa88f49..4db05589b1 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
index f59b742534..e25f903a82 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
index 04a748f455..a0afaabbc7 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 9b96194c87..67dfc4cd35 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
index 0713dfcd99..9001c901cf 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
index 521c3d9219..cb89d3cefd 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
index 231d612d78..91eefba0c1 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
index 165bc3957d..c20798f557 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
index 832fc3b066..3d9ad64b9b 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 99e8712474..cf23d01bf2 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
index f73e3dea84..a541a2d224 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index fd09471958..5e481603ca 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
index 53ad7ba5ff..501ea85f60 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 291a127a66..91ab541bf9 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp
index b62c8b99cb..bc95d2f1b1 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp
index d05b8b592c..fbc8d0bc60 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp
index e3ef95d12e..bed38658a9 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp
index 41be396c24..fc5ec77e42 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_backward_f64_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp
index cd1e05b113..4e38ee13b2 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp
index 073dd583f9..f087eb7982 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp
index be63bd44c6..d0f361401a 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp
index fe87091e8d..710d07b828 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_forward_f64_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp
index 2e695afa97..8801c309f9 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/utility/tuple.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp
index 9ec761e445..b674cfc423 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/utility/tuple.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp
index f0d26c36be..05e3650887 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/utility/tuple.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp
index 9e4066bb06..15a02af021 100644
--- a/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batchnorm/device_batchnorm_infer_f64_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/utility/tuple.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
index ebbff88346..5587db77e0 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
index 980383f3e7..26262855ea 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
index 2d4b6e3489..befc0dcd10 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
index 7caa469f54..e45b47cf94 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
index 093b2f0e98..f437a227d5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
index 0f683e5c28..13fdbeb35c 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
index e384993aed..95ef8c4929 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
index 92e39c173f..290f81d7c9 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
index 5118d0d033..16fd1cb407 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
index 655d4f0061..ff37bf7cce 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
index a9d20be18b..8a1f6f9334 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
index a68f5c9718..d333f59726 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
index 0aa927155a..4c87b51a9e 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
index b84ea274c5..fd3f57c6b6 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
index 578469997a..1e53f0b2f5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
index 8e5a19313e..d02d146a9b 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 // This (ifndef) is a hack to use customized behavior for buffer load rather than using default
 // setting Don't use this hack unless absolutely necessary!
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
index 5a5c838422..e3e90c966d 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
index e0f3d6199f..81e9122d95 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
index 30537d9373..dbc82168f4 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
index 190c39b870..3ac250f3e6 100644
--- a/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv1d_bwd_data/device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
index e14cd55862..fcb858728e 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
index f001b83c17..2baa6ae061 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
index 83ba6a1c6b..28867fe19d 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index 1da9a81d90..40656e382b 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
index 7c33df5e76..422e37e926 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
index a5f8629f2d..5993f6bd7a 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 8076d6d356..bb9b696864 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
index 33503b9f8a..da96c79a6e 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index c5e4bd199e..78e9c893c9 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
index f43d13e309..4663a9ac32 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
index 0ce6b04c42..0b1df52f8c 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 76ab3189d7..9969a8bbc9 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp
index f8c2550888..e34ea06ff4 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu/device_conv2d_fwd_xdl_c_shuffle_bias_relu_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp
index fe7152471e..3254fcfc26 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd_bias_relu_add/device_conv2d_fwd_xdl_c_shuffle_bias_relu_add_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
index 04ce7c0763..94b2a47e50 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
index 0251d9157f..4244ab7b87 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
index c2975727e4..5c7db4ca3b 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
index fc86d73024..ebc56487a1 100644
--- a/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv3d_bwd_data/device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp b/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
index 182037f15c..a62c9e2354 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp
index b160d4fe1a..3e2386ee03 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/elementwise_normalization/device_elementwise_normalization_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_elementwise_normalization_impl.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
index 35df85b7af..ea99a5a30e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
index 7d0863c953..b83acfa8cb 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
index c1b11b19df..d5800e0333 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
index 9a889e8d55..abe52ce1da 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
index d8e7798438..e696bfdcdc 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
index 0034ac59c3..d3ad7c60ec 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
index 0b540b8b34..a56a36b0ab 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
index 4f6ff5111b..63d55e81d6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
index 9f5cebcab0..3d9a265c29 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
index 41afb519f1..240384d19c 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
index d1173095fc..e96905247d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
index efda345a8d..124b818b28 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
index 03eebf4ec3..2e884dfc8a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
index 5d8de04cd9..2ca29b1e6f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
index 7b12b7cf1f..706076098d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
index 730ffd4633..5ac458a7b9 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
index 619473ff0c..a644125447 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
index 8e06f9d26b..44b6848233 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
index f9458b7483..23176269c2 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
index 77a03b746c..31a9abe53a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
index ef8d7d4e40..201fd93110 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
index cb65cc7b68..5d489b207e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
index 5b1014ed82..e09480d57b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
index e6f6add8bf..34065c334d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
index 80b3d03da2..95d7777a79 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
index 93b3df1e57..bf24bc76b1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
index f10365d892..023f987127 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
index a7a9eb62ce..ffb199e58f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
index 9fb45b0036..90e979d891 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
index 18a78674e7..8a81b77891 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
index cef6070af8..e1983add04 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index 1be70d6ca4..47a180e127 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
index 6b8455ffa9..b8e994e91a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
index b9e28e3d7e..a590413acc 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
index 2b1a5a57bb..1d010d1b07 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
index 301d3b55b5..f108b75342 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
index cd16f35ff2..b0b4bc012d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
index 3916669847..df3bd94fca 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
index 0a623034ef..73b4e77666 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
index 5ef8d08de9..76137a1c3e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
index c9557bae89..f0158d8f3d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
index 463e0865c0..7b65f8737a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
index b71ff1b997..a9fef5c60d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
index 9060c9b1b0..c3b1dfccab 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
index 81cf01d6a9..8338b34a4d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
index 4da85cc46e..357dc91aa1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
index ab83e4baab..65b94d0879 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
index a4cd3fadbe..00cd07bab8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
index 207e76ffe5..9955a206b6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
index 3f30937ff5..d9ae781791 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
index d91e6c63be..27c6cbe8ab 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
index 1433215423..ff5f0e94a1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
index 09acc7c0f7..dba625e0ad 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_multiply/device_gemm_add_multiply_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
index 47b8d23424..28a452c1a1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
index efa030ec49..13366238d6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
index f2735020e6..8a4889ee83 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
index 7d4aae928b..fc3cbcf905 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
index e8747af482..bfb95bce85 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
index ed54c3a9bf..d0352339cf 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
index da7eae637b..d5b298ab21 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
index 34345095e0..80c8f018f4 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bias_add_reduce/device_gemm_bias_add_mean_squaremean_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
index 55461dfba7..74ec9e1f8d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
index 405e69975c..eb98b3e7e8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
index 9af31b3a12..5f4a90125a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
index 841b7a1d47..38e3897d6a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
index 9f7f643beb..c5aa59f911 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
index c8e9f35d24..e71b269b8d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
index 5f804d45a5..fdf63f8112 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
index 60cb138f56..4c98a9d5e9 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
index 59e2b2da86..f0e7b6ab43 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
index bb09bf8b8e..56815b9ac5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
index 0a3b566de6..e66d46a269 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
index 2b17e47b1c..fb1dfac69c 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_reduce/device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
index e178d3b0ad..fed2cbbfb9 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
index 52be9fe709..44ac4c08cd 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index c4680db831..30a2bf36d1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index 7fc35c4198..4b2a2dbdc7 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
index f27b2199e0..9d15ccd362 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
index b9a1095570..4e9ad58742 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
index 44e5f597d0..330e5aff90 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
index f3a9063f7f..0db3a15d2a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index 05ba449246..ccbfaeaf4f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index 7a610a747c..e10de67f94 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index 90e074f052..8b47e82f6d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index 74aebf1031..5aa50adb31 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index 361ea8f4ee..333b40c71c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index 3145b71640..506a93ae9b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
index cde93f902c..30084f16d1 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 3d604d42cc..11babea28e 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index ede21f1f4f..c2c0fc553c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 99e556618c..5be7443eca 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 15871a28c3..2828b432ff 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
index b4de825fb6..85a7a5be25 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index f7e575df2b..5eb881549c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_dl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 85300b4e44..4157853c41 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_dl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
index bcda22006b..3d3f9b1793 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "device_grouped_conv2d_fwd_common.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index 40593a0efb..bc7d577b69 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_xdl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 7088028bf5..55bf6da9b9 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_xdl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 919274c503..202cdd6b44 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_xdl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
index 2858671ee9..07bea1c03c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 #include "device_grouped_conv2d_fwd_common.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
index 25caf61df1..82edf896db 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_xdl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
index b997cfb672..4bd3236caf 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_xdl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
index 3256a2a826..4f5bdb2023 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "device_grouped_conv2d_fwd_xdl_instance.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index e48db4a531..7ae87eed5d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index 1655850ec1..ab07341b59 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index aba46b7ebe..15045bedd8 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index b4ae8b6ce5..c3b100ea69 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index 061674bd82..ca488e9dc4 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index ed7e547676..6087b1b187 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
index bf5fa30601..fd8c47deb3 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp
index 8c38493735..5e5dbc53cd 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp
index 487cd22721..a88fe4af63 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp
index d497cd57ed..f6e7e5b28e 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp
index 2e53fbbda5..3d303a3fa7 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
index b550bb2871..aa161e51c9 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
index a3f9c7a9e7..c454deac1a 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index 5f5d6c9b5a..c829e8863d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
index 2ace1b2432..fb30e7a973 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index 764ec06198..8642562fa3 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index a3d73440eb..83b31b07cf 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index cffb0fce17..aa6365cd98 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
index dddfa2aa44..f4460b360b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
index c2f5f00c7a..f4086b6eac 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
index 476d4ce1f8..d68eb76144 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index 1023fa4810..2dfb8caace 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
index 6b065c0f82..598a0b0e29 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
index e9c2112e16..be860f58e0 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
index 79dde38fc9..9a64e555d6 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
index 9f6bf128fa..fe72a27331 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
index 6241e03385..cac8641e13 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
index b64328d5d0..0a9ac84623 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
index d6a2f6f2c1..ad92818ec2 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
index 73097828e3..70e3bbc1c1 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
index 507a683ee7..7c5d2c4a9c 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
index ca1aa0c25c..f5626d4a9a 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "normalization_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
index 9dea41e89d..d9029ac25e 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
index 38338ff997..508ad3873b 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
index 0f4a35deef..ada96a93a2 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
index 6fcb519a9b..62bcad992a 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
index 67ffd47085..47896be911 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
index a41cd0094b..35c8522d9f 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
index fa70569bab..75b7629f24 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
index f3367b9460..dbfc4acfdf 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
index 8477a884da..63b3e8df8e 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "pool_fwd_instance_common.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
index cd508b55b0..8aa707885b 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
index 672cdba65d..711314985a 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/conv2d_quantization_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
index d4b5484d8b..39c4f82fef 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_dl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
index 7db4b8d869..92e73eb2ee 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_dl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
index 7eefbe0381..bb7a570cda 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
index c8f5f7042c..1d8b58fd18 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_dl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
index d7f7384ff7..62826d0d2b 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_dl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
index 658aa8370a..99877c32db 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_xdl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
index 7102e9b259..50ccc69f45 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_xdl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
index 90f8791aa9..caced6c950 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_int8_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
index 9d69377085..526fe73463 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_xdl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
index d6f87335b2..d1d5a66bb9 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_conv2d_xdl_int8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp
index 9cad8d4c8e..0ec498a441 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_quantization_common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
index ffe1efb80b..4c02f8179c 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
index 7f24e5677c..2c635e96d2 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
index 06e66cfe03..c5a52f2cf2 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
index 16635d1e93..fe4a0b4d68 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_dl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp
index dfb8dc29b4..d1b18ac492 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_quantization_common.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
index c153cdf9ea..e3163c32ed 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
index f6cd32026f..f6a0163aba 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
index 45fbacc334..0cfffbed70 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
index 257633fe19..289898257b 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp b/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp
index 213f42b91b..e7c2500fef 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/gemm/gemm_quantization_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp
index cf46059a0d..4771b4aa5d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp
index 0043b19844..f684dabc1f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp
index 6f702ddf1f..86e2ca7c47 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp
index d1f70dc99e..97d09ff834 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp
index a957981a55..87187f493a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp
index 550a9cd76c..d3f5cd8656 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp
index 58cb6ee348..fb4a50d5e3 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp
index 1ac5e79bc1..f74d22f7ac 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp
index b1e1a06800..0982455add 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp
index 1a15b32d23..c6f1c3a633 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp
index 119f384b4e..db9f52a13f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp
index 3f1bd86b8b..1410f51bea 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp
index b507f0d1f9..707a655703 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp
index 04d0ea2e8b..17f45c3327 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp
index 3de561f2b6..d3753d01c4 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp
index 3f45b03134..83fad7826a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp
index 76851d9b72..c627f68636 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp
index 9cef019320..09686d98ff 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp
index ce73ec47e3..44519032f3 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp
index ed6091f924..9e14d80e21 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp
index 4c8375de16..ab217f1160 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp
index 0fa93ab688..3f611da7e0 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp
index 821eec1751..8695db4004 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp
index 0305b4945f..8d15fda786 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp
index 1bda0bcc71..c161858b1f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp
index 7f8018a04e..e6b87fdead 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp
index 887a89cc2b..c2e4acc8d9 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp
index 0cc810363d..e5a5e07e17 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp
index 4c825a9f1b..22c6203345 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp
index bf26913fd3..1890fb78cd 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp
index 629299c7b1..dc5caf297b 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp
index 9a08634498..02bb0c3b1d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp
index 6dc925bd6f..0b0a403038 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp
index 470d68d372..fcf95aab7a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp
index 39303ab580..5151d32a58 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp
index a5481784ed..5463cc2b4c 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp
index aa6e6d3cc9..a15e99fe0f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp
index b1ea551eaa..ee7a4dd4ce 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp
index 2ba83132d8..58f7295a83 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp
index b9018e7c6a..43f2370a2c 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp
index 1abc6de552..eaf84a6de1 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp
index f3a017aeb4..76d63c9a96 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp
index 329617bb43..eb1d389c39 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp
index 1e4d43debd..61225df05d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp
index f9f79675f5..aef231e17a 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp
index d3e7268c5a..6d2939e3e0 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp
index a41a12386d..ff0ddd773b 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp
index 6da1acc4e5..1367022065 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp
index f14b8a4031..f0f9301918 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp
index 5a9f08167c..db9f156600 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp
index d3aff06753..e5644faf29 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp
index 55f7537d83..350259f5f0 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp
index 70f91168d8..89cae55273 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp
index 47f5e67fe0..b0f82ab453 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp
index eae489ff04..fedd0f890b 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp
index 9fb267a201..64ec1ce157 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp
index fecb2691f8..156f182f55 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp
index 232d2b858b..22b278790d 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp
index 07d45c4ca9..f3070d2975 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp
index 596a062f3a..e17476ba75 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp
index 7270cefe8b..2387725b9f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp
index d0f4ef3dff..8be94118c4 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp
index 9c6bce92f4..9ff01615c9 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp
index 5faf8d8283..8a99c7c958 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp
index 8f3c72451a..78909c8dbf 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp
index 8d7794f42e..7f094c28b7 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp
index 4a32543a13..4dfb477c1e 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp
index 26d571c84e..e492833606 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp
index ae56a2a919..de8dce357b 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp
index aae3233c9f..0225e75cb0 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp
index 94d5d3fa2f..6d04d2128f 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp
index dad190a634..046643abe1 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp
index b7ca6998f5..a61302c001 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp
index 22c40187ea..7619e12a8e 100644
--- a/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp
+++ b/library/src/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
index 14d2764529..a86da7cc79 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
index fa334b997c..938fb033a1 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
index 1c9d37d848..3d56593811 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
index 5fbdab5055..d701b4174e 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
index 7dd8640b18..2085aafc56 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
index b32fe6838f..ebe4329f9a 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
index c05048ec56..b8fd5a1e5f 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
index 6a235708bd..112f1940d3 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
index e5bec5e263..ab8a69eec2 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
index 57d3f184a6..5382fec90d 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
index fae3a4dd66..a1a143afa1 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
index b6fb70e8e2..992e0c1ec1 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
index 33c7b6f35f..2be1f45bb1 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
index c22aa574b1..a1da73aa8b 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
index 55f3d2bd20..b5c3b576a6 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
index fb0bcf5ee8..22a0404c05 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
index 608cfcf838..81a2ff80ca 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
index 15552dbae5..3e2cf8d062 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
index 6767402886..c8b038d502 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
index 4b33da93c2..08995d99ec 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
index fe3b823e88..652601ee7c 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
index 8ecdf87d9f..86caac1b62 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
index 3563135204..c46ae1a4ef 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
index aa21a0bf8a..394814ff53 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 
diff --git a/library/src/utility/convolution_parameter.cpp b/library/src/utility/convolution_parameter.cpp
index c8712d2093..57cedd6019 100644
--- a/library/src/utility/convolution_parameter.cpp
+++ b/library/src/utility/convolution_parameter.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/host_utility/io.hpp"
 
diff --git a/library/src/utility/device_memory.cpp b/library/src/utility/device_memory.cpp
index 90f943313b..11166783e8 100644
--- a/library/src/utility/device_memory.cpp
+++ b/library/src/utility/device_memory.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/host_utility/hip_check_error.hpp"
 
diff --git a/library/src/utility/host_tensor.cpp b/library/src/utility/host_tensor.cpp
index e34fbc8f34..7211552641 100644
--- a/library/src/utility/host_tensor.cpp
+++ b/library/src/utility/host_tensor.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cassert>
 
diff --git a/profiler/include/profiler/data_type_enum.hpp b/profiler/include/profiler/data_type_enum.hpp
index afcd6fea22..c046c7fabb 100644
--- a/profiler/include/profiler/data_type_enum.hpp
+++ b/profiler/include/profiler/data_type_enum.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp b/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp
index b16254279c..22dab31100 100644
--- a/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_add_relu_gemm_add_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp b/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp
index 799dccc0ff..5bee67c1ce 100644
--- a/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_bias_softmax_gemm_permute_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp
index 1583c6db21..f3d2c55617 100644
--- a/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_impl.hpp
index c07d7c0555..cdc94aa9a6 100644
--- a/profiler/include/profiler/profile_batched_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp b/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp
index 45b7b77388..901fa338d4 100644
--- a/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp
index f5ec235141..15a21206c5 100644
--- a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp
index 91c28f25fc..f2fcb0b133 100644
--- a/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_softmax_gemm_permute_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batchnorm_backward_impl.hpp b/profiler/include/profiler/profile_batchnorm_backward_impl.hpp
index 79d8862081..3343b5e66e 100644
--- a/profiler/include/profiler/profile_batchnorm_backward_impl.hpp
+++ b/profiler/include/profiler/profile_batchnorm_backward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batchnorm_forward_impl.hpp b/profiler/include/profiler/profile_batchnorm_forward_impl.hpp
index 82fe75bf01..2f9538b16c 100644
--- a/profiler/include/profiler/profile_batchnorm_forward_impl.hpp
+++ b/profiler/include/profiler/profile_batchnorm_forward_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_batchnorm_infer_impl.hpp b/profiler/include/profiler/profile_batchnorm_infer_impl.hpp
index ca65339345..1b31a2aabf 100644
--- a/profiler/include/profiler/profile_batchnorm_infer_impl.hpp
+++ b/profiler/include/profiler/profile_batchnorm_infer_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_bwd_data_impl.hpp b/profiler/include/profiler/profile_conv_bwd_data_impl.hpp
index 86d394daf9..52152a90fe 100644
--- a/profiler/include/profiler/profile_conv_bwd_data_impl.hpp
+++ b/profiler/include/profiler/profile_conv_bwd_data_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp b/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp
index 1aebef8bb2..436fbdbd75 100644
--- a/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp
+++ b/profiler/include/profiler/profile_conv_fwd_bias_relu_add_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp b/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp
index 2bac144334..808c1a1c90 100644
--- a/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp
+++ b/profiler/include/profiler/profile_conv_fwd_bias_relu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_conv_fwd_impl.hpp b/profiler/include/profiler/profile_conv_fwd_impl.hpp
index 1f3ba8f007..bc2eb25797 100644
--- a/profiler/include/profiler/profile_conv_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_conv_fwd_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp b/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
index 7707e16b08..1fd9c81109 100644
--- a/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp b/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp
index 3cc2ea3b92..81b8d8ddbf 100644
--- a/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_add_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp b/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
index d53a6589e0..6f6d881c1e 100644
--- a/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp b/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp
index 40093e774f..25871dfb2e 100644
--- a/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_multiply_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp b/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
index e1c90f0f52..4c3d0a0450 100644
--- a/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp b/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp
index b4ec78cdf3..c0ffea8a32 100644
--- a/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_bias_add_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_bilinear_impl.hpp b/profiler/include/profiler/profile_gemm_bilinear_impl.hpp
index 31bae281c4..b540e938b5 100644
--- a/profiler/include/profiler/profile_gemm_bilinear_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_bilinear_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp b/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp
index f9a544c044..3893f8cdc7 100644
--- a/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_impl.hpp b/profiler/include/profiler/profile_gemm_impl.hpp
index 9b164104b5..eaab5dbcc2 100644
--- a/profiler/include/profiler/profile_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_reduce_impl.hpp b/profiler/include/profiler/profile_gemm_reduce_impl.hpp
index 370121a3cc..ff801e8afd 100644
--- a/profiler/include/profiler/profile_gemm_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index ab1bce258a..6ffa31678f 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
index 4f9aa98376..dc6739773f 100644
--- a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp b/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
index b201a2ed33..9fadfe9699 100644
--- a/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp
index 87e6ae44c7..f05b13b749 100644
--- a/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_fastgelu_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_grouped_gemm_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
index 9abb5e7a53..09a651d77c 100644
--- a/profiler/include/profiler/profile_grouped_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_groupnorm_impl.hpp b/profiler/include/profiler/profile_groupnorm_impl.hpp
index 73343f6bec..ebefe3dad4 100644
--- a/profiler/include/profiler/profile_groupnorm_impl.hpp
+++ b/profiler/include/profiler/profile_groupnorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_layernorm_impl.hpp b/profiler/include/profiler/profile_layernorm_impl.hpp
index 7dd90d0797..2d87c8c8fe 100644
--- a/profiler/include/profiler/profile_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_layernorm_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_pool2d_fwd_impl.hpp b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
index c313a00beb..0c888db1f4 100644
--- a/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_pool3d_fwd_impl.hpp b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
index c9e4c193f8..41b57fd853 100644
--- a/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_reduce_impl.hpp b/profiler/include/profiler/profile_reduce_impl.hpp
index e618200299..b54aa65aef 100644
--- a/profiler/include/profiler/profile_reduce_impl.hpp
+++ b/profiler/include/profiler/profile_reduce_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/include/profiler/profile_softmax_impl.hpp b/profiler/include/profiler/profile_softmax_impl.hpp
index 96816f53bb..65b4be2a60 100644
--- a/profiler/include/profiler/profile_softmax_impl.hpp
+++ b/profiler/include/profiler/profile_softmax_impl.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/profiler/src/profile_avg_pool2d_fwd.cpp b/profiler/src/profile_avg_pool2d_fwd.cpp
index b92288096d..c67897c044 100644
--- a/profiler/src/profile_avg_pool2d_fwd.cpp
+++ b/profiler/src/profile_avg_pool2d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_batched_gemm.cpp b/profiler/src/profile_batched_gemm.cpp
index 907a373794..dc83e25b48 100644
--- a/profiler/src/profile_batched_gemm.cpp
+++ b/profiler/src/profile_batched_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdint>
 #include <iostream>
diff --git a/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp b/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp
index f440a3094e..3d29c4b84a 100644
--- a/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp
+++ b/profiler/src/profile_batched_gemm_add_relu_gemm_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_batched_gemm_gemm.cpp b/profiler/src/profile_batched_gemm_gemm.cpp
index 6015c93be3..9a99874d1c 100644
--- a/profiler/src/profile_batched_gemm_gemm.cpp
+++ b/profiler/src/profile_batched_gemm_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_batched_gemm_reduce.cpp b/profiler/src/profile_batched_gemm_reduce.cpp
index 6b1dfc0142..9620d63caf 100644
--- a/profiler/src/profile_batched_gemm_reduce.cpp
+++ b/profiler/src/profile_batched_gemm_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_batchnorm_bwd.cpp b/profiler/src/profile_batchnorm_bwd.cpp
index 44ce7350ff..1738d53dbe 100644
--- a/profiler/src/profile_batchnorm_bwd.cpp
+++ b/profiler/src/profile_batchnorm_bwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_batchnorm_fwd.cpp b/profiler/src/profile_batchnorm_fwd.cpp
index 902a1fc423..2b3e4eea41 100644
--- a/profiler/src/profile_batchnorm_fwd.cpp
+++ b/profiler/src/profile_batchnorm_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_batchnorm_infer.cpp b/profiler/src/profile_batchnorm_infer.cpp
index 92c16859c1..f1c19bc36e 100644
--- a/profiler/src/profile_batchnorm_infer.cpp
+++ b/profiler/src/profile_batchnorm_infer.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_conv_bwd_data.cpp b/profiler/src/profile_conv_bwd_data.cpp
index 9241ead738..465abacc4a 100644
--- a/profiler/src/profile_conv_bwd_data.cpp
+++ b/profiler/src/profile_conv_bwd_data.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_conv_fwd.cpp b/profiler/src/profile_conv_fwd.cpp
index b57ee7fd94..701999d8a9 100644
--- a/profiler/src/profile_conv_fwd.cpp
+++ b/profiler/src/profile_conv_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_conv_fwd_bias_relu.cpp b/profiler/src/profile_conv_fwd_bias_relu.cpp
index b44007cde4..31055ec1d1 100644
--- a/profiler/src/profile_conv_fwd_bias_relu.cpp
+++ b/profiler/src/profile_conv_fwd_bias_relu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_conv_fwd_bias_relu_add.cpp b/profiler/src/profile_conv_fwd_bias_relu_add.cpp
index 408dd02f78..8c2439a0c7 100644
--- a/profiler/src/profile_conv_fwd_bias_relu_add.cpp
+++ b/profiler/src/profile_conv_fwd_bias_relu_add.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm.cpp b/profiler/src/profile_gemm.cpp
index 61bae6ae70..b3587ea98d 100644
--- a/profiler/src/profile_gemm.cpp
+++ b/profiler/src/profile_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_add_fastgelu.cpp b/profiler/src/profile_gemm_add_add_fastgelu.cpp
index c3c0fb7b67..8af3768a48 100644
--- a/profiler/src/profile_gemm_add_add_fastgelu.cpp
+++ b/profiler/src/profile_gemm_add_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_fastgelu.cpp b/profiler/src/profile_gemm_add_fastgelu.cpp
index 380b25a614..a09bb8340d 100644
--- a/profiler/src/profile_gemm_add_fastgelu.cpp
+++ b/profiler/src/profile_gemm_add_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_multiply.cpp b/profiler/src/profile_gemm_add_multiply.cpp
index 7d6fead402..560467c264 100644
--- a/profiler/src/profile_gemm_add_multiply.cpp
+++ b/profiler/src/profile_gemm_add_multiply.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_add_relu_add_layernorm.cpp b/profiler/src/profile_gemm_add_relu_add_layernorm.cpp
index 5cbc3d21f8..558d255ce1 100644
--- a/profiler/src/profile_gemm_add_relu_add_layernorm.cpp
+++ b/profiler/src/profile_gemm_add_relu_add_layernorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_bias_add_reduce.cpp b/profiler/src/profile_gemm_bias_add_reduce.cpp
index 6d86db0822..76daffbc67 100644
--- a/profiler/src/profile_gemm_bias_add_reduce.cpp
+++ b/profiler/src/profile_gemm_bias_add_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_bilinear.cpp b/profiler/src/profile_gemm_bilinear.cpp
index 3480014ba6..a1a48616b4 100644
--- a/profiler/src/profile_gemm_bilinear.cpp
+++ b/profiler/src/profile_gemm_bilinear.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_fastgelu.cpp b/profiler/src/profile_gemm_fastgelu.cpp
index 2a137224cb..93573002ef 100644
--- a/profiler/src/profile_gemm_fastgelu.cpp
+++ b/profiler/src/profile_gemm_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_reduce.cpp b/profiler/src/profile_gemm_reduce.cpp
index 395bf0627e..48f6f5eb49 100644
--- a/profiler/src/profile_gemm_reduce.cpp
+++ b/profiler/src/profile_gemm_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_gemm_splitk.cpp b/profiler/src/profile_gemm_splitk.cpp
index f636ce718c..cc2da73cb7 100644
--- a/profiler/src/profile_gemm_splitk.cpp
+++ b/profiler/src/profile_gemm_splitk.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index dfd8a099f5..7a062ed519 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <initializer_list>
diff --git a/profiler/src/profile_grouped_conv_fwd.cpp b/profiler/src/profile_grouped_conv_fwd.cpp
index 9ff3c15af0..d0b424cde6 100644
--- a/profiler/src/profile_grouped_conv_fwd.cpp
+++ b/profiler/src/profile_grouped_conv_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_grouped_gemm.cpp b/profiler/src/profile_grouped_gemm.cpp
index 34647adabf..d023db54de 100644
--- a/profiler/src/profile_grouped_gemm.cpp
+++ b/profiler/src/profile_grouped_gemm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_grouped_gemm_fastgelu.cpp b/profiler/src/profile_grouped_gemm_fastgelu.cpp
index 9b6142f015..50ecf25cae 100644
--- a/profiler/src/profile_grouped_gemm_fastgelu.cpp
+++ b/profiler/src/profile_grouped_gemm_fastgelu.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/profiler/src/profile_groupnorm.cpp b/profiler/src/profile_groupnorm.cpp
index d55529a0f0..d55784ff0a 100644
--- a/profiler/src/profile_groupnorm.cpp
+++ b/profiler/src/profile_groupnorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_layernorm.cpp b/profiler/src/profile_layernorm.cpp
index e93fc2dbd2..7bf210e678 100644
--- a/profiler/src/profile_layernorm.cpp
+++ b/profiler/src/profile_layernorm.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_max_pool3d_fwd.cpp b/profiler/src/profile_max_pool3d_fwd.cpp
index 90c6e4e2b9..cf6db2cfc9 100644
--- a/profiler/src/profile_max_pool3d_fwd.cpp
+++ b/profiler/src/profile_max_pool3d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profile_reduce.cpp b/profiler/src/profile_reduce.cpp
index 6925371858..e4af5680a5 100644
--- a/profiler/src/profile_reduce.cpp
+++ b/profiler/src/profile_reduce.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <fstream>
diff --git a/profiler/src/profile_softmax.cpp b/profiler/src/profile_softmax.cpp
index 78b64dda7d..77007ad13b 100644
--- a/profiler/src/profile_softmax.cpp
+++ b/profiler/src/profile_softmax.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/profiler/src/profiler.cpp b/profiler/src/profiler.cpp
index 080117e390..0f528c008f 100644
--- a/profiler/src/profiler.cpp
+++ b/profiler/src/profiler.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/profiler/src/profiler_operation_registry.hpp b/profiler/src/profiler_operation_registry.hpp
index 91ff291233..276b7b38dc 100644
--- a/profiler/src/profiler_operation_registry.hpp
+++ b/profiler/src/profiler_operation_registry.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <functional>
 #include <iostream>
diff --git a/test/batched_gemm/batched_gemm_bf16.cpp b/test/batched_gemm/batched_gemm_bf16.cpp
index 78be540627..fa1652e997 100644
--- a/test/batched_gemm/batched_gemm_bf16.cpp
+++ b/test/batched_gemm/batched_gemm_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm/batched_gemm_fp16.cpp b/test/batched_gemm/batched_gemm_fp16.cpp
index 6cbbedf677..3df4912aa0 100644
--- a/test/batched_gemm/batched_gemm_fp16.cpp
+++ b/test/batched_gemm/batched_gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm/batched_gemm_fp32.cpp b/test/batched_gemm/batched_gemm_fp32.cpp
index c9e565e264..2d808441b9 100644
--- a/test/batched_gemm/batched_gemm_fp32.cpp
+++ b/test/batched_gemm/batched_gemm_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm/batched_gemm_int8.cpp b/test/batched_gemm/batched_gemm_int8.cpp
index 4da941a576..ed233a5ada 100644
--- a/test/batched_gemm/batched_gemm_int8.cpp
+++ b/test/batched_gemm/batched_gemm_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp b/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp
index aa113de219..1a8d5c2e55 100644
--- a/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp
+++ b/test/batched_gemm_gemm/test_batched_gemm_gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_gemm_util.hpp"
diff --git a/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp b/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp
index 53c4d37c44..b0fffc466e 100644
--- a/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp
+++ b/test/batched_gemm_gemm/test_batched_gemm_gemm_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp b/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp
index b150ce50d1..dd2638ce89 100644
--- a/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp
+++ b/test/batched_gemm_reduce/batched_gemm_reduce_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp
index 5df7769d5f..cb46a995c6 100644
--- a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp
+++ b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
index 98debe19c3..d8ee744c60 100644
--- a/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
+++ b/test/batched_gemm_softmax_gemm/test_batched_gemm_softmax_gemm_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp
index fe65a6fb96..ef88ce6d81 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_bias_softmax_gemm_permute_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp
index 7235cd1b0b..b38b10d195 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_permute_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp
index af5f0efec3..d7c39367c8 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_bias_softmax_gemm_permute_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
index defe361240..8e0baede11 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_permute_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
index 293acd6015..81d404109f 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "test_batched_gemm_softmax_gemm_permute_util.hpp"
diff --git a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp
index 912bbc91ed..9df03ffd2a 100644
--- a/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp
+++ b/test/batched_gemm_softmax_gemm_permute/test_batched_gemm_softmax_gemm_permute_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/batchnorm/batchnorm_bwd_rank_4.cpp b/test/batchnorm/batchnorm_bwd_rank_4.cpp
index caa7331ea2..a4696cf2a3 100644
--- a/test/batchnorm/batchnorm_bwd_rank_4.cpp
+++ b/test/batchnorm/batchnorm_bwd_rank_4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/batchnorm/batchnorm_fwd_rank_4.cpp b/test/batchnorm/batchnorm_fwd_rank_4.cpp
index 13aef7d6bf..9b6fbd0f66 100644
--- a/test/batchnorm/batchnorm_fwd_rank_4.cpp
+++ b/test/batchnorm/batchnorm_fwd_rank_4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/batchnorm/batchnorm_infer_rank_4.cpp b/test/batchnorm/batchnorm_infer_rank_4.cpp
index 77fc1daae6..ecb4043b36 100644
--- a/test/batchnorm/batchnorm_infer_rank_4.cpp
+++ b/test/batchnorm/batchnorm_infer_rank_4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/block_to_ctile_map/test_block_to_ctile_map.cpp b/test/block_to_ctile_map/test_block_to_ctile_map.cpp
index 55d9b59f48..b8e349eda1 100644
--- a/test/block_to_ctile_map/test_block_to_ctile_map.cpp
+++ b/test/block_to_ctile_map/test_block_to_ctile_map.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <vector>
diff --git a/test/conv_util/conv_util.cpp b/test/conv_util/conv_util.cpp
index 73797a7169..6922bbbcc7 100644
--- a/test/conv_util/conv_util.cpp
+++ b/test/conv_util/conv_util.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <string>
diff --git a/test/convnd_bwd_data/convnd_bwd_data.cpp b/test/convnd_bwd_data/convnd_bwd_data.cpp
index 70231d42ae..9d2b6cf577 100644
--- a/test/convnd_bwd_data/convnd_bwd_data.cpp
+++ b/test/convnd_bwd_data/convnd_bwd_data.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/convnd_fwd/convnd_fwd.cpp b/test/convnd_fwd/convnd_fwd.cpp
index a1921a9bfb..fe8798ceb8 100644
--- a/test/convnd_fwd/convnd_fwd.cpp
+++ b/test/convnd_fwd/convnd_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/data_type/int4.cpp b/test/data_type/int4.cpp
index 252a450bf9..07549c1c48 100644
--- a/test/data_type/int4.cpp
+++ b/test/data_type/int4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <bitset>
 #include <cinttypes>
diff --git a/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp b/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp
index e80995c4f0..d5ce77dc2b 100644
--- a/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp
+++ b/test/elementwise_normalization/test_elementwise_layernorm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_elementwise_layernorm_impl.hpp"
diff --git a/test/gemm/gemm_bf16.cpp b/test/gemm/gemm_bf16.cpp
index 5290d46632..cde5c45aea 100644
--- a/test/gemm/gemm_bf16.cpp
+++ b/test/gemm/gemm_bf16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_fp16.cpp b/test/gemm/gemm_fp16.cpp
index 92e225def2..cad250c6fb 100644
--- a/test/gemm/gemm_fp16.cpp
+++ b/test/gemm/gemm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_fp32.cpp b/test/gemm/gemm_fp32.cpp
index 5d8c4881b6..c35aa77ea7 100644
--- a/test/gemm/gemm_fp32.cpp
+++ b/test/gemm/gemm_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_fp64.cpp b/test/gemm/gemm_fp64.cpp
index 85d7f95bf4..e67c8ba4f3 100644
--- a/test/gemm/gemm_fp64.cpp
+++ b/test/gemm/gemm_fp64.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_int8.cpp b/test/gemm/gemm_int8.cpp
index e73b22ce9c..6ece05e306 100644
--- a/test/gemm/gemm_int8.cpp
+++ b/test/gemm/gemm_int8.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/gemm/gemm_standalone_xdl_fp16.cpp b/test/gemm/gemm_standalone_xdl_fp16.cpp
index 32a243e0f6..201a49dcd3 100644
--- a/test/gemm/gemm_standalone_xdl_fp16.cpp
+++ b/test/gemm/gemm_standalone_xdl_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gemm_util.hpp"
 
diff --git a/test/gemm/gemm_util.hpp b/test/gemm/gemm_util.hpp
index 9057c0af89..6c46f4ee89 100644
--- a/test/gemm/gemm_util.hpp
+++ b/test/gemm/gemm_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/test/gemm/instance/gemm_f16_nn_instance.cpp b/test/gemm/instance/gemm_f16_nn_instance.cpp
index 4d65c5876c..9016257f13 100644
--- a/test/gemm/instance/gemm_f16_nn_instance.cpp
+++ b/test/gemm/instance/gemm_f16_nn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_nn_instance.hpp b/test/gemm/instance/gemm_f16_nn_instance.hpp
index 5ae3928dc9..e174b99a1d 100644
--- a/test/gemm/instance/gemm_f16_nn_instance.hpp
+++ b/test/gemm/instance/gemm_f16_nn_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_nt_instance.cpp b/test/gemm/instance/gemm_f16_nt_instance.cpp
index 431ff1e62e..27103b88d4 100644
--- a/test/gemm/instance/gemm_f16_nt_instance.cpp
+++ b/test/gemm/instance/gemm_f16_nt_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_nt_instance.hpp b/test/gemm/instance/gemm_f16_nt_instance.hpp
index 99f9ffba45..c624425e69 100644
--- a/test/gemm/instance/gemm_f16_nt_instance.hpp
+++ b/test/gemm/instance/gemm_f16_nt_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tn_instance.cpp b/test/gemm/instance/gemm_f16_tn_instance.cpp
index 6f5dbc311e..5b11f4dad9 100644
--- a/test/gemm/instance/gemm_f16_tn_instance.cpp
+++ b/test/gemm/instance/gemm_f16_tn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tn_instance.hpp b/test/gemm/instance/gemm_f16_tn_instance.hpp
index 62388aeb39..563e10600a 100644
--- a/test/gemm/instance/gemm_f16_tn_instance.hpp
+++ b/test/gemm/instance/gemm_f16_tn_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tt_instance.cpp b/test/gemm/instance/gemm_f16_tt_instance.cpp
index b6ef5b1cd2..9032150f0c 100644
--- a/test/gemm/instance/gemm_f16_tt_instance.cpp
+++ b/test/gemm/instance/gemm_f16_tt_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_f16_tt_instance.hpp b/test/gemm/instance/gemm_f16_tt_instance.hpp
index 9d75b4e48c..62914d7ac2 100644
--- a/test/gemm/instance/gemm_f16_tt_instance.hpp
+++ b/test/gemm/instance/gemm_f16_tt_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp b/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
index 51c014a91a..983af7ecdd 100644
--- a/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
+++ b/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp b/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp
index 110fc5f7d9..ef269d78ee 100644
--- a/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp
+++ b/test/gemm/instance/gemm_wavelet_f16_tn_instance.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <memory>
 #include <vector>
diff --git a/test/gemm/run_gemm_test.inc b/test/gemm/run_gemm_test.inc
index ec27729b3c..d208bb5a7b 100644
--- a/test/gemm/run_gemm_test.inc
+++ b/test/gemm/run_gemm_test.inc
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 int run_gemm_test()
 {
diff --git a/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp b/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp
index 740c63aa7e..3f05996878 100644
--- a/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp
+++ b/test/gemm_layernorm/test_gemm_add_relu_add_layernorm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_gemm_add_relu_add_layernorm_impl.hpp"
diff --git a/test/gemm_reduce/gemm_reduce_fp16.cpp b/test/gemm_reduce/gemm_reduce_fp16.cpp
index 029165ece1..35a149f52c 100644
--- a/test/gemm_reduce/gemm_reduce_fp16.cpp
+++ b/test/gemm_reduce/gemm_reduce_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 
diff --git a/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
index 75f934cc06..207cdab7c7 100644
--- a/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
+++ b/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
index 6df7f9969c..4a804ef7f1 100644
--- a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
+++ b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 #include <iostream>
diff --git a/test/magic_number_division/magic_number_division.cpp b/test/magic_number_division/magic_number_division.cpp
index 680fddf193..253f21e91f 100644
--- a/test/magic_number_division/magic_number_division.cpp
+++ b/test/magic_number_division/magic_number_division.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
diff --git a/test/normalization/test_groupnorm_fp16.cpp b/test/normalization/test_groupnorm_fp16.cpp
index 60d3b13959..325ea75fe5 100644
--- a/test/normalization/test_groupnorm_fp16.cpp
+++ b/test/normalization/test_groupnorm_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_groupnorm_impl.hpp"
diff --git a/test/normalization/test_groupnorm_fp32.cpp b/test/normalization/test_groupnorm_fp32.cpp
index 3542f73a62..ec88442fc0 100644
--- a/test/normalization/test_groupnorm_fp32.cpp
+++ b/test/normalization/test_groupnorm_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_groupnorm_impl.hpp"
diff --git a/test/normalization/test_layernorm2d_fp16.cpp b/test/normalization/test_layernorm2d_fp16.cpp
index d627cbe7f1..2222740fcc 100644
--- a/test/normalization/test_layernorm2d_fp16.cpp
+++ b/test/normalization/test_layernorm2d_fp16.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_layernorm_impl.hpp"
diff --git a/test/normalization/test_layernorm2d_fp32.cpp b/test/normalization/test_layernorm2d_fp32.cpp
index de4133aa83..30fbe06c60 100644
--- a/test/normalization/test_layernorm2d_fp32.cpp
+++ b/test/normalization/test_layernorm2d_fp32.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_layernorm_impl.hpp"
diff --git a/test/pool_fwd/test_avg_pool2d_fwd.cpp b/test/pool_fwd/test_avg_pool2d_fwd.cpp
index 4e5f1e0e91..72749fd6e1 100644
--- a/test/pool_fwd/test_avg_pool2d_fwd.cpp
+++ b/test/pool_fwd/test_avg_pool2d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_pool2d_fwd_impl.hpp"
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool_fwd/test_avg_pool3d_fwd.cpp
index 0d6b105b1b..00cc3740fa 100644
--- a/test/pool_fwd/test_avg_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_avg_pool3d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_pool3d_fwd_impl.hpp"
diff --git a/test/pool_fwd/test_max_pool2d_fwd.cpp b/test/pool_fwd/test_max_pool2d_fwd.cpp
index d16ac7faba..1cf1314f46 100644
--- a/test/pool_fwd/test_max_pool2d_fwd.cpp
+++ b/test/pool_fwd/test_max_pool2d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_pool2d_fwd_impl.hpp"
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool_fwd/test_max_pool3d_fwd.cpp
index f084dd9cbf..0b0de4d90e 100644
--- a/test/pool_fwd/test_max_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_max_pool3d_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "profiler/profile_pool3d_fwd_impl.hpp"
diff --git a/test/pool_fwd/test_pool_fwd_common.hpp b/test/pool_fwd/test_pool_fwd_common.hpp
index a78785171d..f018635170 100644
--- a/test/pool_fwd/test_pool_fwd_common.hpp
+++ b/test/pool_fwd/test_pool_fwd_common.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "gtest/gtest.h"
 #include "ck/ck.hpp"
diff --git a/test/reduce/reduce_no_index.cpp b/test/reduce/reduce_no_index.cpp
index 3f4d0676b4..1ab452442d 100644
--- a/test/reduce/reduce_no_index.cpp
+++ b/test/reduce/reduce_no_index.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <getopt.h>
 
diff --git a/test/reduce/reduce_with_index.cpp b/test/reduce/reduce_with_index.cpp
index c616a68e74..0301669c5a 100644
--- a/test/reduce/reduce_with_index.cpp
+++ b/test/reduce/reduce_with_index.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <getopt.h>
 
diff --git a/test/reference_conv_fwd/reference_conv_fwd.cpp b/test/reference_conv_fwd/reference_conv_fwd.cpp
index 1f9ba0064c..b3328e4b36 100644
--- a/test/reference_conv_fwd/reference_conv_fwd.cpp
+++ b/test/reference_conv_fwd/reference_conv_fwd.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cmath>
 #include <cstdlib>
diff --git a/test/softmax/test_softmax_interface.cpp b/test/softmax/test_softmax_interface.cpp
index 8cac0ba0f5..25f666f0ea 100644
--- a/test/softmax/test_softmax_interface.cpp
+++ b/test/softmax/test_softmax_interface.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <stdexcept>
 #include <vector>
diff --git a/test/softmax/test_softmax_rank3.cpp b/test/softmax/test_softmax_rank3.cpp
index 5691ee3f6c..24ad912d8d 100644
--- a/test/softmax/test_softmax_rank3.cpp
+++ b/test/softmax/test_softmax_rank3.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <stdexcept>
diff --git a/test/softmax/test_softmax_rank4.cpp b/test/softmax/test_softmax_rank4.cpp
index f0b22df25e..b58301fb11 100644
--- a/test/softmax/test_softmax_rank4.cpp
+++ b/test/softmax/test_softmax_rank4.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <stdexcept>
diff --git a/test/softmax/test_softmax_util.hpp b/test/softmax/test_softmax_util.hpp
index 40b300cf99..e36231de85 100644
--- a/test/softmax/test_softmax_util.hpp
+++ b/test/softmax/test_softmax_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/test/space_filling_curve/space_filling_curve.cpp b/test/space_filling_curve/space_filling_curve.cpp
index c7f6759e81..a192ecb28f 100644
--- a/test/space_filling_curve/space_filling_curve.cpp
+++ b/test/space_filling_curve/space_filling_curve.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <vector>
 #include <iostream>
diff --git a/test/wmma_op/wmma_op.cpp b/test/wmma_op/wmma_op.cpp
index 761c15f1dd..47d8c7ed6f 100644
--- a/test/wmma_op/wmma_op.cpp
+++ b/test/wmma_op/wmma_op.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <algorithm>
 #include <cstdlib>
diff --git a/test/wmma_op/wmma_op_util.hpp b/test/wmma_op/wmma_op_util.hpp
index c70e6a407d..49782bce6e 100644
--- a/test/wmma_op/wmma_op_util.hpp
+++ b/test/wmma_op/wmma_op_util.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 

From 9eae73df9b63d04bdfb7117c27dbe06da054c5d6 Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Fri, 2 Jun 2023 05:23:02 +0800
Subject: [PATCH 025/194] Simplify kernel argument of device operator
 Device(Batched)GemmXdl<> (#723)

* Remove M/N/KPad local variables

* Use M/N/KPad to name padded lengths

* Replace duplicated local variable by parameters

* Rename variables M/N/KRaw to M/N/K

* Move AK0/BK0 compute logic into GridwiseGemm

* Use macro to shorten code

* Move CalculateGridSize() logic into GridwiseGemm

* Add comment to credit the implementation source

* Reuse the existing implementation

* Remove no-longer used data members

* Remove elementwise-op objects from interfaces

* Reserve kernel arg as whole object in interfaces

* Remove redundant data member

* Make 3rd type parameter optional

* Remove unnesscary type parameters

* Remove no-longer used descriptor-creation methods

* Move kernel arg type definition into GridwiseGemm

* Add macro to switch between code sections

* Move argument field computing logic into device op side

* Make utility method 'static'

* Declare special methods

* Unify MakeArgument() usage

* Adapt the new GridwiseGemm interface

* Push-down class 'GridwiseGemm::Argument' fields

* Remove no-longer used methods

* Add unused parameters

* Force copying parameters in 'Embed' ctor

* Remove no-longer used descriptors

* Fallback change on BaseArgument

* Remove macro 'INTEGER_DIVIDE_CEIL'

* Make variable naming more consistent

* Make sure methods are only invoked on right place

* Remove tailing underscore in public attribute name

* Remove necessary methods

* Hide computing logic of derived attributes

* Make new 'Embed' ctor only available for device code

* Make sure 'Embed' type args are not references

* Move check for karg.K into CheckValidity()

* Remove more integer division logic form device code

* Undo changes on Embed

* Separate 'Problem' concept out from 'Argument'

* Add overloaded version of __builtin_amdgcn_readfirstlane()

* Remove 'static' specifiers

* Remove more 'static' specifier

* Replace unsigne char by std::byte

* Add 'const' specifier to never changing variable

* Add 'inline' specifier to funcion definition

* Share same name for kernel interfaces

* Fix wrong boundar calculation logic

* Leave the third template arg for compatibility

* Remove unnecessary parameters

* Fix wrong error message (for type name)

* Create descriptor on device side

* Fix wrong debug message

* Remove no-longer used data members

* Rename type trait

* Remove std:: qualifier from standard types

* Replace 'size_t' by 'unsigned'

* Use type alias to hint usage

* Replace static_for<> by ordinary 'for' loop

* Reject unsupported argument

* Rename readfirstlane() to amd_wave_read_first_lane()

* Rename file readfirstlance.hpp as amd_wave_read_first_lane.hpp

* Update function calls

* Reorder statements

* Re-format files

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../convnd_bwd_data_common.hpp                |  41 +-
 .../device/impl/device_batched_gemm_xdl.hpp   | 461 ++++-----------
 ...ice_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp | 180 ++----
 .../device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp  | 132 +----
 ...device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp | 213 ++-----
 .../gpu/device/impl/device_gemm_xdl.hpp       | 325 +----------
 .../gpu/grid/block_to_ctile_map.hpp           |  44 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    | 551 ++++++++++++++++--
 8 files changed, 810 insertions(+), 1137 deletions(-)

diff --git a/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp b/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
index b4b544aabe..4a9d16c5c3 100644
--- a/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
+++ b/example/17_convnd_bwd_data/convnd_bwd_data_common.hpp
@@ -81,32 +81,33 @@ int run_conv_bwd_data(bool do_verification,
     in_device_buf.SetZero();
 
     // do GEMM
-    auto conv     = DeviceConvNdBwdDataInstance{};
-    auto invoker  = conv.MakeInvoker();
-    auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-                                      static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
-                                      static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-                                      conv_param.N_,
-                                      conv_param.K_,
-                                      conv_param.C_,
-                                      conv_param.input_spatial_lengths_,
-                                      conv_param.filter_spatial_lengths_,
-                                      conv_param.GetOutputSpatialLengths(),
-                                      conv_param.conv_filter_strides_,
-                                      conv_param.conv_filter_dilations_,
-                                      conv_param.input_left_pads_,
-                                      conv_param.input_right_pads_,
-                                      in_element_op,
-                                      wei_element_op,
-                                      out_element_op);
+    auto conv    = DeviceConvNdBwdDataInstance{};
+    auto invoker = conv.MakeInvoker();
+    auto argument =
+        conv.MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+                                 static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
+                                 static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+                                 conv_param.N_,
+                                 conv_param.K_,
+                                 conv_param.C_,
+                                 conv_param.input_spatial_lengths_,
+                                 conv_param.filter_spatial_lengths_,
+                                 conv_param.GetOutputSpatialLengths(),
+                                 conv_param.conv_filter_strides_,
+                                 conv_param.conv_filter_dilations_,
+                                 conv_param.input_left_pads_,
+                                 conv_param.input_right_pads_,
+                                 in_element_op,
+                                 wei_element_op,
+                                 out_element_op);
 
-    if(!conv.IsSupportedArgument(argument))
+    if(!conv.IsSupportedArgument(argument.get()))
     {
         std::cout << "Not support,please check parameters or device";
         return 0;
     }
 
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+    float ave_time = invoker.Run(argument.get(), StreamConfig{nullptr, time_kernel});
 
     std::size_t flop      = conv_param.GetFlops();
     std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index b100967065..87c58f371a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -45,75 +45,46 @@ namespace device {
  * realize BatchedGemm and GroupedGemm (and the corresponding GEMM fusion).
  *
  */
-template <typename GridwiseGemm,
-          typename FloatAB,
-          typename FloatC,
-          typename AGridDesc_K0_M_K1,
-          typename BGridDesc_K0_N_K1,
-          typename CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename ComputePtrOffsetOfBatch,
-          typename Block2CTileMap,
-          bool HasMainKBlockLoop>
+template <typename DeviceOp, typename GridwiseGemm, bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_batched_gemm_xdlops_v2r3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const index_t batch_count,
-            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
-            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
-            const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CElementwiseOperation c_element_op,
-            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
-            const Block2CTileMap block_2_ctile_map)
+        kernel_batched_gemm_xdlops_v2r3(const typename DeviceOp::Argument karg)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__))
     const index_t num_blocks_per_batch =
-        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+        __builtin_amdgcn_readfirstlane(get_grid_size() / karg.Batch);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
 
     const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
-        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+        static_cast<long_index_t>(karg.compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
     const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
-        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+        static_cast<long_index_t>(karg.compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
     const long_index_t c_batch_offset = __builtin_amdgcn_readfirstlane(
-        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetCPtrOffset(g_idx)));
+        static_cast<long_index_t>(karg.compute_ptr_offset_of_batch.GetCPtrOffset(g_idx)));
 
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
-                                                  p_b_grid + b_batch_offset,
-                                                  p_c_grid + c_batch_offset,
+    const auto a_grid_desc_k0_m_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeAGridDescriptor_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.K0, karg.StrideA));
+    const auto b_grid_desc_k0_n_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeBGridDescriptor_K0_N_K1(
+            karg.K, karg.N, karg.NPadded, karg.K0, karg.StrideB));
+    const auto c_grid_desc_m_n = amd_wave_read_first_lane(GridwiseGemm::MakeCGridDescriptor_M_N(
+        karg.M, karg.MPadded, karg.N, karg.NPadded, karg.StrideC));
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(karg.p_a_grid + a_batch_offset,
+                                                  karg.p_b_grid + b_batch_offset,
+                                                  karg.p_c_grid + c_batch_offset,
                                                   p_shared,
                                                   a_grid_desc_k0_m_k1,
                                                   b_grid_desc_k0_n_k1,
-                                                  c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  block_2_ctile_map);
+                                                  c_grid_desc_m_n);
 #else
-    ignore = p_a_grid;
-    ignore = p_b_grid;
-    ignore = p_c_grid;
-    ignore = batch_count;
-    ignore = a_grid_desc_k0_m_k1;
-    ignore = b_grid_desc_k0_n_k1;
-    ignore = c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = c_element_op;
-    ignore = compute_ptr_offset_of_batch;
-    ignore = block_2_ctile_map;
+    ignore = karg;
 #endif
 }
 
@@ -171,93 +142,6 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
 
     static constexpr auto K1Number = Number<K1>{};
 
-    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
-    {
-        assert(K % K1 == 0);
-
-        const index_t K0 = K / K1;
-
-        const auto a_grid_desc_m_k = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
-            }
-        }();
-
-        const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-
-        const auto a_grid_desc_k0_mp_k1 =
-            transform_tensor_descriptor(a_grid_desc_m_k,
-                                        make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                                                   make_right_pad_transform(M, PadM)),
-                                        make_tuple(Sequence<1>{}, Sequence<0>{}),
-                                        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-        return a_grid_desc_k0_mp_k1;
-    }
-
-    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
-    {
-        assert(K % K1 == 0);
-
-        const index_t K0 = K / K1;
-
-        const auto b_grid_desc_k_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
-            }
-        }();
-
-        const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-        const auto b_grid_desc_k0_np_k1 =
-            transform_tensor_descriptor(b_grid_desc_k_n,
-                                        make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                                                   make_right_pad_transform(N, PadN)),
-                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
-                                        make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-
-        return b_grid_desc_k0_np_k1;
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
-    {
-        const auto c_grid_desc_m_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
-            }
-        }();
-
-        const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-        const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-        const auto c_grid_desc_mp_np = transform_tensor_descriptor(
-            c_grid_desc_m_n,
-            make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        return c_grid_desc_mp_np;
-    }
-
-    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
-    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
-    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
     struct ComputePtrOffsetOfStridedBatch
     {
         ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
@@ -289,121 +173,82 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
     };
 
     // GridwiseGemm
-    using GridwiseGemm =
-        GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
-                                                ADataType, // TODO: distinguish A/B datatype
-                                                AccDataType,
-                                                CDataType,
-                                                InMemoryDataOperationEnum::Set,
-                                                AGridDesc_K0_M_K1,
-                                                BGridDesc_K0_N_K1,
-                                                CGridDesc_M_N,
-                                                AElementwiseOperation,
-                                                BElementwiseOperation,
-                                                CElementwiseOperation,
-                                                MPerBlock,
-                                                NPerBlock,
-                                                K0PerBlock,
-                                                MPerXDL,
-                                                NPerXDL,
-                                                K1,
-                                                MXdlPerWave,
-                                                NXdlPerWave,
-                                                ABlockTransferThreadClusterLengths_K0_M_K1,
-                                                ABlockTransferThreadClusterArrangeOrder,
-                                                ABlockTransferSrcAccessOrder,
-                                                ABlockTransferSrcVectorDim,
-                                                ABlockTransferSrcScalarPerVector,
-                                                ABlockTransferDstScalarPerVector_K1,
-                                                false, // AThreadTransferSrcResetCoordinateAfterRun,
-                                                ABlockLdsAddExtraM,
-                                                BBlockTransferThreadClusterLengths_K0_N_K1,
-                                                BBlockTransferThreadClusterArrangeOrder,
-                                                BBlockTransferSrcAccessOrder,
-                                                BBlockTransferSrcVectorDim,
-                                                BBlockTransferSrcScalarPerVector,
-                                                BBlockTransferDstScalarPerVector_K1,
-                                                false, // BThreadTransferSrcResetCoordinateAfterRun,
-                                                BBlockLdsAddExtraN,
-                                                Sequence<2, 3, 0, 1, 7, 5, 4, 6>,
-                                                CThreadTransferSrcDstVectorDim,
-                                                CThreadTransferDstScalarPerVector,
-                                                NumGemmKPrefetchStage,
-                                                LoopSched,
-                                                PipelineVer>;
+    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext<
+        BlockSize,
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        CDataType,
+        InMemoryDataOperationEnum::Set,
+        ALayout,
+        BLayout,
+        CLayout,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        GemmSpecialization::MNPadding,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerXDL,
+        NPerXDL,
+        K1,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        Sequence<2, 3, 0, 1, 7, 5, 4, 6>,
+        CThreadTransferSrcDstVectorDim,
+        CThreadTransferDstScalarPerVector,
+        NumGemmKPrefetchStage,
+        LoopSched,
+        PipelineVer>;
 
-    using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 =
-        decltype(GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}));
-    using Block2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
+    using Problem = typename GridwiseGemm::Problem;
 
     // Argument
-    struct Argument : public BaseArgument
+    struct Argument : public Problem, public BaseArgument
     {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t M,
-                 index_t N,
-                 index_t K,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
+        Argument(const ADataType* p_a_grid_,
+                 const BDataType* p_b_grid_,
+                 CDataType* p_c_grid_,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_,
                  index_t BatchStrideA,
                  index_t BatchStrideB,
                  index_t BatchStrideC,
-                 index_t Batch,
-                 index_t M01,
-                 index_t N01,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              Batch_(Batch),
-              a_grid_desc_k0_m_k1_{
-                  DeviceBatchedGemmXdl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA)},
-              b_grid_desc_k0_n_k1_{
-                  DeviceBatchedGemmXdl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB)},
-              c_grid_desc_m_n_{DeviceBatchedGemmXdl::MakeCGridDescriptor_M_N(M, N, StrideC)},
-              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
-              compute_ptr_offset_of_batch_{BatchStrideA, BatchStrideB, BatchStrideC},
-              block_2_ctile_map_{
-                  GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01)},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              kraw_{K}
+                 index_t Batch_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_},
+              Batch(Batch_),
+              compute_ptr_offset_of_batch{BatchStrideA, BatchStrideB, BatchStrideC}
         {
-            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
-                                           b_grid_desc_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
-            }
         }
 
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        index_t Batch_;
-        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
-        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
-        Block2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t kraw_;
+        const ADataType* p_a_grid;
+        const BDataType* p_b_grid;
+        CDataType* p_c_grid;
+        index_t Batch;
+        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch;
     };
 
     // Invoker
@@ -411,107 +256,39 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
     {
         using Argument = DeviceBatchedGemmXdl::Argument;
 
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
         {
-#if DEBUG_LOG
+            if(stream_config.log_level_ > 0)
             {
-                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
-                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
-                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_m_n_{" << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+                karg.Print();
             }
-#endif
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                            arg.b_grid_desc_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(karg))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseBatchedGemm_km_kn_m0m1n0n1_xdlops_v2r3 has invalid setting");
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_) * arg.Batch_;
-
-            const auto K =
-                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+            auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(karg.M, karg.N);
+            gdx *= karg.Batch;
 
             float ave_time = 0;
 
-            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(karg.K))
             {
-                const auto kernel = kernel_batched_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    ComputePtrOffsetOfStridedBatch,
-                    remove_reference_t<Block2CTileMap>,
-                    true>;
+                const auto kernel =
+                    kernel_batched_gemm_xdlops_v2r3<DeviceBatchedGemmXdl, GridwiseGemm, true>;
 
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.Batch_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.compute_ptr_offset_of_batch_,
-                                                  arg.block_2_ctile_map_);
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
             else
             {
-                const auto kernel = kernel_batched_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceBatchedGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceBatchedGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    ComputePtrOffsetOfStridedBatch,
-                    remove_reference_t<Block2CTileMap>,
-                    false>;
+                const auto kernel =
+                    kernel_batched_gemm_xdlops_v2r3<DeviceBatchedGemmXdl, GridwiseGemm, false>;
 
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.Batch_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.compute_ptr_offset_of_batch_,
-                                                  arg.block_2_ctile_map_);
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
 
             return ave_time;
@@ -531,17 +308,14 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
         return true;
     }
 
-    static bool IsSupportedArgument(const Argument& arg)
+    static bool IsSupportedArgument(const Problem& problem)
     {
-        if(arg.kraw_ % K1 != 0)
+        if(problem.K % K1 != 0)
         {
             return false;
         }
 
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                           arg.b_grid_desc_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(problem);
     }
 
     // polymorphic
@@ -562,10 +336,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                              index_t BatchStrideA,
                              index_t BatchStrideB,
                              index_t BatchStrideC,
-                             index_t Batch,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             index_t Batch)
     {
         return Argument{p_a,
                         p_b,
@@ -579,12 +350,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                         BatchStrideA,
                         BatchStrideB,
                         BatchStrideC,
-                        Batch,
-                        1,
-                        1,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+                        Batch};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -603,9 +369,9 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                                                       index_t BatchStrideB,
                                                       index_t BatchStrideC,
                                                       index_t Batch,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op) override
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
@@ -619,12 +385,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
                                           BatchStrideA,
                                           BatchStrideB,
                                           BatchStrideC,
-                                          Batch,
-                                          1,
-                                          1,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          Batch);
     }
 
     // polymorphic
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
index 8cef0eaf95..eb4db6f8cb 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
@@ -379,9 +379,6 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
         InElementwiseOperation,
         WeiElementwiseOperation,
         OutElementwiseOperation,
@@ -428,20 +425,10 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                  std::vector<ck::index_t> conv_filter_strides,
                  std::vector<ck::index_t> conv_filter_dilations,
                  std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+                 std::vector<ck::index_t> input_right_pads)
             : p_a_grid_{p_out_grid},
               p_b_grid_{p_wei_grid},
               p_c_grid_{p_in_grid},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{out_element_op},
-              b_element_op_{wei_element_op},
-              c_element_op_{in_element_op},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
@@ -495,18 +482,6 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                     a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                     b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                     c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                    auto block_2_ctile_map =
-                        GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01, N01);
-
-                    if(GridwiseGemm::CheckValidity(
-                           descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                    {
-                        c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                            GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(descs[I2]));
-
-                        block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                    }
                 }
             }
         }
@@ -517,14 +492,6 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         std::vector<AGridDesc_K0_M_K1> a_grid_desc_k0_m_k1_container_;
         std::vector<BGridDesc_K0_N_K1> b_grid_desc_k0_n_k1_container_;
         std::vector<CGridDesc_M_N> c_grid_desc_m_n_container_;
-        std::vector<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_;
-        std::vector<typename GridwiseGemm::DefaultBlock2CTileMap> block_2_ctile_map_container_;
-        index_t M01_;
-        index_t N01_;
-        OutElementwiseOperation a_element_op_;
-        WeiElementwiseOperation b_element_op_;
-        InElementwiseOperation c_element_op_;
         // for checking IsSupportedArgument()
         index_t Conv_N_;
         index_t Conv_K_;
@@ -567,103 +534,68 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I0) << ", "
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I1) << "}"
                               << std::endl;
-
-                    std::cout << "arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I0)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I1)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I2)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I3)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I4)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I5)
-                              << " ) " << std::endl;
                 }
 #endif
 
                 if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                                 arg.b_grid_desc_k0_n_k1_container_[i],
-                                                arg.c_grid_desc_m_n_container_[i],
-                                                arg.block_2_ctile_map_container_[i]))
+                                                arg.c_grid_desc_m_n_container_[i]))
                 {
                     throw std::runtime_error(
                         "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
                 }
 
-                const index_t grid_size = arg.block_2_ctile_map_container_[i].CalculateGridSize(
-                    arg.c_grid_desc_m_n_container_[i]);
+                const auto [gdx, gdy, gdz] =
+                    GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_container_[i]);
 
                 const auto K = arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I0) *
                                arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I2);
 
                 if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        true>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                true>;
 
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        false>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                false>;
 
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
             }
             return ave_time;
@@ -716,8 +648,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         {
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                             arg.b_grid_desc_k0_n_k1_container_[i],
-                                            arg.c_grid_desc_m_n_container_[i],
-                                            arg.block_2_ctile_map_container_[i]))
+                                            arg.c_grid_desc_m_n_container_[i]))
             {
                 return false;
             }
@@ -742,10 +673,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                              std::vector<ck::index_t> conv_filter_strides,
                              std::vector<ck::index_t> conv_filter_dilations,
                              std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+                             std::vector<ck::index_t> input_right_pads)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -759,12 +687,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
-                        input_right_pads,
-                        1,
-                        1,
-                        in_element_op,
-                        wei_element_op,
-                        out_element_op};
+                        input_right_pads};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -783,9 +706,9 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         std::vector<ck::index_t> conv_filter_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
-                        InElementwiseOperation in_element_op,
-                        WeiElementwiseOperation wei_element_op,
-                        OutElementwiseOperation out_element_op) override
+                        InElementwiseOperation,
+                        WeiElementwiseOperation,
+                        OutElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<InDataType*>(p_in_grid),
                                           static_cast<const WeiDataType*>(p_wei_grid),
@@ -799,12 +722,7 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
-                                          input_right_pads,
-                                          1,
-                                          1,
-                                          in_element_op,
-                                          wei_element_op,
-                                          out_element_op);
+                                          input_right_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
index 710ea91760..88615bba31 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
@@ -329,9 +329,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
         InElementwiseOperation,
         WeiElementwiseOperation,
         OutElementwiseOperation,
@@ -378,25 +375,13 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                  std::vector<ck::index_t> conv_filter_strides,
                  std::vector<ck::index_t> conv_filter_dilations,
                  std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+                 std::vector<ck::index_t> input_right_pads)
             : p_a_grid_{p_in_grid},
               p_b_grid_{p_wei_grid},
               p_c_grid_{p_out_grid},
               a_grid_desc_k0_m_k1_{},
               b_grid_desc_k0_n_k1_{},
               c_grid_desc_m_n_{},
-              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
-              block_2_ctile_map_{},
-              M01_{M01},
-              N01_{N01},
-              in_element_op_{in_element_op},
-              wei_element_op_{wei_element_op},
-              out_element_op_{out_element_op},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
@@ -420,17 +405,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
             a_grid_desc_k0_m_k1_ = descs[I0];
             b_grid_desc_k0_n_k1_ = descs[I1];
             c_grid_desc_m_n_     = descs[I2];
-            block_2_ctile_map_ =
-                GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
-                                           b_grid_desc_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
-            }
         }
 
         //  private:
@@ -440,14 +414,6 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
         BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
         CGridDesc_M_N c_grid_desc_m_n_;
-        typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        InElementwiseOperation in_element_op_;
-        WeiElementwiseOperation wei_element_op_;
-        OutElementwiseOperation out_element_op_;
         // for checking IsSupportedArgument()
         index_t Conv_N_;
         index_t Conv_K_;
@@ -479,17 +445,14 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                           << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
             }
 #endif
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                            arg.b_grid_desc_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(
+                   arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_))
             {
                 throw std::runtime_error(
                     "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
+            const auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_);
 
             const auto K =
                 arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
@@ -498,22 +461,18 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    InElementwiseOperation,
-                    WeiElementwiseOperation,
-                    OutElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    true>;
+                const auto kernel =
+                    kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                            ADataType, // TODO: distiguish A/B datatype
+                                            CDataType,
+                                            DeviceOp::AGridDesc_K0_M_K1,
+                                            DeviceOp::BGridDesc_K0_N_K1,
+                                            DeviceOp::CGridDesc_M_N,
+                                            true>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
-                                                  dim3(grid_size),
+                                                  dim3(gdx, gdy, gdz),
                                                   dim3(BlockSize),
                                                   0,
                                                   arg.p_a_grid_,
@@ -521,30 +480,22 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                                   arg.p_c_grid_,
                                                   arg.a_grid_desc_k0_m_k1_,
                                                   arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.in_element_op_,
-                                                  arg.wei_element_op_,
-                                                  arg.out_element_op_,
-                                                  arg.block_2_ctile_map_);
+                                                  arg.c_grid_desc_m_n_);
             }
             else
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    InElementwiseOperation,
-                    WeiElementwiseOperation,
-                    OutElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    false>;
+                const auto kernel =
+                    kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                            ADataType, // TODO: distiguish A/B datatype
+                                            CDataType,
+                                            DeviceOp::AGridDesc_K0_M_K1,
+                                            DeviceOp::BGridDesc_K0_N_K1,
+                                            DeviceOp::CGridDesc_M_N,
+                                            false>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
-                                                  dim3(grid_size),
+                                                  dim3(gdx, gdy, gdz),
                                                   dim3(BlockSize),
                                                   0,
                                                   arg.p_a_grid_,
@@ -552,11 +503,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                                   arg.p_c_grid_,
                                                   arg.a_grid_desc_k0_m_k1_,
                                                   arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.in_element_op_,
-                                                  arg.wei_element_op_,
-                                                  arg.out_element_op_,
-                                                  arg.block_2_ctile_map_);
+                                                  arg.c_grid_desc_m_n_);
             }
 
             return ave_time;
@@ -616,10 +563,8 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
         }
 
         // Gridwise GEMM size
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                           arg.b_grid_desc_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(
+            arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
     }
 
     bool IsSupportedArgument(const BaseArgument* p_arg) override
@@ -639,10 +584,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                              std::vector<ck::index_t> conv_filter_strides,
                              std::vector<ck::index_t> conv_filter_dilations,
                              std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+                             std::vector<ck::index_t> input_right_pads)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -656,12 +598,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
-                        input_right_pads,
-                        1,
-                        1,
-                        in_element_op,
-                        wei_element_op,
-                        out_element_op};
+                        input_right_pads};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -680,9 +617,9 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                         std::vector<ck::index_t> conv_filter_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
-                        InElementwiseOperation in_element_op,
-                        WeiElementwiseOperation wei_element_op,
-                        OutElementwiseOperation out_element_op) override
+                        InElementwiseOperation,
+                        WeiElementwiseOperation,
+                        OutElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
                                           static_cast<const WeiDataType*>(p_wei_grid),
@@ -696,12 +633,7 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
-                                          input_right_pads,
-                                          1,
-                                          1,
-                                          in_element_op,
-                                          wei_element_op,
-                                          out_element_op);
+                                          input_right_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
index 822e1da4ec..77ad61d7ec 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
@@ -980,9 +980,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
         InElementwiseOperation,
         WeiElementwiseOperation,
         OutElementwiseOperation,
@@ -1029,20 +1026,10 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                  std::vector<ck::index_t> conv_filter_strides,
                  std::vector<ck::index_t> conv_filter_dilations,
                  std::vector<ck::index_t> input_left_pads,
-                 std::vector<ck::index_t> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
-                 InElementwiseOperation in_element_op,
-                 WeiElementwiseOperation wei_element_op,
-                 OutElementwiseOperation out_element_op)
+                 std::vector<ck::index_t> input_right_pads)
             : p_a_grid_{p_out_grid},
               p_b_grid_{p_wei_grid},
               p_c_grid_{p_in_grid},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{out_element_op},
-              b_element_op_{wei_element_op},
-              c_element_op_{in_element_op},
               Conv_N_{N},
               Conv_K_{K},
               Conv_C_{C},
@@ -1092,17 +1079,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                 a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                 b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                 c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                auto block_2_ctile_map =
-                    GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01_, N01_);
-
-                if(GridwiseGemm::CheckValidity(descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                {
-                    c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                        GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(descs[I2]));
-
-                    block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                }
             }
         }
         template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
@@ -1150,18 +1126,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                     a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                     b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                     c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                    auto block_2_ctile_map =
-                        GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01_, N01_);
-
-                    if(GridwiseGemm::CheckValidity(
-                           descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                    {
-                        c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                            GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(descs[I2]));
-
-                        block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                    }
                 }
             }
         }
@@ -1218,19 +1182,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                         a_grid_desc_k0_m_k1_container_.push_back(descs[I0]);
                         b_grid_desc_k0_n_k1_container_.push_back(descs[I1]);
                         c_grid_desc_m_n_container_.push_back(descs[I2]);
-
-                        auto block_2_ctile_map =
-                            GridwiseGemm::MakeDefaultBlock2CTileMap(descs[I2], M01_, N01_);
-
-                        if(GridwiseGemm::CheckValidity(
-                               descs[I0], descs[I1], descs[I2], block_2_ctile_map))
-                        {
-                            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_.push_back(
-                                GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(
-                                    descs[I2]));
-
-                            block_2_ctile_map_container_.push_back(block_2_ctile_map);
-                        }
                     }
                 }
             }
@@ -1242,11 +1193,6 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
         std::vector<AGridDesc_K0_M_K1> a_grid_desc_k0_m_k1_container_;
         std::vector<BGridDesc_K0_N_K1> b_grid_desc_k0_n_k1_container_;
         std::vector<CGridDesc_M_N> c_grid_desc_m_n_container_;
-        std::vector<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_;
-        std::vector<typename GridwiseGemm::DefaultBlock2CTileMap> block_2_ctile_map_container_;
-        index_t M01_;
-        index_t N01_;
         OutElementwiseOperation a_element_op_;
         WeiElementwiseOperation b_element_op_;
         InElementwiseOperation c_element_op_;
@@ -1276,123 +1222,84 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
             {
 #if DEBUG_LOG
                 {
-                    std::cout << "arg.a_grid_desc_k0_m_k1_container_{"
+                    std::cout << "arg.a_grid_desc_k0_m_k1{"
                               << arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I0) << ", "
                               << arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I1) << ", "
                               << arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I2) << "}"
                               << std::endl;
 
-                    std::cout << "arg.b_grid_desc_k0_n_k1_container_{"
+                    std::cout << "arg.b_grid_desc_k0_n_k1{"
                               << arg.b_grid_desc_k0_n_k1_container_[i].GetLength(I0) << ", "
                               << arg.b_grid_desc_k0_n_k1_container_[i].GetLength(I1) << ", "
                               << arg.b_grid_desc_k0_n_k1_container_[i].GetLength(I2) << "}"
                               << std::endl;
 
-                    std::cout << "arg.c_grid_desc_m_n_container_{ "
+                    std::cout << "arg.c_grid_desc_m_n{"
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I0) << ", "
                               << arg.c_grid_desc_m_n_container_[i].GetLength(I1) << "}"
                               << std::endl;
-
-                    std::cout << "arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_( "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I0)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I1)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I2)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I3)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I4)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I5)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I6)
-                              << ", "
-                              << arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i].GetLength(I7)
-                              << " ) " << std::endl;
                 }
 #endif
 
                 if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                                 arg.b_grid_desc_k0_n_k1_container_[i],
-                                                arg.c_grid_desc_m_n_container_[i],
-                                                arg.block_2_ctile_map_container_[i]))
+                                                arg.c_grid_desc_m_n_container_[i]))
                 {
                     throw std::runtime_error(
-                        "wrong! GridwiseGemm_km_kn_m0m1n0n1_xdlops_v3r1 has invalid setting");
+                        "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
                 }
 
-                const index_t grid_size = arg.block_2_ctile_map_container_[i].CalculateGridSize(
-                    arg.c_grid_desc_m_n_container_[i]);
+                const auto [gdx, gdy, gdz] =
+                    GridwiseGemm::CalculateGridSize(arg.c_grid_desc_m_n_container_[i]);
 
                 const auto K = arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I0) *
                                arg.a_grid_desc_k0_m_k1_container_[i].GetLength(I2);
 
                 if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        true>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                true>;
 
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
                 else
                 {
-                    const auto kernel = kernel_gemm_xdlops_v2r3<
-                        GridwiseGemm,
-                        ADataType, // TODO: distiguish A/B datatype
-                        CDataType,
-                        remove_reference_t<DeviceOp::AGridDesc_K0_M_K1>,
-                        remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
-                        remove_reference_t<
-                            typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                        OutElementwiseOperation,
-                        WeiElementwiseOperation,
-                        InElementwiseOperation,
-                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                        false>;
+                    const auto kernel =
+                        kernel_gemm_xdlops_v2r3<GridwiseGemm,
+                                                ADataType, // TODO: distiguish A/B datatype
+                                                CDataType,
+                                                DeviceOp::AGridDesc_K0_M_K1,
+                                                DeviceOp::BGridDesc_K0_N_K1,
+                                                DeviceOp::CGridDesc_M_N,
+                                                false>;
 
-                    ave_time += launch_and_time_kernel(
-                        stream_config,
-                        kernel,
-                        dim3(grid_size),
-                        dim3(BlockSize),
-                        0,
-                        arg.p_a_grid_,
-                        arg.p_b_grid_,
-                        arg.p_c_grid_,
-                        arg.a_grid_desc_k0_m_k1_container_[i],
-                        arg.b_grid_desc_k0_n_k1_container_[i],
-                        arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_container_[i],
-                        arg.a_element_op_,
-                        arg.b_element_op_,
-                        arg.c_element_op_,
-                        arg.block_2_ctile_map_container_[i]);
+                    ave_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(gdx, gdy, gdz),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.p_a_grid_,
+                                                       arg.p_b_grid_,
+                                                       arg.p_c_grid_,
+                                                       arg.a_grid_desc_k0_m_k1_container_[i],
+                                                       arg.b_grid_desc_k0_n_k1_container_[i],
+                                                       arg.c_grid_desc_m_n_container_[i]);
                 }
             }
             return ave_time;
@@ -1446,8 +1353,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
         {
             if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_container_[i],
                                             arg.b_grid_desc_k0_n_k1_container_[i],
-                                            arg.c_grid_desc_m_n_container_[i],
-                                            arg.block_2_ctile_map_container_[i]))
+                                            arg.c_grid_desc_m_n_container_[i]))
             {
                 return false;
             }
@@ -1472,10 +1378,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                              std::vector<ck::index_t> conv_filter_strides,
                              std::vector<ck::index_t> conv_filter_dilations,
                              std::vector<ck::index_t> input_left_pads,
-                             std::vector<ck::index_t> input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op)
+                             std::vector<ck::index_t> input_right_pads)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -1489,12 +1392,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
-                        input_right_pads,
-                        1,
-                        1,
-                        in_element_op,
-                        wei_element_op,
-                        out_element_op};
+                        input_right_pads};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -1513,9 +1411,9 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                         std::vector<ck::index_t> conv_filter_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
-                        InElementwiseOperation in_element_op,
-                        WeiElementwiseOperation wei_element_op,
-                        OutElementwiseOperation out_element_op) override
+                        InElementwiseOperation,
+                        WeiElementwiseOperation,
+                        OutElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<InDataType*>(p_in_grid),
                                           static_cast<const WeiDataType*>(p_wei_grid),
@@ -1529,12 +1427,7 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
-                                          input_right_pads,
-                                          1,
-                                          1,
-                                          in_element_op,
-                                          wei_element_op,
-                                          out_element_op);
+                                          input_right_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
index 528d9bf42f..ebbb1d38c0 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
@@ -75,132 +75,20 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
 
     static constexpr auto K1Number = Number<K1>{};
 
-    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
-    {
-        const index_t K0 = K / K1;
-
-        const auto a_grid_desc_m_k = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-
-            return transform_tensor_descriptor(
-                a_grid_desc_m_k,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_right_pad_transform(M, PadM)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                a_grid_desc_m_k,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_pass_through_transform(M)),
-                make_tuple(Sequence<1>{}, Sequence<0>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-    }
-
-    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
-    {
-        const index_t K0 = K / K1;
-
-        const auto b_grid_desc_k_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-            return transform_tensor_descriptor(
-                b_grid_desc_k_n,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-        else
-        {
-            return transform_tensor_descriptor(
-                b_grid_desc_k_n,
-                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
-                           make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-        }
-    }
-
-    static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
-    {
-        const auto c_grid_desc_m_n = [&]() {
-            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
-            }
-            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
-            {
-                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
-            }
-        }();
-
-        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
-        {
-            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
-            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-        else
-        {
-
-            return transform_tensor_descriptor(
-                c_grid_desc_m_n,
-                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-        }
-    }
-
-    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
-    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
-    using CGridDesc_M_N     = decltype(MakeCGridDescriptor_M_N(1, 1, 1));
-
     // GridwiseGemm
-    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<
+    using GridwiseGemm = GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext<
         BlockSize,
         ADataType, // TODO: distinguish A/B datatype
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::Set,
-        AGridDesc_K0_M_K1,
-        BGridDesc_K0_N_K1,
-        CGridDesc_M_N,
+        ALayout,
+        BLayout,
+        CLayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
+        GemmSpec,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -232,173 +120,41 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
         LoopSched,
         PipelineVer>;
 
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(const ADataType* p_a_grid,
-                 const BDataType* p_b_grid,
-                 CDataType* p_c_grid,
-                 index_t M,
-                 index_t N,
-                 index_t K,
-                 index_t StrideA,
-                 index_t StrideB,
-                 index_t StrideC,
-                 index_t M01,
-                 index_t N01,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CElementwiseOperation c_element_op)
-            : p_a_grid_{p_a_grid},
-              p_b_grid_{p_b_grid},
-              p_c_grid_{p_c_grid},
-              a_grid_desc_k0_m_k1_{},
-              b_grid_desc_k0_n_k1_{},
-              c_grid_desc_m_n_{},
-              c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_{},
-              block_2_ctile_map_{},
-              M01_{M01},
-              N01_{N01},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              c_element_op_{c_element_op},
-              kraw_{K}
-        {
-            a_grid_desc_k0_m_k1_ = DeviceGemmXdl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
-            b_grid_desc_k0_n_k1_ = DeviceGemmXdl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
-            c_grid_desc_m_n_     = DeviceGemmXdl::MakeCGridDescriptor_M_N(M, N, StrideC);
-
-            block_2_ctile_map_ =
-                GridwiseGemm::MakeDefaultBlock2CTileMap(c_grid_desc_m_n_, M01, N01);
-
-            if(GridwiseGemm::CheckValidity(a_grid_desc_k0_m_k1_,
-                                           b_grid_desc_k0_n_k1_,
-                                           c_grid_desc_m_n_,
-                                           block_2_ctile_map_))
-            {
-                c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_ =
-                    GridwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n_);
-            }
-        }
-
-        //  private:
-        const ADataType* p_a_grid_;
-        const BDataType* p_b_grid_;
-        CDataType* p_c_grid_;
-        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
-        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
-        CGridDesc_M_N c_grid_desc_m_n_;
-        typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2
-            c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_;
-        typename GridwiseGemm::DefaultBlock2CTileMap block_2_ctile_map_;
-        index_t M01_;
-        index_t N01_;
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CElementwiseOperation c_element_op_;
-        index_t kraw_;
-    };
+    using Argument = typename GridwiseGemm::Argument;
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        using Argument = DeviceGemmXdl::Argument;
-
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
         {
-#if DEBUG_LOG
+            if(stream_config.log_level_ > 0)
             {
-                std::cout << "arg.a_grid_desc_k0_m_k1_{" << arg.a_grid_desc_k0_m_k1_.GetLength(I0)
-                          << ", " << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
-                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.b_grid_desc_k0_n_k1_{" << arg.b_grid_desc_k0_n_k1_.GetLength(I0)
-                          << ", " << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
-                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
-
-                std::cout << "arg.c_grid_desc_m_n_{ " << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
-                          << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+                karg.Print();
             }
-#endif
 
-            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                            arg.b_grid_desc_k0_n_k1_,
-                                            arg.c_grid_desc_m_n_,
-                                            arg.block_2_ctile_map_))
+            if(!GridwiseGemm::CheckValidity(karg))
             {
                 throw std::runtime_error(
-                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext has invalid setting");
             }
 
-            const index_t grid_size =
-                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_);
-
-            const auto K =
-                arg.a_grid_desc_k0_m_k1_.GetLength(I0) * arg.a_grid_desc_k0_m_k1_.GetLength(I2);
+            const auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(karg.M, karg.N);
 
             float ave_time = 0;
 
-            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(karg.K))
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    true>;
+                const auto kernel = kernel_gemm_xdlops_v2r3<GridwiseGemm, true>;
 
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.block_2_ctile_map_);
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
             else
             {
-                const auto kernel = kernel_gemm_xdlops_v2r3<
-                    GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
-                    CDataType,
-                    remove_reference_t<DeviceGemmXdl::AGridDesc_K0_M_K1>,
-                    remove_reference_t<DeviceGemmXdl::BGridDesc_K0_N_K1>,
-                    remove_reference_t<typename GridwiseGemm::CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2>,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CElementwiseOperation,
-                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
-                    false>;
+                const auto kernel = kernel_gemm_xdlops_v2r3<GridwiseGemm, false>;
 
-                ave_time = launch_and_time_kernel(stream_config,
-                                                  kernel,
-                                                  dim3(grid_size),
-                                                  dim3(BlockSize),
-                                                  0,
-                                                  arg.p_a_grid_,
-                                                  arg.p_b_grid_,
-                                                  arg.p_c_grid_,
-                                                  arg.a_grid_desc_k0_m_k1_,
-                                                  arg.b_grid_desc_k0_n_k1_,
-                                                  arg.c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2_,
-                                                  arg.a_element_op_,
-                                                  arg.b_element_op_,
-                                                  arg.c_element_op_,
-                                                  arg.block_2_ctile_map_);
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             }
 
             return ave_time;
@@ -418,7 +174,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
         return true;
     }
 
-    static bool IsSupportedArgument(const Argument& arg)
+    static bool IsSupportedArgument(const Argument& karg)
     {
         if(ck::get_device_name() == "gfx908")
         {
@@ -441,15 +197,12 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
             return false;
         }
 
-        if(arg.kraw_ % K1 != 0)
+        if(karg.K % K1 != 0)
         {
             return false;
         }
 
-        return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
-                                           arg.b_grid_desc_k0_n_k1_,
-                                           arg.c_grid_desc_m_n_,
-                                           arg.block_2_ctile_map_);
+        return GridwiseGemm::CheckValidity(karg);
     }
 
     // polymorphic
@@ -467,24 +220,11 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CElementwiseOperation c_element_op)
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
     {
-        return Argument{p_a,
-                        p_b,
-                        p_c,
-                        M,
-                        N,
-                        K,
-                        StrideA,
-                        StrideB,
-                        StrideC,
-                        1,
-                        1,
-                        a_element_op,
-                        b_element_op,
-                        c_element_op};
+        return Argument{p_a, p_b, p_c, M, N, K, StrideA, StrideB, StrideC};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -499,9 +239,9 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation a_element_op,
-                                                      BElementwiseOperation b_element_op,
-                                                      CElementwiseOperation c_element_op) override
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
                                           static_cast<const BDataType*>(p_b),
@@ -511,12 +251,7 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                                           K,
                                           StrideA,
                                           StrideB,
-                                          StrideC,
-                                          1,
-                                          1,
-                                          a_element_op,
-                                          b_element_op,
-                                          c_element_op);
+                                          StrideC);
     }
 
     // polymorphic
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index 91ed1112f6..5c8b9f419a 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -134,6 +134,14 @@ struct BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
     {
     }
 
+    template <typename CGridDesc_M_N>
+    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
+                                                        index_t M01 = 8)
+        : BlockToCTileMap_M00_N0_M01Adapt(
+              c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1), M01)
+    {
+    }
+
     __host__ static constexpr index_t CalculateGridSize(index_t M, index_t N)
     {
         const auto M0 = math::integer_divide_ceil(M, MPerBlock);
@@ -142,6 +150,18 @@ struct BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
         return M0 * N0;
     }
 
+    template <typename CGridDesc_M_N>
+    __host__ static constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        return CalculateGridSize(c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1));
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const
+    {
+        return true;
+    }
+
     template <typename TopIdx>
     __host__ __device__ constexpr auto CalculateBottomIndex(const TopIdx& idx_top) const
     {
@@ -222,30 +242,12 @@ struct BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
     index_t M01_;
 };
 
+// keep the redundant type argument for backward compatibility
 template <index_t MPerBlock, index_t NPerBlock, typename CGridDesc_M_N>
 struct BlockToCTileMap_M00_N0_M01Adapt : BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>
 {
-    using Parent = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>;
-
-    using Parent::I0;
-    using Parent::I1;
-
-    using Parent::Parent;
-    using Parent::operator=;
-
-    __host__ __device__ BlockToCTileMap_M00_N0_M01Adapt(const CGridDesc_M_N& c_grid_desc_m_n,
-                                                        index_t M01 = 8)
-        : Parent(c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1), M01)
-    {
-    }
-
-    __host__ static constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
-    {
-        return Parent::CalculateGridSize(c_grid_desc_m_n.GetLength(I0),
-                                         c_grid_desc_m_n.GetLength(I1));
-    }
-
-    __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const { return true; }
+    using BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, void>::
+        BlockToCTileMap_M00_N0_M01Adapt;
 };
 
 // 2D slices of column-vectors in 3D space
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index f4504a9402..bd18fdb10d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -7,6 +7,7 @@
 #include "ck/tensor_description/multi_index_transform_helper.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
@@ -21,27 +22,18 @@ template <typename GridwiseGemm,
           typename FloatC,
           typename AGridDesc_K0_M_K1,
           typename BGridDesc_K0_N_K1,
-          typename CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CElementwiseOperation,
-          typename Block2CTileMap,
+          typename CGridDesc_M_N,
           bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r3(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
-            FloatC* __restrict__ p_c_grid,
-            const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
-            const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
-            const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CElementwiseOperation c_element_op,
-            const Block2CTileMap block_2_ctile_map)
+        kernel_gemm_xdlops_v2r3(const FloatAB* __restrict__ p_a_grid,
+                                const FloatAB* __restrict__ p_b_grid,
+                                FloatC* __restrict__ p_c_grid,
+                                const AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1,
+                                const BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1,
+                                const CGridDesc_M_N c_grid_desc_m_n)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__))
@@ -53,22 +45,46 @@ __global__ void
                                                   p_shared,
                                                   a_grid_desc_k0_m_k1,
                                                   b_grid_desc_k0_n_k1,
-                                                  c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  c_element_op,
-                                                  block_2_ctile_map);
+                                                  c_grid_desc_m_n);
 #else
     ignore                = p_a_grid;
     ignore                = p_b_grid;
     ignore                = p_c_grid;
     ignore                = a_grid_desc_k0_m_k1;
     ignore                = b_grid_desc_k0_n_k1;
-    ignore                = c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2;
-    ignore                = a_element_op;
-    ignore                = b_element_op;
-    ignore                = c_element_op;
-    ignore                = block_2_ctile_map;
+    ignore                = c_grid_desc_m_n;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <typename GridwiseGemm, bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_xdlops_v2r3(const typename GridwiseGemm::Argument karg)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    const auto a_grid_desc_k0_m_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeAGridDescriptor_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.K0, karg.StrideA));
+    const auto b_grid_desc_k0_n_k1 =
+        amd_wave_read_first_lane(GridwiseGemm::MakeBGridDescriptor_K0_N_K1(
+            karg.K, karg.N, karg.NPadded, karg.K0, karg.StrideB));
+    const auto c_grid_desc_m_n = amd_wave_read_first_lane(GridwiseGemm::MakeCGridDescriptor_M_N(
+        karg.M, karg.MPadded, karg.N, karg.NPadded, karg.StrideC));
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(karg.p_a_grid,
+                                                  karg.p_b_grid,
+                                                  karg.p_c_grid,
+                                                  p_shared,
+                                                  a_grid_desc_k0_m_k1,
+                                                  b_grid_desc_k0_n_k1,
+                                                  c_grid_desc_m_n);
+#else
+    ignore                = karg;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -77,9 +93,6 @@ template <index_t BlockSize,
           typename FloatAcc,
           typename FloatC,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename AGridDesc_K0_M_K1,
-          typename BGridDesc_K0_N_K1,
-          typename CGridDesc_M_N,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
@@ -129,6 +142,105 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ static auto CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(c_grid_desc_m_n), 1, 1);
+    }
+
+    template <typename>
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    __host__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_divide_ceil(M, MPerBlock) * MPerBlock;
+    }
+
+    __host__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_divide_ceil(N, NPerBlock) * NPerBlock;
+    }
+
+    __host__ static auto CalculateK0(index_t K) { return math::integer_divide_floor(K, K1Value); }
+
+    // Argument
+    struct Problem
+    {
+        __host__ Problem(index_t M_,
+                         index_t N_,
+                         index_t K_,
+                         index_t StrideA_,
+                         index_t StrideB_,
+                         index_t StrideC_)
+            : M{M_},
+              N{N_},
+              K{K_},
+              StrideA{StrideA_},
+              StrideB{StrideB_},
+              StrideC{StrideC_},
+              MPadded{CalculateMPadded(M_)},
+              NPadded{CalculateNPadded(N_)},
+              K0{CalculateK0(K)}
+        {
+        }
+
+        __host__ void Print() const
+        {
+            std::cout << "problem {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "K0:" << K0 << "}" << std::endl;
+        }
+
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t K0;
+    };
+
+    // Argument
+    struct Argument : public Problem, public tensor_operation::device::BaseArgument
+    {
+        __host__ Argument(const FloatAB* p_a_grid_,
+                          const FloatAB* p_b_grid_,
+                          FloatC* p_c_grid_,
+                          index_t M_,
+                          index_t N_,
+                          index_t K_,
+                          index_t StrideA_,
+                          index_t StrideB_,
+                          index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_}
+        {
+        }
+
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+    };
+
     using GridwiseGemmPipe = remove_cvref_t<decltype(
         GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
@@ -204,13 +316,11 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         return (a_block_space_size_aligned + b_block_space_size_aligned) * sizeof(FloatAB);
     }
 
-    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
-    template <typename Block2CTileMap>
+    template <typename AGridDesc_K0_M_K1, typename BGridDesc_K0_N_K1, typename CGridDesc_M_N>
     __host__ __device__ static constexpr bool
     CheckValidity(const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
                   const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
-                  const CGridDesc_M_N& c_grid_desc_m_n,
-                  const Block2CTileMap& block_2_ctile_map)
+                  const CGridDesc_M_N& c_grid_desc_m_n)
     {
         static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
                       "wrong! K1 need to be known at compile-time");
@@ -239,7 +349,24 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
             return false;
         }
 
-        if(!block_2_ctile_map.CheckValidity(c_grid_desc_m_n))
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        return true;
+    }
+
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
+                      "wrong! K1 need to be known at compile-time");
+
+        static_assert((MPerBlock % (MPerXDL * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXDL)) == 0,
+                      "Invalid tuning param!");
+
+        // check gridwise gemm pipeline
+        const index_t K0      = problem.K / K1Value;
+        const auto num_k_loop = K0 / K0PerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {
             return false;
         }
@@ -248,15 +375,16 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         return true;
     }
 
-    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    __host__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
     {
         const index_t num_loop = K / (K0PerBlock * K1);
 
         return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
 
+    template <typename CGridDesc>
     __host__ __device__ static constexpr auto
-    MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(const CGridDesc_M_N& c_grid_desc_m_n)
+    MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(const CGridDesc& c_grid_desc_m_n)
     {
         constexpr auto max_lds_align = K1;
 
@@ -306,31 +434,23 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
     }
 
     // return block_id to C matrix tile idx (m0, n0) mapping
-    __host__ __device__ static constexpr auto MakeDefaultBlock2CTileMap(
-        const CGridDesc_M_N& c_grid_desc_m_n, index_t /* M01 */, index_t /* N01 */)
-    {
-        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>(
-            c_grid_desc_m_n);
-    }
+    using Block2CTileMap = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>;
 
-    using CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2 =
-        decltype(MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(CGridDesc_M_N{}));
-    using DefaultBlock2CTileMap = decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}, 1, 1));
-
-    template <bool HasMainKBlockLoop, typename Block2CTileMap = DefaultBlock2CTileMap>
-    __device__ static void
-    Run(const FloatAB* __restrict__ p_a_grid,
-        const FloatAB* __restrict__ p_b_grid,
-        FloatC* __restrict__ p_c_grid,
-        void* __restrict__ p_shared,
-        const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
-        const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
-        const CGridDesc_M0_N0_M1_N1_M2_M3_M4_N2& c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2,
-        const AElementwiseOperation& a_element_op,
-        const BElementwiseOperation& b_element_op,
-        const CElementwiseOperation& c_element_op,
-        const Block2CTileMap& block_2_ctile_map)
+    template <bool HasMainKBlockLoop,
+              typename AGridDesc_K0_M_K1,
+              typename BGridDesc_K0_N_K1,
+              typename CGridDesc_M_N>
+    __device__ static void Run(const FloatAB* p_a_grid,
+                               const FloatAB* p_b_grid,
+                               FloatC* p_c_grid,
+                               void* __restrict__ p_shared,
+                               const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                               const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                               const CGridDesc_M_N& c_grid_desc_m_n)
     {
+        const auto c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+            MakeCGridDescriptor_M0_N0_M1_N1_M2_M3_M4_N2(c_grid_desc_m_n);
+
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_grid_desc_k0_m_k1.GetElementSpaceSize());
         const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
@@ -338,7 +458,12 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_c_grid, c_grid_desc_m0_n0_m1_n1_m2_m3_m4_n2.GetElementSpaceSize());
 
-        const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
+        const AElementwiseOperation a_element_op{};
+        const BElementwiseOperation b_element_op{};
+        const CElementwiseOperation c_element_op{};
+
+        const auto block_2_ctile_map =
+            Block2CTileMap{c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1)};
 
         // divide block work by [M, N]
         const auto block_work_idx =
@@ -467,6 +592,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         constexpr auto b_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
 
         // gridwise GEMM pipeline
+        const auto K0                       = a_grid_desc_k0_m_k1.GetLength(I0);
         const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(K0 / K0PerBlock);
 
         GridwiseGemmPipe::template Run<HasMainKBlockLoop>(a_grid_desc_k0_m_k1,
@@ -565,4 +691,309 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
     }
 };
 
+template <index_t BlockSize,
+          typename FloatAB,
+          typename FloatAcc,
+          typename FloatC,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t K1Value,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_K1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_K1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          bool BBlockLdsExtraN,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          index_t NumGemmKPrefetchStage = 1,
+          LoopScheduler LoopSched       = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer   = PipelineVersion::v1>
+struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext
+    : GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
+                                              FloatAB,
+                                              FloatAcc,
+                                              FloatC,
+                                              CGlobalMemoryDataOperation,
+                                              AElementwiseOperation,
+                                              BElementwiseOperation,
+                                              CElementwiseOperation,
+                                              MPerBlock,
+                                              NPerBlock,
+                                              K0PerBlock,
+                                              MPerXDL,
+                                              NPerXDL,
+                                              K1Value,
+                                              MXdlPerWave,
+                                              NXdlPerWave,
+                                              ABlockTransferThreadClusterLengths_K0_M_K1,
+                                              ABlockTransferThreadClusterArrangeOrder,
+                                              ABlockTransferSrcAccessOrder,
+                                              ABlockTransferSrcVectorDim,
+                                              ABlockTransferSrcScalarPerVector,
+                                              ABlockTransferDstScalarPerVector_K1,
+                                              AThreadTransferSrcResetCoordinateAfterRun,
+                                              ABlockLdsExtraM,
+                                              BBlockTransferThreadClusterLengths_K0_N_K1,
+                                              BBlockTransferThreadClusterArrangeOrder,
+                                              BBlockTransferSrcAccessOrder,
+                                              BBlockTransferSrcVectorDim,
+                                              BBlockTransferSrcScalarPerVector,
+                                              BBlockTransferDstScalarPerVector_K1,
+                                              BThreadTransferSrcResetCoordinateAfterRun,
+                                              BBlockLdsExtraN,
+                                              CThreadTransferSrcDstAccessOrder,
+                                              CThreadTransferSrcDstVectorDim,
+                                              CThreadTransferDstScalarPerVector,
+                                              NumGemmKPrefetchStage,
+                                              LoopSched,
+                                              PipelineVer>
+{
+    using Parent =
+        GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3<BlockSize,
+                                                FloatAB,
+                                                FloatAcc,
+                                                FloatC,
+                                                CGlobalMemoryDataOperation,
+                                                AElementwiseOperation,
+                                                BElementwiseOperation,
+                                                CElementwiseOperation,
+                                                MPerBlock,
+                                                NPerBlock,
+                                                K0PerBlock,
+                                                MPerXDL,
+                                                NPerXDL,
+                                                K1Value,
+                                                MXdlPerWave,
+                                                NXdlPerWave,
+                                                ABlockTransferThreadClusterLengths_K0_M_K1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ABlockTransferSrcAccessOrder,
+                                                ABlockTransferSrcVectorDim,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_K1,
+                                                AThreadTransferSrcResetCoordinateAfterRun,
+                                                ABlockLdsExtraM,
+                                                BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                BBlockTransferSrcAccessOrder,
+                                                BBlockTransferSrcVectorDim,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_K1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                BBlockLdsExtraN,
+                                                CThreadTransferSrcDstAccessOrder,
+                                                CThreadTransferSrcDstVectorDim,
+                                                CThreadTransferDstScalarPerVector,
+                                                NumGemmKPrefetchStage,
+                                                LoopSched,
+                                                PipelineVer>;
+
+    using typename Parent::GridwiseGemmPipe;
+    using typename Parent::Problem;
+
+    using Parent::I1;
+
+    using Parent::K1;
+
+    __device__ static auto
+    MakeAGridDescriptor_K0_M_K1(index_t M, index_t MPad, index_t K, index_t K0, index_t StrideA)
+    {
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    __device__ static auto
+    MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t NPad, index_t K0, index_t StrideB)
+    {
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Value)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    __device__ static auto
+    MakeCGridDescriptor_M_N(index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        {
+            return transform_tensor_descriptor(c_grid_desc_m_n,
+                                               make_tuple(make_right_pad_transform(M, MPad - M),
+                                                          make_right_pad_transform(N, NPad - N)),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(K1)>>::value,
+                      "wrong! K1 need to be known at compile-time");
+
+        static_assert((MPerBlock % (MPerXDL * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXDL)) == 0,
+                      "Invalid tuning param!");
+
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.M % MPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.N % NPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(problem.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(problem.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        // check gridwise gemm pipeline
+        const index_t K0      = problem.K / K1;
+        const auto num_k_loop = K0 / K0PerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        return true;
+    }
+};
+
 } // namespace ck

From e2ebc8e79599531e2b837c27b8aaa0404c5c58ce Mon Sep 17 00:00:00 2001
From: who who who <fsx950223@outlook.com>
Date: Fri, 2 Jun 2023 05:23:41 +0800
Subject: [PATCH 026/194] replace hipMemcpy with hipMemcpyWithStream (#734)

---
 ...grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  9 +++++----
 ...ce_grouped_contraction_multiple_d_xdl_cshuffle.hpp | 11 ++++++-----
 .../device/impl/device_grouped_gemm_multiple_d_dl.hpp |  9 +++++----
 .../gpu/device/impl/device_grouped_gemm_xdl.hpp       |  9 +++++----
 .../impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp  |  9 +++++----
 5 files changed, 26 insertions(+), 21 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 9c8b5f462e..a2940fa755 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -611,10 +611,11 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
                 some_has_main_k_block_loop |= y;
             }
 
-            hipGetErrorString(hipMemcpy(arg.p_workspace_,
-                                        arg.group_kernel_args_.data(),
-                                        arg.group_kernel_args_.size() * sizeof(GroupKernelArg),
-                                        hipMemcpyHostToDevice));
+            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
+                                                  arg.group_kernel_args_.data(),
+                                                  arg.group_kernel_args_.size() * sizeof(GroupKernelArg),
+                                                  hipMemcpyHostToDevice,
+                                                  stream_config.stream_id_));
 
             float ave_time = 0;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index a275ee1023..efe3a69ac6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -652,11 +652,12 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
                 }
             }
 
-            hipGetErrorString(hipMemcpy(arg.p_workspace_,
-                                        arg.contraction_multi_d_kernel_args_.data(),
-                                        arg.contraction_multi_d_kernel_args_.size() *
-                                            sizeof(ContractionMultiDKernelArg),
-                                        hipMemcpyHostToDevice));
+            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
+                                                  arg.contraction_multi_d_kernel_args_.data(),
+                                                  arg.contraction_multi_d_kernel_args_.size() *
+                                                  sizeof(ContractionMultiDKernelArg),
+                                                  hipMemcpyHostToDevice,
+                                                  stream_config.stream_id_));
 
             float ave_time = 0;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
index d1f1b7fcc9..2a3e2b6cf2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -597,10 +597,11 @@ struct DeviceGroupedGemmMultipleD_Dl : public DeviceGroupedGemm<ALayout,
                 }
             }
 
-            hipGetErrorString(hipMemcpy(arg.p_workspace_,
-                                        arg.gemm_desc_kernel_arg_.data(),
-                                        arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
-                                        hipMemcpyHostToDevice));
+            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
+                                                  arg.gemm_desc_kernel_arg_.data(),
+                                                  arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
+                                                  hipMemcpyHostToDevice,
+                                                  stream_config.stream_id_));
 
             auto launch_kernel = [&](auto has_main_k_block_loop,
                                      auto has_double_tail_k_block_loop) {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index 77a8e6ecbc..e5bf395be9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -549,10 +549,11 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
             }
 
             hipGetErrorString(
-                hipMemcpy(arg.p_workspace_,
-                          arg.gemm_desc_kernel_arg_.data(),
-                          arg.gemm_desc_kernel_arg_.size() * sizeof(GemmBiasTransKernelArg),
-                          hipMemcpyHostToDevice));
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.gemm_desc_kernel_arg_.data(),
+                                    arg.gemm_desc_kernel_arg_.size() * sizeof(GemmBiasTransKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
 
             float ave_time = 0;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index cd39cc983b..ebeea71db6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -406,10 +406,11 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 }
             }
 
-            hip_check_error(hipMemcpy(arg.p_workspace_,
-                                      arg.gemm_kernel_args_.data(),
-                                      arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
-                                      hipMemcpyHostToDevice));
+            hip_check_error(hipMemcpyWithStream(arg.p_workspace_,
+                                                arg.gemm_kernel_args_.data(),
+                                                arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                                                hipMemcpyHostToDevice,
+                                                stream_config.stream_id_));
 
             float ave_time = 0;
 

From 403659040105e6877cba1d454ccfd85bece7fc09 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 2 Jun 2023 14:10:02 -0700
Subject: [PATCH 027/194] fix clang format (#740)

---
 ...rouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp | 11 ++++++-----
 ...e_grouped_contraction_multiple_d_xdl_cshuffle.hpp |  2 +-
 .../impl/device_grouped_gemm_multiple_d_dl.hpp       | 11 ++++++-----
 .../gpu/device/impl/device_grouped_gemm_xdl.hpp      | 12 ++++++------
 .../impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp | 11 ++++++-----
 5 files changed, 25 insertions(+), 22 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index a2940fa755..30e29cc8e2 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -611,11 +611,12 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
                 some_has_main_k_block_loop |= y;
             }
 
-            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
-                                                  arg.group_kernel_args_.data(),
-                                                  arg.group_kernel_args_.size() * sizeof(GroupKernelArg),
-                                                  hipMemcpyHostToDevice,
-                                                  stream_config.stream_id_));
+            hipGetErrorString(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.group_kernel_args_.data(),
+                                    arg.group_kernel_args_.size() * sizeof(GroupKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
 
             float ave_time = 0;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index efe3a69ac6..5775ff3971 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -655,7 +655,7 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
             hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
                                                   arg.contraction_multi_d_kernel_args_.data(),
                                                   arg.contraction_multi_d_kernel_args_.size() *
-                                                  sizeof(ContractionMultiDKernelArg),
+                                                      sizeof(ContractionMultiDKernelArg),
                                                   hipMemcpyHostToDevice,
                                                   stream_config.stream_id_));
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
index 2a3e2b6cf2..22be58259a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -597,11 +597,12 @@ struct DeviceGroupedGemmMultipleD_Dl : public DeviceGroupedGemm<ALayout,
                 }
             }
 
-            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
-                                                  arg.gemm_desc_kernel_arg_.data(),
-                                                  arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
-                                                  hipMemcpyHostToDevice,
-                                                  stream_config.stream_id_));
+            hipGetErrorString(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.gemm_desc_kernel_arg_.data(),
+                                    arg.gemm_desc_kernel_arg_.size() * sizeof(GemmKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
 
             auto launch_kernel = [&](auto has_main_k_block_loop,
                                      auto has_double_tail_k_block_loop) {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index e5bf395be9..3900047567 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -548,12 +548,12 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
                 }
             }
 
-            hipGetErrorString(
-                hipMemcpyWithStream(arg.p_workspace_,
-                                    arg.gemm_desc_kernel_arg_.data(),
-                                    arg.gemm_desc_kernel_arg_.size() * sizeof(GemmBiasTransKernelArg),
-                                    hipMemcpyHostToDevice,
-                                    stream_config.stream_id_));
+            hipGetErrorString(hipMemcpyWithStream(arg.p_workspace_,
+                                                  arg.gemm_desc_kernel_arg_.data(),
+                                                  arg.gemm_desc_kernel_arg_.size() *
+                                                      sizeof(GemmBiasTransKernelArg),
+                                                  hipMemcpyHostToDevice,
+                                                  stream_config.stream_id_));
 
             float ave_time = 0;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index ebeea71db6..1ac9969f86 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -406,11 +406,12 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 }
             }
 
-            hip_check_error(hipMemcpyWithStream(arg.p_workspace_,
-                                                arg.gemm_kernel_args_.data(),
-                                                arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
-                                                hipMemcpyHostToDevice,
-                                                stream_config.stream_id_));
+            hip_check_error(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.gemm_kernel_args_.data(),
+                                    arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
 
             float ave_time = 0;
 

From 1dd455d6337cf7ed3ec6cb82a1618701a5c17351 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 7 Jun 2023 09:35:14 -0700
Subject: [PATCH 028/194] Update docker (#744)

* update dockerfile to build rocm5.6 rc3

* fix couple of docker issues
---
 Dockerfile | 28 +++++++++++++++-------------
 1 file changed, 15 insertions(+), 13 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index 8e6ddb1eba..710db05c06 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -1,5 +1,5 @@
 FROM ubuntu:20.04
-
+ARG DEBIAN_FRONTEND=noninteractive
 ARG ROCMVERSION=5.6
 ARG compiler_version=""
 ARG compiler_commit=""
@@ -9,23 +9,30 @@ RUN set -xe
 ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
 RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 # Add rocm repository
+RUN chmod 1777 /tmp
 RUN apt-get update
-RUN apt-get install -y wget gnupg curl
-RUN --mount=type=ssh if [ "$ROCMVERSION" != "5.6"]; then \
-	wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
+RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
+RUN --mount=type=ssh if [ "$ROCMVERSION" != "5.6" ]; then \
+        wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
         sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"; \
-    else sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
+    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "" ]; then \
+         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
          apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914 && \
-         DEBIAN_FRONTEND=noninteractive amdgpu-install -y --usecase=rocm ; \
+         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914; \
+    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc3" ]; then \
+         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.6-20.04-1_all.deb" && \
+         apt update && apt-get install -y ./amdgpu-install-internal_5.6-20.04-1_all.deb && \
+         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-45  > /etc/apt/sources.list.d/rocm-build.list' && \
+         amdgpu-repo --amdgpu-build=1602498; \
     fi
+RUN amdgpu-install -y --usecase=rocm --no-dkms
+
 RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
 RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
-    apt-utils \
     build-essential \
     ccache \
     cmake \
@@ -38,16 +45,11 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     libpthread-stubs0-dev \
     llvm-amdgpu \
     pkg-config \
-    python \
     python3 \
-    python-dev \
     python3-dev \
     python3-pip \
     sshpass \
     software-properties-common \
-    rocm-dev \
-    rocm-device-libs \
-    rocm-cmake \
     vim \
     nano \
     zlib1g-dev \

From 016ebaa7f33d2c3e86cd617210bd636fe7c99b42 Mon Sep 17 00:00:00 2001
From: carlushuang <carlus.huang@amd.com>
Date: Thu, 8 Jun 2023 20:40:29 +0800
Subject: [PATCH 029/194] support dynamic buffer using memory coherence glc_slc
 bit from template (#725)

---
 include/ck/utility/amd_buffer_addressing.hpp | 284 ++++++++++++-------
 include/ck/utility/dynamic_buffer.hpp        |  23 +-
 2 files changed, 201 insertions(+), 106 deletions(-)

diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index ef3d2032c5..38ee76d883 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -286,7 +286,22 @@ llvm_amdgcn_raw_buffer_atomic_max_fp64(double vdata,
                                        int soffset,    // dst_wave_addr_offset
                                        int glc_slc) __asm("llvm.amdgcn.raw.buffer.atomic.fmax.f64");
 
-template <typename T, index_t N>
+// memory coherency bit for buffer store/load instruction
+// check ISA manual for each GFX target
+// e.g. for
+// https://www.amd.com/system/files/TechDocs/instinct-mi200-cdna2-instruction-set-architecture.pdf,
+// page 67~68
+enum struct AmdBufferCoherenceEnum
+{
+    DefaultCoherence = 0, // default value
+    GLC              = 1,
+    SLC              = 2,
+    GLC_SLC          = 3,
+};
+
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_wave_buffer_resource,
                                                                  index_t src_thread_addr_offset,
                                                                  index_t src_wave_addr_offset)
@@ -305,28 +320,37 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
         // use fp32 load to mimic fp64 load
         if constexpr(N == 1)
         {
-            const float2_t tmp = llvm_amdgcn_raw_buffer_load_fp32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            const float2_t tmp =
+                llvm_amdgcn_raw_buffer_load_fp32x2(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             return bit_cast<double>(tmp);
         }
         else if constexpr(N == 2)
         {
-            const float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            const float4_t tmp =
+                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             return bit_cast<double2_t>(tmp);
         }
         else if constexpr(N == 4)
         {
-            const float4_t f32_0 = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            const float4_t f32_0 =
+                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             const float4_t f32_1 =
                 llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
                                                    src_thread_addr_offset,
                                                    src_wave_addr_offset + 4 * sizeof(float),
-                                                   0);
+                                                   static_cast<index_t>(coherence));
             vector_type<double, 4> tmp;
 
             tmp.AsType<double2_t>()(Number<0>{}) = bit_cast<double2_t>(f32_0);
@@ -339,31 +363,40 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_fp32(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp32(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_fp32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp32x2(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
             vector_type<float, 8> tmp;
 
-            tmp.AsType<float4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<float4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
 
             tmp.AsType<float4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
                                                    src_thread_addr_offset,
                                                    src_wave_addr_offset + 4 * sizeof(float),
-                                                   0);
+                                                   static_cast<index_t>(coherence));
 
             return tmp.AsType<float8_t>()(Number<0>{});
         }
@@ -372,24 +405,32 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_fp16(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp16(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_fp16x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp16x2(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_fp16x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_fp16x4(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
             // use fp32 load to mimic fp16 load
-            float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<half8_t>(tmp);
         }
@@ -398,23 +439,31 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_i16(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_i16x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i16x2(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_i16x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i16x4(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
-            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<bhalf8_t>(tmp);
         }
@@ -423,31 +472,40 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_i32(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
+                                                   src_thread_addr_offset,
+                                                   src_wave_addr_offset,
+                                                   static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
-            return llvm_amdgcn_raw_buffer_load_i32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
-            return llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                     src_thread_addr_offset,
+                                                     src_wave_addr_offset,
+                                                     static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
             vector_type<int32_t, 8> tmp;
 
-            tmp.AsType<int32x4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<int32x4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                  src_thread_addr_offset,
+                                                  src_wave_addr_offset,
+                                                  static_cast<index_t>(coherence));
 
             tmp.AsType<int32x4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
                                                   src_thread_addr_offset,
                                                   src_wave_addr_offset + 4 * sizeof(int32_t),
-                                                  0);
+                                                  static_cast<index_t>(coherence));
             return tmp.AsType<int32x8_t>()(Number<0>{});
         }
     }
@@ -455,17 +513,23 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     {
         if constexpr(N == 1)
         {
-            return llvm_amdgcn_raw_buffer_load_i8(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i8(src_wave_buffer_resource,
+                                                  src_thread_addr_offset,
+                                                  src_wave_addr_offset,
+                                                  static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            return llvm_amdgcn_raw_buffer_load_i8x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i8x2(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
 #else
-            int16_t tmp = llvm_amdgcn_raw_buffer_load_i16(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int16_t tmp = llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
+                                                          src_thread_addr_offset,
+                                                          src_wave_addr_offset,
+                                                          static_cast<index_t>(coherence));
 
             return bit_cast<int8x2_t>(tmp);
 #endif
@@ -473,11 +537,15 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
         else if constexpr(N == 4)
         {
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            return llvm_amdgcn_raw_buffer_load_i8x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            return llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
+                                                    src_thread_addr_offset,
+                                                    src_wave_addr_offset,
+                                                    static_cast<index_t>(coherence));
 #else
-            int32_t tmp = llvm_amdgcn_raw_buffer_load_i32(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32_t tmp = llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
+                                                          src_thread_addr_offset,
+                                                          src_wave_addr_offset,
+                                                          static_cast<index_t>(coherence));
 
             return bit_cast<int8x4_t>(tmp);
 #endif
@@ -487,19 +555,24 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
             vector_type<int8_t, 8> tmp;
 
-            tmp.AsType<int8x4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_i8x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<int8x4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
+                                                 src_thread_addr_offset,
+                                                 src_wave_addr_offset,
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 4 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             return tmp.AsType<int8x8_t>()(Number<0>{});
 #else
-            int32x2_t tmp = llvm_amdgcn_raw_buffer_load_i32x2(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32x2_t tmp = llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<int8x8_t>(tmp);
 #endif
@@ -509,31 +582,36 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
 #if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
             vector_type<int8_t, 16> tmp;
 
-            tmp.AsType<int8x4_t>()(Number<0>{}) = llvm_amdgcn_raw_buffer_load_i8x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            tmp.AsType<int8x4_t>()(Number<0>{}) =
+                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
+                                                 src_thread_addr_offset,
+                                                 src_wave_addr_offset,
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<1>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 4 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<2>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 8 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             tmp.AsType<int8x4_t>()(Number<3>{}) =
                 llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
                                                  src_thread_addr_offset,
                                                  src_wave_addr_offset + 12 * sizeof(int8_t),
-                                                 0);
+                                                 static_cast<index_t>(coherence));
 
             return tmp.AsType<int8x16_t>()(Number<0>{});
 #else
-            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(
-                src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset, 0);
+            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                              src_thread_addr_offset,
+                                                              src_wave_addr_offset,
+                                                              static_cast<index_t>(coherence));
 
             return bit_cast<int8x16_t>(tmp);
 #endif
@@ -541,7 +619,9 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
     }
 }
 
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src_thread_data,
                                       int32x4_t dst_wave_buffer_resource,
                                       index_t dst_thread_addr_offset,
@@ -565,7 +645,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -573,7 +653,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, float>::value)
@@ -584,7 +664,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -592,7 +672,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -600,7 +680,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, half_t>::value)
@@ -611,7 +691,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -619,7 +699,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -627,7 +707,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
@@ -638,19 +718,19 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
 
             llvm_amdgcn_raw_buffer_store_fp16x4(tmp.AsType<half4_t>()[Number<1>{}],
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset + 4 * sizeof(half_t),
-                                                0);
+                                                static_cast<index_t>(coherence));
 #else
             llvm_amdgcn_raw_buffer_store_fp32x4(bit_cast<float4_t>(src_thread_data),
                                                 dst_wave_buffer_resource,
                                                 dst_thread_addr_offset,
                                                 dst_wave_addr_offset,
-                                                0);
+                                                static_cast<index_t>(coherence));
 #endif
         }
     }
@@ -662,7 +742,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -670,7 +750,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -678,7 +758,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 8)
         {
@@ -688,13 +768,13 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
 
             llvm_amdgcn_raw_buffer_store_i16x4(tmp.AsType<bhalf4_t>()[Number<1>{}],
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset + 4 * sizeof(bhalf_t),
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, int32_t>::value)
@@ -705,7 +785,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -713,7 +793,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 4)
         {
@@ -721,7 +801,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
     }
     else if constexpr(is_same<T, int8_t>::value)
@@ -732,7 +812,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                             dst_wave_buffer_resource,
                                             dst_thread_addr_offset,
                                             dst_wave_addr_offset,
-                                            0);
+                                            static_cast<index_t>(coherence));
         }
         else if constexpr(N == 2)
         {
@@ -741,13 +821,13 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
 #else
             llvm_amdgcn_raw_buffer_store_i16(bit_cast<int16_t>(src_thread_data),
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
 #endif
         }
         else if constexpr(N == 4)
@@ -757,13 +837,13 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                               dst_wave_buffer_resource,
                                               dst_thread_addr_offset,
                                               dst_wave_addr_offset,
-                                              0);
+                                              static_cast<index_t>(coherence));
 #else
             llvm_amdgcn_raw_buffer_store_i32(bit_cast<int32_t>(src_thread_data),
                                              dst_wave_buffer_resource,
                                              dst_thread_addr_offset,
                                              dst_wave_addr_offset,
-                                             0);
+                                             static_cast<index_t>(coherence));
 #endif
         }
         else if constexpr(N == 8)
@@ -772,7 +852,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
         else if constexpr(N == 16)
         {
@@ -780,7 +860,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                dst_wave_buffer_resource,
                                                dst_thread_addr_offset,
                                                dst_wave_addr_offset,
-                                               0);
+                                               static_cast<index_t>(coherence));
         }
     }
 }
@@ -1012,7 +1092,9 @@ __device__ void amd_buffer_atomic_max_impl(const typename vector_type<T, N>::typ
 //   1) p_src_wave must point to global memory space
 //   2) p_src_wave must be a wavewise pointer.
 // It is user's responsibility to make sure that is true.
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ typename vector_type_maker<T, N>::type::type
 amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
                                             index_t src_thread_element_offset,
@@ -1032,10 +1114,10 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 #if CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK
     uint32_t src_addr_shift = src_thread_element_valid ? 0 : 0x80000000;
 
-    return amd_buffer_load_impl<scalar_t, vector_size>(
+    return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
         src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
 #else
-    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size>(
+    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
         src_wave_buffer_resource, src_thread_addr_offset, 0);
 
     return src_thread_element_valid ? tmp : vector_t(0);
@@ -1046,7 +1128,9 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 //   1) p_src_wave must point to global memory space
 //   2) p_src_wave must be a wavewise pointer.
 // It is user's responsibility to make sure that is true.
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ typename vector_type_maker<T, N>::type::type
 amd_buffer_load_invalid_element_return_customized_value(const T* p_src_wave,
                                                         index_t src_thread_element_offset,
@@ -1064,7 +1148,7 @@ amd_buffer_load_invalid_element_return_customized_value(const T* p_src_wave,
 
     constexpr index_t vector_size = scalar_type<vector_t>::vector_size;
 
-    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size>(
+    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
         src_wave_buffer_resource, src_thread_addr_offset, 0);
 
     return src_thread_element_valid ? tmp : vector_t(customized_value);
@@ -1074,7 +1158,9 @@ amd_buffer_load_invalid_element_return_customized_value(const T* p_src_wave,
 //   1) p_dst_wave must point to global memory
 //   2) p_dst_wave must be a wavewise pointer.
 // It is user's responsibility to make sure that is true.
-template <typename T, index_t N>
+template <typename T,
+          index_t N,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::type src_thread_data,
                                  T* p_dst_wave,
                                  const index_t dst_thread_element_offset,
@@ -1093,12 +1179,12 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
 #if CK_EXPERIMENTAL_USE_BUFFER_STORE_OOB_CHECK_OFFSET_TRICK
     uint32_t dst_addr_shift = dst_thread_element_valid ? 0 : 0x80000000;
 
-    amd_buffer_store_impl<scalar_t, vector_size>(
+    amd_buffer_store_impl<scalar_t, vector_size, coherence>(
         src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
 #else
     if(dst_thread_element_valid)
     {
-        amd_buffer_store_impl<scalar_t, vector_size>(
+        amd_buffer_store_impl<scalar_t, vector_size, coherence>(
             src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
     }
 #endif
diff --git a/include/ck/utility/dynamic_buffer.hpp b/include/ck/utility/dynamic_buffer.hpp
index 9ea0d6c001..02d61f34ed 100644
--- a/include/ck/utility/dynamic_buffer.hpp
+++ b/include/ck/utility/dynamic_buffer.hpp
@@ -19,7 +19,8 @@ namespace ck {
 template <AddressSpaceEnum BufferAddressSpace,
           typename T,
           typename ElementSpaceSize,
-          bool InvalidElementUseNumericalZeroValue>
+          bool InvalidElementUseNumericalZeroValue,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
 struct DynamicBuffer
 {
     using type = T;
@@ -77,13 +78,16 @@ struct DynamicBuffer
 
             if constexpr(InvalidElementUseNumericalZeroValue)
             {
-                return amd_buffer_load_invalid_element_return_zero<remove_cvref_t<T>, t_per_x>(
+                return amd_buffer_load_invalid_element_return_zero<remove_cvref_t<T>,
+                                                                   t_per_x,
+                                                                   coherence>(
                     p_data_, i, is_valid_element, element_space_size_);
             }
             else
             {
                 return amd_buffer_load_invalid_element_return_customized_value<remove_cvref_t<T>,
-                                                                               t_per_x>(
+                                                                               t_per_x,
+                                                                               coherence>(
                     p_data_, i, is_valid_element, element_space_size_, invalid_element_value_);
             }
         }
@@ -173,7 +177,7 @@ struct DynamicBuffer
         {
             constexpr index_t t_per_x = scalar_per_x_vector / scalar_per_t_vector;
 
-            amd_buffer_store<remove_cvref_t<T>, t_per_x>(
+            amd_buffer_store<remove_cvref_t<T>, t_per_x, coherence>(
                 x, p_data_, i, is_valid_element, element_space_size_);
         }
         else if constexpr(GetAddressSpace() == AddressSpaceEnum::Lds &&
@@ -376,14 +380,19 @@ struct DynamicBuffer
     __host__ __device__ static constexpr bool IsDynamicBuffer() { return true; }
 };
 
-template <AddressSpaceEnum BufferAddressSpace, typename T, typename ElementSpaceSize>
+template <AddressSpaceEnum BufferAddressSpace,
+          AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence,
+          typename T,
+          typename ElementSpaceSize>
 __host__ __device__ constexpr auto make_dynamic_buffer(T* p, ElementSpaceSize element_space_size)
 {
-    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, true>{p, element_space_size};
+    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, true, coherence>{
+        p, element_space_size};
 }
 
 template <
     AddressSpaceEnum BufferAddressSpace,
+    AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence,
     typename T,
     typename ElementSpaceSize,
     typename X,
@@ -391,7 +400,7 @@ template <
 __host__ __device__ constexpr auto
 make_dynamic_buffer(T* p, ElementSpaceSize element_space_size, X invalid_element_value)
 {
-    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, false>{
+    return DynamicBuffer<BufferAddressSpace, T, ElementSpaceSize, false, coherence>{
         p, element_space_size, invalid_element_value};
 }
 

From 0ede66de54477a2c723e314f1ae934f8b51c40f8 Mon Sep 17 00:00:00 2001
From: ltqin <letao.qin@amd.com>
Date: Mon, 12 Jun 2023 21:35:31 +0800
Subject: [PATCH 030/194] Fix flash attn mask bug (#733)

* add check input parameter

* add instance for vector load = 1

* move gerneral instance to first pos

* fix read bias code

* regular code for bias load

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../gemm_bias_softmax_gemm_permute.cpp        |   3 +-
 ...gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  26 ++++-
 ...ultiple_d_softmax_gemm_xdl_cshuffle_v1.hpp | 104 +++++++-----------
 ...f16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp |  14 ++-
 ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp |  10 +-
 ...f16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp |   4 +-
 ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp |   4 +-
 7 files changed, 86 insertions(+), 79 deletions(-)

diff --git a/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp b/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
index cfb42c6e17..a90a6340a4 100644
--- a/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
+++ b/example/47_gemm_bias_softmax_gemm_permute/gemm_bias_softmax_gemm_permute.cpp
@@ -121,7 +121,8 @@ using DeviceOpInstance =
         2,              // CShuffleNXdlPerWavePerShuffle
         S<1, 32, 1, 8>, // CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
         8,              // CShuffleBlockTransferScalarPerVector_NPerBlock
-        MaskingSpec>;   // MaskingSpecialization
+        MaskingSpec,    // MaskingSpecialization
+        1>;
 
 // Ref Gemm0: fp16 in, fp32 out
 using ReferenceGemm0Instance = ck::tensor_operation::host::ReferenceBatchedGemm<ADataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index e54c013cf9..3fad319e90 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -197,7 +197,8 @@ template <index_t NumDimG,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           MaskingSpecialization MaskingSpec,
-          LoopScheduler LoopSched = LoopScheduler::Default>
+          int D0sTransferSrcScalarPerVector = 4,
+          LoopScheduler LoopSched           = LoopScheduler::Default>
 struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
     : public DeviceBatchedGemmSoftmaxGemmPermute<NumDimG,
                                                  NumDimM,
@@ -438,7 +439,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched,
         Transform::matrix_padder.PadN,
-        MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle>;
+        MaskingSpec == MaskingSpecialization::MaskOutUpperTriangle,
+        D0sTransferSrcScalarPerVector>;
 
     // Argument
     // FIXME: constness
@@ -530,6 +532,11 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
                 using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
                 // D0 pointer
                 p_d0s_grid_(i) = static_cast<const D0DataType*>(p_acc0_biases[i]);
+                // for  check
+                d0s_nl_ns_lengths_strides_[i].push_back(
+                    acc0_biases_gs_ms_ns_lengths[i][NumDimG + NumDimM]);
+                d0s_nl_ns_lengths_strides_[i].push_back(
+                    acc0_biases_gs_ms_ns_strides[i][NumDimG + NumDimM]);
             });
 
             if(GridwiseGemm::CheckValidity(a_grid_desc_ak0_m_ak1_,
@@ -608,6 +615,7 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         std::vector<index_t> b_nz_kz_strides_;
         std::vector<index_t> b1_nz_kz_strides_;
         std::vector<index_t> c_mz_gemm1nz_strides_;
+        std::array<std::vector<ck::index_t>, NumD0Tensor> d0s_nl_ns_lengths_strides_;
 
         index_t batch_count_;
         ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch_;
@@ -772,6 +780,20 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         {
             return false;
         }
+        for(int i = 0; i < NumD0Tensor; i++)
+        {
+            if(arg.d0s_nl_ns_lengths_strides_[i][1] == 1 &&
+               arg.d0s_nl_ns_lengths_strides_[i][0] % D0sTransferSrcScalarPerVector != 0)
+            {
+                std::cout << "first" << std::endl;
+                return false;
+            }
+            if(arg.d0s_nl_ns_lengths_strides_[i][1] != 1 && D0sTransferSrcScalarPerVector != 1)
+            {
+                std::cout << "second" << std::endl;
+                return false;
+            }
+        }
 
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
                                            arg.b_grid_desc_bk0_n_bk1_,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
index 74171ea9d9..135b9da6a0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -80,7 +80,8 @@ template <typename FloatAB,
           LoopScheduler LoopSched,
           bool PadN,
           bool MaskOutUpperTriangle,
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          int D0sTransferSrcScalarPerVector = 4,
+          PipelineVersion PipelineVer       = PipelineVersion::v1>
 struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
 {
     static_assert(LoopSched == LoopScheduler::Default,
@@ -621,13 +622,13 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
         constexpr auto d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5 =
             make_naive_tensor_descriptor_packed(make_tuple(I1,   // MBlockId
                                                            I1,   // NBlockID
-                                                           I1,   // MRepeat
-                                                           I1,   // NRepeat
-                                                           I1,   // MWaveId
-                                                           I1,   // NWaveId
-                                                           I1,   // MPerXdl
-                                                           I1,   // NGroupNum
-                                                           I1,   // NInputNum
+                                                           m0,   // MRepeat
+                                                           n0,   // NRepeat
+                                                           m1,   // MWaveId
+                                                           n1,   // NWaveId
+                                                           m2,   // MPerXdl
+                                                           n2,   // NGroupNum
+                                                           n3,   // NInputNum
                                                            n4)); // registerNum
 
         auto d0s_thread_buf = generate_tuple(
@@ -644,9 +645,6 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
         const auto wave_id     = GetGemm0WaveIdx();
         const auto wave_m_n_id = GetGemm0WaveMNIdx(wave_id[I2]); // I2: 0~63
 
-        constexpr auto acc0_thread_desc = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<MXdlPerWave>{}, Number<NXdlPerWave>{}, n2, n4));
-
         auto d0s_threadwise_copy = generate_tuple(
             [&](auto i) {
                 using D0DataType = remove_cvref_t<tuple_element_t<i.value, D0sDataType>>;
@@ -655,10 +653,19 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
                     D0DataType,
                     decltype(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i]),
                     decltype(d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5),
-                    Sequence<I1, I1, I1, I1, I1, I1, I1, I1, I1, n4>,
+                    Sequence<I1, // MBlockId
+                             I1, // NBlockID
+                             m0, // MRepeat
+                             n0, // NRepeat
+                             m1, // MWaveId
+                             n1, // NWaveId
+                             m2, // MPerXdl
+                             n2, // NGroupNum
+                             n3, // NInputNum
+                             n4>,
                     Sequence<0, 1, 2, 3, 4, 5, 6, 7, 8, 9>,
                     9,
-                    n4,
+                    D0sTransferSrcScalarPerVector,
                     1,
                     false>(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
                            make_multi_index(block_work_idx[I0], // MBlockId
@@ -884,62 +891,35 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
             // multiple d
             if constexpr(NumD0Tensor)
             {
-                static_for<0, MXdlPerWave, 1>{}([&](auto mr) {
-                    static_for<0, NXdlPerWave, 1>{}([&](auto nr) {
-                        static_for<0, n2, 1>{}([&](auto groupid) {
-                            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                                d0s_threadwise_copy(i).Run(
-                                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                    d0s_grid_buf[i],
-                                    d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
-                                    make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                    d0s_thread_buf(i));
-                            });
+                static_assert(NXdlPerWave == n0);
+                static_assert(MXdlPerWave == m0);
 
-                            static_for<0, n4, 1>{}([&](auto i) {
-                                constexpr index_t c_offset = acc0_thread_desc.CalculateOffset(
-                                    make_tuple(mr, nr, groupid, i));
+                static_for<0, NumD0Tensor, 1>{}([&](auto i) {
+                    d0s_threadwise_copy(i).Run(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
+                                               d0s_grid_buf[i],
+                                               d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
+                                               make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
+                                               d0s_thread_buf(i));
+                });
+                static_for<0, m0 * n0 * n2 * n4, 1>{}([&](auto i) {
+                    // get reference to src data
+                    const auto src_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iSrc) -> const auto& { return d0s_thread_buf[iSrc][i]; },
+                        Number<NumD0Tensor>{});
 
-                                // get reference to src data
-                                const auto src_data_refs = generate_tie(
-                                    // return type should be lvalue
-                                    [&](auto iSrc) -> const auto& {
-                                        return d0s_thread_buf[iSrc][i];
-                                    },
-                                    Number<NumD0Tensor>{});
+                    // get reference to dst data
+                    auto dst_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto) -> auto& { return acc_thread_buf(i); },
+                        Number<2>{});
 
-                                // get reference to dst data
-                                auto dst_data_refs = generate_tie(
-                                    // return type should be lvalue
-                                    [&](auto) -> auto& {
-                                        return acc_thread_buf(Number<c_offset>{});
-                                    },
-                                    Number<2>{});
-
-                                unpack2(c0de_element_op, dst_data_refs, src_data_refs);
-                            });
-                            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                                d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                    make_multi_index(0, 0, 0, 0, 0, 0, 0, 1, 0, 0));
-                            });
-                        });
-                        static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                            d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                                d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                make_multi_index(0, 0, 0, 1, 0, 0, 0, -n2.value, 0, 0));
-                        });
-                    });
-                    static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                        d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                            d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                            make_multi_index(0, 0, 1, -NXdlPerWave, 0, 0, 0, 0, 0, 0));
-                    });
+                    unpack2(c0de_element_op, dst_data_refs, src_data_refs);
                 });
                 static_for<0, NumD0Tensor, 1>{}([&](auto i) {
                     d0s_threadwise_copy(i).MoveSrcSliceWindow(
                         d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                        make_multi_index(0, 1, -MXdlPerWave, 0, 0, 0, 0, 0, 0, 0));
+                        make_multi_index(0, 1, 0, 0, 0, 0, 0, 0, 0, 0));
                 });
             }
             else
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
index a541a2d224..498bf58fb3 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
@@ -41,10 +41,11 @@ template <index_t NumDimG,
 using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances =
     std::tuple<
         // clang-format off
-        // #############################################|  NumDimG| NumDimM| NumDimN| NumDimK| NumDimO|  AData|  B0Data|  B1Data|  CData|     Acc0BiasData| Acc1BiasData| AccData| CShuffle|           A|          B0|        Acc0|          B1|           C|           GEMM|   ATensorSpec|  B0TensorSpec|  B1TensorSpec|   CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| MaskingSpec|
-        // #############################################|         |        |        |        |        |   Type|    Type|    Type|   Type|             Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
-        // #############################################|         |        |        |        |        |       |        |        |       |                 |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
-        // #############################################|         |        |        |        |        |       |        |        |       |                 |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |
+        // #############################################|  NumDimG| NumDimM| NumDimN| NumDimK| NumDimO|  AData|  B0Data|  B1Data|  CData|     Acc0BiasData| Acc1BiasData| AccData| CShuffle|           A|          B0|        Acc0|          B1|           C|           GEMM|   ATensorSpec|  B0TensorSpec|  B1TensorSpec|   CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| MaskingSpec|  D0s Bias|
+        // #############################################|         |        |        |        |        |   Type|    Type|    Type|   Type|             Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            | SrcScalar|
+        // #############################################|         |        |        |        |        |       |        |        |       |                 |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            | PerVector|
+        // #############################################|         |        |        |        |        |       |        |        |       |                 |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |          |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,        1>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
@@ -58,8 +59,9 @@ using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         // Padded fallback kernel
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,        1>,  
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<BF16>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 5e481603ca..744bd6456d 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -41,10 +41,11 @@ template <index_t NumDimG,
 using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances =
     std::tuple<
         // clang-format off
-        // #############################################|  NumDimG| NumDimM| NumDimN| NumDimK| NumDimO| AData| B0Data| B1Data| CData|    Acc0BiasData| Acc1BiasData| AccData| CShuffle|           A|          B0|        Acc0|          B1|           C|           GEMM|   ATensorSpec|  B0TensorSpec|  B1TensorSpec|   CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| MaskingSpec|
-        // #############################################|         |        |        |        |        |  Type|   Type|   Type|  Type|            Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
-        // #############################################|         |        |        |        |        |      |       |       |      |                |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
-        // #############################################|         |        |        |        |        |      |       |       |      |                |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |
+        // #############################################|  NumDimG| NumDimM| NumDimN| NumDimK| NumDimO| AData| B0Data| B1Data| CData|    Acc0BiasData| Acc1BiasData| AccData| CShuffle|           A|          B0|        Acc0|          B1|           C|           GEMM|   ATensorSpec|  B0TensorSpec|  B1TensorSpec|   CTensorSpec| NumGemmK| Block| Gemm01| Gemm0| Gemm0| Gemm1| Gemm1| AK1| BK1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| MaskingSpec|  D0s Bias|
+        // #############################################|         |        |        |        |        |  Type|   Type|   Type|  Type|            Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            | SrcScalar|
+        // #############################################|         |        |        |        |        |      |       |       |      |                |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            | PerVector|
+        // #############################################|         |        |        |        |        |      |       |       |      |                |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |          |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,       1>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
  #if CK_WORKAROUND_SWDEV_388832
@@ -60,6 +61,7 @@ using device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         // Padded fallback kernel
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec,       1>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<F16>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,    ScaleAdd, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
         // clang-format on
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
index 501ea85f60..b342612d1c 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
@@ -45,6 +45,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_
         // #############################################|         |        |        |        |        |   Type|    Type|    Type|   Type|         Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
         // #############################################|         |        |        |        |        |       |        |        |       |             |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
         // #############################################|         |        |        |        |        |       |        |        |       |             |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   256,    32,    64,    32,   8,   8,    2,   32,   32,     1,     8,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
@@ -58,8 +59,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         // Padded fallback kernel
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   BF16,    BF16,    BF16,   BF16,  ck::Tuple<>,  ck::Tuple<>,     F32,     BF16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 91ab541bf9..3fd0c07370 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -45,6 +45,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_
         // #############################################|         |        |        |        |        |  Type|   Type|   Type|  Type|         Type|         Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Elementwise| Elementwise| Specialization|              |              |              |              | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
         // #############################################|         |        |        |        |        |      |       |       |      |             |             |        |         |   Operation|   Operation|   Operation|   Operation|   Operation|               |              |              |              |              |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
         // #############################################|         |        |        |        |        |      |       |       |      |             |             |        |         |            |            |            |            |            |               |              |              |              |              |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |            |
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,    64,    32,   8,   8,    2,   32,   32,     2,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    256,   128,    32,   128,    32,   8,   8,    2,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
  #if CK_WORKAROUND_SWDEV_388832
@@ -60,8 +61,7 @@ using device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,               S<1, 16, 1,16>,               8, MaskingSpec>,
         DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,    GemmDefault, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,               S<1, 32, 1, 8>,               8, MaskingSpec>,
         // Padded fallback kernel
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>,
-        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
+        DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle<  NumDimG, NumDimM, NumDimN, NumDimK, NumDimO,   F16,    F16,    F16,   F16,  ck::Tuple<>,  ck::Tuple<>,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded, TensorDefault, TensorDefault, TensorDefault, TensorDefault,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8, MaskingSpec>
         // clang-format on
         >;
 

From 7c24654c248bbcc003062302a64c3f08d8422906 Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Mon, 12 Jun 2023 21:36:40 +0800
Subject: [PATCH 031/194] Fix incomplete object size (=4n + 3) support of
 amd_wave_read_first_lane() (#738)

* Fix wrong pointer type

* Rename type trait get_unsigned_int<> to get_carrier<>

* Add 3-bytes carrier type

* Add missing __device__ specifier

* Rename template non-type parameter

* Leave the rest byte uninitialized

* Avoid invoking (host) STL algorithms

* Remove unnecessary 'inline' specifier

* Extract common logic out as helper method

* Hide dummy member function

* Add missing __device__ specifier
---
 .../ck/utility/amd_wave_read_first_lane.hpp   | 75 ++++++++++++++++---
 1 file changed, 65 insertions(+), 10 deletions(-)

diff --git a/include/ck/utility/amd_wave_read_first_lane.hpp b/include/ck/utility/amd_wave_read_first_lane.hpp
index 4652ce7a74..741b2975af 100644
--- a/include/ck/utility/amd_wave_read_first_lane.hpp
+++ b/include/ck/utility/amd_wave_read_first_lane.hpp
@@ -7,6 +7,7 @@
 #include "ck/utility/functional2.hpp"
 #include "ck/utility/math.hpp"
 
+#include <array>
 #include <cstddef>
 #include <cstdint>
 #include <type_traits>
@@ -14,29 +15,83 @@
 namespace ck {
 namespace detail {
 
-template <unsigned Size>
-struct get_unsigned_int;
+template <unsigned SizeInBytes>
+struct get_carrier;
 
 template <>
-struct get_unsigned_int<1>
+struct get_carrier<1>
 {
     using type = uint8_t;
 };
 
 template <>
-struct get_unsigned_int<2>
+struct get_carrier<2>
 {
     using type = uint16_t;
 };
 
 template <>
-struct get_unsigned_int<4>
+struct get_carrier<3>
+{
+    using type = class carrier
+    {
+        using value_type = uint32_t;
+
+        std::array<std::byte, 3> bytes;
+        static_assert(sizeof(bytes) <= sizeof(value_type));
+
+        // replacement of host std::copy_n()
+        template <typename InputIterator, typename Size, typename OutputIterator>
+        __device__ static OutputIterator copy_n(InputIterator from, Size size, OutputIterator to)
+        {
+            if(0 < size)
+            {
+                *to = *from;
+                ++to;
+                for(Size count = 1; count < size; ++count)
+                {
+                    *to = *++from;
+                    ++to;
+                }
+            }
+
+            return to;
+        }
+
+        // method to trigger template substitution failure
+        __device__ carrier(const carrier& other) noexcept
+        {
+            copy_n(other.bytes.begin(), bytes.size(), bytes.begin());
+        }
+
+        public:
+        __device__ carrier& operator=(value_type value) noexcept
+        {
+            copy_n(reinterpret_cast<const std::byte*>(&value), bytes.size(), bytes.begin());
+
+            return *this;
+        }
+
+        __device__ operator value_type() const noexcept
+        {
+            std::byte result[sizeof(value_type)];
+
+            copy_n(bytes.begin(), bytes.size(), result);
+
+            return *reinterpret_cast<const value_type*>(result);
+        }
+    };
+};
+static_assert(sizeof(get_carrier<3>::type) == 3);
+
+template <>
+struct get_carrier<4>
 {
     using type = uint32_t;
 };
 
-template <unsigned Size>
-using get_unsigned_int_t = typename get_unsigned_int<Size>::type;
+template <unsigned SizeInBytes>
+using get_carrier_t = typename get_carrier<SizeInBytes>::type;
 
 } // namespace detail
 
@@ -61,7 +116,7 @@ __device__ auto amd_wave_read_first_lane(const Object& obj)
     constexpr Size CompleteSgprCopyBoundary = ObjectSize - RemainedSize;
     for(Size offset = 0; offset < CompleteSgprCopyBoundary; offset += SgprSize)
     {
-        using Sgpr = detail::get_unsigned_int_t<SgprSize>;
+        using Sgpr = detail::get_carrier_t<SgprSize>;
 
         *reinterpret_cast<Sgpr*>(to_obj + offset) =
             amd_wave_read_first_lane(*reinterpret_cast<const Sgpr*>(from_obj + offset));
@@ -69,9 +124,9 @@ __device__ auto amd_wave_read_first_lane(const Object& obj)
 
     if constexpr(0 < RemainedSize)
     {
-        using Carrier = detail::get_unsigned_int_t<RemainedSize>;
+        using Carrier = detail::get_carrier_t<RemainedSize>;
 
-        *reinterpret_cast<Carrier>(to_obj + CompleteSgprCopyBoundary) = amd_wave_read_first_lane(
+        *reinterpret_cast<Carrier*>(to_obj + CompleteSgprCopyBoundary) = amd_wave_read_first_lane(
             *reinterpret_cast<const Carrier*>(from_obj + CompleteSgprCopyBoundary));
     }
 

From fc9f97568ffcecf36c3f59f4b0d6680720ddd099 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Mon, 12 Jun 2023 15:37:15 +0200
Subject: [PATCH 032/194] Add DeviceBatchedGemmMultipleD_Dl (#732)

* Add DeviceBatchedGemmMultipleD_Dl

* Fix batched_gemm tests

* Fix comments

* test_batched_gemm_multi_d fixes

* Fix args for isSupported batchedGemmMultipleDDl

* Disable tests for gfx90a
---
 .../device_batched_gemm_multiple_d_dl.hpp     | 796 ++++++++++++++++++
 .../gpu/batched_gemm_multi_d.hpp              | 337 ++++++++
 .../gpu/batched_gemm_multi_d/CMakeLists.txt   |  18 +
 ..._d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp |  95 +++
 ...f16_f16_gkm_gkn_gmn_irregular_instance.cpp |  84 ++
 ..._d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp |  95 +++
 ...f16_f16_gkm_gnk_gmn_irregular_instance.cpp |  83 ++
 ..._d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp |  95 +++
 ...f16_f16_gmk_gkn_gmn_irregular_instance.cpp |  83 ++
 ..._d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp |  95 +++
 ...f16_f16_gmk_gnk_gmn_irregular_instance.cpp |  83 ++
 ...lti_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp |  93 ++
 ...8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp |  90 ++
 ...lti_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp |  93 ++
 ...8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp |  90 ++
 ...lti_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp |  93 ++
 ...8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp |  90 ++
 ...lti_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp |  93 ++
 ...8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp |  90 ++
 profiler/README.md                            |  27 +
 .../profiler/profile_batched_gemm_impl.hpp    |  95 ++-
 profiler/src/CMakeLists.txt                   |   3 +-
 profiler/src/profile_batched_gemm.cpp         | 103 ++-
 profiler/src/profile_batched_gemm_multi_d.cpp | 190 +++++
 test/CMakeLists.txt                           |   1 +
 test/batched_gemm/batched_gemm_bf16.cpp       |  94 ++-
 test/batched_gemm/batched_gemm_fp16.cpp       |  94 ++-
 test/batched_gemm/batched_gemm_fp32.cpp       |  94 ++-
 test/batched_gemm/batched_gemm_int8.cpp       |  94 ++-
 test/batched_gemm_multi_d/CMakeLists.txt      |   5 +
 .../test_batched_gemm_multi_d.cpp             |  74 ++
 31 files changed, 3347 insertions(+), 123 deletions(-)
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
 create mode 100644 profiler/src/profile_batched_gemm_multi_d.cpp
 create mode 100644 test/batched_gemm_multi_d/CMakeLists.txt
 create mode 100644 test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
new file mode 100644
index 0000000000..0df365ceb2
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
@@ -0,0 +1,796 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+/*
+ * \tparam ComputePtrOffsetOfBatch Class that computes the base pointer offsets of A, B, C matrix
+ * given the batch. For example, ComputePtrOffsetOfStridedBatch() computes the offsets of evenly
+ * strided batched, but we can easily extend to other layouts. The returned offset can be either \p
+ * index_t or \p long_index_t. If it returns \p long_index_t, we are not subject to the 2GB
+ * limitations.
+ *
+ * \note Using \p ComputePtrOffsetOfBatch gives us the flexibility that 2 workgroups can compute 2
+ * tiles from different matrices. Keep in mind that these 2 matrices can share the same grid
+ * descriptor (like in BatchedGEMM), or use their own grid descriptors (in GroupedGemm). \link
+ * impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp kernel_gemm_xdlops_v2r3_for_conv3d \endlink for
+ * \link DeviceConv3d \endlink uses the same concept, but currently does NOT encapsulate the
+ * computing of pointer offset into \p ComputePtrOffsetOfStridedBatch.
+ */
+
+template <typename GridwiseGemm,
+          typename ABDataType,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AGridDesc_K0_M0_M1_K1,
+          typename BGridDesc_K0_N0_N1_K1,
+          typename DsGridDesc_M0_M10_M11_N0_N10_N11,
+          typename CGridDesc_M0_M10_M11_N0_N10_N11,
+          typename ComputePtrOffsetOfBatch,
+          typename Block2CTileMap,
+          bool HasMainKBlockLoop,
+          bool HasDoubleTailKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_dl_multiple_d(
+            const ABDataType* __restrict__ p_a_grid,
+            const ABDataType* __restrict__ p_b_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const index_t batch_count,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1,
+            const BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1,
+            const DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11,
+            const CGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11,
+            const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
+            const Block2CTileMap block_2_ctile_map)
+{
+// TODO: Enable for gfx90a after complier fix
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||              \
+    defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || defined(__gfx1101__) || \
+    defined(__gfx1102__))
+
+    const index_t num_blocks_per_batch =
+        __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+    const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+    const long_index_t a_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx)));
+    const long_index_t b_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetBPtrOffset(g_idx)));
+    const long_index_t e_batch_offset = __builtin_amdgcn_readfirstlane(
+        static_cast<long_index_t>(compute_ptr_offset_of_batch.GetEPtrOffset(g_idx)));
+
+    const auto ds_batch_offset = compute_ptr_offset_of_batch.GetDsPtrOffset(g_idx);
+
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    DsPointer p_ds_grid_grp;
+
+    static constexpr index_t NumDTensor = DsGridDesc_M0_M10_M11_N0_N10_N11::Size();
+
+    static_for<0, NumDTensor, 1>{}(
+        [&](auto i) { p_ds_grid_grp(i) = p_ds_grid[i] + ds_batch_offset[i]; });
+
+    GridwiseGemm::Run(p_a_grid + a_batch_offset,
+                      p_b_grid + b_batch_offset,
+                      p_ds_grid_grp,
+                      p_e_grid + e_batch_offset,
+                      p_shared,
+                      a_element_op,
+                      b_element_op,
+                      cde_element_op,
+                      a_grid_desc_k0_m0_m1_k1,
+                      b_grid_desc_k0_n0_n1_k1,
+                      ds_grid_desc_m0_m10_m11_n0_n10_n11,
+                      e_grid_desc_m0_m10_m11_n0_n10_n11,
+                      block_2_ctile_map,
+                      integral_constant<bool, HasMainKBlockLoop>{},
+                      integral_constant<bool, HasDoubleTailKBlockLoop>{});
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = batch_count;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = a_grid_desc_k0_m0_m1_k1;
+    ignore = b_grid_desc_k0_n0_n1_k1;
+    ignore = ds_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = e_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = compute_ptr_offset_of_batch;
+    ignore = block_2_ctile_map;
+
+#endif
+}
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t K1,
+          index_t M1PerThread,
+          index_t N1PerThread,
+          index_t KPerThread,
+          typename M1N1ThreadClusterM1Xs,
+          typename M1N1ThreadClusterN1Xs,
+          typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+          typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+          typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+          typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+          typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+          typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          enable_if_t<
+              is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+                  is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+              bool> = false>
+struct DeviceBatchedGemmMultipleD_Dl : public DeviceBatchedGemmMultiD<ALayout,
+                                                                      BLayout,
+                                                                      DsLayout,
+                                                                      ELayout,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      DsDataType,
+                                                                      EDataType,
+                                                                      AElementwiseOperation,
+                                                                      BElementwiseOperation,
+                                                                      CDEElementwiseOperation>
+
+{
+    using DeviceOp                      = DeviceBatchedGemmMultipleD_Dl;
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto K1Number = Number<K1>{};
+
+    static auto MakeAGridDescriptor_K0_M_K1(index_t M, index_t K, index_t StrideA)
+    {
+        const index_t K0 = K / K1;
+
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(M, PadM)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeBGridDescriptor_K0_N_K1(index_t K, index_t N, index_t StrideB)
+    {
+        const index_t K0 = K / K1;
+
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0, K1Number)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+    }
+
+    template <typename ELay>
+    static auto MakeEGridDescriptor_M_N(index_t M, index_t N, index_t StrideE)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideE));
+            }
+        }();
+
+        if constexpr(GemmSpec == GemmSpecialization::MNPadding)
+        {
+            const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_right_pad_transform(M, PadM), make_right_pad_transform(N, PadN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+        else
+        {
+
+            return transform_tensor_descriptor(
+                c_grid_desc_m_n,
+                make_tuple(make_pass_through_transform(M), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                                         const std::array<index_t, NumDTensor>& NRaws,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    using AGridDesc_K0_M_K1 = decltype(MakeAGridDescriptor_K0_M_K1(1, 1, 1));
+    using BGridDesc_K0_N_K1 = decltype(MakeBGridDescriptor_K0_N_K1(1, 1, 1));
+    using DsGridDesc_M_N    = decltype(MakeDsGridDescriptor_M_N({}, {}, {}));
+    using EGridDesc_M_N     = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
+
+    struct ComputePtrOffsetOfStridedBatch
+    {
+        ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
+                                       index_t BatchStrideB,
+                                       std::array<ck::index_t, NumDTensor> BatchStrideDs,
+                                       index_t BatchStrideE)
+            : BatchStrideA_(BatchStrideA),
+              BatchStrideB_(BatchStrideB),
+              BatchStrideDs_(BatchStrideDs),
+              BatchStrideE_(BatchStrideE)
+        {
+        }
+
+        __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideA_);
+        }
+
+        __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideB_);
+        }
+
+        __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
+        {
+            std::array<long_index_t, NumDTensor> ds_offset;
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                ds_offset[i] = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]);
+            });
+            return ds_offset;
+        }
+
+        __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+        {
+            return g_idx * static_cast<long_index_t>(BatchStrideE_);
+        }
+
+        private:
+        index_t BatchStrideA_;
+        index_t BatchStrideB_;
+        std::array<ck::index_t, NumDTensor> BatchStrideDs_;
+        index_t BatchStrideE_;
+    };
+
+    // GridwiseGemm
+    using GridwiseGemm =
+        GridwiseGemmDlMultipleD_km_kn_mn<BlockSize,
+                                         ADataType,
+                                         AccDataType,
+                                         DsDataType,
+                                         EDataType,
+                                         AElementwiseOperation,
+                                         BElementwiseOperation,
+                                         CDEElementwiseOperation,
+                                         InMemoryDataOperationEnum::Set,
+                                         AGridDesc_K0_M_K1,
+                                         BGridDesc_K0_N_K1,
+                                         EGridDesc_M_N,
+                                         MPerBlock,
+                                         NPerBlock,
+                                         K0PerBlock,
+                                         K1,
+                                         M1PerThread,
+                                         N1PerThread,
+                                         KPerThread,
+                                         M1N1ThreadClusterM1Xs,
+                                         M1N1ThreadClusterN1Xs,
+                                         ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+                                         ABlockTransferThreadClusterArrangeOrder,
+                                         ABlockTransferSrcAccessOrder,
+                                         ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+                                         ABlockTransferSrcVectorTensorContiguousDimOrder,
+                                         ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+                                         BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+                                         BBlockTransferThreadClusterArrangeOrder,
+                                         BBlockTransferSrcAccessOrder,
+                                         BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+                                         BBlockTransferSrcVectorTensorContiguousDimOrder,
+                                         BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+                                         CThreadTransferSrcDstAccessOrder,
+                                         CThreadTransferSrcDstVectorDim,
+                                         CThreadTransferDstScalarPerVector>;
+
+    using AGridDesc_K0_M0_M1_K1 =
+        decltype(GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
+    using BGridDesc_K0_N0_N1_K1 =
+        decltype(GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
+    using DsGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(DsGridDesc_M_N{}));
+    using EGridDesc_M0_M10_M11_N0_N10_N11 =
+        decltype(GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(EGridDesc_M_N{}));
+    using DefaultBlock2CTileMap =
+        decltype(GridwiseGemm::MakeDefaultBlock2CTileMap(EGridDesc_M_N{}));
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a_grid,
+                 const void* p_b_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 index_t M,
+                 index_t N,
+                 index_t K,
+                 index_t Batch,
+                 index_t StrideA,
+                 index_t StrideB,
+                 std::array<index_t, NumDTensor> StrideDs,
+                 index_t StrideE,
+                 index_t BatchStrideA,
+                 index_t BatchStrideB,
+                 const std::array<ck::index_t, NumDTensor>& BatchStrideDs,
+                 index_t BatchStrideE,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a_grid)},
+              p_b_grid_{static_cast<const BDataType*>(p_b_grid)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              K_(K),
+              Batch_(Batch),
+              a_grid_desc_k0_m0_m1_k1_{},
+              b_grid_desc_k0_n0_n1_k1_{},
+              e_grid_desc_m0_m10_m11_n0_n10_n11_{},
+              compute_ptr_offset_of_batch_{BatchStrideA, BatchStrideB, BatchStrideDs, BatchStrideE},
+              block_2_ctile_map_{},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+            a_grid_desc_k0_m_k1_ =
+                DeviceBatchedGemmMultipleD_Dl::MakeAGridDescriptor_K0_M_K1(M, K, StrideA);
+            b_grid_desc_k0_n_k1_ =
+                DeviceBatchedGemmMultipleD_Dl::MakeBGridDescriptor_K0_N_K1(K, N, StrideB);
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    DeviceOp::MakeEGridDescriptor_M_N<DLayout>(M, N, StrideDs[i]);
+            });
+            e_grid_desc_m_n_ =
+                DeviceBatchedGemmMultipleD_Dl::MakeEGridDescriptor_M_N<ELayout>(M, N, StrideE);
+
+            if(GridwiseGemm::CheckValidity(
+                   a_grid_desc_k0_m_k1_, b_grid_desc_k0_n_k1_, e_grid_desc_m_n_))
+            {
+                a_grid_desc_k0_m0_m1_k1_ =
+                    GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(a_grid_desc_k0_m_k1_);
+                b_grid_desc_k0_n0_n1_k1_ =
+                    GridwiseGemm::MakeBGridDescriptor_K0_N0_N1_K1(b_grid_desc_k0_n_k1_);
+
+                ds_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                    GridwiseGemm::MakeDsGridDescriptor_M0_M10_M11_N0_N10_N11(ds_grid_desc_m_n_);
+
+                e_grid_desc_m0_m10_m11_n0_n10_n11_ =
+                    GridwiseGemm::MakeCGridDescriptor_M0_M10_M11_N0_N10_N11(e_grid_desc_m_n_);
+
+                block_2_ctile_map_ = GridwiseGemm::MakeDefaultBlock2CTileMap(e_grid_desc_m_n_);
+            }
+        }
+
+        //  private:
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        index_t K_;
+
+        // Batch
+        index_t Batch_;
+
+        AGridDesc_K0_M_K1 a_grid_desc_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_k0_n_k1_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        AGridDesc_K0_M0_M1_K1 a_grid_desc_k0_m0_m1_k1_;
+        BGridDesc_K0_N0_N1_K1 b_grid_desc_k0_n0_n1_k1_;
+        DsGridDesc_M0_M10_M11_N0_N10_N11 ds_grid_desc_m0_m10_m11_n0_n10_n11_;
+        EGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11_;
+
+        // for calculating batch offset
+        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
+
+        DefaultBlock2CTileMap block_2_ctile_map_;
+
+        // TODO: unused since gridwise_gemm_dl_v1r3 does NOT support prologue for the time being.
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceBatchedGemmMultipleD_Dl::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            {
+                std::cout << "arg.a_grid_desc_k0_m0_m1_k1_{"
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I0) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I1) << ", "
+                          << arg.a_grid_desc_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.b_grid_desc_k0_n0_n1_k1_{"
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I0) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I1) << ", "
+                          << arg.b_grid_desc_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+                std::cout << "arg.e_grid_desc_m_n_{ " << arg.e_grid_desc_m_n_.GetLength(I0) << ", "
+                          << arg.e_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+            }
+
+            if(!GridwiseGemm::CheckValidity(
+                   arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemmDlMultipleD_km_kn_mn has invalid setting");
+            }
+
+            const index_t grid_size =
+                GridwiseGemm::CalculateGridSize(arg.e_grid_desc_m_n_.GetLength(I0),
+                                                arg.e_grid_desc_m_n_.GetLength(I1)) *
+                arg.Batch_;
+
+            auto launch_kernel = [&](auto has_main_k_block_loop,
+                                     auto has_double_tail_k_block_loop) {
+                constexpr bool has_main_loop   = has_main_k_block_loop.value;
+                constexpr bool has_double_loop = has_double_tail_k_block_loop.value;
+
+                const auto kernel =
+                    kernel_gemm_dl_multiple_d<GridwiseGemm,
+                                              ADataType,
+                                              typename GridwiseGemm::DsGridPointer,
+                                              EDataType,
+                                              AElementwiseOperation,
+                                              BElementwiseOperation,
+                                              CDEElementwiseOperation,
+                                              DeviceOp::AGridDesc_K0_M0_M1_K1,
+                                              DeviceOp::BGridDesc_K0_N0_N1_K1,
+                                              DeviceOp::DsGridDesc_M0_M10_M11_N0_N10_N11,
+                                              DeviceOp::EGridDesc_M0_M10_M11_N0_N10_N11,
+                                              ComputePtrOffsetOfStridedBatch,
+                                              DefaultBlock2CTileMap,
+                                              has_main_loop,
+                                              has_double_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.Batch_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.a_grid_desc_k0_m0_m1_k1_,
+                                              arg.b_grid_desc_k0_n0_n1_k1_,
+                                              arg.ds_grid_desc_m0_m10_m11_n0_n10_n11_,
+                                              arg.e_grid_desc_m0_m10_m11_n0_n10_n11_,
+                                              arg.compute_ptr_offset_of_batch_,
+                                              arg.block_2_ctile_map_);
+            };
+
+            const auto K0                    = arg.a_grid_desc_k0_m0_m1_k1_.GetLength(I0);
+            const bool has_main_k_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+            const bool has_double_tail_k_block_loop =
+                GridwiseGemm::CalculateHasDoubleTailKBlockLoop(K0);
+
+            if(has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, true>{});
+            }
+            else if(has_main_k_block_loop && !has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{},
+                                     integral_constant<bool, false>{});
+            }
+            else if(!has_main_k_block_loop && has_double_tail_k_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{},
+                                     integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        // TODO: Enable for gfx90a after complier fix
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
+           ck::get_device_name() == "gfx1030" || ck::get_device_name() == "gfx940" ||
+           ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            bool pass = true;
+            pass      = pass && arg.K_ % K1 == 0;
+
+            pass = pass && GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
+                                                       arg.b_grid_desc_k0_n_k1_,
+                                                       arg.e_grid_desc_m_n_);
+
+            return pass;
+        }
+        else
+        {
+            return false;
+        }
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_a,
+                             const void* p_b,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t Batch,
+                             index_t StrideA,
+                             index_t StrideB,
+                             std::array<ck::index_t, NumDTensor> StrideDs,
+                             index_t StrideE,
+                             index_t BatchStrideA,
+                             index_t BatchStrideB,
+                             const std::array<ck::index_t, NumDTensor>& BatchStrideDs,
+                             index_t BatchStrideE,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        M,
+                        N,
+                        K,
+                        Batch,
+                        StrideA,
+                        StrideB,
+                        StrideDs,
+                        StrideE,
+                        BatchStrideA,
+                        BatchStrideB,
+                        BatchStrideDs,
+                        BatchStrideE,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_a,
+                        const void* p_b,
+                        const std::array<const void*, NumDTensor>& p_ds,
+                        void* p_e,
+                        index_t M,
+                        index_t N,
+                        index_t K,
+                        index_t Batch,
+                        index_t StrideA,
+                        index_t StrideB,
+                        const std::array<ck::index_t, NumDTensor>& StrideDs,
+                        index_t StrideE,
+                        index_t BatchStrideA,
+                        index_t BatchStrideB,
+                        const std::array<ck::index_t, NumDTensor>& BatchStrideDs,
+                        index_t BatchStrideE,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          M,
+                                          N,
+                                          K,
+                                          Batch,
+                                          StrideA,
+                                          StrideB,
+                                          StrideDs,
+                                          StrideE,
+                                          BatchStrideA,
+                                          BatchStrideB,
+                                          BatchStrideDs,
+                                          BatchStrideE,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceBatchedGemmMultipleD_Dl"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << M1PerThread << ", "
+            << N1PerThread << ", "
+            << KPerThread
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
new file mode 100644
index 0000000000..ae12f4c7ac
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
@@ -0,0 +1,337 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatchedGemmMultiD<
+    ALayout,
+    BLayout,
+    Empty_Tuple,
+    ELayout,
+    ADataType,
+    BDataType,
+    Empty_Tuple,
+    EDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough>>
+{
+    using DeviceOp = DeviceBatchedGemmMultiD<ALayout,
+                                             BLayout,
+                                             Empty_Tuple,
+                                             ELayout,
+                                             ADataType,
+                                             BDataType,
+                                             Empty_Tuple,
+                                             EDataType,
+                                             ck::tensor_operation::element_wise::PassThrough,
+                                             ck::tensor_operation::element_wise::PassThrough,
+                                             ck::tensor_operation::element_wise::PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
+                          is_same_v<EDataType, int8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(op_ptrs);
+                add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
new file mode 100644
index 0000000000..fda55a9303
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
@@ -0,0 +1,18 @@
+add_instance_library(device_batched_gemm_multi_d_instance
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
+   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000..3fe9f78b2a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..4ab22bb03a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000..80c890cdb3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..647c583036
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64       
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000..3ce582f1f9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 8, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..34c29d0380
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000..e0da8e3d5a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<4, 2>,       S<8, 2>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,   128,    16,  2,          4,          4,      1,       S<2, 4>,       S<2, 8>,      S<8, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 2, 2>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<1, 4>,       S<1, 4>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..3edd3f78c6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout| AData| BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |  Type|  Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |      |      |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |      |      |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row,   F16,   F16,     F32, Empty_Tuple, F16, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        F16,
+                                                        F16,
+                                                        Empty_Tuple,
+                                                        F16,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000..33234edc17
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  4,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  4,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,     S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..16107e1bde
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 4, 4>,      S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,      S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000..3e4bdb0172
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..8107717162
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Col,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
new file mode 100644
index 0000000000..0e943c88c3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,        S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..ea5e7c562d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Row, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Row,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
new file mode 100644
index 0000000000..000a4b0130
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  4,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=16, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        // MPerBlock=64, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    16,    16,  2,          4,          1,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  2>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
new file mode 100644
index 0000000000..24fb67619a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
@@ -0,0 +1,90 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Empty_Tuple = ck::Tuple<>;
+
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances = std::tuple<
+    // clang-format off
+        // ##########################| ALayout| BLayout|    DsLayout| CLayout|  AData|  BData| AccData|      DsData| CData|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|     CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########################|        |        |            |        |   Type|   Type|    Type|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|        SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########################|        |        |            |        |       |       |        |            |      |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
+        // ##########################|        |        |            |        |       |       |        |            |      |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
+        // MPerBlock=128, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=128, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // // MPerBlock=64, NPerBlock=128
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=32, NPerBlock=32
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=16, NPerBlock=16
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=64
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=64, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        // MPerBlock=8, NPerBlock=8
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceBatchedGemmMultipleD_Dl<     Row,     Col, Empty_Tuple,     Row, int8_t, int8_t, int32_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,        S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+    // clang-format on
+    >;
+
+void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances(
+    std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Row,
+                                                        Col,
+                                                        Empty_Tuple,
+                                                        Row,
+                                                        int8_t,
+                                                        int8_t,
+                                                        Empty_Tuple,
+                                                        int8_t,
+                                                        PassThrough,
+                                                        PassThrough,
+                                                        PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/README.md b/profiler/README.md
index 400a64a396..7a1fb29113 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -76,3 +76,30 @@ e_m_n: dim 4, lengths {128, 128, 128, 128}, strides {2097152, 16384, 128, 1}
 ....
 Best Perf: 211.405 ms, 41.6077 TFlops, 15.2372 GB/s
 ```
+
+## Profile batched gemm multiple D kernels
+```bash
+#arg1: tensor operation (batched_gemm_multi_d=Batched GEMM multi D);
+#arg2: data type (0: fp16; 1: int8)
+#arg3: matrix layout (0: A[g, m, k] * B[g, k, n] = C[g, m, n];
+#                     1: A[g, m, k] * B[g, n, k] = C[g, m, n];
+#                     2: A[g, k, m] * B[g, k, n] = C[g, m, n];
+#                     3: A[g, k, m] * B[g, n, k] = C[g, m, n])
+#arg4: verification (0: no; 1: yes)
+#arg5: initialization (0: no init; 1: integer value; 2: decimal value)
+#arg6: print tensor value (0: no; 1: yes)
+#arg7: time kernel (0=n0, 1=yes)
+#arg8 to 17: M, N, K, StrideA, StrideB, StrideC, BatchStrideA, BatchStrideB, BatchStrideC, BatchCount
+
+################                   op  datatype  layout  verify  init  log  time    M    N    K StrideA StrideB StrideC BatchStrideA BatchStrideB BatchStrideC BatchCount
+./bin/ckProfiler batched_gemm_multi_d         0       1       0     0    0     1 4096 4096 4096    4096    4096    4096     16777216     16777216     16777216         16
+```
+
+Result (Radeon RX 6800 XT)
+```bash
+arg.a_grid_desc_k0_m0_m1_k1_{2048, 4096, 2}
+arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
+arg.e_grid_desc_m_n_{ 4096, 4096}
+....
+Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
+```
diff --git a/profiler/include/profiler/profile_batched_gemm_impl.hpp b/profiler/include/profiler/profile_batched_gemm_impl.hpp
index cdc94aa9a6..936c22f5d8 100644
--- a/profiler/include/profiler/profile_batched_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_batched_gemm_impl.hpp
@@ -8,9 +8,11 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_batched_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/device_batched_gemm_multi_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp"
 
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
@@ -27,7 +29,11 @@ template <typename ADataType,
           typename CDataType,
           typename ALayout,
           typename BLayout,
-          typename CLayout>
+          typename CLayout,
+          typename AElementOp,
+          typename BElementOp,
+          typename CElementOp,
+          typename DeviceOp>
 bool profile_batched_gemm_impl(int do_verification,
                                int init_method,
                                bool do_log,
@@ -88,10 +94,6 @@ bool profile_batched_gemm_impl(int do_verification,
         b_g_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
     }
 
-    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
-    using CElementOp = ck::tensor_operation::element_wise::PassThrough;
-
     const auto a_element_op = AElementOp{};
     const auto b_element_op = BElementOp{};
     const auto c_element_op = CElementOp{};
@@ -124,16 +126,6 @@ bool profile_batched_gemm_impl(int do_verification,
     b_device_buf.ToDevice(b_g_k_n.mData.data());
     c_device_buf.ToDevice(c_g_m_n_device_result.mData.data());
 
-    using DeviceOp = ck::tensor_operation::device::DeviceBatchedGemm<ALayout,
-                                                                     BLayout,
-                                                                     CLayout,
-                                                                     ADataType,
-                                                                     BDataType,
-                                                                     CDataType,
-                                                                     AElementOp,
-                                                                     BElementOp,
-                                                                     CElementOp>;
-
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
@@ -148,23 +140,62 @@ bool profile_batched_gemm_impl(int do_verification,
     // profile device op instances
     for(auto& op_ptr : op_ptrs)
     {
-        auto argument_ptr =
-            op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
-                                        static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
-                                        static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
-                                        M,
-                                        N,
-                                        K,
-                                        StrideA,
-                                        StrideB,
-                                        StrideC,
-                                        BatchStrideA,
-                                        BatchStrideB,
-                                        BatchStrideC,
-                                        BatchCount,
-                                        ck::tensor_operation::element_wise::PassThrough{},
-                                        ck::tensor_operation::element_wise::PassThrough{},
-                                        ck::tensor_operation::element_wise::PassThrough{});
+        std::unique_ptr<tensor_operation::device::BaseArgument> argument_ptr;
+        // false branch for multi d dl kernel
+        if constexpr(std::is_same<
+                         DeviceOp,
+                         ck::tensor_operation::device::DeviceBatchedGemm<ALayout,
+                                                                         BLayout,
+                                                                         CLayout,
+                                                                         ADataType,
+                                                                         BDataType,
+                                                                         CDataType,
+                                                                         AElementOp,
+                                                                         BElementOp,
+                                                                         CElementOp>>::value)
+        {
+
+            argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                            M,
+                                            N,
+                                            K,
+                                            StrideA,
+                                            StrideB,
+                                            StrideC,
+                                            BatchStrideA,
+                                            BatchStrideB,
+                                            BatchStrideC,
+                                            BatchCount,
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{});
+        }
+        else
+        {
+            argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                            {},
+                                            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                            M,
+                                            N,
+                                            K,
+                                            BatchCount,
+                                            StrideA,
+                                            StrideB,
+                                            {},
+                                            StrideC,
+                                            BatchStrideA,
+                                            BatchStrideB,
+                                            {},
+                                            BatchStrideC,
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{},
+                                            ck::tensor_operation::element_wise::PassThrough{});
+        }
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index c9fccc2588..6f768e0ae1 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -34,6 +34,7 @@ set(PROFILER_SOURCES
     profile_grouped_gemm_fastgelu.cpp
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
+    profile_batched_gemm_multi_d.cpp
 )
 
 set(PROFILER_EXECUTABLE ckProfiler)
@@ -77,5 +78,5 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgel
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
-
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_batched_gemm.cpp b/profiler/src/profile_batched_gemm.cpp
index dc83e25b48..222532b7bb 100644
--- a/profiler/src/profile_batched_gemm.cpp
+++ b/profiler/src/profile_batched_gemm.cpp
@@ -10,6 +10,8 @@
 #include "profiler/profile_batched_gemm_impl.hpp"
 #include "profiler_operation_registry.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 enum struct GemmMatrixLayout
 {
     MK_KN_MN, // 0
@@ -78,55 +80,72 @@ int profile_batched_gemm(int argc, char* argv[])
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
 
-    auto profile = [&](auto a_type,
-                       auto b_type,
-                       auto c_type,
-                       auto a_layout,
-                       auto b_layout,
-                       auto c_layout) {
-        using ADataType = decltype(a_type);
-        using BDataType = decltype(b_type);
-        using CDataType = decltype(c_type);
+    auto profile =
+        [&](auto a_type, auto b_type, auto c_type, auto a_layout, auto b_layout, auto c_layout) {
+            using ADataType = decltype(a_type);
+            using BDataType = decltype(b_type);
+            using CDataType = decltype(c_type);
 
-        using ALayout = decltype(a_layout);
-        using BLayout = decltype(b_layout);
-        using CLayout = decltype(c_layout);
+            using ALayout = decltype(a_layout);
+            using BLayout = decltype(b_layout);
+            using CLayout = decltype(c_layout);
 
-        const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
-        const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
-        const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
+            const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
+            const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
+            const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
 
-        const int StrideA_ = (StrideA < 0) ? DefaultStrideA : StrideA;
-        const int StrideB_ = (StrideB < 0) ? DefaultStrideB : StrideB;
-        const int StrideC_ = (StrideC < 0) ? DefaultStrideC : StrideC;
+            const int StrideA_ = (StrideA < 0) ? DefaultStrideA : StrideA;
+            const int StrideB_ = (StrideB < 0) ? DefaultStrideB : StrideB;
+            const int StrideC_ = (StrideC < 0) ? DefaultStrideC : StrideC;
 
-        const int DefaultBatchStrideA = (ck::is_same_v<ALayout, Row> ? M : K) * StrideA_;
-        const int DefaultBatchStrideB = (ck::is_same_v<BLayout, Row> ? K : N) * StrideB_;
-        const int DefaultBatchStrideC = (ck::is_same_v<CLayout, Row> ? M : N) * StrideC_;
+            const int DefaultBatchStrideA = (ck::is_same_v<ALayout, Row> ? M : K) * StrideA_;
+            const int DefaultBatchStrideB = (ck::is_same_v<BLayout, Row> ? K : N) * StrideB_;
+            const int DefaultBatchStrideC = (ck::is_same_v<CLayout, Row> ? M : N) * StrideC_;
 
-        const int BatchStrideA_ = (BatchStrideA < 0) ? DefaultBatchStrideA : BatchStrideA;
-        const int BatchStrideB_ = (BatchStrideB < 0) ? DefaultBatchStrideB : BatchStrideB;
-        const int BatchStrideC_ = (BatchStrideC < 0) ? DefaultBatchStrideC : BatchStrideC;
+            const int BatchStrideA_ = (BatchStrideA < 0) ? DefaultBatchStrideA : BatchStrideA;
+            const int BatchStrideB_ = (BatchStrideB < 0) ? DefaultBatchStrideB : BatchStrideB;
+            const int BatchStrideC_ = (BatchStrideC < 0) ? DefaultBatchStrideC : BatchStrideC;
 
-        bool pass = ck::profiler::
-            profile_batched_gemm_impl<ADataType, BDataType, CDataType, ALayout, BLayout, CLayout>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                M,
-                N,
-                K,
-                BatchStrideA_,
-                BatchStrideB_,
-                BatchStrideC_,
-                StrideA_,
-                StrideB_,
-                StrideC_,
-                BatchCount);
+            using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using CElementOp = ck::tensor_operation::element_wise::PassThrough;
 
-        return pass ? 0 : 1;
-    };
+            using DeviceOp = ck::tensor_operation::device::DeviceBatchedGemm<ALayout,
+                                                                             BLayout,
+                                                                             CLayout,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             AElementOp,
+                                                                             BElementOp,
+                                                                             CElementOp>;
+
+            bool pass = ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                ALayout,
+                                                                BLayout,
+                                                                CLayout,
+                                                                AElementOp,
+                                                                BElementOp,
+                                                                CElementOp,
+                                                                DeviceOp>(do_verification,
+                                                                          init_method,
+                                                                          do_log,
+                                                                          time_kernel,
+                                                                          M,
+                                                                          N,
+                                                                          K,
+                                                                          BatchStrideA_,
+                                                                          BatchStrideB_,
+                                                                          BatchStrideC_,
+                                                                          StrideA_,
+                                                                          StrideB_,
+                                                                          StrideC_,
+                                                                          BatchCount);
+
+            return pass ? 0 : 1;
+        };
 
     if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
     {
diff --git a/profiler/src/profile_batched_gemm_multi_d.cpp b/profiler/src/profile_batched_gemm_multi_d.cpp
new file mode 100644
index 0000000000..98b462d950
--- /dev/null
+++ b/profiler/src/profile_batched_gemm_multi_d.cpp
@@ -0,0 +1,190 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdint>
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_batched_gemm_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp"
+
+enum struct GemmMatrixLayout
+{
+    MK_KN_MN, // 0
+    MK_NK_MN, // 1
+    KM_KN_MN, // 2
+    KM_NK_MN, // 3
+};
+
+enum struct GemmDataType
+{
+    F16_F16_F16,    // 0
+    INT8_INT8_INT8, // 1
+};
+
+#define OP_NAME "batched_gemm_multi_d"
+#define OP_DESC "Batched GEMM multi D"
+
+int profile_batched_gemm_multi_d(int argc, char* argv[])
+{
+    if(argc != 18)
+    {
+        // clang-format off
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp16; 1: int8)\n");
+        printf("arg3: matrix layout (0: A[g, m, k] * B[g, k, n] = C[g, m, n];\n");
+        printf("                     1: A[g, m, k] * B[g, n, k] = C[g, m, n];\n");
+        printf("                     2: A[g, k, m] * B[g, k, n] = C[g, m, n];\n");
+        printf("                     3: A[g, k, m] * B[g, n, k] = C[g, m, n])\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=n0, 1=yes)\n");
+        printf("arg8 to 17: M, N, K, StrideA, StrideB, StrideC, BatchStrideA, BatchStrideB, BatchStrideC, BatchCount\n");
+        // clang-format on
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<GemmDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<GemmMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const int M = std::stoi(argv[8]);
+    const int N = std::stoi(argv[9]);
+    const int K = std::stoi(argv[10]);
+
+    const int StrideA = std::stoi(argv[11]);
+    const int StrideB = std::stoi(argv[12]);
+    const int StrideC = std::stoi(argv[13]);
+
+    const int BatchStrideA = std::stoi(argv[14]);
+    const int BatchStrideB = std::stoi(argv[15]);
+    const int BatchStrideC = std::stoi(argv[16]);
+
+    const int BatchCount = std::stoi(argv[17]);
+
+    using F16  = ck::half_t;
+    using INT8 = int8_t;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    auto profile =
+        [&](auto a_type, auto b_type, auto c_type, auto a_layout, auto b_layout, auto c_layout) {
+            using ADataType  = decltype(a_type);
+            using BDataType  = decltype(b_type);
+            using CDataType  = decltype(c_type);
+            using DsDataType = ck::Tuple<>;
+
+            using ALayout  = decltype(a_layout);
+            using BLayout  = decltype(b_layout);
+            using CLayout  = decltype(c_layout);
+            using DsLayout = ck::Tuple<>;
+
+            const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
+            const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
+            const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
+
+            const int StrideA_ = (StrideA < 0) ? DefaultStrideA : StrideA;
+            const int StrideB_ = (StrideB < 0) ? DefaultStrideB : StrideB;
+            const int StrideC_ = (StrideC < 0) ? DefaultStrideC : StrideC;
+
+            const int DefaultBatchStrideA = (ck::is_same_v<ALayout, Row> ? M : K) * StrideA_;
+            const int DefaultBatchStrideB = (ck::is_same_v<BLayout, Row> ? K : N) * StrideB_;
+            const int DefaultBatchStrideC = (ck::is_same_v<CLayout, Row> ? M : N) * StrideC_;
+
+            const int BatchStrideA_ = (BatchStrideA < 0) ? DefaultBatchStrideA : BatchStrideA;
+            const int BatchStrideB_ = (BatchStrideB < 0) ? DefaultBatchStrideB : BatchStrideB;
+            const int BatchStrideC_ = (BatchStrideC < 0) ? DefaultBatchStrideC : BatchStrideC;
+
+            using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+            using CElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+            using DeviceOp = ck::tensor_operation::device::DeviceBatchedGemmMultiD<ALayout,
+                                                                                   BLayout,
+                                                                                   DsLayout,
+                                                                                   CLayout,
+                                                                                   ADataType,
+                                                                                   BDataType,
+                                                                                   DsDataType,
+                                                                                   CDataType,
+                                                                                   AElementOp,
+                                                                                   BElementOp,
+                                                                                   CElementOp>;
+
+            bool pass = ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                ALayout,
+                                                                BLayout,
+                                                                CLayout,
+                                                                AElementOp,
+                                                                BElementOp,
+                                                                CElementOp,
+                                                                DeviceOp>(do_verification,
+                                                                          init_method,
+                                                                          do_log,
+                                                                          time_kernel,
+                                                                          M,
+                                                                          N,
+                                                                          K,
+                                                                          BatchStrideA_,
+                                                                          BatchStrideB_,
+                                                                          BatchStrideC_,
+                                                                          StrideA_,
+                                                                          StrideB_,
+                                                                          StrideC_,
+                                                                          BatchCount);
+
+            return pass ? 0 : 1;
+        };
+
+    if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F16{}, F16{}, F16{}, Col{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(INT8{}, INT8{}, INT8{}, Col{}, Col{}, Row{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_batched_gemm_multi_d);
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index dad9b53cec..e3385b9dd4 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -58,6 +58,7 @@ add_subdirectory(elementwise_normalization)
 add_subdirectory(batchnorm)
 add_subdirectory(contraction)
 add_subdirectory(pool_fwd)
+add_subdirectory(batched_gemm_multi_d)
 if(GPU_TARGETS MATCHES "gfx1100")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/batched_gemm/batched_gemm_bf16.cpp b/test/batched_gemm/batched_gemm_bf16.cpp
index fa1652e997..5d12a1e956 100644
--- a/test/batched_gemm/batched_gemm_bf16.cpp
+++ b/test/batched_gemm/batched_gemm_bf16.cpp
@@ -5,6 +5,8 @@
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = ck::bhalf_t;
 using BDataType = ck::bhalf_t;
@@ -12,6 +14,8 @@ using CDataType = ck::bhalf_t;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM bf16: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm/batched_gemm_fp16.cpp b/test/batched_gemm/batched_gemm_fp16.cpp
index 3df4912aa0..a2b61d951a 100644
--- a/test/batched_gemm/batched_gemm_fp16.cpp
+++ b/test/batched_gemm/batched_gemm_fp16.cpp
@@ -5,6 +5,8 @@
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = ck::half_t;
 using BDataType = ck::half_t;
@@ -12,6 +14,8 @@ using CDataType = ck::half_t;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM fp16: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm/batched_gemm_fp32.cpp b/test/batched_gemm/batched_gemm_fp32.cpp
index 2d808441b9..2b18d166e6 100644
--- a/test/batched_gemm/batched_gemm_fp32.cpp
+++ b/test/batched_gemm/batched_gemm_fp32.cpp
@@ -5,6 +5,8 @@
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = float;
 using BDataType = float;
@@ -12,6 +14,8 @@ using CDataType = float;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM fp32: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm/batched_gemm_int8.cpp b/test/batched_gemm/batched_gemm_int8.cpp
index ed233a5ada..f607eaa84b 100644
--- a/test/batched_gemm/batched_gemm_int8.cpp
+++ b/test/batched_gemm/batched_gemm_int8.cpp
@@ -5,6 +5,8 @@
 
 #include "profiler/profile_batched_gemm_impl.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
 namespace {
 using ADataType = int8_t;
 using BDataType = int8_t;
@@ -12,6 +14,8 @@ using CDataType = int8_t;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 } // namespace
 
 int main()
@@ -23,21 +27,87 @@ int main()
 
     bool pass = true;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Row, Row>(
-               true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+    using namespace ck::tensor_operation::device;
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Row, Col, Row>(
-               true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Row, Row>(
-               true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Row,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Row,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
 
-    pass = pass &&
-           ck::profiler::profile_batched_gemm_impl<ADataType, BDataType, CDataType, Col, Col, Row>(
-               true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Row,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Row,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+
+    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
+                                                           BDataType,
+                                                           CDataType,
+                                                           Col,
+                                                           Col,
+                                                           Row,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           DeviceBatchedGemm<Col,
+                                                                             Col,
+                                                                             Row,
+                                                                             ADataType,
+                                                                             BDataType,
+                                                                             CDataType,
+                                                                             PassThrough,
+                                                                             PassThrough,
+                                                                             PassThrough>>(
+                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
 
     std::cout << "test BatchedGEMM int8: " << (pass ? "Pass" : "Fail") << std::endl;
     return pass ? 0 : 1;
diff --git a/test/batched_gemm_multi_d/CMakeLists.txt b/test/batched_gemm_multi_d/CMakeLists.txt
new file mode 100644
index 0000000000..45a306551f
--- /dev/null
+++ b/test/batched_gemm_multi_d/CMakeLists.txt
@@ -0,0 +1,5 @@
+# TODO: Enable for gfx90a after complier fix
+if(NOT GPU_TARGETS MATCHES "gfx90a")
+    add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
+    target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
+endif()
diff --git a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
new file mode 100644
index 0000000000..4a82654034
--- /dev/null
+++ b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <gtest/gtest.h>
+
+#include "profiler/profile_batched_gemm_impl.hpp"
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp"
+
+namespace {
+using F16 = ck::half_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+template <typename Tuple>
+class TestBatchedGemmMultiD : public ::testing::Test
+{
+    protected:
+    using ALayout = std::tuple_element_t<0, Tuple>;
+    using BLayout = std::tuple_element_t<1, Tuple>;
+    using CLayout = std::tuple_element_t<2, Tuple>;
+
+    static constexpr int M          = 512;
+    static constexpr int N          = 256;
+    static constexpr int K          = 128;
+    static constexpr int BatchCount = 3;
+
+    template <typename DataType>
+    void Run()
+    {
+        using namespace ck::tensor_operation::device;
+
+        const bool pass =
+            ck::profiler::profile_batched_gemm_impl<DataType,
+                                                    DataType,
+                                                    DataType,
+                                                    ALayout,
+                                                    BLayout,
+                                                    CLayout,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    DeviceBatchedGemmMultiD<ALayout,
+                                                                            BLayout,
+                                                                            Empty_Tuple,
+                                                                            CLayout,
+                                                                            DataType,
+                                                                            DataType,
+                                                                            Empty_Tuple,
+                                                                            DataType,
+                                                                            PassThrough,
+                                                                            PassThrough,
+                                                                            PassThrough>>(
+                true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+        EXPECT_TRUE(pass);
+    }
+};
+
+using KernelTypes = ::testing::Types<std::tuple<Row, Row, Row>,
+                                     std::tuple<Row, Col, Row>,
+                                     std::tuple<Col, Row, Row>,
+                                     std::tuple<Col, Col, Row>>;
+} // namespace
+
+TYPED_TEST_SUITE(TestBatchedGemmMultiD, KernelTypes);
+
+TYPED_TEST(TestBatchedGemmMultiD, f16) { this->template Run<F16>(); }
+
+TYPED_TEST(TestBatchedGemmMultiD, int8) { this->template Run<int8_t>(); }

From a35456a3f4ab8565c6a263e9383b1da84c33a3ed Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Mon, 12 Jun 2023 08:38:46 -0500
Subject: [PATCH 033/194] Fix arg order (#751)

---
 client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp | 2 +-
 client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp     | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
index 5965e9d1d6..8d2a8c234a 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu.cpp
@@ -76,7 +76,7 @@ int main(int argc, char* argv[])
         StrideA  = std::stoi(argv[4]);
         StrideB  = std::stoi(argv[5]);
         StrideD0 = std::stoi(argv[6]);
-        StrideE  = std::stoi(argv[8]);
+        StrideE  = std::stoi(argv[7]);
     }
     else
     {
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
index 319fdb0b03..c02df018fd 100644
--- a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu.cpp
@@ -72,7 +72,7 @@ int main(int argc, char* argv[])
 
         StrideA = std::stoi(argv[4]);
         StrideB = std::stoi(argv[5]);
-        StrideE = std::stoi(argv[8]);
+        StrideE = std::stoi(argv[6]);
     }
     else
     {

From 54b68eb343e9b29af024f4a5c4ac57e4e225c8fc Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Wed, 14 Jun 2023 16:06:56 -0500
Subject: [PATCH 034/194] Add generic kernel instances for
 ck::tensor_operation::device::DeviceGemmMultipleD (#741)

* Add generic instance gemm_add_add_fastgelu

* Add a client example for generic gemm_add_add_fastgelu

* Update CMakeLists

* Format

* Format

* Add generic instance gemm_add_fastgelu

* Format

* Add a gemm_add_fastgelu client example

* Format

* Add generic instance gemm_fastgelu

* Format

* Fix argument order

* Add gemm_fastgelu client example

* Add exceptions if argument is not supported
---
 .../02_gemm_add_add_fastgelu/CMakeLists.txt   |  14 ++
 .../gemm_add_add_fastgelu_generic.cpp         | 176 ++++++++++++++++++
 .../gemm_add_fastgelu_generic.cpp             | 169 +++++++++++++++++
 .../gemm_fastgelu_generic.cpp                 | 162 ++++++++++++++++
 ...16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp |  14 ++
 ...16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp |  14 ++
 ...16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp |  14 ++
 ...16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp |  14 ++
 ...e_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp |  14 ++
 ...e_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp |  14 ++
 ...e_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp |  14 ++
 ...e_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp |  14 ++
 ..._shuffle_f16_f16_f16_km_kn_mn_instance.cpp |  12 ++
 ..._shuffle_f16_f16_f16_km_nk_mn_instance.cpp |  12 ++
 ..._shuffle_f16_f16_f16_mk_kn_mn_instance.cpp |  12 ++
 ..._shuffle_f16_f16_f16_mk_nk_mn_instance.cpp |  12 ++
 16 files changed, 681 insertions(+)
 create mode 100644 client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp
 create mode 100644 client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp
 create mode 100644 client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp

diff --git a/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt b/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
index b7b724ccc4..ba29520222 100644
--- a/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/client_example/02_gemm_add_add_fastgelu/CMakeLists.txt
@@ -11,3 +11,17 @@ target_link_libraries(client_gemm_fastgelu PRIVATE composable_kernel::device_ope
 
 add_dependencies(client_gemm_fastgelu_examples client_gemm_add_add_fastgelu client_gemm_add_fastgelu
                  client_gemm_fastgelu)
+
+add_custom_target(client_gemm_fastgelu_generic_examples)
+
+add_executable(client_gemm_add_add_fastgelu_generic gemm_add_add_fastgelu_generic.cpp)
+target_link_libraries(client_gemm_add_add_fastgelu_generic PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_add_fastgelu_generic gemm_add_fastgelu_generic.cpp)
+target_link_libraries(client_gemm_add_fastgelu_generic PRIVATE composable_kernel::device_operations)
+
+add_executable(client_gemm_fastgelu_generic gemm_fastgelu_generic.cpp)
+target_link_libraries(client_gemm_fastgelu_generic PRIVATE composable_kernel::device_operations)
+
+add_dependencies(client_gemm_fastgelu_generic_examples client_gemm_add_add_fastgelu_generic 
+                 client_gemm_add_fastgelu_generic client_gemm_fastgelu_generic)
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp
new file mode 100644
index 0000000000..2ed942f0ad
--- /dev/null
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_add_fastgelu_generic.cpp
@@ -0,0 +1,176 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_add_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough    = ck::tensor_operation::element_wise::PassThrough;
+using AddAddFastGelu = ck::tensor_operation::element_wise::AddAddFastGelu;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddAddFastGelu;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using D0DataType = F16;
+using D1DataType = F16;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using D1Layout = Row;
+using ELayout  = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA  = 4096;
+    ck::index_t StrideB  = 4096;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideD1 = 4096;
+    ck::index_t StrideE  = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 9)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA  = std::stoi(argv[4]);
+        StrideB  = std::stoi(argv[5]);
+        StrideD0 = std::stoi(argv[6]);
+        StrideD1 = std::stoi(argv[7]);
+        StrideE  = std::stoi(argv[8]);
+    }
+    else
+    {
+        printf("arg1 to 8: M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) *
+                                      f_matrix_space_size(M, N, StrideD0, D0Layout{}));
+    SimpleDeviceMem d1_m_n_device_buf(sizeof(D1DataType) *
+                                      f_matrix_space_size(M, N, StrideD1, D1Layout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
+        ALayout,
+        BLayout,
+        ck::Tuple<D0Layout, D1Layout>,
+        ELayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<D0DataType, D1DataType>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::AddAddFastGelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    // get generic instance
+    auto& op_ptr = op_ptrs[0];
+
+    std::cout << "Run the generic instance without timing: " << op_ptr->GetTypeString()
+              << std::endl;
+
+    // run the generic instance
+    auto argument_ptr =
+        op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 2>{d0_m_n_device_buf.GetDeviceBuffer(),
+                                                               d1_m_n_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    M,
+                                    N,
+                                    K,
+                                    StrideA,
+                                    StrideB,
+                                    std::array<ck::index_t, 2>{StrideD0, StrideD1},
+                                    StrideE,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+    if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+    {
+        invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+    }
+    else
+    {
+        throw std::runtime_error(
+            "Generic instance should be suitable for various input lengths/strides");
+    }
+
+    std::cout << "Done" << std::endl;
+
+    return 0;
+}
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp
new file mode 100644
index 0000000000..644b428fc9
--- /dev/null
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_add_fastgelu_generic.cpp
@@ -0,0 +1,169 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_add_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using AddFastGelu = ck::tensor_operation::element_wise::AddFastGelu;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AddFastGelu;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using D0DataType = F16;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using D0Layout = Row;
+using ELayout  = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA  = 4096;
+    ck::index_t StrideB  = 4096;
+    ck::index_t StrideD0 = 0;
+    ck::index_t StrideE  = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 8)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA  = std::stoi(argv[4]);
+        StrideB  = std::stoi(argv[5]);
+        StrideD0 = std::stoi(argv[6]);
+        StrideE  = std::stoi(argv[7]);
+    }
+    else
+    {
+        printf("arg1 to 7: M, N, K, StrideA, StrideB, StrideD0, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem d0_m_n_device_buf(sizeof(D0DataType) *
+                                      f_matrix_space_size(M, N, StrideD0, D0Layout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
+        ALayout,
+        BLayout,
+        ck::Tuple<D0Layout>,
+        ELayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<D0DataType>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::AddFastGelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    // get generic instance
+    auto& op_ptr = op_ptrs[0];
+
+    std::cout << "Run the generic instance without timing: " << op_ptr->GetTypeString()
+              << std::endl;
+
+    // run the generic instance
+    auto argument_ptr =
+        op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d0_m_n_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    M,
+                                    N,
+                                    K,
+                                    StrideA,
+                                    StrideB,
+                                    std::array<ck::index_t, 1>{StrideD0},
+                                    StrideE,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+    if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+    {
+        invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+    }
+    else
+    {
+        throw std::runtime_error(
+            "Generic instance should be suitable for various input lengths/strides");
+    }
+
+    std::cout << "Done" << std::endl;
+
+    return 0;
+}
diff --git a/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp
new file mode 100644
index 0000000000..482e93b421
--- /dev/null
+++ b/client_example/02_gemm_add_add_fastgelu/gemm_fastgelu_generic.cpp
@@ -0,0 +1,162 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+#include <stdexcept>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using FastGelu    = ck::tensor_operation::element_wise::FastGelu;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = FastGelu;
+
+using ADataType = F16;
+using BDataType = F16;
+using EDataType = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using ELayout = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideE = 4096;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 7)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA = std::stoi(argv[4]);
+        StrideB = std::stoi(argv[5]);
+        StrideE = std::stoi(argv[6]);
+    }
+    else
+    {
+        printf("arg1 to 6: M, N, K, StrideA, StrideB, StrideE\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem e_device_buf(sizeof(EDataType) * f_matrix_space_size(M, N, StrideE, ELayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmMultipleD<
+        ALayout,
+        BLayout,
+        ck::Tuple<>,
+        ELayout,
+        ADataType,
+        BDataType,
+        ck::Tuple<>,
+        EDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::FastGelu>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    // get generic instance
+    auto& op_ptr = op_ptrs[0];
+
+    std::cout << "Run the generic instance without timing: " << op_ptr->GetTypeString()
+              << std::endl;
+
+    // run the generic instance
+    auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                    b_device_buf.GetDeviceBuffer(),
+                                                    {},
+                                                    e_device_buf.GetDeviceBuffer(),
+                                                    M,
+                                                    N,
+                                                    K,
+                                                    StrideA,
+                                                    StrideB,
+                                                    {},
+                                                    StrideE,
+                                                    a_element_op,
+                                                    b_element_op,
+                                                    cde_element_op);
+
+    auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+    if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+    {
+        invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+    }
+    else
+    {
+        throw std::runtime_error(
+            "Generic instance should be suitable for various input lengths/strides");
+    }
+
+    std::cout << "Done" << std::endl;
+
+    return 0;
+}
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
index 7b65f8737a..125fbc21a2 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[k, m], b[k, n], d0[m, n], d1[m, n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -139,6 +150,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
index a9fef5c60d..cc33692d73 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[k, m], b[n, k], d0[m, n], d1[m, n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -139,6 +150,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
index c3b1dfccab..704787a080 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[k, n], d0[m, n], d1[m, n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -139,6 +150,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
index 8338b34a4d..d64c9ec5e1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -36,6 +36,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[n, k], d0[m, n], d1[m ,n]
+using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Row_Tuple,    Row,   F16,   F16,     F32,      F32, F16_F16_Tuple,   F16, PassThrough, PassThrough, AddAddFastGelu, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -130,6 +141,9 @@ void add_device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn
                                                     PassThrough,
                                                     AddAddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances,
         device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
index 357dc91aa1..e68bd8e7e4 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0)
 // outout: e[m, n]
 // input: a[k, m], b[k, n], d0[m, n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -123,6 +134,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
index 65b94d0879..5aaa2e8fe5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[k, m], b[n, k], d0[m, n], d1[m, n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -123,6 +134,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
index 00cd07bab8..7a2a3dbaf3 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[k, n], d0[m, n], d1[m, n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -123,6 +134,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
index 9955a206b6..fa33609978 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
@@ -21,6 +21,17 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b), d0, d1)
 // outout: e[m, n]
 // input: a[m, k], b[n, k], d0[m, n], d1[m ,n]
+using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_generic_instance =
+    std::tuple<
+        // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col,     Row_Tuple,    Row,   F16,   F16,     F32,      F32,     F16_Tuple,   F16, PassThrough, PassThrough,    AddFastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
 using device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -114,6 +125,9 @@ void add_device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_inst
                                                     PassThrough,
                                                     AddFastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
index c5aa59f911..803c44c7f5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[k, m], b[k, n]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Row,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -122,6 +132,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
index e71b269b8d..9b9ef3db24 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[k, m], b[k, n]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Col,    Col,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -122,6 +132,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
index fdf63f8112..1a0b6c9d1c 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[m, k], b[k, n]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -122,6 +132,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances{});
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
index 4c98a9d5e9..18b1c0e993 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -21,6 +21,16 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 // e = elementwise((a * b))
 // outout: e[m, n]
 // input: a[m, k], b[n, k]
+using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_generic_instance = std::tuple<
+    // clang-format off
+        //##############################|      A|      B|            Ds|      E| AData| BData| AccData| CShuffle|        DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
+        //##############################| Layout| Layout|        Layout| Layout|  Type|  Type|    Type| DataType|          Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
+        //##############################|       |       |              |       |      |      |        |         |              |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
+        // pipeline v1, 1 wave
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col,   Empty_Tuple,    Row,   F16,   F16,     F32,      F32,   Empty_Tuple,   F16, PassThrough, PassThrough,       FastGelu, GemmMNKPadding,        1,    64,    16,    16,    32,   8,   8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
 using device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // clang-format off
         //##############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -113,6 +123,8 @@ void add_device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(
                                                     PassThrough,
                                                     FastGelu>>>& instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_generic_instance{});
     add_device_operation_instances(
         instances, device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances{});
     add_device_operation_instances(

From d1838d328c2bf7119af5dddc522ccf37dba33c3f Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 14 Jun 2023 16:44:13 -0700
Subject: [PATCH 035/194] Fix the daily CI job with latest staging compiler.
 (#753)

* fix CI builds with latest staging compiler

* remove mount flags from dockerfile
---
 Dockerfile | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index 710db05c06..bc47768060 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -12,20 +12,20 @@ RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 RUN chmod 1777 /tmp
 RUN apt-get update
 RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
-RUN --mount=type=ssh if [ "$ROCMVERSION" != "5.6" ]; then \
+RUN if [ "$ROCMVERSION" != "5.6" ]; then \
         wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
         sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"; \
     elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "" ]; then \
          sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
          apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc3" ]; then \
+         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914 && \
+         amdgpu-install -y --usecase=rocm --no-dkms; \
+    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc3" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
          sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.6-20.04-1_all.deb" && \
          apt update && apt-get install -y ./amdgpu-install-internal_5.6-20.04-1_all.deb && \
          sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-45  > /etc/apt/sources.list.d/rocm-build.list' && \
-         amdgpu-repo --amdgpu-build=1602498; \
+         amdgpu-repo --amdgpu-build=1602498 && amdgpu-install -y --usecase=rocm --no-dkms; \
     fi
-RUN amdgpu-install -y --usecase=rocm --no-dkms
 
 RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
@@ -105,7 +105,7 @@ ENV compiler_commit=$compiler_commit
 RUN sh -c "echo compiler version = '$compiler_version'"
 RUN sh -c "echo compiler commit = '$compiler_commit'"
 
-RUN --mount=type=ssh if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; then \
+RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && mkdir build && cd build && \
         cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
@@ -113,7 +113,7 @@ RUN --mount=type=ssh if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler
     else echo "using the release compiler"; \
     fi
 
-RUN --mount=type=ssh if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ]; then \
+RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
         cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \

From c5f6ec842c83a386b78dc0098fcdf081586df309 Mon Sep 17 00:00:00 2001
From: Qianfeng <qianfeng.zhang@amd.com>
Date: Thu, 15 Jun 2023 23:13:59 +0800
Subject: [PATCH 036/194] Using number of compute units to set  gridSize (#754)

* Add getAvailableComputeUnitCount() interface

* Use available number of compute units to set kernel grid size
---
 include/ck/host_utility/stream_utility.hpp    | 43 ++++++++++++
 .../impl/device_elementwise_2d_impl.hpp       | 69 +++++++++----------
 .../device/impl/device_elementwise_impl.hpp   | 43 ++++++------
 3 files changed, 97 insertions(+), 58 deletions(-)
 create mode 100644 include/ck/host_utility/stream_utility.hpp

diff --git a/include/ck/host_utility/stream_utility.hpp b/include/ck/host_utility/stream_utility.hpp
new file mode 100644
index 0000000000..ef05f2e26b
--- /dev/null
+++ b/include/ck/host_utility/stream_utility.hpp
@@ -0,0 +1,43 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <hip/hip_runtime.h>
+
+#include "ck/stream_config.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
+
+static int getAvailableComputeUnitCount(const StreamConfig& stream_config)
+{
+    constexpr int MAX_MASK_DWORDS = 64;
+
+    // assume at most 64*32 = 2048 CUs
+    uint32_t cuMask[MAX_MASK_DWORDS];
+
+    for(int i = 0; i < MAX_MASK_DWORDS; i++)
+        cuMask[i] = 0;
+
+    auto countSetBits = [](uint32_t dword) {
+        int count = 0;
+
+        while(dword != 0)
+        {
+            if(dword & 0x1)
+                count++;
+
+            dword = dword >> 1;
+        };
+
+        return (count);
+    };
+
+    hip_check_error(hipExtStreamGetCUMask(stream_config.stream_id_, MAX_MASK_DWORDS, &cuMask[0]));
+
+    int ret = 0;
+
+    for(int i = 0; i < MAX_MASK_DWORDS; i++)
+        ret += countSetBits(cuMask[i]);
+
+    return (ret);
+};
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
index c5f90e40fe..02ef29e32d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_2d_impl.hpp
@@ -13,6 +13,7 @@
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 
 #include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -171,10 +172,7 @@ struct DeviceElementwise2dImpl : public DeviceElementwise<InDataTypeTuple,
               inStridesArray_(inStridesArray),
               outStridesArray_(outStridesArray),
               elementwise_op_(elementwise_op),
-              blockSize_(256),
-              gridSize_(120), // FIXME - Calculate the grid size by number of CU in the future
-              num_threads_m_((gridSize_ * blockSize_) / 16),
-              num_threads_n_(16)
+              blockSize_(256)
         {
             static_assert(NumDim_m > 0, "");
             static_assert(NumDim_n > 0, "");
@@ -192,34 +190,10 @@ struct DeviceElementwise2dImpl : public DeviceElementwise<InDataTypeTuple,
                     return static_cast<DataType*>(out_dev_buffers[I.value]);
                 },
                 Number<NumOutput>{});
-
-            in_grid_2d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_MN(lengths,
-                                             inStridesArray[I.value],
-                                             gridSize_,
-                                             blockSize_,
-                                             num_threads_m_,
-                                             num_threads_n_);
-                },
-                Number<NumInput>{});
-
-            out_grid_2d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_MN(lengths,
-                                             outStridesArray[I.value],
-                                             gridSize_,
-                                             blockSize_,
-                                             num_threads_m_,
-                                             num_threads_n_);
-                },
-                Number<NumOutput>{});
         }
 
         InDataTypePointerTuple in_dev_buffers_;
         OutDataTypePointerTuple out_dev_buffers_;
-        InGrid2dDescTuple in_grid_2d_desc_tuple_;
-        OutGrid2dDescTuple out_grid_2d_desc_tuple_;
 
         std::array<index_t, NumDim> lengths_;
         std::array<std::array<index_t, NumDim>, NumInput> inStridesArray_;
@@ -227,15 +201,38 @@ struct DeviceElementwise2dImpl : public DeviceElementwise<InDataTypeTuple,
 
         ElementwiseOperation elementwise_op_;
         index_t blockSize_;
-        index_t gridSize_;
-        index_t num_threads_m_;
-        index_t num_threads_n_;
     };
 
     struct Invoker : public BaseInvoker
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+            index_t gridSize      = getAvailableComputeUnitCount(stream_config);
+            index_t num_threads_m = (gridSize * arg.blockSize_) / 16;
+            index_t num_threads_n = 16;
+
+            auto in_grid_2d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_MN(arg.lengths_,
+                                             arg.inStridesArray_[I.value],
+                                             gridSize,
+                                             arg.blockSize_,
+                                             num_threads_m,
+                                             num_threads_n);
+                },
+                Number<NumInput>{});
+
+            auto out_grid_2d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_MN(arg.lengths_,
+                                             arg.outStridesArray_[I.value],
+                                             gridSize,
+                                             arg.blockSize_,
+                                             num_threads_m,
+                                             num_threads_n);
+                },
+                Number<NumOutput>{});
+
             const auto kernel = kernel_elementwise_2d<GridwiseElementwise,
                                                       InGrid2dDescTuple,
                                                       OutGrid2dDescTuple,
@@ -245,16 +242,16 @@ struct DeviceElementwise2dImpl : public DeviceElementwise<InDataTypeTuple,
 
             float elapsed_time = launch_and_time_kernel(stream_config,
                                                         kernel,
-                                                        dim3(arg.gridSize_),
+                                                        dim3(gridSize),
                                                         dim3(arg.blockSize_),
                                                         0,
-                                                        arg.in_grid_2d_desc_tuple_,
-                                                        arg.out_grid_2d_desc_tuple_,
+                                                        in_grid_2d_desc_tuple,
+                                                        out_grid_2d_desc_tuple,
                                                         arg.in_dev_buffers_,
                                                         arg.out_dev_buffers_,
                                                         arg.elementwise_op_,
-                                                        arg.num_threads_m_,
-                                                        arg.num_threads_n_);
+                                                        num_threads_m,
+                                                        num_threads_n);
             return elapsed_time;
         }
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
index 5618fba517..e203468c61 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
@@ -13,6 +13,7 @@
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 
 #include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -144,8 +145,7 @@ struct DeviceElementwiseImpl
               inStridesArray_(inStridesArray),
               outStridesArray_(outStridesArray),
               elementwise_op_(elementwise_op),
-              blockSize_(256),
-              gridSize_(120) // FIXME - Calculate the grid size by number of CU in the future
+              blockSize_(256)
         {
             in_dev_buffers_ = generate_tuple(
                 [&](auto I) {
@@ -160,26 +160,10 @@ struct DeviceElementwiseImpl
                     return static_cast<DataType*>(out_dev_buffers[I.value]);
                 },
                 Number<NumOutput>{});
-
-            in_grid_1d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_M(
-                        lengths, inStridesArray[I.value], gridSize_, blockSize_);
-                },
-                Number<NumInput>{});
-
-            out_grid_1d_desc_tuple_ = generate_tuple(
-                [&](auto I) {
-                    return MakeDescriptor_M(
-                        lengths, outStridesArray[I.value], gridSize_, blockSize_);
-                },
-                Number<NumOutput>{});
         }
 
         InDataTypePointerTuple in_dev_buffers_;
         OutDataTypePointerTuple out_dev_buffers_;
-        InGrid1dDescTuple in_grid_1d_desc_tuple_;
-        OutGrid1dDescTuple out_grid_1d_desc_tuple_;
 
         std::array<index_t, NumDim> lengths_;
         std::array<std::array<index_t, NumDim>, NumInput> inStridesArray_;
@@ -187,13 +171,28 @@ struct DeviceElementwiseImpl
 
         ElementwiseOperation elementwise_op_;
         index_t blockSize_;
-        index_t gridSize_;
     };
 
     struct Invoker : public BaseInvoker
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+            index_t gridSize = getAvailableComputeUnitCount(stream_config);
+
+            auto in_grid_1d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_M(
+                        arg.lengths_, arg.inStridesArray_[I.value], gridSize, arg.blockSize_);
+                },
+                Number<NumInput>{});
+
+            auto out_grid_1d_desc_tuple = generate_tuple(
+                [&](auto I) {
+                    return MakeDescriptor_M(
+                        arg.lengths_, arg.outStridesArray_[I.value], gridSize, arg.blockSize_);
+                },
+                Number<NumOutput>{});
+
             const auto kernel = kernel_elementwise_1d<GridwiseElementwise,
                                                       InGrid1dDescTuple,
                                                       OutGrid1dDescTuple,
@@ -203,11 +202,11 @@ struct DeviceElementwiseImpl
 
             float elapsed_time = launch_and_time_kernel(stream_config,
                                                         kernel,
-                                                        dim3(arg.gridSize_),
+                                                        dim3(gridSize),
                                                         dim3(arg.blockSize_),
                                                         0,
-                                                        arg.in_grid_1d_desc_tuple_,
-                                                        arg.out_grid_1d_desc_tuple_,
+                                                        in_grid_1d_desc_tuple,
+                                                        out_grid_1d_desc_tuple,
                                                         arg.in_dev_buffers_,
                                                         arg.out_dev_buffers_,
                                                         arg.elementwise_op_);

From 309b1c64618c714b7f47ceb038bba68b61fa4e4e Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Thu, 15 Jun 2023 08:19:33 -0700
Subject: [PATCH 037/194] Fixed Weight layout of grouped_conv 3d fwd (#743)

* Changed wei layout

* changed layout for examples

* fixed client example

---------

Co-authored-by: root <root@ctr-ubbsmc15.amd.com>
---
 client_example/16_convnd_fwd/common.hpp       |  4 -
 .../16_convnd_fwd/conv3d_fwd_fp16.cpp         |  4 +-
 .../16_convnd_fwd/conv3d_fwd_fp32.cpp         |  4 +-
 .../gpu/grouped_convolution_forward.hpp       | 28 +++---
 .../gpu/grouped_conv3d_fwd/CMakeLists.txt     |  8 +-
 ...dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp} | 88 +++++++++----------
 ...xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp} | 88 +++++++++----------
 ...xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp} | 88 +++++++++----------
 ...dl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp} | 88 +++++++++----------
 9 files changed, 198 insertions(+), 202 deletions(-)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp => device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp} (91%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp => device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp} (91%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp => device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp} (91%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp => device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp} (91%)

diff --git a/client_example/16_convnd_fwd/common.hpp b/client_example/16_convnd_fwd/common.hpp
index a6bb5aa65b..449c9466e8 100644
--- a/client_example/16_convnd_fwd/common.hpp
+++ b/client_example/16_convnd_fwd/common.hpp
@@ -141,14 +141,10 @@ bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialD
                 std::next(rbegin(in_strides)),
                 std::next(rbegin(in_strides), NumDimSpatial + 1));
 
-    std::rotate(
-        std::next(rbegin(wei_lengths)), std::next(rbegin(wei_lengths), 2), rend(wei_lengths));
     std::rotate(rbegin(wei_lengths),
                 std::next(rbegin(wei_lengths)),
                 std::next(rbegin(wei_lengths), NumDimSpatial + 1));
 
-    std::rotate(
-        std::next(rbegin(wei_strides)), std::next(rbegin(wei_strides), 2), rend(wei_strides));
     std::rotate(rbegin(wei_strides),
                 std::next(rbegin(wei_strides)),
                 std::next(rbegin(wei_strides), NumDimSpatial + 1));
diff --git a/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp b/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp
index 10f914bbee..d4455df628 100644
--- a/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp
+++ b/client_example/16_convnd_fwd/conv3d_fwd_fp16.cpp
@@ -11,7 +11,7 @@ using WeiDataType = ck::half_t;
 using OutDataType = ck::half_t;
 
 using InLayout  = ck::tensor_layout::convolution::NDHWGC;
-using WeiLayout = ck::tensor_layout::convolution::KZYXGC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
 using OutLayout = ck::tensor_layout::convolution::NDHWGK;
 
 static constexpr ck::index_t NumDimSpatial = 3;
@@ -38,7 +38,7 @@ int main()
                                 InLayout,
                                 WeiLayout,
                                 OutLayout>(
-               {N, Di, Hi, Wi, G, C}, {K, Z, Y, X, G, C}, {N, Do, Ho, Wo, G, K})
+               {N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp b/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp
index 43c98f1e9b..7e8c98b603 100644
--- a/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp
+++ b/client_example/16_convnd_fwd/conv3d_fwd_fp32.cpp
@@ -11,7 +11,7 @@ using WeiDataType = float;
 using OutDataType = float;
 
 using InLayout  = ck::tensor_layout::convolution::NDHWGC;
-using WeiLayout = ck::tensor_layout::convolution::KZYXGC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
 using OutLayout = ck::tensor_layout::convolution::NDHWGK;
 
 static constexpr ck::index_t NumDimSpatial = 3;
@@ -38,7 +38,7 @@ int main()
                                 InLayout,
                                 WeiLayout,
                                 OutLayout>(
-               {N, Di, Hi, Wi, G, C}, {K, Z, Y, X, G, C}, {N, Do, Ho, Wo, G, K})
+               {N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index a82ec543c6..627a5ae2aa 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -245,11 +245,11 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-// grouped conv3d forward, NDHWGC/KZYXGC/NDHWGK
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances(
+// grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               BF16,
@@ -260,10 +260,10 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               F16,
@@ -274,10 +274,10 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               F32,
@@ -288,10 +288,10 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               int8_t,
@@ -433,28 +433,28 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
-                          is_same_v<WeiLayout, KZYXGC> && is_same_v<OutLayout, NDHWGK>)
+                          is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
         {
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
-                add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
             {
-                add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&
                               is_same_v<OutDataType, ck::bhalf_t>)
             {
-                add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
-                add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(op_ptrs);
             }
         }
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
index 90efc09ee7..cd209dbf9e 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
@@ -4,8 +4,8 @@ add_instance_library(device_grouped_conv3d_fwd_instance
    device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
    device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
 
-   device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp
-   device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp
-   device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp
-   device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
similarity index 91%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
index 5e5dbc53cd..d2cacf4539 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -26,7 +26,7 @@ template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
 using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using KZYXGC = ck::tensor_layout::convolution::KZYXGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
 using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
@@ -43,7 +43,7 @@ static constexpr auto ConvFwd1x1S1P0 =
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances =
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances =
     std::tuple<
         // clang-format off
         // Default
@@ -51,64 +51,64 @@ using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances =
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
 
         // Filter1x1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
 
         // Filter1x1Stride1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
         // clang-format on
         >;
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               BF16,
@@ -120,7 +120,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_bf16_instances{});
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
similarity index 91%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
index a88fe4af63..6354f01112 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -26,7 +26,7 @@ template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
 using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using KZYXGC = ck::tensor_layout::convolution::KZYXGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
 using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
@@ -43,7 +43,7 @@ static constexpr auto ConvFwd1x1S1P0 =
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances =
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances =
     std::tuple<
         // clang-format off
         // Default
@@ -51,64 +51,64 @@ using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances =
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
 
         // Filter1x1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
 
         // Filter1x1Stride1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
         // clang-format on
         >;
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               F16,
@@ -120,7 +120,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f16_instances{});
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
similarity index 91%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
index f6e7e5b28e..381ac5ef08 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -25,7 +25,7 @@ template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
 using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using KZYXGC = ck::tensor_layout::convolution::KZYXGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
 using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
@@ -42,7 +42,7 @@ static constexpr auto ConvFwd1x1S1P0 =
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances =
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances =
     std::tuple<
         // clang-format off
         // Default
@@ -50,64 +50,64 @@ using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances =
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
 
         // Filter1x1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
 
         // Filter1x1Stride1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
         // clang-format on
         >;
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               F32,
@@ -119,7 +119,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_f32_instances{});
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
similarity index 91%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
index 3d303a3fa7..6bd53d8693 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
@@ -23,7 +23,7 @@ template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
 
 using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using KZYXGC = ck::tensor_layout::convolution::KZYXGC;
+using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
 using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
@@ -40,71 +40,71 @@ static constexpr auto ConvFwd1x1S1P0 =
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instances = std::tuple<
+using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances = std::tuple<
     // clang-format off
         // Default
         //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
 
         // Filter1x1Pad0
         //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
                                                                                                                                                                                                                 
         // Filter1x1Stride1Pad0                                                                                                                                                                                 
         //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
         //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, KZYXGC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
     // clang-format on
     >;
 
-void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instances(
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
-                                                              KZYXGC,
+                                                              GKZYXC,
                                                               Empty_Tuple,
                                                               NDHWGK,
                                                               int8_t,
@@ -116,7 +116,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_kzyxgc_ndhwgk_int8_instances{});
+        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances{});
 }
 
 } // namespace instance

From 027e46ee82cdb8a02313985361333c78f281b887 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 15 Jun 2023 08:20:59 -0700
Subject: [PATCH 038/194] Enable gfx941 and gfx942 architectures. (#752)

* enable gfx941/942 targets

* fix clang format

* fix the cmake logic for multiple targets

* fix cmake syntax for looping over targets

* add gfx941/942 support for gemm_xdl instances
---
 Jenkinsfile                                   |  4 ++--
 example/02_gemm_bilinear/CMakeLists.txt       | 19 +++++++++++----
 example/03_gemm_bias_relu/CMakeLists.txt      |  9 +++++--
 .../04_gemm_add_add_fastgelu/CMakeLists.txt   |  9 +++++--
 example/09_convnd_fwd/CMakeLists.txt          | 11 ++++++---
 .../CMakeLists.txt                            |  9 +++++--
 example/14_gemm_quantization/CMakeLists.txt   |  9 +++++--
 .../CMakeLists.txt                            |  9 +++++--
 example/17_convnd_bwd_data/CMakeLists.txt     |  9 +++++--
 example/18_batched_gemm_reduce/CMakeLists.txt | 10 +++++---
 .../20_grouped_conv_bwd_weight/CMakeLists.txt |  9 +++++--
 example/21_gemm_layernorm/CMakeLists.txt      |  9 +++++--
 .../CMakeLists.txt                            | 24 ++++++++++++++-----
 example/31_batched_gemm_gemm/CMakeLists.txt   | 21 ++++++++++++----
 example/35_splitK_gemm/CMakeLists.txt         |  9 +++++--
 .../CMakeLists.txt                            |  9 +++++--
 .../40_conv2d_fwd_quantization/CMakeLists.txt |  9 +++++--
 .../41_grouped_conv_conv_fwd/CMakeLists.txt   | 21 ++++++++++++----
 .../CMakeLists.txt                            |  9 +++++--
 include/ck/ck.hpp                             | 18 ++++++++------
 .../device_gemm_xdl_waveletmodel_cshuffle.hpp |  5 ++--
 ..._conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp |  5 ++--
 ...gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  5 ++--
 ...tk_contraction_multiple_d_xdl_cshuffle.hpp |  5 ++--
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |  5 ++--
 .../device_batched_gemm_e_permute_xdl.hpp     |  2 +-
 .../device_batched_gemm_gemm_xdl_cshuffle.hpp |  5 ++--
 .../impl/device_batched_gemm_multi_d_xdl.hpp  |  5 ++--
 ...ultiple_d_gemm_multiple_d_xdl_cshuffle.hpp |  5 ++--
 ...evice_batched_gemm_reduce_xdl_cshuffle.hpp |  2 +-
 ...gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  5 ++--
 ...batched_gemm_softmax_gemm_xdl_cshuffle.hpp |  5 ++--
 .../device/impl/device_batched_gemm_xdl.hpp   |  2 +-
 ...ce_contraction_multiple_d_xdl_cshuffle.hpp |  5 ++--
 ...evice_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp |  2 +-
 .../device/impl/device_gemm_multiple_d_dl.hpp |  5 ++--
 ...gemm_multiple_d_layernorm_xdl_cshuffle.hpp |  5 ++--
 ...emm_multiple_d_multiple_r_xdl_cshuffle.hpp |  5 ++--
 .../device_gemm_multiple_d_xdl_cshuffle.hpp   |  5 ++--
 .../device/impl/device_gemm_xdl_cshuffle.hpp  |  3 ++-
 .../device_gemm_xdl_layernorm_cshuffle.hpp    |  3 ++-
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |  5 ++--
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  2 +-
 ...bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp |  2 +-
 ...fwd_multiple_d_multiple_r_xdl_cshuffle.hpp |  5 ++--
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |  5 ++--
 .../device_grouped_gemm_multiple_d_dl.hpp     |  2 +-
 .../device/impl/device_grouped_gemm_xdl.hpp   |  2 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp |  2 +-
 ...e_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp |  2 +-
 .../gridwise_gemm_reduce_xdl_cshuffle_v1.hpp  |  2 +-
 .../grid/gridwise_gemm_xdl_cshuffle_v1.hpp    |  4 ++--
 ...ridwise_gemm_xdl_layernorm_cshuffle_v1.hpp |  2 +-
 .../grid/gridwise_gemm_xdlops_bwd_weight.hpp  |  2 +-
 .../gridwise_gemm_xdlops_skip_b_lds_v1.hpp    |  2 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |  4 ++--
 .../gpu/grid/gridwise_gemm_xdlops_v2r4.hpp    |  2 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  2 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r1.hpp    |  2 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r2.hpp    |  2 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r3.hpp    |  2 +-
 include/ck/utility/amd_xdlops.hpp             |  2 +-
 script/cmake-ck-dev.sh                        |  2 +-
 script/cmake-ck-release.sh                    |  2 +-
 test/batched_gemm/CMakeLists.txt              |  9 +++++--
 test/batched_gemm_gemm/CMakeLists.txt         | 10 +++++---
 test/batched_gemm_reduce/CMakeLists.txt       |  9 +++++--
 test/batched_gemm_softmax_gemm/CMakeLists.txt | 10 +++++---
 .../CMakeLists.txt                            |  9 +++++--
 test/contraction/CMakeLists.txt               |  9 +++++--
 test/convnd_bwd_data/CMakeLists.txt           |  9 +++++--
 test/convnd_fwd/CMakeLists.txt                |  9 +++++--
 test/gemm_layernorm/CMakeLists.txt            |  9 +++++--
 test/gemm_split_k/CMakeLists.txt              |  9 +++++--
 test/grouped_convnd_bwd_weight/CMakeLists.txt | 11 ++++++---
 test/grouped_gemm/CMakeLists.txt              |  9 +++++--
 76 files changed, 346 insertions(+), 150 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index fbff349fc3..8cfc5f8048 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -695,8 +695,8 @@ pipeline {
                     }
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
diff --git a/example/02_gemm_bilinear/CMakeLists.txt b/example/02_gemm_bilinear/CMakeLists.txt
index eecec2437c..7ea4564d1b 100644
--- a/example/02_gemm_bilinear/CMakeLists.txt
+++ b/example/02_gemm_bilinear/CMakeLists.txt
@@ -1,6 +1,17 @@
-if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
     add_example_executable(example_gemm_bilinear_wmma_fp16 gemm_bilinear_wmma_fp16.cpp)
-endif()
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+   set(target 1)
+ endif()
+endforeach()
+
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
     add_example_executable(example_gemm_bilinear_xdl_fp16 gemm_bilinear_xdl_fp16.cpp)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/03_gemm_bias_relu/CMakeLists.txt b/example/03_gemm_bias_relu/CMakeLists.txt
index 8834a910f0..2f5cba924d 100644
--- a/example/03_gemm_bias_relu/CMakeLists.txt
+++ b/example/03_gemm_bias_relu/CMakeLists.txt
@@ -1,3 +1,8 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_example_executable(example_gemm_bias_relu_xdl_fp16 gemm_bias_relu_xdl_fp16.cpp)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/04_gemm_add_add_fastgelu/CMakeLists.txt b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
index a706830b68..447fa9871b 100644
--- a/example/04_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_custom_target(example_gemm_add_add_fastgelu_xdl)
 
     add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
@@ -16,4 +19,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
         add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
     endif(USE_BITINT_EXTENSION_INT4)
     add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/09_convnd_fwd/CMakeLists.txt b/example/09_convnd_fwd/CMakeLists.txt
index 1bcf2d148d..90104c1630 100644
--- a/example/09_convnd_fwd/CMakeLists.txt
+++ b/example/09_convnd_fwd/CMakeLists.txt
@@ -1,12 +1,17 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
     add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
     add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
     add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
     # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
     add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
-endif()
+   set(target 1)
+ endif()
+endforeach()
+
 add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
 add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
 add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
-
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
index de26462f69..9577b4569a 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_convnd_fwd_reduce_xdl)
    add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
    add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
@@ -12,4 +15,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
       add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
       add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
    endif(USE_BITINT_EXTENSION_INT4)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/14_gemm_quantization/CMakeLists.txt b/example/14_gemm_quantization/CMakeLists.txt
index 584333e7bb..72bdff5ab4 100644
--- a/example/14_gemm_quantization/CMakeLists.txt
+++ b/example/14_gemm_quantization/CMakeLists.txt
@@ -2,7 +2,12 @@
 add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
 
 # xdlops
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_gemm_xdl_bias_relu_quantization_int8 gemm_xdl_bias_relu_quantization_int8.cpp)
    add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
index 0f1ca777c5..a42b427c61 100644
--- a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+++ b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_gemm_reduce_xdl)
    add_custom_target(example_gemm_reduce_xdl_max)
    add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
@@ -39,4 +42,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
       add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
       add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
    endif()
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/17_convnd_bwd_data/CMakeLists.txt b/example/17_convnd_bwd_data/CMakeLists.txt
index ed95946c42..8ab9f37bb3 100644
--- a/example/17_convnd_bwd_data/CMakeLists.txt
+++ b/example/17_convnd_bwd_data/CMakeLists.txt
@@ -1,6 +1,11 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
    target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
-endif()
+   set(target 1)
+ endif()
+endforeach()
 add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
 target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
diff --git a/example/18_batched_gemm_reduce/CMakeLists.txt b/example/18_batched_gemm_reduce/CMakeLists.txt
index 0c3648dbfd..94ed129dc0 100644
--- a/example/18_batched_gemm_reduce/CMakeLists.txt
+++ b/example/18_batched_gemm_reduce/CMakeLists.txt
@@ -1,4 +1,8 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_batched_gemm_reduce_xdl_fp16 batched_gemm_reduce_xdl_fp16.cpp)
-endif()
-
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index 0ee39ac84e..db170decce 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_weight)
 
    add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
@@ -6,7 +9,9 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
 
    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
                                                  example_grouped_conv_bwd_weight_xdl_bf16)
-endif()
+   set(target 1)
+ endif()
+endforeach()
 
 add_custom_target(example_grouped_conv_bwd_weight_dl)
 
diff --git a/example/21_gemm_layernorm/CMakeLists.txt b/example/21_gemm_layernorm/CMakeLists.txt
index 7f974221b7..ff870e7c6b 100644
--- a/example/21_gemm_layernorm/CMakeLists.txt
+++ b/example/21_gemm_layernorm/CMakeLists.txt
@@ -1,6 +1,11 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_welford_fp16 gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp)
    add_example_executable(example_gemm_bias_relu_add_layernorm_xdl_naive_fp16 gemm_bias_relu_add_layernorm_xdl_naive_fp16.cpp)
    add_example_executable(example_gemm_layernorm_xdl_naive_fp16 gemm_layernorm_xdl_naive_fp16.cpp)
    add_example_executable(example_gemm_xdl_layernorm_naive_single_kernel_fp16 gemm_xdl_layernorm_naive_single_kernel_fp16.cpp)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index 9780a64cc2..1c07538b09 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -1,4 +1,9 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
+
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
    add_custom_target(example_grouped_conv_fwd_multiple_d)
 
    add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
@@ -17,8 +22,15 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
    endif() # USE_BITINT_EXTENSION_INT4
    add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
    add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
-endif()
-if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
-  add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
-  add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
-endif()
+   set(target 1)
+ endif()
+endforeach()
+
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index dd9aef94a0..f8d139275d 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -1,12 +1,23 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list2 gfx908 gfx90a)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
    add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
    add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
    add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
-   if(NOT GPU_TARGETS MATCHES "gfx940")
-      add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
-   endif()
 
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
+
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+   add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index f5a6ccb241..57ac33fc9a 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_splitK_gemm_xdl)
    add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
    add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
@@ -15,4 +18,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
       add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
       add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
    endif()
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index 472d59c77d..de48093ac2 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -1,8 +1,13 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_data)
    add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
    add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
 
    add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
    add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/example/40_conv2d_fwd_quantization/CMakeLists.txt b/example/40_conv2d_fwd_quantization/CMakeLists.txt
index c12ab7a340..b82a013b58 100644
--- a/example/40_conv2d_fwd_quantization/CMakeLists.txt
+++ b/example/40_conv2d_fwd_quantization/CMakeLists.txt
@@ -1,9 +1,14 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_conv2d_fwd_xdl_perlayer_quantization_int8 conv2d_fwd_xdl_perlayer_quantization_int8.cpp)
    add_example_executable(example_conv2d_fwd_xdl_perchannel_quantization_int8 conv2d_fwd_xdl_perchannel_quantization_int8.cpp)
    add_example_executable(example_conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8 conv2d_fwd_xdl_bias_relu_perlayer_quantization_int8.cpp)
    add_example_executable(example_conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8 conv2d_fwd_xdl_bias_relu_perchannel_quantization_int8.cpp)
-endif()
+   set(target 1)
+ endif()
+endforeach()
 # Conv perlayer quantization
 add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
 
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index 14f5c284aa..0c9df707b3 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -1,11 +1,22 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list2 gfx908 gfx90a)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
    add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
    add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
    add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
-   if(NOT GPU_TARGETS MATCHES "gfx940")
-      add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
-   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
-endif()
+   set(target 1)
+ endif()
+endforeach()
+
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt b/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
index b60789ea37..14432f6e23 100644
--- a/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
+++ b/example/47_gemm_bias_softmax_gemm_permute/CMakeLists.txt
@@ -1,3 +1,8 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_gemm_bias_softmax_gemm_permute gemm_bias_softmax_gemm_permute.cpp)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 314e6a8134..d7ce449bb6 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -31,7 +31,8 @@
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_BUFFER_RESOURCE_3RD_DWORD -1
 #elif defined(__gfx803__) || defined(__gfx900__) || defined(__gfx906__) || defined(__gfx908__) || \
-    defined(__gfx90a__) || defined(__gfx940__) // for GPU code
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) ||                          \
+    defined(__gfx942__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
 #elif defined(__gfx1030__) // for GPU code
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
@@ -44,7 +45,7 @@
 #elif defined(__gfx803__) || defined(__gfx900__) // for GPU code
 #define CK_USE_AMD_V_MAC_F32
 #elif defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx1030__) || \
-    defined(__gfx940__) // for GPU code
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__) // for GPU code
 #define CK_USE_AMD_V_FMAC_F32
 #define CK_USE_AMD_V_DOT2_F32_F16
 #define CK_USE_AMD_V_DOT4_I32_I8
@@ -53,15 +54,16 @@
 // MFMA instruction
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_MFMA
-#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) // for GPU code
+#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
+    defined(__gfx942__) // for GPU code
 #define CK_USE_AMD_MFMA
 #endif
 
-#if(defined(__gfx90a__) || defined(__gfx940__))
+#if(defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
 #define CK_USE_AMD_MFMA_BF16_1K_OP
 #endif
 
-#if defined(__gfx940__)
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
 #define CK_USE_AMD_MFMA_GFX940
 #endif
 
@@ -84,13 +86,15 @@
 // buffer atomic add: floating point
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
-#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) // for GPU code
+#elif defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
+    defined(__gfx942__) // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1
 #else // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 0
 #endif
 
-#if(defined(__gfx90a__) || defined(__gfx940__)) // for GPU code
+#if(defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || \
+    defined(__gfx942__)) // for GPU code
 #define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 1
 #else
 #define CK_USE_AMD_BUFFER_ATOMIC_MAX_FLOAT64 0
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
index d00e19b440..d0de101c49 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -48,7 +48,7 @@ __global__ void
             const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -418,7 +418,8 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index f26974ccb9..ea958a4eb5 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -136,7 +136,7 @@ __global__ void
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) ||           \
     defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__) || defined(__gfx1100__) || \
-    defined(__gfx1101__) || defined(__gfx1102__))
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -713,7 +713,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
         if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
              ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx908" ||
              ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
-             ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102"))
+             ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102" ||
+             ck::get_device_name() == "gfx941" || ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 30e29cc8e2..ab6d2716c0 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -44,7 +44,7 @@ __global__ void
             const CElementwiseOperation c_element_op)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -682,7 +682,8 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index 95517b1074..f849ac799d 100644
--- a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -57,7 +57,7 @@ __global__ void
             const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t num_blocks_per_batch =
@@ -940,7 +940,8 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index bc86e78b62..46e71240c1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -57,7 +57,7 @@ __global__ void
             const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t num_blocks_per_batch =
@@ -841,7 +841,8 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
index 2018445869..fc080df5f5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -75,7 +75,7 @@ __global__ void
                                           const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
index 09220813bf..7012584aab 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
@@ -61,7 +61,7 @@ __global__ void
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -590,7 +590,8 @@ struct DeviceBatchedGemmGemm_Xdl_CShuffle : public DeviceBatchedGemmGemm<ALayout
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index 2d91c620c6..e00deaf000 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -84,7 +84,7 @@ __global__ void
 {
 
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -581,7 +581,8 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index e39d8f0696..d259ebffae 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -69,7 +69,7 @@ __global__ void
             const ComputeBasePtrOfStridedBatch compute_base_ptr_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -806,7 +806,8 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
index eff503f650..1d52416d4d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
@@ -60,7 +60,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 3fad319e90..ca72bcdd23 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -68,7 +68,7 @@ __global__ void
             const C0MatrixMask c0_matrix_mask)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -724,7 +724,8 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
 #endif
 
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
index 9310d0752a..2405640ca9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
@@ -63,7 +63,7 @@ __global__ void
             const C0MatrixMask c0_matrix_mask)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -614,7 +614,8 @@ struct DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index 87c58f371a..b6a0567fee 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -53,7 +53,7 @@ __global__ void
         kernel_batched_gemm_xdlops_v2r3(const typename DeviceOp::Argument karg)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / karg.Batch);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index b6f38698cd..dd57c90898 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -53,7 +53,7 @@ __global__ void
             const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -583,7 +583,8 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
index fd8c88da9a..bc6582835e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
@@ -56,7 +56,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / num_batches);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
index 22f66931f0..ad51096db7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
@@ -52,7 +52,7 @@ __global__ void
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
     defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
-    defined(__gfx1101__) || defined(__gfx1102__))
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
 
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);
@@ -555,7 +555,8 @@ struct DeviceGemmMultipleD_Dl : public DeviceGemmMultipleD<ALayout,
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
            ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx1030" ||
            ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
-           ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102")
+           ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102" ||
+           ck::get_device_name() == "gfx941" || ck::get_device_name() == "gfx942")
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.e_grid_desc_m_n_);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
index 1ab8362472..5afeaf98c4 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -64,7 +64,7 @@ __global__ void
             index_t NRaw)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemmWelford::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemmWelford::template Run<HasMainKBlockLoop>(
@@ -856,7 +856,8 @@ struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index be174e599e..3a0be837b7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -61,7 +61,7 @@ __global__ void
             const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -556,7 +556,8 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index bd5be99f8b..d4ecea7bb7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -52,7 +52,7 @@ __global__ void
                                             const Block2ETileMap block_2_etile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -492,7 +492,8 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index a3dda82aa6..b68ecfe7b1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -189,7 +189,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
index 14ac5420ae..a2403f109c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
@@ -649,7 +649,8 @@ struct DeviceGemmLayerNorm_Xdl_CShuffle : public BaseOperator
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 5775ff3971..4b1cc65ba2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -38,7 +38,7 @@ __global__ void
             const CDEElementwiseOperation cde_element_op)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
@@ -706,7 +706,8 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
     static bool IsSupportedArgument(const Argument& arg)
     {
         if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940"))
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 3f6238d21f..81a4d69275 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -131,7 +131,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
index 3bcc7bd64b..4f4e0d576a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
@@ -79,7 +79,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
index 71e4e28bf2..af8e0e0279 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -156,7 +156,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -811,7 +811,8 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
                 return false;
             }
         }
-        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940")
+        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940" ||
+                get_device_name() == "gfx941" || get_device_name() == "gfx942")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index 02458bf02a..a9294f0344 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -136,7 +136,7 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     // offset base pointer for each work-group
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -685,7 +685,8 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
                 return false;
             }
         }
-        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940")
+        else if(get_device_name() == "gfx90a" || get_device_name() == "gfx940" ||
+                get_device_name() == "gfx941" || get_device_name() == "gfx942")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
index 22be58259a..0190b3cee6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
@@ -41,7 +41,7 @@ __global__ void
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||              \
     defined(__gfx90a__) || defined(__gfx1030__) || defined(__gfx1100__) || defined(__gfx1101__) || \
-    defined(__gfx1102__) || defined(__gfx940__))
+    defined(__gfx1102__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index 3900047567..bc9b3a4dcc 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -39,7 +39,7 @@ __global__ void
                                 const CDEElementwiseOperation c_element_op)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const index_t block_id = get_block_1d_id();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 1ac9969f86..74f38b9db2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -35,7 +35,7 @@ __global__ void
                                        const index_t group_count)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
     __shared__ uint8_t p_shared[shared_size];
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
index c5c09e909e..533559adf1 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
@@ -67,7 +67,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
index bb3e6a80bf..3d1fbb73bd 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
@@ -55,7 +55,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940))
+    defined(__gfx940) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index d5552656e3..7f1fb21d33 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -25,7 +25,7 @@ __global__ void
         kernel_gemm_xdl_cshuffle_v1(typename GridwiseGemm::Argument karg)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
@@ -46,7 +46,7 @@ __global__ void
                                     typename GridwiseGemm::Problem problem)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid, p_b_grid, p_c_grid, p_shared, problem);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
index d805c9fa2b..8675350094 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
@@ -58,7 +58,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     // TODO ANT: separate into MMA + Epilogue
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index da7ad1cacf..0fc18bb923 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -166,7 +166,7 @@ __global__ void
                                       const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
index e9881d645b..b12bcee0f4 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_skip_b_lds_v1.hpp
@@ -45,7 +45,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(p_a_grid,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index bd18fdb10d..7b8bbd301e 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -36,7 +36,7 @@ __global__ void
                                 const CGridDesc_M_N c_grid_desc_m_n)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid,
@@ -64,7 +64,7 @@ __global__ void
         kernel_gemm_xdlops_v2r3(const typename GridwiseGemm::Argument karg)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const auto a_grid_desc_k0_m_k1 =
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
index 33a4f2b2c5..19fbee727f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4.hpp
@@ -43,7 +43,7 @@ __global__ void
                                 const CBlockClusterAdaptor c_block_cluster_adaptor)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index a4b320ddfa..4cd70cc902 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -31,7 +31,7 @@ __global__ void
                                              const Block2CTileMap& b2c_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
 
     __shared__ uint8_t p_shared[shared_size];
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
index df543c063f..d090ba54d8 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
@@ -47,7 +47,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainK0BlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
index ec98fc9c93..2c09e80fdc 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
@@ -50,7 +50,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
index 3a752dd749..c1bc6a8fec 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
@@ -54,7 +54,7 @@ __global__ void
             const Block2CTileMap block_2_ctile_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__))
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index f4a04e2814..d00b7cd078 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -344,7 +344,7 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
     template <class FloatC>
     __device__ static void Run(const double& reg_a, const double& reg_b, FloatC& reg_c)
     {
-#if defined(__gfx90a__) || defined(__gfx940__)
+#if defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
         reg_c.template AsType<double4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f64_16x16x4f64(
             reg_a, reg_b, reg_c.template AsType<double4_t>()[Number<0>{}], 0, 0, 0);
 #else
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index 426f68d443..33352a6220 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -12,7 +12,7 @@ cmake
 -save-temps=$PWD"                                                                                 \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=ON                                                                                   \
--D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942"                                               \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
diff --git a/script/cmake-ck-release.sh b/script/cmake-ck-release.sh
index 787eabbf96..03d422478e 100755
--- a/script/cmake-ck-release.sh
+++ b/script/cmake-ck-release.sh
@@ -11,7 +11,7 @@ cmake
 -D CMAKE_CXX_FLAGS="-O3"                                                                          \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=OFF                                                                                  \
--D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942"                                               \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
diff --git a/test/batched_gemm/CMakeLists.txt b/test/batched_gemm/CMakeLists.txt
index 5552a6c98e..12142c33e7 100644
--- a/test/batched_gemm/CMakeLists.txt
+++ b/test/batched_gemm/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
    target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
    target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
@@ -14,4 +17,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
    add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
    target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
    target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_gemm/CMakeLists.txt b/test/batched_gemm_gemm/CMakeLists.txt
index eff3391750..858efd8379 100644
--- a/test/batched_gemm_gemm/CMakeLists.txt
+++ b/test/batched_gemm_gemm/CMakeLists.txt
@@ -1,7 +1,11 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(test_batched_gemm_gemm)
-
    add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
    target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
    add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_reduce/CMakeLists.txt b/test/batched_gemm_reduce/CMakeLists.txt
index 1a8d7112a3..0710f4647a 100644
--- a/test/batched_gemm_reduce/CMakeLists.txt
+++ b/test/batched_gemm_reduce/CMakeLists.txt
@@ -1,5 +1,10 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
    target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
    target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/batched_gemm_softmax_gemm/CMakeLists.txt b/test/batched_gemm_softmax_gemm/CMakeLists.txt
index dee0bb56ae..984cc3c160 100644
--- a/test/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -1,7 +1,11 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(test_batched_gemm_softmax_gemm)
-
    add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
    target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
    add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index cb68af06c6..4d04be3faa 100644
--- a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(test_batched_gemm_softmax_gemm_permute)
 
    add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
@@ -14,4 +17,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
    target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
    add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
    add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/contraction/CMakeLists.txt b/test/contraction/CMakeLists.txt
index ec44151f53..1f6e0ed341 100644
--- a/test/contraction/CMakeLists.txt
+++ b/test/contraction/CMakeLists.txt
@@ -1,6 +1,11 @@
 add_gtest_executable(test_contraction test_contraction.cpp)
 target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
     target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/convnd_bwd_data/CMakeLists.txt b/test/convnd_bwd_data/CMakeLists.txt
index 4bfd21945d..f734b46f53 100644
--- a/test/convnd_bwd_data/CMakeLists.txt
+++ b/test/convnd_bwd_data/CMakeLists.txt
@@ -1,4 +1,9 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_gtest_executable(test_convnd_bwd_data convnd_bwd_data.cpp)
    target_link_libraries(test_convnd_bwd_data PRIVATE utility device_conv1d_bwd_data_instance device_conv2d_bwd_data_instance device_conv3d_bwd_data_instance)
-endif()
\ No newline at end of file
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/convnd_fwd/CMakeLists.txt b/test/convnd_fwd/CMakeLists.txt
index 058f990266..745aceffc9 100644
--- a/test/convnd_fwd/CMakeLists.txt
+++ b/test/convnd_fwd/CMakeLists.txt
@@ -1,4 +1,9 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_gtest_executable(test_convnd_fwd convnd_fwd.cpp)
    target_link_libraries(test_convnd_fwd PRIVATE utility device_conv2d_fwd_instance)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/gemm_layernorm/CMakeLists.txt b/test/gemm_layernorm/CMakeLists.txt
index b2a5178ac4..56b8d7737a 100644
--- a/test/gemm_layernorm/CMakeLists.txt
+++ b/test/gemm_layernorm/CMakeLists.txt
@@ -1,6 +1,11 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(test_gemm_layernorm)
    add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
    target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
    add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/gemm_split_k/CMakeLists.txt b/test/gemm_split_k/CMakeLists.txt
index 2274854f88..caf30fca59 100644
--- a/test/gemm_split_k/CMakeLists.txt
+++ b/test/gemm_split_k/CMakeLists.txt
@@ -1,4 +1,9 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_gtest_executable(test_gemm_splitk test_gemm_splitk.cpp)
    target_link_libraries(test_gemm_splitk PRIVATE utility device_gemm_splitk_instance)
-endif()
+   set(target 1)
+ endif()
+endforeach()
diff --git a/test/grouped_convnd_bwd_weight/CMakeLists.txt b/test/grouped_convnd_bwd_weight/CMakeLists.txt
index da554f677a..8be872bc69 100644
--- a/test/grouped_convnd_bwd_weight/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_weight/CMakeLists.txt
@@ -1,4 +1,9 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_gtest_executable(test_grouped_convnd_bwd_weight grouped_convnd_bwd_weight.cpp)
-    target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
-endif()
\ No newline at end of file
+   target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+   set(target 1)
+ endif()
+endforeach()
\ No newline at end of file
diff --git a/test/grouped_gemm/CMakeLists.txt b/test/grouped_gemm/CMakeLists.txt
index 40f634d8b3..8c57b667e2 100644
--- a/test/grouped_gemm/CMakeLists.txt
+++ b/test/grouped_gemm/CMakeLists.txt
@@ -1,4 +1,7 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(test_grouped_gemm)
    add_gtest_executable(test_grouped_gemm_splitk test_grouped_gemm_splitk.cpp)
    add_gtest_executable(test_grouped_gemm_interface test_grouped_gemm_interface.cpp)
@@ -6,4 +9,6 @@ if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS M
    target_link_libraries(test_grouped_gemm_interface PRIVATE utility device_grouped_gemm_instance)
    
    add_dependencies(test_grouped_gemm test_grouped_gemm_splitk test_grouped_gemm_interface)
-endif()
+   set(target 1)
+ endif()
+endforeach()

From d140bdc9fa251d9519055c932e169e510d7f6785 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 16 Jun 2023 12:13:16 -0700
Subject: [PATCH 039/194] do not build gfx941/942 targets during daily QA runs
 (#758)

---
 Jenkinsfile                | 2 +-
 script/cmake-ck-dev.sh     | 2 +-
 script/cmake-ck-release.sh | 2 +-
 3 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 8cfc5f8048..ad2baa00e0 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -695,7 +695,7 @@ pipeline {
                     }
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" """
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940" """
                         execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
                     }
                     steps{
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index 33352a6220..426f68d443 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -12,7 +12,7 @@ cmake
 -save-temps=$PWD"                                                                                 \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=ON                                                                                   \
--D GPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942"                                               \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
diff --git a/script/cmake-ck-release.sh b/script/cmake-ck-release.sh
index 03d422478e..787eabbf96 100755
--- a/script/cmake-ck-release.sh
+++ b/script/cmake-ck-release.sh
@@ -11,7 +11,7 @@ cmake
 -D CMAKE_CXX_FLAGS="-O3"                                                                          \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=OFF                                                                                  \
--D GPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942"                                               \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}

From 0d9118226b167b8cc49360560850d4cf79936c8c Mon Sep 17 00:00:00 2001
From: Qianfeng <qianfeng.zhang@amd.com>
Date: Sat, 17 Jun 2023 12:43:11 +0800
Subject: [PATCH 040/194] Padded Generic Kernel Instance  (#730)

* Add NumReduceDim template parameter to DeviceSoftmax and Softmax client API to simplify instances collecting

* Move the generic kernel instance to be the first of the instance list for elementwise op of normalization

* Add GetGenericInstance() interface for DeviceOperationInstanceFactory class of DeviceSoftmax

* Add testing of GetGenericInstance() in client_example of Softmax

* Revert "Add testing of GetGenericInstance() in client_example of Softmax"

This reverts commit f629cd9a93ce38dfed4886d849f3c38d2e5379c8.

* Revert "Add GetGenericInstance() interface for DeviceOperationInstanceFactory class of DeviceSoftmax"

This reverts commit a9f0d000eb9fd240404112a526ef125429a351df.

* Support generic kernel instance to be the first instance returned by GetInstances() for GroupNorm

* Move generic kernel instance to separate tuple for elementwise op of normalization

* Remove un-used files for softmax instance

* Store generic kernel instance to separate tuple for softmax

* Add IsSupported checking for generic instance to client example of softmax

* Replace the get_device_normalize_from_mean_meansquare_instances() by the DeviceOperationInstanceFactory class for elementwise-normalization

* clang-format fix

* Remove int8 from softmax instances

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../gemm_add_add_layernorm_naive.cpp          |  19 +-
 client_example/06_softmax/softmax4d.cpp       |  32 ++-
 .../18_groupnorm/groupnorm_swish.cpp          |  24 +++
 .../gpu/device/device_softmax.hpp             |  17 +-
 .../gpu/device/impl/device_softmax_impl.hpp   |  17 +-
 .../gpu/device_elementwise_instance.hpp       |  39 ++--
 .../tensor_operation_instance/gpu/softmax.hpp |  89 +++++---
 .../device_softmax_f16_f16_instance.hpp       |  22 --
 ...softmax_f16_f16_instance_rank3_reduce1.hpp |   2 +-
 ...softmax_f16_f16_instance_rank3_reduce2.hpp |   2 +-
 ...softmax_f16_f16_instance_rank3_reduce3.hpp |   2 +-
 ...softmax_f16_f16_instance_rank4_reduce1.hpp |   2 +-
 ...softmax_f16_f16_instance_rank4_reduce2.hpp |   2 +-
 ...softmax_f16_f16_instance_rank4_reduce3.hpp |   2 +-
 ...softmax_f16_f16_instance_rank4_reduce4.hpp |   2 +-
 .../device_softmax_f16_f16_instance_type.hpp  |   8 +-
 .../device_softmax_f32_f32_instance.hpp       |  22 --
 ...softmax_f32_f32_instance_rank3_reduce1.hpp |   2 +-
 ...softmax_f32_f32_instance_rank3_reduce2.hpp |   2 +-
 ...softmax_f32_f32_instance_rank3_reduce3.hpp |   2 +-
 ...softmax_f32_f32_instance_rank4_reduce1.hpp |   2 +-
 ...softmax_f32_f32_instance_rank4_reduce2.hpp |   2 +-
 ...softmax_f32_f32_instance_rank4_reduce3.hpp |   2 +-
 ...softmax_f32_f32_instance_rank4_reduce4.hpp |   2 +-
 .../device_softmax_f32_f32_instance_type.hpp  |   9 +-
 .../softmax/device_softmax_i8_i8_instance.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank3_reduce1.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank3_reduce2.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank3_reduce3.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank4_reduce1.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank4_reduce2.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank4_reduce3.hpp |  22 --
 ...e_softmax_i8_i8_instance_rank4_reduce4.hpp |  22 --
 .../device_softmax_i8_i8_instance_type.hpp    |  40 ----
 .../gpu/softmax/device_softmax_instance.hpp   |  17 +-
 .../elementwise/device_normalize_instance.cpp |  10 +-
 .../device_groupnorm_f16_instance.cpp         |   2 +
 .../device_groupnorm_f32_instance.cpp         |   2 +
 ...oupnorm_swish_f16_f32_f32_f16_instance.cpp |   2 +
 .../device_groupnorm_swish_f16_instance.cpp   |   2 +
 .../device_groupnorm_swish_f32_instance.cpp   |   2 +
 .../device_layernorm2d_f16_instance.cpp       |   2 +
 .../device_layernorm2d_f32_instance.cpp       |   2 +
 .../device_layernorm4d_f16_instance.cpp       |   2 +
 .../device_layernorm4d_f32_instance.cpp       |   2 +
 .../normalization_instance_common.hpp         |  21 ++
 .../gpu/softmax/CMakeLists.txt                |  10 -
 .../device_softmax_f16_f16_instance.cpp       |  40 ----
 ...softmax_f16_f16_instance_rank3_reduce1.cpp |   7 +-
 ...softmax_f16_f16_instance_rank3_reduce2.cpp |   7 +-
 ...softmax_f16_f16_instance_rank3_reduce3.cpp |   7 +-
 ...softmax_f16_f16_instance_rank4_reduce1.cpp |   7 +-
 ...softmax_f16_f16_instance_rank4_reduce2.cpp |   7 +-
 ...softmax_f16_f16_instance_rank4_reduce3.cpp |   7 +-
 ...softmax_f16_f16_instance_rank4_reduce4.cpp |   7 +-
 .../device_softmax_f32_f32_instance.cpp       |  40 ----
 ...softmax_f32_f32_instance_rank3_reduce1.cpp |   7 +-
 ...softmax_f32_f32_instance_rank3_reduce2.cpp |   7 +-
 ...softmax_f32_f32_instance_rank3_reduce3.cpp |   7 +-
 ...softmax_f32_f32_instance_rank4_reduce1.cpp |   7 +-
 ...softmax_f32_f32_instance_rank4_reduce2.cpp |   7 +-
 ...softmax_f32_f32_instance_rank4_reduce3.cpp |   7 +-
 ...softmax_f32_f32_instance_rank4_reduce4.cpp |   7 +-
 .../softmax/device_softmax_i8_i8_instance.cpp |  40 ----
 ...e_softmax_i8_i8_instance_rank3_reduce1.cpp |  27 ---
 ...e_softmax_i8_i8_instance_rank3_reduce2.cpp |  27 ---
 ...e_softmax_i8_i8_instance_rank3_reduce3.cpp |  27 ---
 ...e_softmax_i8_i8_instance_rank4_reduce1.cpp |  27 ---
 ...e_softmax_i8_i8_instance_rank4_reduce2.cpp |  27 ---
 ...e_softmax_i8_i8_instance_rank4_reduce3.cpp |  27 ---
 ...e_softmax_i8_i8_instance_rank4_reduce4.cpp |  27 ---
 .../include/profiler/profile_softmax_impl.hpp |  30 ++-
 profiler/src/profile_softmax.cpp              | 191 ++++++++++++++----
 test/softmax/test_softmax_rank3.cpp           |   4 +-
 test/softmax/test_softmax_rank4.cpp           |   4 +-
 test/softmax/test_softmax_util.hpp            |  88 +++++++-
 76 files changed, 552 insertions(+), 790 deletions(-)
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp

diff --git a/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp b/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
index 1129dfa6b4..58c91f903b 100644
--- a/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
+++ b/client_example/03_gemm_layernorm/gemm_add_add_layernorm_naive.cpp
@@ -172,18 +172,19 @@ int main()
             BLayout,
             CLayout>();
 
-    const auto normalize_ptrs =
-        ck::tensor_operation::device::instance::get_device_normalize_from_mean_meansquare_instances<
-            CDataType,
-            ReduceDataType,
-            ReduceDataType,
-            GammaDataType,
-            BetaDataType,
-            LayerNormOutDataType>();
-
     std::cout << "found " << gemm_reduce_ptrs.size()
               << " gemm_reduceMean_reduceSquareMean instances" << std::endl;
 
+    using NormalizeDeviceOp = ck::tensor_operation::device::DeviceElementwise<
+        ck::Tuple<CDataType, ReduceDataType, ReduceDataType, GammaDataType, BetaDataType>,
+        ck::Tuple<LayerNormOutDataType>,
+        ck::tensor_operation::element_wise::Normalize,
+        2>;
+
+    const auto normalize_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            NormalizeDeviceOp>::GetInstances();
+
     std::cout << "found " << normalize_ptrs.size() << " normalize instances" << std::endl;
 
     auto f_matrix_space_size =
diff --git a/client_example/06_softmax/softmax4d.cpp b/client_example/06_softmax/softmax4d.cpp
index 987ac95690..2ccad27a88 100644
--- a/client_example/06_softmax/softmax4d.cpp
+++ b/client_example/06_softmax/softmax4d.cpp
@@ -53,12 +53,35 @@ int main(int argc, char* argv[])
     SimpleDeviceMem in(sizeof(InDataType) * num_elements);
     SimpleDeviceMem out(sizeof(OutDataType) * num_elements);
 
-    using DeviceOp = ck::tensor_operation::device::
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>;
+    using DeviceOp = ck::tensor_operation::device::DeviceSoftmax<InDataType,
+                                                                 AccDataType,
+                                                                 OutDataType,
+                                                                 PassThrough,
+                                                                 PassThrough,
+                                                                 Rank,
+                                                                 NumReduceDim>;
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
 
+    auto& generic_op_ptr = op_ptrs[0];
+
+    auto generic_argument_ptr = generic_op_ptr->MakeArgumentPointer(in_lengths,
+                                                                    in_strides,
+                                                                    reduce_dims,
+                                                                    alpha,
+                                                                    beta,
+                                                                    in.GetDeviceBuffer(),
+                                                                    out.GetDeviceBuffer(),
+                                                                    PassThrough{},
+                                                                    PassThrough{});
+
+    if(!generic_op_ptr->IsSupportedArgument(generic_argument_ptr.get()))
+    {
+        throw std::runtime_error(
+            "The generic kernel instance should be able to support any input shapes");
+    };
+
     std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
 
     std::string best_op_name;
@@ -74,11 +97,6 @@ int main(int argc, char* argv[])
     {
         auto& op_ptr = op_ptrs[i];
 
-        if(op_ptr->GetRank() != Rank || op_ptr->GetNumReduceDim() != NumReduceDim)
-        {
-            continue;
-        }
-
         auto argument_ptr   = op_ptr->MakeArgumentPointer(in_lengths,
                                                         in_strides,
                                                         reduce_dims,
diff --git a/client_example/18_groupnorm/groupnorm_swish.cpp b/client_example/18_groupnorm/groupnorm_swish.cpp
index 308061a324..e1d198d228 100644
--- a/client_example/18_groupnorm/groupnorm_swish.cpp
+++ b/client_example/18_groupnorm/groupnorm_swish.cpp
@@ -72,6 +72,30 @@ int main(int argc, char* argv[])
 
     std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
 
+    const auto& generic_op_ptr = op_ptrs[0];
+
+    auto generic_argument_ptr =
+        generic_op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
+                                            xy_strides,         // xStrides
+                                            gamma_beta_strides, // gammaStrides
+                                            gamma_beta_strides, // betaStrides
+                                            xy_strides,         // yStrides
+                                            {1, 2, 4},          // reduceDims
+                                            1e-6,
+                                            x_device_buf.GetDeviceBuffer(),
+                                            gamma_device_buf.GetDeviceBuffer(),
+                                            beta_device_buf.GetDeviceBuffer(),
+                                            y_device_buf.GetDeviceBuffer(),
+                                            nullptr,
+                                            nullptr,
+                                            Swish{});
+
+    if(!generic_op_ptr->IsSupportedArgument(generic_argument_ptr.get()))
+    {
+        throw std::runtime_error(
+            "The generic kernel instance should be able to support any input shapes");
+    };
+
     std::string best_op_name;
     bool found            = false;
     int best_op_id        = -1;
diff --git a/include/ck/tensor_operation/gpu/device/device_softmax.hpp b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
index a96ba89e24..1902fd09ee 100644
--- a/include/ck/tensor_operation/gpu/device/device_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
@@ -18,7 +18,8 @@ template <typename InDataType,
           typename OutDataType,
           typename InElementwiseOp,
           typename AccElementwiseOp,
-          index_t Rank>
+          index_t Rank,
+          index_t NumReduceDim>
 struct DeviceSoftmax : public BaseOperator
 {
     //
@@ -49,8 +50,6 @@ struct DeviceSoftmax : public BaseOperator
                         AccElementwiseOp acc_elementwise_op) = 0;
 
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
-    virtual index_t GetRank() const                           = 0;
-    virtual index_t GetNumReduceDim() const                   = 0;
 };
 
 template <typename InDataType,
@@ -58,9 +57,15 @@ template <typename InDataType,
           typename OutDataType,
           typename InElementwiseOp,
           typename AccElementwiseOp,
-          index_t Rank>
-using DeviceSoftmaxPtr = std::unique_ptr<
-    DeviceSoftmax<InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank>>;
+          index_t Rank,
+          index_t NumReduceDim>
+using DeviceSoftmaxPtr = std::unique_ptr<DeviceSoftmax<InDataType,
+                                                       AccDataType,
+                                                       OutDataType,
+                                                       InElementwiseOp,
+                                                       AccElementwiseOp,
+                                                       Rank,
+                                                       NumReduceDim>>;
 
 } // namespace device
 } // namespace tensor_operation
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
index 4aa02dfd32..8eff9d2415 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
@@ -38,16 +38,9 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
                                                 OutDataType,
                                                 InElementwiseOp,
                                                 AccElementwiseOp,
-                                                Rank>
+                                                Rank,
+                                                NumReduceDim>
 {
-    static constexpr index_t kRank            = Rank;
-    static constexpr index_t kNumReduceDim    = NumReduceDim;
-    static constexpr index_t kNumInvariantDim = Rank - NumReduceDim;
-
-    virtual index_t GetRank() const override { return kRank; }
-
-    virtual index_t GetNumReduceDim() const override { return kNumReduceDim; }
-
     static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
 
     static constexpr index_t NumSrcDim = Rank;
@@ -287,13 +280,13 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
     {
         if constexpr(InSrcVectorDim == 0)
         {
-            if constexpr(kNumInvariantDim == 0)
+            if constexpr(NumInvariantDim == 0)
             {
                 return false;
             }
             else
             {
-                if(arg.inStrides_[kNumInvariantDim - 1] != 1 && InSrcVectorSize != 1)
+                if(arg.inStrides_[NumInvariantDim - 1] != 1 && InSrcVectorSize != 1)
                 {
                     return false;
                 }
@@ -316,7 +309,7 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
         }
 
         // To improve
-        if(kNumInvariantDim > 0 && arg.invariant_lowest_length_ % OutDstVectorSize != 0)
+        if(NumInvariantDim > 0 && arg.invariant_lowest_length_ % OutDstVectorSize != 0)
         {
             return false;
         }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
index 7e6267c87b..b03693b00a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp
@@ -5,11 +5,10 @@
 
 #include <vector>
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp"
+#include "ck/tensor_operation/gpu/device/device_elementwise.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -29,20 +28,34 @@ template <typename InputType,
           typename GammaDataType,
           typename BetaDataType,
           typename OutputType>
-auto get_device_normalize_from_mean_meansquare_instances()
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElementwise<
+    ck::Tuple<InputType, MeanType, MeanSquareType, GammaDataType, BetaDataType>,
+    ck::Tuple<OutputType>,
+    Normalize,
+    2>>
 {
-    std::vector<DeviceNormalizeFromMeanMeanSquarePtr> op_ptrs;
+    using DeviceOp = DeviceElementwise<
+        ck::Tuple<InputType, MeanType, MeanSquareType, GammaDataType, BetaDataType>,
+        ck::Tuple<OutputType>,
+        Normalize,
+        2>;
 
-    if constexpr(is_same<InputType, half_t>::value && is_same<MeanType, float>::value &&
-                 is_same<MeanSquareType, float>::value && is_same<GammaDataType, half_t>::value &&
-                 is_same<BetaDataType, half_t>::value && is_same<OutputType, half_t>::value)
+    static auto GetInstances()
     {
-        ck::tensor_operation::device::instance::
-            add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(op_ptrs);
-    }
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-    return op_ptrs;
-}
+        if constexpr(is_same<InputType, half_t>::value && is_same<MeanType, float>::value &&
+                     is_same<MeanSquareType, float>::value &&
+                     is_same<GammaDataType, half_t>::value &&
+                     is_same<BetaDataType, half_t>::value && is_same<OutputType, half_t>::value)
+        {
+            ck::tensor_operation::device::instance::
+                add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(op_ptrs);
+        }
+
+        return op_ptrs;
+    };
+};
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
index c5c2d2cdd7..26815f1447 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
@@ -9,34 +9,33 @@
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 #include "ck/tensor_operation/gpu/device/device_softmax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_softmax_f16_f16_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>&);
-void add_device_softmax_f16_f16_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>&);
-
-void add_device_softmax_f32_f32_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>&);
-void add_device_softmax_f32_f32_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>&);
-
-void add_device_softmax_i8_i8_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>&);
-void add_device_softmax_i8_i8_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>&);
-
-template <typename InDataType, typename AccDataType, typename OutDataType, index_t Rank>
-struct DeviceOperationInstanceFactory<
-    ck::tensor_operation::device::
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>>
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
+          index_t NumReduceDim>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftmax<InDataType,
+                                                                                  AccDataType,
+                                                                                  OutDataType,
+                                                                                  PassThrough,
+                                                                                  PassThrough,
+                                                                                  Rank,
+                                                                                  NumReduceDim>>
 {
-    using DeviceOp =
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>;
+    using DeviceOp = DeviceSoftmax<InDataType,
+                                   AccDataType,
+                                   OutDataType,
+                                   PassThrough,
+                                   PassThrough,
+                                   Rank,
+                                   NumReduceDim>;
 
     static auto GetInstances()
     {
@@ -46,25 +45,49 @@ struct DeviceOperationInstanceFactory<
                      std::is_same_v<OutDataType, F16>)
         {
             if constexpr(Rank == 3)
-                add_device_softmax_f16_f16_rank3_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f16_f16_rank3_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f16_f16_rank3_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f16_f16_rank3_reduce3_instances(op_ptrs);
+            }
             else if constexpr(Rank == 4)
-                add_device_softmax_f16_f16_rank4_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f16_f16_rank4_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f16_f16_rank4_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f16_f16_rank4_reduce3_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 4)
+                    add_device_softmax_f16_f16_rank4_reduce4_instances(op_ptrs);
+            }
         }
         else if constexpr(std::is_same_v<InDataType, F32> && std::is_same_v<AccDataType, F32> &&
                           std::is_same_v<OutDataType, F32>)
         {
             if constexpr(Rank == 3)
-                add_device_softmax_f32_f32_rank3_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f32_f32_rank3_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f32_f32_rank3_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f32_f32_rank3_reduce3_instances(op_ptrs);
+            }
             else if constexpr(Rank == 4)
-                add_device_softmax_f32_f32_rank4_instances(op_ptrs);
-        }
-        else if constexpr(std::is_same_v<InDataType, I8> && std::is_same_v<AccDataType, F32> &&
-                          std::is_same_v<OutDataType, I8>)
-        {
-            if constexpr(Rank == 3)
-                add_device_softmax_i8_i8_rank3_instances(op_ptrs);
-            else if constexpr(Rank == 4)
-                add_device_softmax_i8_i8_rank4_instances(op_ptrs);
+            {
+                if constexpr(NumReduceDim == 1)
+                    add_device_softmax_f32_f32_rank4_reduce1_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 2)
+                    add_device_softmax_f32_f32_rank4_reduce2_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 3)
+                    add_device_softmax_f32_f32_rank4_reduce3_instances(op_ptrs);
+                else if constexpr(NumReduceDim == 4)
+                    add_device_softmax_f32_f32_rank4_reduce4_instances(op_ptrs);
+            }
         }
 
         return op_ptrs;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
deleted file mode 100644
index 7c6f189cb9..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f16_f16_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
-void add_device_softmax_f16_f16_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
index 33d5cc6839..3fd2bd089e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
index 7668248c3f..210fdc0a58 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
index 20eb7bbc94..894fb034d0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
index e8356a9292..708ef0ce13 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
index b3f7d4890b..6754e5ceff 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
index 4190f50a35..5e111176e1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
index b7f3344905..a3cecb32f8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f16_f16_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 4>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
index 53c142f612..8c0782daa5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_type.hpp
@@ -16,7 +16,6 @@ template <index_t Rank, index_t Reduce>
 using device_softmax_f16_f16_instances = std::tuple<
     // clang-format off
     //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
-    // fallback kernel
     DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,              1>,
     DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               8,              8>,
     DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,                8,              1,               8,              8>,
@@ -33,6 +32,13 @@ using device_softmax_f16_f16_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t Rank, index_t Reduce>
+using device_softmax_f16_f16_generic_instance = std::tuple<
+    // clang-format off
+    DeviceSoftmaxImpl<       F16,         F32,         F16,     PassThrough,      PassThrough, Rank,       Reduce,        64,                  8,                  8,                1,                1,              1,               1,              1>
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
deleted file mode 100644
index 41c67af7ad..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f32_f32_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
-void add_device_softmax_f32_f32_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
index 2d791ff979..4cc4690253 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
index eb9cc1ee2b..65724d7888 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
index 68af443a54..13bd45598e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
index 3bf8704b45..d58b424ee9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 1>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
index 43e54aaca7..378e45eeb7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 2>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
index 32c4cd74b5..293df08c7e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 3>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
index f8f5caddb8..e503a9fec1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp
@@ -14,7 +14,7 @@ namespace device {
 namespace instance {
 
 void add_device_softmax_f32_f32_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances);
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 4>>& instances);
 
 } // namespace instance
 } // namespace device
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
index a034e41a07..90c5ddc8a0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_type.hpp
@@ -16,7 +16,7 @@ template <index_t Rank, index_t Reduce>
 using device_softmax_f32_f32_instances = std::tuple<
     // clang-format off
     //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
-    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,               1>, // fallback kernel
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               1,               1>,
     DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,                8,              1,               4,               4>,
     DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,                8,              1,               4,               4>,
     DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,                8,              1,               4,               4>,
@@ -32,6 +32,13 @@ using device_softmax_f32_f32_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t Rank, index_t Reduce>
+using device_softmax_f32_f32_generic_instance = std::tuple<
+    // clang-format off
+    DeviceSoftmaxImpl<       F32,         F32,         F32,     PassThrough,      PassThrough, Rank,       Reduce,        64,                  8,                  8,                1,                1,              1,               1,               1>
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
deleted file mode 100644
index 3cd3742093..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances);
-void add_device_softmax_i8_i8_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
deleted file mode 100644
index f7d4dd045a..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
deleted file mode 100644
index c49dd4d858..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
deleted file mode 100644
index 4074ee3b1f..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
deleted file mode 100644
index 479fcc92fe..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
deleted file mode 100644
index 0dd644fab7..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
deleted file mode 100644
index 50f39396ab..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
deleted file mode 100644
index defa2dbda6..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp
+++ /dev/null
@@ -1,22 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/device_softmax.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances);
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
deleted file mode 100644
index 6ff07de236..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <tuple>
-
-#include "ck/ck.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp"
-#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-template <index_t Rank, index_t Reduce>
-using device_softmax_i8_i8_instances = std::tuple<
-    // clang-format off
-    //                InDataType, AccDataType, OutDataType, InElementwiseOp, AccElementwiseOp, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, InSrcVectorDim, InSrcVectorSize, OutDstVectorSize>
-    // fallback kernel
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,               16,              1,               1,              1>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  4,                 64,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               32,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  2,                128,                1,               64,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               16,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               32,              1,              16,             16>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  1,                256,                1,               64,              1,              16,             16>,
-    // Reduction on middle dimensions
-    // InSrcVectorDim is 0 since we want to coalesce reads on M dimension
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                  8,                 32,                8,                8,              0,               1,              1>,
-    DeviceSoftmaxImpl<        I8,         F32,          I8,     PassThrough,      PassThrough, Rank,       Reduce,       256,                 32,                  8,               32,                8,              0,              16,              8>
-    // clang-format on
-    >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
index 206980cf10..10f99acb8d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax/device_softmax_instance.hpp
@@ -3,6 +3,17 @@
 
 #pragma once
 
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp"
+#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp b/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
index a62c9e2354..f2a5f0728a 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/elementwise/device_normalize_instance.cpp
@@ -30,7 +30,12 @@ using device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances = std:
     //###################|<in, mean, square_mean, gamma, beta>| <out>|  functor| NDim| MPerThread| <in, mean, square_mean, gamma, beta ScalarPerVector>| <out ScalarPerVector>|
     DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   8,       Sequence<8, 1, 1, 8, 8>,      Sequence<8>                >,
     DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   4,       Sequence<4, 1, 1, 4, 4>,      Sequence<4>                >,
-    DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   2,       Sequence<2, 1, 1, 2, 2>,      Sequence<2>                >,
+    DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   2,       Sequence<2, 1, 1, 2, 2>,      Sequence<2>                >
+    // clang-format on
+    >;
+
+using device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_generic_instance = std::tuple<
+    // clang-format off
     DeviceElementwiseImpl<Tuple<F16, F32, F32, F16, F16>,  Tuple<F16>,  Normalize,  2,   1,       Sequence<1, 1, 1, 1, 1>,      Sequence<1>                >
     // clang-format on
     >;
@@ -39,6 +44,9 @@ void add_device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances(
     std::vector<DeviceElementwisePtr<Tuple<F16, F32, F32, F16, F16>, Tuple<F16>, Normalize, 2>>&
         instances)
 {
+    add_device_operation_instances(
+        instances, device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_generic_instance{});
+
     add_device_operation_instances(
         instances, device_normalize_from_mean_squaremean_f16_f32_f32_f16_f16_instances{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
index be860f58e0..e3820462cf 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_5_3_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 5, 3>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Pass, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 5, 3>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
index 9a64e555d6..d85817aad3 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_5_3_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 5, 3>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Pass, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 5, 3>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
index fe72a27331..a81f776c0f 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&
         instances)
 {
+    add_device_operation_instances(
+        instances, device_normalization_f16_f32_f32_f16_generic_instance<Swish, 5, 3>{});
     add_device_operation_instances(instances,
                                    device_normalization_f16_f32_f32_f16_instances<Swish, 5, 3>{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
index cac8641e13..f4bb8bda81 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_5_3_swish_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Swish, 5, 3>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Swish, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Swish, 5, 3>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
index 0a9ac84623..bbb9bd0fe8 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_5_3_swish_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Swish, 5, 3>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Swish, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Swish, 5, 3>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
index ad92818ec2..3f7e4aff1a 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_2_1_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 2, 1>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Pass, 2, 1>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 2, 1>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
index 70e3bbc1c1..1f0db3a036 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_2_1_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 2, 1>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Pass, 2, 1>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 2, 1>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
index 7c5d2c4a9c..cb9d72e614 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_4_3_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 4, 3>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_generic_instance<Pass, 4, 3>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 4, 3>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
index f5626d4a9a..ed555b840d 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
@@ -14,6 +14,8 @@ void add_device_normalization_rank_4_3_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Pass, 4, 3>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_normalization_f32_generic_instance<Pass, 4, 3>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 4, 3>{});
 }
 
diff --git a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
index d9029ac25e..b0684962f9 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
@@ -43,6 +43,13 @@ using device_normalization_f16_instances =
         // clang-format on
         >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_generic_instance = std::tuple<
+    // clang-format off
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+    // clang-format on
+    >;
+
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f32_instances = std::tuple<
     // clang-format off
@@ -69,6 +76,13 @@ using device_normalization_f32_instances = std::tuple<
     // clang-format on
     >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f32_generic_instance = std::tuple<
+    // clang-format off
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+    // clang-format on
+    >;
+
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_f32_f32_f16_instances = std::tuple<
     // clang-format off
@@ -95,6 +109,13 @@ using device_normalization_f16_f32_f32_f16_instances = std::tuple<
     // clang-format on
     >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_f32_f32_f16_generic_instance = std::tuple<
+    // clang-format off
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
index fc13261a6a..202ad12b97 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
@@ -1,13 +1,4 @@
 add_instance_library(device_softmax_instance
-    device_softmax_i8_i8_instance.cpp
-    device_softmax_i8_i8_instance_rank3_reduce1.cpp
-    device_softmax_i8_i8_instance_rank3_reduce2.cpp
-    device_softmax_i8_i8_instance_rank3_reduce3.cpp
-    device_softmax_i8_i8_instance_rank4_reduce1.cpp
-    device_softmax_i8_i8_instance_rank4_reduce2.cpp
-    device_softmax_i8_i8_instance_rank4_reduce3.cpp
-    device_softmax_i8_i8_instance_rank4_reduce4.cpp
-    device_softmax_f16_f16_instance.cpp
     device_softmax_f16_f16_instance_rank3_reduce1.cpp
     device_softmax_f16_f16_instance_rank3_reduce2.cpp
     device_softmax_f16_f16_instance_rank3_reduce3.cpp
@@ -15,7 +6,6 @@ add_instance_library(device_softmax_instance
     device_softmax_f16_f16_instance_rank4_reduce2.cpp
     device_softmax_f16_f16_instance_rank4_reduce3.cpp
     device_softmax_f16_f16_instance_rank4_reduce4.cpp
-    device_softmax_f32_f32_instance.cpp
     device_softmax_f32_f32_instance_rank3_reduce1.cpp
     device_softmax_f32_f32_instance_rank3_reduce2.cpp
     device_softmax_f32_f32_instance_rank3_reduce3.cpp
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
deleted file mode 100644
index a86da7cc79..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance.cpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f16_f16_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3>>& instances)
-{
-    add_device_softmax_f16_f16_rank3_reduce1_instances(instances);
-    add_device_softmax_f16_f16_rank3_reduce2_instances(instances);
-    add_device_softmax_f16_f16_rank3_reduce3_instances(instances);
-}
-
-void add_device_softmax_f16_f16_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4>>& instances)
-{
-    add_device_softmax_f16_f16_rank4_reduce1_instances(instances);
-    add_device_softmax_f16_f16_rank4_reduce2_instances(instances);
-    add_device_softmax_f16_f16_rank4_reduce3_instances(instances);
-    add_device_softmax_f16_f16_rank4_reduce4_instances(instances);
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
index 938fb033a1..36867d993f 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce1.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f16_f16_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<3, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<3, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
index 3d56593811..373f33ad59 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce2.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f16_f16_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<3, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<3, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
index d701b4174e..d26b92b4f4 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank3_reduce3.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f16_f16_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 3, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<3, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<3, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
index 2085aafc56..bbb735b6fe 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce1.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 1>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
index ebe4329f9a..92dbe67760 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce2.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 2>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
index b8fd5a1e5f..354cda85d7 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce3.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 3>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
index 112f1940d3..edb5e42c10 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f16_f16_instance_rank4_reduce4.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f16_f16_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F16, F32, F16, PassThrough, PassThrough, 4, 4>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f16_f16_instances<RANK, 4>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_generic_instance<4, 4>{});
+    add_device_operation_instances(instances, device_softmax_f16_f16_instances<4, 4>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
deleted file mode 100644
index ab8a69eec2..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance.cpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_f32_f32_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3>>& instances)
-{
-    add_device_softmax_f32_f32_rank3_reduce1_instances(instances);
-    add_device_softmax_f32_f32_rank3_reduce2_instances(instances);
-    add_device_softmax_f32_f32_rank3_reduce3_instances(instances);
-}
-
-void add_device_softmax_f32_f32_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4>>& instances)
-{
-    add_device_softmax_f32_f32_rank4_reduce1_instances(instances);
-    add_device_softmax_f32_f32_rank4_reduce2_instances(instances);
-    add_device_softmax_f32_f32_rank4_reduce3_instances(instances);
-    add_device_softmax_f32_f32_rank4_reduce4_instances(instances);
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
index 5382fec90d..566be8fc22 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce1.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f32_f32_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<3, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<3, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
index a1a143afa1..f9c76e3116 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce2.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f32_f32_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<3, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<3, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
index 992e0c1ec1..541e0d71a9 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank3_reduce3.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 3;
-
 void add_device_softmax_f32_f32_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 3, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<3, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<3, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
index 2be1f45bb1..95a38df283 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce1.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 1>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 1>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
index a1da73aa8b..a29b88891d 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce2.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 2>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 2>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 2>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
index b5c3b576a6..0da46ea1b4 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce3.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 3>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 3>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
index 22a0404c05..fa217dc3f5 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
+++ b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_f32_f32_instance_rank4_reduce4.cpp
@@ -13,12 +13,11 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-static constexpr index_t RANK = 4;
-
 void add_device_softmax_f32_f32_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, RANK>>& instances)
+    std::vector<DeviceSoftmaxPtr<F32, F32, F32, PassThrough, PassThrough, 4, 4>>& instances)
 {
-    add_device_operation_instances(instances, device_softmax_f32_f32_instances<RANK, 4>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_generic_instance<4, 4>{});
+    add_device_operation_instances(instances, device_softmax_f32_f32_instances<4, 4>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
deleted file mode 100644
index 81a2ff80ca..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance.cpp
+++ /dev/null
@@ -1,40 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-void add_device_softmax_i8_i8_rank3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 3>>& instances)
-{
-    add_device_softmax_i8_i8_rank3_reduce1_instances(instances);
-    add_device_softmax_i8_i8_rank3_reduce2_instances(instances);
-    add_device_softmax_i8_i8_rank3_reduce3_instances(instances);
-}
-
-void add_device_softmax_i8_i8_rank4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, 4>>& instances)
-{
-    add_device_softmax_i8_i8_rank4_reduce1_instances(instances);
-    add_device_softmax_i8_i8_rank4_reduce2_instances(instances);
-    add_device_softmax_i8_i8_rank4_reduce3_instances(instances);
-    add_device_softmax_i8_i8_rank4_reduce4_instances(instances);
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
deleted file mode 100644
index 3e2cf8d062..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 3;
-
-void add_device_softmax_i8_i8_rank3_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 1>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
deleted file mode 100644
index c8b038d502..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 3;
-
-void add_device_softmax_i8_i8_rank3_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 2>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
deleted file mode 100644
index 08995d99ec..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank3_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 3;
-
-void add_device_softmax_i8_i8_rank3_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
deleted file mode 100644
index 652601ee7c..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce1.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce1_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 1>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
deleted file mode 100644
index 86caac1b62..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce2_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 2>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
deleted file mode 100644
index c46ae1a4ef..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce3.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce3_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 3>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp b/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
deleted file mode 100644
index 394814ff53..0000000000
--- a/library/src/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.cpp
+++ /dev/null
@@ -1,27 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <vector>
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_rank4_reduce4.hpp"
-#include "ck/library/tensor_operation_instance/gpu/softmax/device_softmax_i8_i8_instance_type.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr index_t RANK = 4;
-
-void add_device_softmax_i8_i8_rank4_reduce4_instances(
-    std::vector<DeviceSoftmaxPtr<I8, F32, I8, PassThrough, PassThrough, RANK>>& instances)
-{
-    add_device_operation_instances(instances, device_softmax_i8_i8_instances<RANK, 4>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/profiler/include/profiler/profile_softmax_impl.hpp b/profiler/include/profiler/profile_softmax_impl.hpp
index 65b4be2a60..daaf565149 100644
--- a/profiler/include/profiler/profile_softmax_impl.hpp
+++ b/profiler/include/profiler/profile_softmax_impl.hpp
@@ -40,7 +40,11 @@ template <> std::string type_to_string<int8_t>()  { return "int8"; }
 template <> std::string type_to_string<int32_t>() { return "int32"; }
 // clang-format on
 
-template <typename InDataType, typename AccDataType, typename OutDataType, index_t Rank>
+template <typename InDataType,
+          typename AccDataType,
+          typename OutDataType,
+          index_t Rank,
+          index_t NumReduceDim>
 bool profile_softmax_impl(int do_verification,
                           int init_method,
                           bool do_log,
@@ -54,7 +58,13 @@ bool profile_softmax_impl(int do_verification,
     if(Rank != in_length.size())
     {
         throw std::runtime_error("Input tensor rank is different from template argument Rank!");
-    }
+    };
+
+    if(NumReduceDim != reduce_dims.size())
+    {
+        throw std::runtime_error(
+            "Input reduce_dims rank is different from template argument NumReduceDim!");
+    };
 
     Tensor<InDataType> in = in_strides.empty() ? Tensor<InDataType>(in_length)
                                                : Tensor<InDataType>(in_length, in_strides);
@@ -92,8 +102,13 @@ bool profile_softmax_impl(int do_verification,
 
     // add device softmax instances
     using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-    using DeviceOp    = tensor_operation::device::
-        DeviceSoftmax<InDataType, AccDataType, OutDataType, PassThrough, PassThrough, Rank>;
+    using DeviceOp    = tensor_operation::device::DeviceSoftmax<InDataType,
+                                                             AccDataType,
+                                                             OutDataType,
+                                                             PassThrough,
+                                                             PassThrough,
+                                                             Rank,
+                                                             NumReduceDim>;
 
     // get device op instances
     const auto instances = tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -112,13 +127,6 @@ bool profile_softmax_impl(int do_verification,
 
     for(auto& inst_ptr : instances)
     {
-        // Is this user's responsibility to check if problem mismatches kernel instance (ie. rank 3
-        // problem to rank 4 kernel) other than invoking IsSupportedArgument()?
-        if(!(inst_ptr->GetNumReduceDim() == static_cast<index_t>(reduce_dims.size())))
-        {
-            continue;
-        }
-
         auto argument_ptr = inst_ptr->MakeArgumentPointer(in_tensor_lengths,
                                                           in_tensor_strides,
                                                           reduce_dims,
diff --git a/profiler/src/profile_softmax.cpp b/profiler/src/profile_softmax.cpp
index 77007ad13b..dfe8d95c90 100644
--- a/profiler/src/profile_softmax.cpp
+++ b/profiler/src/profile_softmax.cpp
@@ -92,27 +92,76 @@ int profile_softmax(int argc, char* argv[])
     {
         if(data_type == SoftmaxDataType::F16_F16)
         {
-            ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3>(do_verification,
-                                                                                 init_method,
-                                                                                 do_log,
-                                                                                 time_kernel,
-                                                                                 length,
-                                                                                 stride,
-                                                                                 reduce,
-                                                                                 double(alpha),
-                                                                                 double(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3, 1>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3, 2>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 3, 3>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else if(data_type == SoftmaxDataType::F32_F32)
         {
-            ck::profiler::profile_softmax_impl<float, float, float, 3>(do_verification,
-                                                                       init_method,
-                                                                       do_log,
-                                                                       time_kernel,
-                                                                       length,
-                                                                       stride,
-                                                                       reduce,
-                                                                       double(alpha),
-                                                                       double(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<float, float, float, 3, 1>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<float, float, float, 3, 2>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<float, float, float, 3, 3>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else
         {
@@ -124,27 +173,97 @@ int profile_softmax(int argc, char* argv[])
     {
         if(data_type == SoftmaxDataType::F16_F16)
         {
-            ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4>(do_verification,
-                                                                                 init_method,
-                                                                                 do_log,
-                                                                                 time_kernel,
-                                                                                 length,
-                                                                                 stride,
-                                                                                 reduce,
-                                                                                 double(alpha),
-                                                                                 double(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 1>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 2>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 3>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else if(reduce.size() == 4)
+                ck::profiler::profile_softmax_impl<ck::half_t, float, ck::half_t, 4, 4>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    length,
+                    stride,
+                    reduce,
+                    double(alpha),
+                    double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else if(data_type == SoftmaxDataType::F32_F32)
         {
-            ck::profiler::profile_softmax_impl<float, float, float, 4>(do_verification,
-                                                                       init_method,
-                                                                       do_log,
-                                                                       time_kernel,
-                                                                       length,
-                                                                       stride,
-                                                                       reduce,
-                                                                       double(alpha),
-                                                                       double(beta));
+            if(reduce.size() == 1)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 1>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 2)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 2>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 3)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 3>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else if(reduce.size() == 4)
+                ck::profiler::profile_softmax_impl<float, float, float, 4, 4>(do_verification,
+                                                                              init_method,
+                                                                              do_log,
+                                                                              time_kernel,
+                                                                              length,
+                                                                              stride,
+                                                                              reduce,
+                                                                              double(alpha),
+                                                                              double(beta));
+            else
+                throw std::runtime_error("invalid number of dimensions to reduce");
         }
         else
         {
diff --git a/test/softmax/test_softmax_rank3.cpp b/test/softmax/test_softmax_rank3.cpp
index 24ad912d8d..43ae11bf1f 100644
--- a/test/softmax/test_softmax_rank3.cpp
+++ b/test/softmax/test_softmax_rank3.cpp
@@ -13,7 +13,6 @@ using I = ck::Number<N>;
 
 using F16 = ck::half_t;
 using F32 = float;
-using I8  = int8_t;
 
 template <typename Tuple>
 class TestSoftmax : public ck::TestSoftmax<Tuple>
@@ -24,8 +23,7 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
     std::tuple<       F16,         F32,         F16,    I<3>>,
-    std::tuple<       F32,         F32,         F32,    I<3>>,
-    std::tuple<        I8,         F32,          I8,    I<3>>
+    std::tuple<       F32,         F32,         F32,    I<3>>
     >;
 // clang-format on
 
diff --git a/test/softmax/test_softmax_rank4.cpp b/test/softmax/test_softmax_rank4.cpp
index b58301fb11..5cf96bbaa8 100644
--- a/test/softmax/test_softmax_rank4.cpp
+++ b/test/softmax/test_softmax_rank4.cpp
@@ -13,7 +13,6 @@ using I = ck::Number<N>;
 
 using F16 = ck::half_t;
 using F32 = float;
-using I8  = int8_t;
 
 template <typename Tuple>
 class TestSoftmax : public ck::TestSoftmax<Tuple>
@@ -24,8 +23,7 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
     std::tuple<       F16,         F32,         F16,    I<4>>,
-    std::tuple<       F32,         F32,         F32,    I<4>>,
-    std::tuple<        I8,         F32,          I8,    I<4>>
+    std::tuple<       F32,         F32,         F32,    I<4>>
     >;
 // clang-format on
 
diff --git a/test/softmax/test_softmax_util.hpp b/test/softmax/test_softmax_util.hpp
index e36231de85..1409af8453 100644
--- a/test/softmax/test_softmax_util.hpp
+++ b/test/softmax/test_softmax_util.hpp
@@ -61,8 +61,92 @@ class TestSoftmax : public ::testing::Test
         int init_method = 1; // integer value initialization
         bool log        = false;
         std::vector<ck::index_t> strides; // intenionally empty, to get packed layout.
-        bool pass = ck::profiler::profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank>(
-            verify_, init_method, log, bench_, in_length, strides, reduce_dims, alpha, beta);
+        bool pass = false;
+
+        if constexpr(Rank == 3)
+        {
+            if(reduce_dims.size() == 1)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 1>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 2)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 2>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 3)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 3>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+        }
+        else if constexpr(Rank == 4)
+        {
+            if(reduce_dims.size() == 1)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 1>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 2)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 2>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 3)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 3>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+            else if(reduce_dims.size() == 4)
+                pass = ck::profiler::
+                    profile_softmax_impl<InDataType, AccDataType, OutDataType, Rank, 4>(verify_,
+                                                                                        init_method,
+                                                                                        log,
+                                                                                        bench_,
+                                                                                        in_length,
+                                                                                        strides,
+                                                                                        reduce_dims,
+                                                                                        alpha,
+                                                                                        beta);
+        };
+
         EXPECT_TRUE(pass);
     }
 

From 341ad956657a0ad3501af0dcceeddb5018449de6 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Mon, 19 Jun 2023 22:44:22 +0800
Subject: [PATCH 041/194] Maxpool bwd (#750)

* Add maxpool f32 kernel and example

* Revise copyright

* Add device pool bwd device op

* Support f16 and bf16

* Add compute datatype for reference code.
Prevent error in bf16

* Fix type error

* Remove layout

* Fix bf16 error

* Add f16 and bf16 example

* Add more operations

* Implement IsSupportedArgument

* Add changelog

* Add comment

* Add comment

* Remove useless header

* Move initialize of workspace to the run

* Move set din zero to the device operator

* Save din_length_raw

* Remove useless header

* Calculate gridsize according to the number of CU

* Calculate gridSize according to the number of CU.
Remove useless header

* Add put example

* Remove useless header

* Fix CI fail
---
 CHANGELOG.md                                  |   2 +
 example/49_maxpool2d_bwd/CMakeLists.txt       |   3 +
 .../49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp   |  62 ++++
 .../49_maxpool2d_bwd/maxpool2d_bwd_common.hpp | 222 ++++++++++++
 .../49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp   |  62 ++++
 .../49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp   |  62 ++++
 example/50_put_element/CMakeLists.txt         |   1 +
 example/50_put_element/put_element_fp16.cpp   |  88 +++++
 include/ck/host_utility/stream_utility.hpp    |   2 +-
 .../gpu/device/device_index_pool_bwd.hpp      |  32 ++
 .../gpu/device/device_put_element.hpp         |  36 ++
 .../impl/device_index_pool_bwd_impl.hpp       | 316 ++++++++++++++++++
 .../device/impl/device_put_element_impl.hpp   | 155 +++++++++
 .../gpu/grid/gridwise_put_element_1d.hpp      | 155 +++++++++
 .../cpu/reference_maxpool_bwd.hpp             | 103 ++++++
 .../cpu/reference_pool_fwd.hpp                |  20 +-
 16 files changed, 1310 insertions(+), 11 deletions(-)
 create mode 100644 example/49_maxpool2d_bwd/CMakeLists.txt
 create mode 100644 example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
 create mode 100644 example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
 create mode 100644 example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
 create mode 100644 example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
 create mode 100644 example/50_put_element/CMakeLists.txt
 create mode 100644 example/50_put_element/put_element_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_put_element.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
 create mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 0188350046..3898b5ce2d 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -20,6 +20,8 @@ Full documentation for Composable Kernel is not yet available.
 - Added multi-embeddings support (#542).
 - Added Navi3x blockwise GEMM and real GEMM support (#541).
 - Added Navi grouped ConvBwdWeight support (#505).
+- Added pool3d forward (#697).
+- Added maxpool backward (#750).
 
 ### Changed
 - Changed ...
diff --git a/example/49_maxpool2d_bwd/CMakeLists.txt b/example/49_maxpool2d_bwd/CMakeLists.txt
new file mode 100644
index 0000000000..b29cf9ccbc
--- /dev/null
+++ b/example/49_maxpool2d_bwd/CMakeLists.txt
@@ -0,0 +1,3 @@
+add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
+add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
+add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
new file mode 100644
index 0000000000..08a8009c6c
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "maxpool2d_bwd_common.hpp"
+
+using InDataType      = ck::bhalf_t;
+using OutDataType     = ck::bhalf_t;
+using IndexDataType   = int32_t;
+using ComputeDataType = float;
+using DInDataType     = ck::bhalf_t;
+using DOutDataType    = ck::bhalf_t;
+
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 1;
+    ck::index_t C               = 1;
+    ck::index_t Y               = 3;
+    ck::index_t X               = 3;
+    ck::index_t Hi              = 32;
+    ck::index_t Wi              = 32;
+    ck::index_t window_stride_h = 1;
+    ck::index_t window_stride_w = 1;
+    ck::index_t in_left_pad_h   = 0;
+    ck::index_t in_left_pad_w   = 0;
+    ck::index_t in_right_pad_h  = 0;
+    ck::index_t in_right_pad_w  = 0;
+
+    bool pass = maxpool_bwd_test<InDataType,
+                                 OutDataType,
+                                 IndexDataType,
+                                 ComputeDataType,
+                                 DInDataType,
+                                 DOutDataType,
+                                 PropagateNan>(do_verification,
+                                               time_kernel,
+                                               N,
+                                               C,
+                                               Y,
+                                               X,
+                                               Hi,
+                                               Wi,
+                                               window_stride_h,
+                                               window_stride_w,
+                                               in_left_pad_h,
+                                               in_left_pad_w,
+                                               in_right_pad_h,
+                                               in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
new file mode 100644
index 0000000000..045793cc2a
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
@@ -0,0 +1,222 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/utility/reduction_enums.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp"
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          typename DInDataType,
+          typename DOutDataType,
+          bool PropagateNan>
+bool maxpool_bwd_test(bool do_verification,
+                      bool time_kernel,
+                      ck::index_t N,
+                      ck::index_t C,
+                      ck::index_t Y,
+                      ck::index_t X,
+                      ck::index_t Hi,
+                      ck::index_t Wi,
+                      ck::index_t window_stride_h,
+                      ck::index_t window_stride_w,
+                      ck::index_t in_left_pad_h,
+                      ck::index_t in_left_pad_w,
+                      ck::index_t in_right_pad_h,
+                      ck::index_t in_right_pad_w)
+{
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    using DevicePoolFwdInstance =
+        ck::tensor_operation::device::DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<
+            InDataType,      // InDataType
+            OutDataType,     // OutDataType
+            IndexDataType,   // IndexDataType
+            ComputeDataType, // ComputeDataType
+            ck::ReduceTensorOp::MAX,
+            true, // OutputIndex
+            64,   // BlockSize
+            64,   // ReduceMThreadClusterSize
+            1,    // ReduceKThreadClusterSize
+            4,    // ReduceMThreadSliceSize
+            1,    // ReduceKThreadSliceSize
+            1>;   // InSrcOutDstVectorSize
+
+    using DeviceMaxPoolBwdInstance = ck::tensor_operation::device::
+        DeviceIndexPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 4>;
+
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+
+    const std::vector<ck::index_t> window_spatial_lengths{Y, X};
+    const std::vector<ck::index_t> window_strides{window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
+    const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+            // reference need Tensor with NCHW order
+            return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
+        };
+
+    // in
+    Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
+
+    // out
+    Tensor<OutDataType> out_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<OutDataType> out_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    // indices
+    Tensor<IndexDataType> indices_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
+    Tensor<IndexDataType> indices_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    // dout
+    Tensor<DOutDataType> dout_n_c_ho_wo(f_host_tensor_descriptor(N, C, Ho, Wo));
+
+    // din
+    Tensor<DInDataType> din_n_c_hi_wi_host(f_host_tensor_descriptor(N, C, Hi, Wi));
+    Tensor<DInDataType> din_n_c_hi_wi_device(f_host_tensor_descriptor(N, C, Hi, Wi));
+
+    std::cout << "in_n_c_hi_wi: " << in_n_c_hi_wi.mDesc << std::endl;
+    std::cout << "out_n_c_ho_wo: " << out_n_c_ho_wo_host.mDesc << std::endl;
+    std::cout << "indices_n_c_ho_wo: " << indices_n_c_ho_wo_host.mDesc << std::endl;
+    std::cout << "dout_n_c_ho_wo: " << dout_n_c_ho_wo.mDesc << std::endl;
+    std::cout << "din_n_c_hi_wi: " << din_n_c_hi_wi_host.mDesc << std::endl;
+
+    in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
+    dout_n_c_ho_wo.GenerateTensorValue(GeneratorTensor_3<DOutDataType>{-1.0, 1.0});
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) *
+                             out_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem indices_device_buf(sizeof(IndexDataType) *
+                                 indices_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
+    DeviceMem dout_device_buf(sizeof(DOutDataType) * dout_n_c_ho_wo.mDesc.GetElementSpaceSize());
+    DeviceMem din_device_buf(sizeof(DInDataType) *
+                             din_n_c_hi_wi_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
+    dout_device_buf.ToDevice(dout_n_c_ho_wo.mData.data());
+
+    auto pool_fwd              = DevicePoolFwdInstance{};
+    auto pool_fwd_invoker_ptr  = pool_fwd.MakeInvokerPointer();
+    auto pool_fwd_argument_ptr = pool_fwd.MakeArgumentPointer(
+        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+        static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+        {N, C, Hi, Wi},
+        window_spatial_lengths,
+        {N, C, Ho, Wo},
+        {C * Hi * Wi, 1, Wi * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
+        {C * Ho * Wo, 1, Wo * C, C},
+        window_strides,
+        input_left_pads,
+        input_right_pads,
+        {2, 3});
+
+    if(!pool_fwd.IsSupportedArgument(pool_fwd_argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! pool_fwd with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    float ave_time_fwd =
+        pool_fwd_invoker_ptr->Run(pool_fwd_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    auto pool_bwd              = DeviceMaxPoolBwdInstance{};
+    auto pool_bwd_invoker_ptr  = pool_bwd.MakeInvokerPointer();
+    auto pool_bwd_argument_ptr = pool_bwd.MakeArgumentPointer(
+        static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+        static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+        dout_n_c_ho_wo.mDesc.GetElementSpaceSize(),
+        din_n_c_hi_wi_device.mDesc.GetElementSpaceSize(),
+        window_spatial_lengths,
+        window_strides);
+
+    if(!pool_bwd.IsSupportedArgument(pool_bwd_argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! pool_bwd with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    size_t pool_bwd_workspace_sz = pool_bwd.GetWorkSpaceSize(pool_bwd_argument_ptr.get());
+    DeviceMem pool_bwd_workspace_device_buf(pool_bwd_workspace_sz);
+    pool_bwd.SetWorkSpacePointer(pool_bwd_argument_ptr.get(),
+                                 pool_bwd_workspace_device_buf.GetDeviceBuffer());
+
+    float ave_time_bwd =
+        pool_bwd_invoker_ptr->Run(pool_bwd_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    std::cout << "Pool fwd perf: " << ave_time_fwd << " ms" << std::endl;
+    std::cout << "Pool bwd perf: " << ave_time_bwd << " ms" << std::endl;
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<4,
+                                                            2,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            PropagateNan,
+                                                            true>;
+
+        auto ref_pooling_fwd          = ReferencePoolingFwdInstance{};
+        auto ref_pooling_fwd_invoker  = ref_pooling_fwd.MakeInvoker();
+        auto ref_pooling_fwd_argument = ref_pooling_fwd.MakeArgument(in_n_c_hi_wi,
+                                                                     out_n_c_ho_wo_host,
+                                                                     indices_n_c_ho_wo_host,
+                                                                     window_spatial_lengths,
+                                                                     window_strides,
+                                                                     input_left_pads,
+                                                                     input_right_pads);
+        ref_pooling_fwd_invoker.Run(ref_pooling_fwd_argument);
+
+        using ReferencePoolingBwdInstance =
+            ck::tensor_operation::host::ReferenceMaxPoolBwd<DOutDataType,
+                                                            IndexDataType,
+                                                            ComputeDataType,
+                                                            DInDataType,
+                                                            PassThrough>;
+
+        auto ref_pooling_bwd          = ReferencePoolingBwdInstance{};
+        auto ref_pooling_bwd_invoker  = ref_pooling_bwd.MakeInvoker();
+        auto ref_pooling_bwd_argument = ref_pooling_bwd.MakeArgument(
+            dout_n_c_ho_wo, indices_n_c_ho_wo_host, din_n_c_hi_wi_host, PassThrough{});
+
+        ref_pooling_bwd_invoker.Run(ref_pooling_bwd_argument);
+
+        out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
+        indices_device_buf.FromDevice(indices_n_c_ho_wo_device.mData.data());
+        din_device_buf.FromDevice(din_n_c_hi_wi_device.mData.data());
+
+        pass = pass && ck::utils::check_err(out_n_c_ho_wo_device, out_n_c_ho_wo_host);
+        pass = pass && ck::utils::check_err(indices_n_c_ho_wo_device, indices_n_c_ho_wo_host);
+        pass = pass && ck::utils::check_err(din_n_c_hi_wi_device, din_n_c_hi_wi_host);
+    }
+
+    return (pass);
+};
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
new file mode 100644
index 0000000000..c6b0b2c860
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "maxpool2d_bwd_common.hpp"
+
+using InDataType      = ck::half_t;
+using OutDataType     = ck::half_t;
+using IndexDataType   = int32_t;
+using ComputeDataType = float;
+using DInDataType     = ck::half_t;
+using DOutDataType    = ck::half_t;
+
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 1;
+    ck::index_t C               = 1;
+    ck::index_t Y               = 3;
+    ck::index_t X               = 3;
+    ck::index_t Hi              = 32;
+    ck::index_t Wi              = 32;
+    ck::index_t window_stride_h = 1;
+    ck::index_t window_stride_w = 1;
+    ck::index_t in_left_pad_h   = 0;
+    ck::index_t in_left_pad_w   = 0;
+    ck::index_t in_right_pad_h  = 0;
+    ck::index_t in_right_pad_w  = 0;
+
+    bool pass = maxpool_bwd_test<InDataType,
+                                 OutDataType,
+                                 IndexDataType,
+                                 ComputeDataType,
+                                 DInDataType,
+                                 DOutDataType,
+                                 PropagateNan>(do_verification,
+                                               time_kernel,
+                                               N,
+                                               C,
+                                               Y,
+                                               X,
+                                               Hi,
+                                               Wi,
+                                               window_stride_h,
+                                               window_stride_w,
+                                               in_left_pad_h,
+                                               in_left_pad_w,
+                                               in_right_pad_h,
+                                               in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
new file mode 100644
index 0000000000..c79b84c489
--- /dev/null
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+#include "maxpool2d_bwd_common.hpp"
+
+using InDataType      = float;
+using OutDataType     = float;
+using IndexDataType   = int32_t;
+using ComputeDataType = float;
+using DInDataType     = float;
+using DOutDataType    = float;
+
+static constexpr bool PropagateNan = false;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    // Pool shape
+    ck::index_t N               = 1;
+    ck::index_t C               = 1;
+    ck::index_t Y               = 2;
+    ck::index_t X               = 2;
+    ck::index_t Hi              = 32;
+    ck::index_t Wi              = 32;
+    ck::index_t window_stride_h = 2;
+    ck::index_t window_stride_w = 2;
+    ck::index_t in_left_pad_h   = 0;
+    ck::index_t in_left_pad_w   = 0;
+    ck::index_t in_right_pad_h  = 0;
+    ck::index_t in_right_pad_w  = 0;
+
+    bool pass = maxpool_bwd_test<InDataType,
+                                 OutDataType,
+                                 IndexDataType,
+                                 ComputeDataType,
+                                 DInDataType,
+                                 DOutDataType,
+                                 PropagateNan>(do_verification,
+                                               time_kernel,
+                                               N,
+                                               C,
+                                               Y,
+                                               X,
+                                               Hi,
+                                               Wi,
+                                               window_stride_h,
+                                               window_stride_w,
+                                               in_left_pad_h,
+                                               in_left_pad_w,
+                                               in_right_pad_h,
+                                               in_right_pad_w);
+
+    return (pass ? 0 : 1);
+}
diff --git a/example/50_put_element/CMakeLists.txt b/example/50_put_element/CMakeLists.txt
new file mode 100644
index 0000000000..1b0020ebcf
--- /dev/null
+++ b/example/50_put_element/CMakeLists.txt
@@ -0,0 +1 @@
+add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
diff --git a/example/50_put_element/put_element_fp16.cpp b/example/50_put_element/put_element_fp16.cpp
new file mode 100644
index 0000000000..d4b6831bc4
--- /dev/null
+++ b/example/50_put_element/put_element_fp16.cpp
@@ -0,0 +1,88 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+using XDataType     = ck::half_t;
+using YDataType     = ck::half_t;
+using IndexDataType = int32_t;
+
+using YElementwiseOp = ck::tensor_operation::element_wise::PassThrough;
+
+using DeviceInstance =
+    ck::tensor_operation::device::DevicePutElementImpl<XDataType,     // XDataType
+                                                       IndexDataType, // IndexDataType
+                                                       YDataType,     // YDataType
+                                                       YElementwiseOp,
+                                                       ck::InMemoryDataOperationEnum::Set,
+                                                       1>;
+
+int main()
+{
+    bool do_verification = true;
+    bool time_kernel     = false;
+
+    int N = 1024;
+
+    Tensor<XDataType> x(HostTensorDescriptor{N, 1});
+    Tensor<IndexDataType> indices(HostTensorDescriptor{N, 1});
+    Tensor<YDataType> y(HostTensorDescriptor{N, 1});
+
+    x.GenerateTensorValue(GeneratorTensor_3<XDataType>{-1.0, 1.0});
+    for(int i = 0; i < N; ++i)
+        indices(i) = i;
+
+    DeviceMem x_device_buf(sizeof(XDataType) * x.mDesc.GetElementSpaceSize());
+    DeviceMem y_device_buf(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem indices_device_buf(sizeof(IndexDataType) * indices.mDesc.GetElementSpaceSize());
+
+    x_device_buf.ToDevice(x.mData.data());
+    indices_device_buf.ToDevice(indices.mData.data());
+
+    auto put_instance     = DeviceInstance{};
+    auto put_invoker_ptr  = put_instance.MakeInvokerPointer();
+    auto put_argument_ptr = put_instance.MakeArgumentPointer(
+        static_cast<XDataType*>(x_device_buf.GetDeviceBuffer()),
+        static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+        static_cast<YDataType*>(y_device_buf.GetDeviceBuffer()),
+        N,
+        N,
+        YElementwiseOp{});
+
+    if(!put_instance.IsSupportedArgument(put_argument_ptr.get()))
+    {
+        throw std::runtime_error("argument is not supported!");
+    }
+
+    float ave_time =
+        put_invoker_ptr->Run(put_argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    std::cout << "perf: " << ave_time << " ms" << std::endl;
+
+    bool pass = true;
+    if(do_verification)
+    {
+        Tensor<YDataType> y_host(HostTensorDescriptor{N, 1});
+
+        for(int i = 0; i < N; ++i)
+        {
+            IndexDataType idx = indices(i);
+            y_host(idx)       = x(i);
+        }
+
+        y_device_buf.FromDevice(y.mData.data());
+        pass = ck::utils::check_err(y, y_host);
+    }
+
+    return (pass ? 0 : 1);
+}
diff --git a/include/ck/host_utility/stream_utility.hpp b/include/ck/host_utility/stream_utility.hpp
index ef05f2e26b..9ab49489bb 100644
--- a/include/ck/host_utility/stream_utility.hpp
+++ b/include/ck/host_utility/stream_utility.hpp
@@ -8,7 +8,7 @@
 #include "ck/stream_config.hpp"
 #include "ck/host_utility/hip_check_error.hpp"
 
-static int getAvailableComputeUnitCount(const StreamConfig& stream_config)
+static inline int getAvailableComputeUnitCount(const StreamConfig& stream_config)
 {
     constexpr int MAX_MASK_DWORDS = 64;
 
diff --git a/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp b/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
new file mode 100644
index 0000000000..bf81ed9f5b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// For pooling which used indexable operation, such as MaxPool, MinPool...etc
+template <typename DOutDataType, typename IndexDataType, typename DInDataType>
+struct DeviceIndexPoolBwd : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_dout,
+                        const void* p_indices,
+                        void* p_din,
+                        index_t dout_length,
+                        index_t din_length,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> window_strides) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_put_element.hpp b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
new file mode 100644
index 0000000000..9186827491
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/utility/reduction_enums.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// output[indices] = input
+template <typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation,
+          InMemoryDataOperationEnum Op>
+struct DevicePutElement : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_input,
+                        const void* p_indices,
+                        void* p_output,
+                        index_t input_length,
+                        index_t output_length,
+                        ElementwiseOperation elementwise_op) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
new file mode 100644
index 0000000000..175994d49f
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
@@ -0,0 +1,316 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// output[indices] = input
+template <typename DOutDataType,
+          typename IndexDataType,
+          typename DInDataType,
+          ck::index_t InOutVectorSize>
+struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDataType, DInDataType>
+{
+    using DInDataType_AutomicAddPreCast =
+        conditional_t<is_same_v<DInDataType, float> || is_same_v<DInDataType, double>,
+                      DInDataType,
+                      float>;
+
+    using PassThrough  = ck::tensor_operation::element_wise::PassThrough;
+    using UnaryConvert = ck::tensor_operation::element_wise::UnaryConvert;
+
+    static constexpr auto I0 = Number<0>{};
+
+    template <typename Desc_M>
+    static auto PadDescriptor_M_1d(Desc_M desc_m, index_t loop_step)
+    {
+        const auto m   = desc_m.GetLength(I0);
+        const auto pad = math::integer_least_multiple(m, loop_step) - m;
+        const auto desc_m_pad =
+            transform_tensor_descriptor(desc_m,
+                                        make_tuple(make_right_pad_transform(m, pad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return desc_m_pad;
+    }
+
+    static auto MakeDescriptor_M(index_t length, index_t loop_step)
+    {
+        const auto desc_m = make_naive_tensor_descriptor_packed(make_tuple(length));
+        return PadDescriptor_M_1d(desc_m, loop_step);
+    }
+
+    using InOutGrid1dDesc = decltype(MakeDescriptor_M(1, 1));
+
+    using GridwisePutElementSet = GridwisePutElement_1D<InOutGrid1dDesc,
+                                                        DOutDataType,
+                                                        IndexDataType,
+                                                        DInDataType,
+                                                        PassThrough,
+                                                        InMemoryDataOperationEnum::Set,
+                                                        InOutVectorSize>;
+
+    using GridwisePutElementAtomicAdd = GridwisePutElement_1D<InOutGrid1dDesc,
+                                                              DOutDataType,
+                                                              IndexDataType,
+                                                              DInDataType_AutomicAddPreCast,
+                                                              PassThrough,
+                                                              InMemoryDataOperationEnum::AtomicAdd,
+                                                              InOutVectorSize>;
+
+    using GridwiseCasting = GridwiseElementwise_1D<Tuple<InOutGrid1dDesc>,
+                                                   Tuple<InOutGrid1dDesc>,
+                                                   Tuple<const DInDataType_AutomicAddPreCast*>,
+                                                   Tuple<DInDataType*>,
+                                                   UnaryConvert,
+                                                   InOutVectorSize,
+                                                   Sequence<InOutVectorSize>,
+                                                   Sequence<InOutVectorSize>>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const DOutDataType* p_dout,
+                 const IndexDataType* p_indices,
+                 DInDataType* p_din,
+                 index_t dout_length,
+                 index_t din_length,
+                 const std::vector<ck::index_t>& window_lengths,
+                 const std::vector<ck::index_t>& window_strides)
+            : p_dout_{p_dout},
+              p_indices_{p_indices},
+              p_din_{p_din},
+              dout_length_raw_{dout_length},
+              din_length_raw_{din_length},
+              blockSize_{256},
+              windowOverlap_{false}
+        {
+            for(size_t i = 0; i < window_lengths.size(); ++i)
+            {
+                windowOverlap_ |= window_lengths.at(i) > window_strides.at(i);
+            }
+        }
+
+        const DOutDataType* p_dout_;
+        const IndexDataType* p_indices_;
+        DInDataType* p_din_;
+        index_t dout_length_raw_;
+        index_t din_length_raw_;
+        index_t blockSize_;
+        bool windowOverlap_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            index_t gridSize               = getAvailableComputeUnitCount(stream_config);
+            index_t loop_step              = gridSize * arg.blockSize_ * InOutVectorSize;
+            InOutGrid1dDesc din_grid_desc  = MakeDescriptor_M(arg.din_length_raw_, loop_step);
+            InOutGrid1dDesc dout_grid_desc = MakeDescriptor_M(arg.dout_length_raw_, loop_step);
+
+            if constexpr(is_same_v<DInDataType, float> || is_same_v<DInDataType, double>)
+            {
+                hip_check_error(hipMemsetAsync(arg.p_din_,
+                                               0,
+                                               arg.din_length_raw_ * sizeof(DInDataType),
+                                               stream_config.stream_id_));
+
+                if(arg.windowOverlap_)
+                {
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementAtomicAdd,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType,
+                                                                  PassThrough>;
+
+                    return launch_and_time_kernel(stream_config,
+                                                  put_kernel,
+                                                  dim3(gridSize),
+                                                  dim3(arg.blockSize_),
+                                                  0,
+                                                  dout_grid_desc,
+                                                  arg.p_dout_,
+                                                  arg.p_indices_,
+                                                  arg.p_din_,
+                                                  PassThrough{});
+                }
+                else
+                {
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementSet,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType,
+                                                                  PassThrough>;
+
+                    return launch_and_time_kernel(stream_config,
+                                                  put_kernel,
+                                                  dim3(gridSize),
+                                                  dim3(arg.blockSize_),
+                                                  0,
+                                                  dout_grid_desc,
+                                                  arg.p_dout_,
+                                                  arg.p_indices_,
+                                                  arg.p_din_,
+                                                  PassThrough{});
+                }
+            }
+            else
+            {
+                if(arg.windowOverlap_)
+                {
+                    if(arg.p_workspace_ == nullptr)
+                        throw std::runtime_error("wrong! WorkSpace pointer has not been set");
+
+                    hip_check_error(
+                        hipMemsetAsync(arg.p_workspace_,
+                                       0,
+                                       arg.din_length_raw_ * sizeof(DInDataType_AutomicAddPreCast),
+                                       stream_config.stream_id_));
+
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementAtomicAdd,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType_AutomicAddPreCast,
+                                                                  PassThrough>;
+
+                    const auto cast_kernel =
+                        kernel_elementwise_1d<GridwiseCasting,
+                                              Tuple<InOutGrid1dDesc>,
+                                              Tuple<InOutGrid1dDesc>,
+                                              Tuple<const DInDataType_AutomicAddPreCast*>,
+                                              Tuple<DInDataType*>,
+                                              UnaryConvert>;
+
+                    float elapsed_time = launch_and_time_kernel(
+                        stream_config,
+                        put_kernel,
+                        dim3(gridSize),
+                        dim3(arg.blockSize_),
+                        0,
+                        dout_grid_desc,
+                        arg.p_dout_,
+                        arg.p_indices_,
+                        static_cast<DInDataType_AutomicAddPreCast*>(arg.p_workspace_),
+                        PassThrough{});
+
+                    elapsed_time += launch_and_time_kernel(
+                        stream_config,
+                        cast_kernel,
+                        dim3(gridSize),
+                        dim3(arg.blockSize_),
+                        0,
+                        ck::make_tuple(din_grid_desc),
+                        ck::make_tuple(din_grid_desc),
+                        static_cast<DInDataType_AutomicAddPreCast*>(arg.p_workspace_),
+                        arg.p_din_,
+                        UnaryConvert{});
+
+                    return elapsed_time;
+                }
+                else
+                {
+                    const auto put_kernel = kernel_put_element_1d<GridwisePutElementSet,
+                                                                  InOutGrid1dDesc,
+                                                                  DOutDataType,
+                                                                  IndexDataType,
+                                                                  DInDataType,
+                                                                  PassThrough>;
+
+                    hip_check_error(hipMemsetAsync(arg.p_din_,
+                                                   0,
+                                                   arg.din_length_raw_ * sizeof(DInDataType),
+                                                   stream_config.stream_id_));
+
+                    return launch_and_time_kernel(stream_config,
+                                                  put_kernel,
+                                                  dim3(gridSize),
+                                                  dim3(arg.blockSize_),
+                                                  0,
+                                                  dout_grid_desc,
+                                                  arg.p_dout_,
+                                                  arg.p_indices_,
+                                                  arg.p_din_,
+                                                  PassThrough{});
+                }
+            }
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        bool needCast = pArg_->windowOverlap_ &&
+                        !(is_same_v<DInDataType, float> || is_same_v<DInDataType, double>);
+
+        if(!needCast)
+            return 0;
+        else
+            return pArg_->din_length_raw_ * sizeof(DInDataType_AutomicAddPreCast);
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
+        if(pArg->din_length_raw_ % InOutVectorSize != 0 ||
+           pArg->dout_length_raw_ % InOutVectorSize != 0)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_dout,
+                        const void* p_indices,
+                        void* p_din,
+                        index_t dout_length,
+                        index_t din_length,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> window_strides) override
+    {
+        // Assume p_dout, p_indices, p_din are packed memory space, dout_length and din_length are
+        // physical size of the packed tensor
+        return std::make_unique<Argument>(static_cast<const DOutDataType*>(p_dout),
+                                          static_cast<const IndexDataType*>(p_indices),
+                                          static_cast<DInDataType*>(p_din),
+                                          dout_length,
+                                          din_length,
+                                          window_lengths,
+                                          window_strides);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp
new file mode 100644
index 0000000000..7334da0e35
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_put_element_impl.hpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/device_put_element.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/stream_utility.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// output[indices] = input
+template <typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation,
+          InMemoryDataOperationEnum MemOp,
+          ck::index_t InVectorSize>
+struct DevicePutElementImpl
+    : public DevicePutElement<InDataType, IndexDataType, OutDataType, ElementwiseOperation, MemOp>
+{
+    template <typename Desc_M>
+    static auto PadDescriptor_M_1d(Desc_M desc_m, index_t gridSize, index_t blockSize)
+    {
+        constexpr auto I0 = Number<0>{};
+
+        const auto m            = desc_m.GetLength(I0);
+        const index_t loop_step = gridSize * blockSize * InVectorSize;
+        const auto pad          = math::integer_least_multiple(m, loop_step) - m;
+        const auto desc_m_pad =
+            transform_tensor_descriptor(desc_m,
+                                        make_tuple(make_right_pad_transform(m, pad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return desc_m_pad;
+    }
+
+    static auto MakeDescriptor_M(index_t length, index_t gridSize, index_t blockSize)
+    {
+        const auto desc_m = make_naive_tensor_descriptor_packed(make_tuple(length));
+        return PadDescriptor_M_1d(desc_m, gridSize, blockSize);
+    }
+
+    using InGrid1dDesc = decltype(MakeDescriptor_M(1, 1, 1));
+
+    using GridwisePutElement = GridwisePutElement_1D<InGrid1dDesc,
+                                                     InDataType,
+                                                     IndexDataType,
+                                                     OutDataType,
+                                                     ElementwiseOperation,
+                                                     MemOp,
+                                                     InVectorSize>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const InDataType* p_input,
+                 const IndexDataType* p_indices,
+                 OutDataType* p_output,
+                 index_t input_length,
+                 ElementwiseOperation elementwise_op)
+            : p_input_{p_input},
+              p_indices_{p_indices},
+              p_output_{p_output},
+              input_length_raw_{input_length},
+              elementwise_op_{elementwise_op},
+              blockSize_{256}
+        {
+        }
+
+        const InDataType* p_input_;
+        const IndexDataType* p_indices_;
+        OutDataType* p_output_;
+        index_t input_length_raw_;
+        ElementwiseOperation elementwise_op_;
+        index_t blockSize_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            index_t gridSize = getAvailableComputeUnitCount(stream_config);
+            InGrid1dDesc in_grid_desc =
+                MakeDescriptor_M(arg.input_length_raw_, gridSize, arg.blockSize_);
+
+            const auto kernel = kernel_put_element_1d<GridwisePutElement,
+                                                      InGrid1dDesc,
+                                                      InDataType,
+                                                      IndexDataType,
+                                                      OutDataType,
+                                                      ElementwiseOperation>;
+
+            float elapsed_time = launch_and_time_kernel(stream_config,
+                                                        kernel,
+                                                        dim3(gridSize),
+                                                        dim3(arg.blockSize_),
+                                                        0,
+                                                        in_grid_desc,
+                                                        arg.p_input_,
+                                                        arg.p_indices_,
+                                                        arg.p_output_,
+                                                        arg.elementwise_op_);
+            return elapsed_time;
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
+
+        if(pArg->input_length_raw_ % InVectorSize != 0)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_input,
+                                                      const void* p_indices,
+                                                      void* p_output,
+                                                      index_t input_length,
+                                                      index_t,
+                                                      ElementwiseOperation elementwise_op) override
+    {
+        return std::make_unique<Argument>(static_cast<const InDataType*>(p_input),
+                                          static_cast<const IndexDataType*>(p_indices),
+                                          static_cast<OutDataType*>(p_output),
+                                          input_length,
+                                          elementwise_op);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
new file mode 100644
index 0000000000..8f72f88b06
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+
+namespace ck {
+
+template <typename GridwisePutElementwise1dFunctor,
+          typename InGrid1dDesc,
+          typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation>
+__global__ void kernel_put_element_1d(const InGrid1dDesc in_grid_1d_desc,
+                                      const InDataType* __restrict__ p_in_global,
+                                      const IndexDataType* __restrict__ p_indices_global,
+                                      OutDataType* __restrict__ p_out_global,
+                                      const ElementwiseOperation elementwise_op)
+{
+    GridwisePutElementwise1dFunctor::Run(
+        in_grid_1d_desc, p_in_global, p_indices_global, p_out_global, elementwise_op);
+}
+
+// output[indices] = input
+template <typename InGrid1dDesc,
+          typename InDataType,
+          typename IndexDataType,
+          typename OutDataType,
+          typename ElementwiseOperation,
+          InMemoryDataOperationEnum MemOp,
+          index_t InVectorSize>
+struct GridwisePutElement_1D
+{
+    static constexpr auto I0 = Number<0>{};
+
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<InVectorSize>{}));
+
+    __device__ static void Run(const InGrid1dDesc& in_grid_1d_desc,
+                               const InDataType* __restrict__ p_in_global,
+                               const IndexDataType* __restrict__ p_indices_global,
+                               OutDataType* __restrict__ p_out_global,
+                               const ElementwiseOperation& elementwise_op)
+    {
+        // Global Memory
+        const auto in_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_global, in_grid_1d_desc.GetElementSpaceSize());
+
+        const auto indices_global_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global>(p_indices_global,
+                                                          in_grid_1d_desc.GetElementSpaceSize(),
+                                                          NumericLimits<IndexDataType>::Lowest());
+
+        // VGPR
+        StaticBuffer<AddressSpaceEnum::Vgpr, InDataType, InVectorSize, true> in_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, IndexDataType, InVectorSize, true> indices_thread_buf;
+
+        // Thread id, Block id and index
+        const index_t thread_global_id  = get_thread_global_1d_id();
+        const auto thread_global_offset = make_multi_index(thread_global_id * InVectorSize);
+        const index_t blockSize         = get_block_size();
+        const index_t blockPerGrid      = get_grid_size();
+        const auto M                    = in_grid_1d_desc.GetLength(I0);
+        const index_t loop_step         = blockPerGrid * blockSize * InVectorSize;
+        const auto loop_step_index      = make_multi_index(loop_step);
+
+        auto in_global_load =
+            ThreadwiseTensorSliceTransfer_v2<InDataType,
+                                             InDataType,
+                                             decltype(in_grid_1d_desc),
+                                             decltype(thread_buffer_desc_m),
+                                             Sequence<InVectorSize>, // SliceLengths
+                                             Sequence<0>,            // DimAccessOrder
+                                             0,                      // SrcVectorDim
+                                             InVectorSize,           // ScalarPerVector
+                                             1,                      // SrcScalarStrideInVector
+                                             false>{in_grid_1d_desc, thread_global_offset};
+
+        auto indices_global_load =
+            ThreadwiseTensorSliceTransfer_v2<IndexDataType,
+                                             IndexDataType,
+                                             decltype(in_grid_1d_desc),
+                                             decltype(thread_buffer_desc_m),
+                                             Sequence<InVectorSize>, // SliceLengths
+                                             Sequence<0>,            // DimAccessOrder
+                                             0,                      // SrcVectorDim
+                                             InVectorSize,           // ScalarPerVector
+                                             1,                      // SrcScalarStrideInVector
+                                             false>{in_grid_1d_desc, thread_global_offset};
+
+        index_t num_iter = M / loop_step;
+        do
+        {
+            in_global_load.Run(in_grid_1d_desc,
+                               in_global_buf,
+                               thread_buffer_desc_m,
+                               make_tuple(I0),
+                               in_thread_buf);
+
+            in_global_load.MoveSrcSliceWindow(in_grid_1d_desc, loop_step_index);
+
+            static_for<0, InVectorSize, 1>{}(
+                [&](auto iM) { elementwise_op(in_thread_buf(iM), in_thread_buf[iM]); });
+
+            indices_global_load.Run(in_grid_1d_desc,
+                                    indices_global_buf,
+                                    thread_buffer_desc_m,
+                                    make_tuple(I0),
+                                    indices_thread_buf);
+
+            indices_global_load.MoveSrcSliceWindow(in_grid_1d_desc, loop_step_index);
+
+            static_for<0, InVectorSize, 1>{}([&](auto iM) {
+                if(indices_thread_buf[iM] >= 0)
+                {
+                    if constexpr(MemOp == InMemoryDataOperationEnum::Set)
+                    {
+                        // User should guarantee each index in p_indices_global is different
+                        *(p_out_global + indices_thread_buf[iM]) =
+                            ck::type_convert<OutDataType>(in_thread_buf[iM]);
+                    }
+                    else if constexpr(MemOp == InMemoryDataOperationEnum::AtomicAdd)
+                    {
+                        atomic_add<OutDataType>(p_out_global + indices_thread_buf[iM],
+                                                ck::type_convert<OutDataType>(in_thread_buf[iM]));
+                    }
+                    else if constexpr(MemOp == InMemoryDataOperationEnum::AtomicMax)
+                    {
+                        atomic_max<OutDataType>(p_out_global + indices_thread_buf[iM],
+                                                ck::type_convert<OutDataType>(in_thread_buf[iM]));
+                    }
+                    else if constexpr(MemOp == InMemoryDataOperationEnum::Add)
+                    {
+                        // User should guarantee each index in p_indices_global is different
+                        *(p_out_global + indices_thread_buf[iM]) +=
+                            ck::type_convert<OutDataType>(in_thread_buf[iM]);
+                    }
+                    else
+                    {
+                        static_assert(MemOp == InMemoryDataOperationEnum::Set ||
+                                      MemOp == InMemoryDataOperationEnum::AtomicAdd ||
+                                      MemOp == InMemoryDataOperationEnum::AtomicMax ||
+                                      MemOp == InMemoryDataOperationEnum::Add);
+                    }
+                }
+            });
+
+        } while(--num_iter);
+    }
+};
+
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
new file mode 100644
index 0000000000..3f1fc6165c
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
@@ -0,0 +1,103 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+using namespace std;
+
+template <typename DOutDataType,
+          typename IndexDataType,
+          typename ConputeDataType,
+          typename DInDataType,
+          typename ElementwiseOperation>
+struct ReferenceMaxPoolBwd : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(const Tensor<DOutDataType>& dout,
+                 const Tensor<IndexDataType>& indices,
+                 Tensor<DInDataType>& din,
+                 ElementwiseOperation elementwise_op)
+            : dout_(dout), indices_(indices), din_(din), elementwise_op_(elementwise_op)
+        {
+        }
+
+        const Tensor<DOutDataType>& dout_;
+        const Tensor<IndexDataType>& indices_;
+        Tensor<DInDataType>& din_;
+        ElementwiseOperation elementwise_op_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        float Run(const Argument& arg)
+        {
+            int din_length  = arg.din_.GetElementSpaceSize();
+            int dout_length = arg.dout_.GetElementSpaceSize();
+            std::vector<ConputeDataType> buf(din_length, 0);
+
+            for(int i = 0; i < dout_length; ++i)
+            {
+                int index = arg.indices_.mData[i];
+                if(index >= 0 && index < din_length)
+                    buf[index] += ck::type_convert<ConputeDataType>(arg.dout_.mData[i]);
+            }
+
+            for(int i = 0; i < din_length; ++i)
+                arg.din_.mData[i] = ck::type_convert<DInDataType>(buf[i]);
+            return 0;
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(const Tensor<DOutDataType>& dout,
+                             const Tensor<IndexDataType>& indices,
+                             Tensor<DInDataType>& din,
+                             ElementwiseOperation elementwise_op)
+    {
+        return Argument{dout, indices, din, elementwise_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceMaxPoolBwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
index b4b7a5a032..696fb5eaf1 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
@@ -100,8 +100,8 @@ struct ReferencePoolingFwd : public device::BaseOperator
                                    wi >= 0 &&
                                    wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[4]))
                                 {
-                                    ComputeDataType currVal =
-                                        static_cast<ComputeDataType>(arg.in_(n, c, di, hi, wi));
+                                    ComputeDataType currVal = ck::type_convert<ComputeDataType>(
+                                        arg.in_(n, c, di, hi, wi));
 
                                     in_elementwise_op(currVal, currVal);
 
@@ -112,7 +112,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                     }
                     acc_elementwise_op(accuVal, accuVal);
 
-                    arg.out_(n, c, do_, ho, wo) = accuVal;
+                    arg.out_(n, c, do_, ho, wo) = ck::type_convert<OutDataType>(accuVal);
                 };
 
                 make_ParallelTensorFunctor(f_ncdhw,
@@ -151,8 +151,8 @@ struct ReferencePoolingFwd : public device::BaseOperator
                                    wi >= 0 &&
                                    wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[4]))
                                 {
-                                    ComputeDataType currVal =
-                                        static_cast<ComputeDataType>(arg.in_(n, c, di, hi, wi));
+                                    ComputeDataType currVal = ck::type_convert<ComputeDataType>(
+                                        arg.in_(n, c, di, hi, wi));
                                     IndexDataType currIndex =
                                         arg.in_.GetOffsetFromMultiIndex(n, c, di, hi, wi);
 
@@ -166,7 +166,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
 
                     acc_elementwise_op(accuVal, accuVal);
 
-                    arg.out_(n, c, do_, ho, wo)         = accuVal;
+                    arg.out_(n, c, do_, ho, wo)         = ck::type_convert<OutDataType>(accuVal);
                     arg.out_indices_(n, c, do_, ho, wo) = accuIndex;
                 };
 
@@ -212,7 +212,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                                wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]))
                             {
                                 ComputeDataType currVal =
-                                    static_cast<ComputeDataType>(arg.in_(n, c, hi, wi));
+                                    ck::type_convert<ComputeDataType>(arg.in_(n, c, hi, wi));
 
                                 in_elementwise_op(currVal, currVal);
 
@@ -222,7 +222,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                     }
 
                     acc_elementwise_op(accuVal, accuVal);
-                    arg.out_(n, c, ho, wo) = accuVal;
+                    arg.out_(n, c, ho, wo) = ck::type_convert<OutDataType>(accuVal);
                 };
 
                 make_ParallelTensorFunctor(f_nchw,
@@ -255,7 +255,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                                wi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[3]))
                             {
                                 ComputeDataType currVal =
-                                    static_cast<ComputeDataType>(arg.in_(n, c, hi, wi));
+                                    ck::type_convert<ComputeDataType>(arg.in_(n, c, hi, wi));
 
                                 IndexDataType currIndex =
                                     arg.in_.GetOffsetFromMultiIndex(n, c, hi, wi);
@@ -268,7 +268,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                     }
 
                     acc_elementwise_op(accuVal, accuVal);
-                    arg.out_(n, c, ho, wo)         = accuVal;
+                    arg.out_(n, c, ho, wo)         = ck::type_convert<OutDataType>(accuVal);
                     arg.out_indices_(n, c, ho, wo) = accuIndex;
                 };
 

From f0c620c42e753432ded96040abdac1bbae89aab5 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Mon, 19 Jun 2023 11:20:35 -0500
Subject: [PATCH 042/194] FP8 enablement - add a pseudorandom number generator,
 add conversion methods (#708)

* Add basic fp8 definitions and prn-generator

* Format

* Add fp8<->fp32 type_convert

* Format

* Split type_convert and cast_to/from_f8

* Format

* Minor fix

* Minor fix

* Move fp8 utils to a separate header

* Add elementwise ops

* Add fp8_convert_sr

* Format

* Add element op

* Eliminate magic numbers

* Split f8_convert_sr in host and device

* Format

* Add some constexpr

* Add a datatype test

* Format

* Another format

* Add fp8<->fp16 tests

* Update type_converts

* Format

* Add fp16 casting functions

* Format

* Use seed as a runtime arg

* Use element location for PRNG

* Format

* Add fp8<->fp16 to PassThrough element op

* Clean up

* Merge host and device implementations

* Add comments on rounding modes

* Remove leftover code

* Put type_converts into a separate header

* Put random number gen to a separate header

* Rearrange f8_utils' namespaces

* Refactor type_convert.hpp

* Move f8_t definition
---
 .../element/unary_element_wise_operation.hpp  |  48 ++++
 include/ck/utility/common_header.hpp          |   1 +
 include/ck/utility/data_type.hpp              | 177 +++----------
 include/ck/utility/f8_utils.hpp               | 250 ++++++++++++++++++
 include/ck/utility/inner_product.hpp          |   1 +
 include/ck/utility/random_gen.hpp             |  53 ++++
 include/ck/utility/reduction_operator.hpp     |   1 +
 include/ck/utility/type_convert.hpp           | 230 ++++++++++++++++
 .../ck/library/utility/host_tensor.hpp        |   1 +
 test/data_type/CMakeLists.txt                 |   3 +
 test/data_type/fp8.cpp                        | 123 +++++++++
 11 files changed, 743 insertions(+), 145 deletions(-)
 create mode 100644 include/ck/utility/f8_utils.hpp
 create mode 100644 include/ck/utility/random_gen.hpp
 create mode 100644 include/ck/utility/type_convert.hpp
 create mode 100644 test/data_type/fp8.cpp

diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index c3e7706ef3..4fb061fadb 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -6,6 +6,7 @@
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/math.hpp"
 #include "ck/utility/math_v2.hpp"
+#include "ck/utility/type_convert.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -81,6 +82,36 @@ struct PassThrough
         y = x;
     }
 #endif
+
+    template <>
+    __host__ __device__ void operator()<f8_t, f8_t>(f8_t& y, const f8_t& x) const
+    {
+        y = x;
+    }
+
+    template <>
+    __host__ __device__ void operator()<float, f8_t>(float& y, const f8_t& x) const
+    {
+        y = type_convert<float>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<f8_t, float>(f8_t& y, const float& x) const
+    {
+        y = type_convert<f8_t>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<half_t, f8_t>(half_t& y, const f8_t& x) const
+    {
+        y = type_convert<half_t>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<f8_t, half_t>(f8_t& y, const half_t& x) const
+    {
+        y = type_convert<f8_t>(x);
+    }
 };
 
 struct UnaryConvert
@@ -109,6 +140,23 @@ struct ConvertBF16RTN
     }
 };
 
+struct ConvertF8SR
+{
+    // convert to fp8 using stochastic rounding (SR)
+    template <typename Y, typename X>
+    __host__ __device__ void operator()(Y& y, const X& x) const
+    {
+        // check Y datatype
+        static_assert(is_same<Y, f8_t>::value, "Data type is not supported by this operation!");
+
+        // check X datatype
+        static_assert(is_same<X, float>::value || is_same<X, half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        y = f8_convert_sr<Y>(x);
+    }
+};
+
 struct Scale
 {
     __host__ __device__ Scale(float scale) : scale_(scale) {}
diff --git a/include/ck/utility/common_header.hpp b/include/ck/utility/common_header.hpp
index 41a9d0b585..f95660a8a4 100644
--- a/include/ck/utility/common_header.hpp
+++ b/include/ck/utility/common_header.hpp
@@ -24,6 +24,7 @@
 #include "ck/utility/tuple.hpp"
 #include "ck/utility/tuple_helper.hpp"
 #include "ck/utility/type.hpp"
+#include "ck/utility/type_convert.hpp"
 #include "ck/utility/magic_division.hpp"
 #include "ck/utility/c_style_pointer_cast.hpp"
 #include "ck/utility/is_known_at_compile_time.hpp"
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index d43af8a2e3..c240afa2b8 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -12,6 +12,7 @@ using half_t  = _Float16;
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 using int4_t = _BitInt(4);
 #endif
+using f8_t = uint8_t;
 
 // vector_type
 template <typename T, index_t N>
@@ -142,6 +143,13 @@ struct scalar_type<int4_t>
 };
 #endif
 
+template <>
+struct scalar_type<f8_t>
+{
+    using type                           = f8_t;
+    static constexpr index_t vector_size = 1;
+};
+
 //
 template <typename T>
 struct vector_type<T, 1>
@@ -944,151 +952,13 @@ using int8x16_t = typename vector_type<int8_t, 16>::type;
 using int8x32_t = typename vector_type<int8_t, 32>::type;
 using int8x64_t = typename vector_type<int8_t, 64>::type;
 
-// Convert X to Y
-template <typename Y, typename X>
-__host__ __device__ constexpr Y type_convert(X x)
-{
-    static_assert(!std::is_reference_v<Y> && !std::is_reference_v<X>);
-
-    return static_cast<Y>(x);
-}
-
-// convert bfp16 to fp32
-template <>
-inline __host__ __device__ constexpr float type_convert<float, bhalf_t>(bhalf_t x)
-{
-    union
-    {
-        uint32_t int32;
-        float fp32;
-    } u = {uint32_t(x) << 16};
-
-    return u.fp32;
-}
-
-// convert fp32 to bfp16
-template <>
-inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, float>(float x)
-{
-    union
-    {
-        float fp32;
-        uint32_t int32;
-    } u = {x};
-
-    return uint16_t(u.int32 >> 16);
-}
-
-// convert bfp16 to fp16 via fp32
-template <>
-inline __host__ __device__ constexpr half_t type_convert<half_t, bhalf_t>(bhalf_t x)
-{
-    float x_fp32 = type_convert<float>(x);
-
-    return static_cast<half_t>(x_fp32);
-}
-
-// convert fp16 to bfp16 via fp32
-template <>
-inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, half_t>(half_t x)
-{
-    float x_fp32 = static_cast<float>(x);
-
-    return type_convert<bhalf_t>(x_fp32);
-}
-
-// convert bfp16 to int32 via fp32
-template <>
-inline __host__ __device__ constexpr int32_t type_convert<int32_t, bhalf_t>(bhalf_t x)
-{
-    float x_fp32 = type_convert<float>(x);
-
-    return static_cast<int32_t>(x_fp32);
-}
-
-// convert int32 to bfp16 via fp32
-template <>
-inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int32_t>(int32_t x)
-{
-    float x_fp32 = static_cast<float>(x);
-
-    return type_convert<bhalf_t>(x_fp32);
-}
-
-// convert bfp16 to int8 via fp32
-template <>
-inline __host__ __device__ constexpr int8_t type_convert<int8_t, bhalf_t>(bhalf_t x)
-{
-    float x_fp32 = type_convert<float>(x);
-
-    return static_cast<int8_t>(x_fp32);
-}
-
-// convert int8 to bfp16 via fp32
-template <>
-inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_t x)
-{
-    float x_fp32 = static_cast<float>(x);
-
-    return type_convert<bhalf_t>(x_fp32);
-}
-
-// Declare a template function for bf16 conversion using RTN
-template <typename Y, typename X>
-__host__ __device__ constexpr Y bf16_convert_rtn(X x);
-
-// Convert fp32 to bf16 with RTN if higher precision is needed
-template <>
-inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, float>(float x)
-{
-    union
-    {
-        float fp32;
-        uint32_t int32;
-    } u = {x};
-
-    // When the exponent bits are not all 1s, then the value is zero, normal,
-    // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
-    // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
-    // This causes the bfloat16's mantissa to be incremented by 1 if the 16
-    // least significant bits of the float mantissa are greater than 0x8000,
-    // or if they are equal to 0x8000 and the least significant bit of the
-    // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
-    // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
-    // has the value 0x7f, then incrementing it causes it to become 0x00 and
-    // the exponent is incremented by one, which is the next higher FP value
-    // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
-    // with an exponent of 0x00 and a mantissa of 0x7f, it may be rounded up
-    // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
-    // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
-    // incrementing it causes it to become an exponent of 0xFF and a mantissa
-    // of 0x00, which is Inf, the next higher value to the unrounded value.
-    bool flag0 = ~u.int32 & 0x7f800000;
-
-    // When all of the exponent bits are 1, the value is Inf or NaN.
-    // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
-    // mantissa bit. Quiet NaN is indicated by the most significant mantissa
-    // bit being 1. Signaling NaN is indicated by the most significant
-    // mantissa bit being 0 but some other bit(s) being 1. If any of the
-    // lower 16 bits of the mantissa are 1, we set the least significant bit
-    // of the bfloat16 mantissa, in order to preserve signaling NaN in case
-    // the bfloat16's mantissa bits are all 0.
-    bool flag1 = !flag0 && (u.int32 & 0xffff);
-
-    u.int32 += flag0 ? 0x7fff + ((u.int32 >> 16) & 1) : 0; // Round to nearest, round to even
-    u.int32 |= flag1 ? 0x10000 : 0x0;                      // Preserve signaling NaN
-
-    return uint16_t(u.int32 >> 16);
-}
-
-// convert fp16 to bfp16 via fp32 with RTN if higher precision is needed
-template <>
-inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(half_t x)
-{
-    float x_fp32 = static_cast<float>(x);
-
-    return bf16_convert_rtn<bhalf_t>(x_fp32);
-}
+// f8
+using f8x2_t  = typename vector_type<f8_t, 2>::type;
+using f8x4_t  = typename vector_type<f8_t, 4>::type;
+using f8x8_t  = typename vector_type<f8_t, 8>::type;
+using f8x16_t = typename vector_type<f8_t, 16>::type;
+using f8x32_t = typename vector_type<f8_t, 32>::type;
+using f8x64_t = typename vector_type<f8_t, 64>::type;
 
 template <typename T>
 struct NumericLimits
@@ -1136,4 +1006,21 @@ struct NumericLimits<int4_t>
 };
 #endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 
+template <>
+struct NumericLimits<f8_t>
+{
+    static constexpr uint8_t binary_min    = 0x08; // 0b00001000
+    static constexpr uint8_t binary_max    = 0x77; // 0b01110111
+    static constexpr uint8_t binary_lowest = 0xF7; // 0b11110111
+    static constexpr uint8_t binary_qnan   = 0x80; // 0b10000000
+
+    __host__ __device__ static constexpr f8_t Min() { return bit_cast<f8_t>(binary_min); }
+
+    __host__ __device__ static constexpr f8_t Max() { return bit_cast<f8_t>(binary_max); }
+
+    __host__ __device__ static constexpr f8_t Lowest() { return bit_cast<f8_t>(binary_lowest); }
+
+    __host__ __device__ static constexpr f8_t QuietNaN() { return bit_cast<f8_t>(binary_qnan); }
+};
+
 } // namespace ck
diff --git a/include/ck/utility/f8_utils.hpp b/include/ck/utility/f8_utils.hpp
new file mode 100644
index 0000000000..bb13f98154
--- /dev/null
+++ b/include/ck/utility/f8_utils.hpp
@@ -0,0 +1,250 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+
+namespace ck {
+
+// fp8 rounding modes
+// use standard for rounding to nearest, the faster one
+// use stochastic for stochastic rounding, helps to avoid error accumulation
+enum class f8_rounding_mode
+{
+    standard,
+    stochastic
+};
+
+} // namespace ck
+
+namespace ck::utils {
+
+namespace {
+
+template <typename T, bool negative_zero_nan, bool clip, bool stoch>
+__host__ __device__ f8_t run_cast_to_f8(T x, uint32_t rng)
+{
+    // check data type
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+
+    // fp8 exponent/mantissa layout
+    constexpr int f8_exp  = 4;
+    constexpr int f8_mant = 3;
+
+    // resulting type exponent/mantissa layout
+    constexpr int type_exp  = is_half ? 5 : 8;
+    constexpr int type_mant = is_half ? 10 : 23;
+
+    int exponent;
+    uint32_t head, mantissa, sign;
+    // nan code is same for float and half
+    constexpr uint8_t nan_code  = 0x80;
+    constexpr uint32_t nan_mask = is_half ? 0x7C00 : 0x7F800000;
+
+    // convert to bitwise
+    typedef typename std::conditional<std::is_same<T, half_t>::value, uint16_t, uint32_t>::type
+        T_bitwise;
+    T_bitwise x_bitwise = *(reinterpret_cast<T_bitwise*>(&x));
+
+    // unpack the input, depends on datatype
+    if constexpr(is_float)
+    {
+        head     = x_bitwise & 0xFF800000;
+        mantissa = x_bitwise & 0x7FFFFF;
+        exponent = (head >> type_mant) & 0xFF;
+        sign     = head >> (type_exp + type_mant);
+    }
+    else if constexpr(is_half)
+    {
+        head     = x_bitwise & 0xFC00;
+        mantissa = x_bitwise & 0x3FF;
+        exponent = (head >> type_mant) & 0x1F;
+        sign     = head >> (type_exp + type_mant);
+    }
+
+    uint32_t signed_inf   = (sign << (type_exp + type_mant)) + (((1 << type_exp) - 1) << type_mant);
+    uint32_t drop_mask    = (1 << (type_mant - f8_mant)) - 1;
+    constexpr int max_exp = (1 << f8_exp) - (negative_zero_nan ? 1 : 2);
+    constexpr int exp_low_cutoff =
+        (1 << (type_exp - 1)) - (1 << (f8_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
+
+    if constexpr(negative_zero_nan)
+    {
+        if((x_bitwise & nan_mask) == nan_mask)
+            return nan_code;
+    }
+    else
+    {
+        if((x_bitwise & nan_mask) == nan_mask)
+            return signed_inf + (mantissa != 0 ? 1 : 0);
+    }
+
+    // check if x is 0.0
+    if(x_bitwise == 0)
+        return 0;
+
+    exponent -= exp_low_cutoff - 1;
+    if(exponent <= 0)
+        drop_mask = (1 << (type_mant - f8_mant + 1 - exponent)) - 1;
+    mantissa += 1 << type_mant;
+    // apply random number if needed
+    mantissa += (stoch ? rng : mantissa) & drop_mask;
+    if(mantissa >= (2 << type_mant))
+    {
+        mantissa >>= 1;
+        exponent++;
+    }
+    mantissa >>= (type_mant - f8_mant);
+
+    // check negative exponent
+    if(exponent <= 0)
+    {
+        if(x_bitwise == 0)
+            return 0;
+        else
+        {
+            // subnormal range; represented by a subnormal float8 (exponent 0)
+            // and involves loss of accuracy
+            mantissa >>= 1 - exponent;
+            exponent = 0;
+        }
+    }
+    // above range: quantize to maximum possible float of the same sign
+    else if(exponent > max_exp)
+    {
+        if(clip)
+        {
+            mantissa = (1 << f8_mant) - 1;
+            exponent = max_exp;
+        }
+        else
+        {
+            return signed_inf;
+        }
+    }
+
+    // check if x is 0.0 or -0.0
+    if(exponent == 0 && mantissa == 0)
+        return negative_zero_nan ? 0 : (sign << (f8_exp + f8_mant));
+    mantissa &= (1 << f8_mant) - 1;
+    return (sign << (f8_exp + f8_mant)) | (exponent << f8_mant) | mantissa;
+}
+
+template <typename T, bool negative_zero_nan>
+__host__ __device__ T run_cast_from_f8(f8_t x)
+{
+    // check data type
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+
+    // fp8 exponent/mantissa layout
+    constexpr int f8_exp  = 4;
+    constexpr int f8_mant = 3;
+
+    // resulting type exponent/mantissa layout
+    constexpr int type_exp  = is_half ? 5 : 8;
+    constexpr int type_mant = is_half ? 10 : 23;
+
+    // prepare the codes
+    constexpr uint8_t nan_code = 0x80;
+    T fInf, fNegInf, fNaN, fNeg0;
+    if constexpr(is_half)
+    {
+        constexpr uint16_t ihInf    = 0x7C00;
+        constexpr uint16_t ihNegInf = 0xFC00;
+        constexpr uint16_t ihNaN    = 0x7C01;
+        constexpr uint16_t ihNeg0   = 0x8000;
+        fInf                        = *(reinterpret_cast<const half_t*>(&ihInf));
+        fNegInf                     = *(reinterpret_cast<const half_t*>(&ihNegInf));
+        fNaN                        = *(reinterpret_cast<const half_t*>(&ihNaN));
+        fNeg0                       = *(reinterpret_cast<const half_t*>(&ihNeg0));
+    }
+    else if constexpr(is_float)
+    {
+        constexpr uint32_t ifInf    = 0x7F800000;
+        constexpr uint32_t ifNegInf = 0xFF800000;
+        constexpr uint32_t ifNaN    = 0x7F800001;
+        constexpr uint32_t ifNeg0   = 0x80000000;
+        fInf                        = *(reinterpret_cast<const float*>(&ifInf));
+        fNegInf                     = *(reinterpret_cast<const float*>(&ifNegInf));
+        fNaN                        = *(reinterpret_cast<const float*>(&ifNaN));
+        fNeg0                       = *(reinterpret_cast<const float*>(&ifNeg0));
+    }
+
+    // unpack the input
+    uint32_t sign     = x >> (f8_exp + f8_mant);
+    uint32_t mantissa = x & ((1 << f8_mant) - 1);
+    int exponent      = (x & 0x7F) >> f8_mant;
+
+    constexpr int exp_low_cutoff =
+        (1 << (type_exp - 1)) - (1 << (f8_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
+    typename std::conditional<std::is_same<T, half_t>::value, uint16_t, uint32_t>::type retval;
+
+    if constexpr(negative_zero_nan)
+    {
+        if(x == nan_code)
+            return fNaN;
+    }
+    else
+    {
+        if(x == nan_code)
+            return fNeg0;
+        if(exponent == ((1 << f8_exp) - 1))
+            return (mantissa == 0) ? (sign ? fNegInf : fInf) : fNaN;
+    }
+
+    // subnormal input
+    if(exponent == 0)
+    {
+        // guaranteed mantissa!=0 since cases 0x0 and 0x80 are handled above
+        int sh = 1 + __builtin_clz(mantissa) - ((1 + type_exp + type_mant) - f8_mant);
+        mantissa <<= sh;
+        mantissa &= ((1 << f8_mant) - 1);
+        exponent += 1 - sh;
+    }
+    exponent += exp_low_cutoff - 1;
+    mantissa <<= type_mant - f8_mant;
+
+    // subnormal output (occurs when T=half, we=5, negative_zero_nan=true)
+    if(exponent <= 0)
+    {
+        mantissa |= 1 << type_mant;
+        mantissa >>= 1 - exponent;
+        exponent = 0;
+    }
+
+    retval = (sign << (type_exp + type_mant)) | (exponent << type_mant) | mantissa;
+    return *(reinterpret_cast<const T*>(&retval));
+}
+
+} // namespace
+
+template <typename T, bool negative_zero_nan, bool clip, bool stoch>
+__host__ __device__ f8_t cast_to_f8(T x, uint32_t rng)
+{
+    // check datatype
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+    static_assert(is_half || is_float, "Only half and float can be casted to f8.");
+
+    return run_cast_to_f8<T, negative_zero_nan, clip, stoch>(x, rng);
+}
+
+template <typename T, bool negative_zero_nan>
+__host__ __device__ T cast_from_f8(f8_t x)
+{
+    // check datatype
+    constexpr bool is_half  = std::is_same<T, half_t>::value;
+    constexpr bool is_float = std::is_same<T, float>::value;
+    static_assert(is_half || is_float, "only half and float are supported.");
+
+    // check if x is 0.0
+    if(x == 0)
+        return static_cast<T>(0);
+
+    return run_cast_from_f8<T, negative_zero_nan>(x);
+}
+
+} // namespace ck::utils
diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index 7828d21d7f..b13bccb5ab 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -3,6 +3,7 @@
 
 #pragma once
 #include "data_type.hpp"
+#include "type_convert.hpp"
 
 namespace ck {
 
diff --git a/include/ck/utility/random_gen.hpp b/include/ck/utility/random_gen.hpp
new file mode 100644
index 0000000000..b7edf26507
--- /dev/null
+++ b/include/ck/utility/random_gen.hpp
@@ -0,0 +1,53 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+
+// Pseudo random number generator
+// version for fp32
+template <typename T, uint32_t seed_t, std::enable_if_t<std::is_same<float, T>{}, bool> = false>
+__host__ __device__ uint32_t prand_generator(index_t id, T val, uint32_t seed = seed_t)
+{
+    uint32_t x         = *(reinterpret_cast<uint32_t*>(&val));
+    uint32_t drop_bits = uint32_t(x) & 0xFFFFu;
+    drop_bits ^= x >> 16;
+    drop_bits = ((drop_bits & 31) << 11) | (drop_bits >> 5);
+    drop_bits *= 0x7000149;
+    // NOTE: If id is in 64 bit, we are only using lower 32 bit.
+    //       So, it can have an effect of using same id for multiple elements when the id is very
+    //       large!
+    uint32_t rng = (drop_bits ^ 0x13371337 ^ (id * 229791) ^ seed);
+    return rng;
+}
+
+// version for fp16
+template <typename T, uint32_t seed_t, std::enable_if_t<std::is_same<half_t, T>{}, bool> = false>
+__host__ __device__ uint32_t prand_generator(index_t id, T val, uint32_t seed = seed_t)
+{
+    uint16_t x         = *(reinterpret_cast<uint16_t*>(&val));
+    uint32_t drop_bits = uint32_t(x) & 0xFFFFu;
+    drop_bits          = ((drop_bits & 31) << 11) | (drop_bits >> 5);
+    drop_bits *= 0x7000149;
+    // NOTE: If id is in 64 bit, we are only using lower 32 bit.
+    //       So, it can have an effect of using same id for multiple elements when the id is very
+    //       large!
+    uint32_t rng = (drop_bits ^ 0x13371337 ^ (id * 229791) ^ seed);
+    return rng;
+}
+
+// return 0 if data is not fp16 or fp32
+template <typename T,
+          uint32_t seed_t,
+          std::enable_if_t<!(std::is_same<float, T>{} || std::is_same<half_t, T>{}), bool> = false>
+__host__ __device__ uint32_t prand_generator(int id, T val, uint32_t seed = seed_t)
+{
+    std::ignore = id;
+    std::ignore = val;
+    std::ignore = seed;
+
+    return 0;
+}
+
+} // namespace ck
diff --git a/include/ck/utility/reduction_operator.hpp b/include/ck/utility/reduction_operator.hpp
index 0f5b73cb03..36c25203ea 100644
--- a/include/ck/utility/reduction_operator.hpp
+++ b/include/ck/utility/reduction_operator.hpp
@@ -6,6 +6,7 @@
 #include "ck/ck.hpp"
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/type.hpp"
+#include "ck/utility/type_convert.hpp"
 
 namespace ck {
 
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
new file mode 100644
index 0000000000..ed83964936
--- /dev/null
+++ b/include/ck/utility/type_convert.hpp
@@ -0,0 +1,230 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/f8_utils.hpp"
+#include "ck/utility/random_gen.hpp"
+
+namespace ck {
+
+// Convert X to Y
+template <typename Y, typename X>
+__host__ __device__ constexpr Y type_convert(X x)
+{
+    static_assert(!std::is_reference_v<Y> && !std::is_reference_v<X>);
+
+    return static_cast<Y>(x);
+}
+
+// convert bfp16 to fp32
+template <>
+inline __host__ __device__ constexpr float type_convert<float, bhalf_t>(bhalf_t x)
+{
+    union
+    {
+        uint32_t int32;
+        float fp32;
+    } u = {uint32_t(x) << 16};
+
+    return u.fp32;
+}
+
+// convert fp32 to bfp16
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, float>(float x)
+{
+    union
+    {
+        float fp32;
+        uint32_t int32;
+    } u = {x};
+
+    return uint16_t(u.int32 >> 16);
+}
+
+// convert bfp16 to fp16 via fp32
+template <>
+inline __host__ __device__ constexpr half_t type_convert<half_t, bhalf_t>(bhalf_t x)
+{
+    float x_fp32 = type_convert<float>(x);
+
+    return static_cast<half_t>(x_fp32);
+}
+
+// convert fp16 to bfp16 via fp32
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, half_t>(half_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return type_convert<bhalf_t>(x_fp32);
+}
+
+// convert bfp16 to int32 via fp32
+template <>
+inline __host__ __device__ constexpr int32_t type_convert<int32_t, bhalf_t>(bhalf_t x)
+{
+    float x_fp32 = type_convert<float>(x);
+
+    return static_cast<int32_t>(x_fp32);
+}
+
+// convert int32 to bfp16 via fp32
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int32_t>(int32_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return type_convert<bhalf_t>(x_fp32);
+}
+
+// convert bfp16 to int8 via fp32
+template <>
+inline __host__ __device__ constexpr int8_t type_convert<int8_t, bhalf_t>(bhalf_t x)
+{
+    float x_fp32 = type_convert<float>(x);
+
+    return static_cast<int8_t>(x_fp32);
+}
+
+// convert int8 to bfp16 via fp32
+template <>
+inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return type_convert<bhalf_t>(x_fp32);
+}
+
+// convert fp32 to fp8
+template <>
+inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+// convert fp8 to fp32
+template <>
+inline __host__ __device__ float type_convert<float, f8_t>(f8_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<float, negative_zero_nan>(x);
+}
+
+// convert fp16 to fp8
+template <>
+inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+// convert fp8 to fp16
+template <>
+inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<half_t, negative_zero_nan>(x);
+}
+
+// Declare a template function for bf16 conversion using RTN
+template <typename Y, typename X>
+__host__ __device__ constexpr Y bf16_convert_rtn(X x);
+
+// Convert fp32 to bf16 with RTN if higher precision is needed
+template <>
+inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, float>(float x)
+{
+    union
+    {
+        float fp32;
+        uint32_t int32;
+    } u = {x};
+
+    // When the exponent bits are not all 1s, then the value is zero, normal,
+    // or subnormal. We round the bfloat16 mantissa up by adding 0x7FFF, plus
+    // 1 if the least significant bit of the bfloat16 mantissa is 1 (odd).
+    // This causes the bfloat16's mantissa to be incremented by 1 if the 16
+    // least significant bits of the float mantissa are greater than 0x8000,
+    // or if they are equal to 0x8000 and the least significant bit of the
+    // bfloat16 mantissa is 1 (odd). This causes it to be rounded to even when
+    // the lower 16 bits are exactly 0x8000. If the bfloat16 mantissa already
+    // has the value 0x7f, then incrementing it causes it to become 0x00 and
+    // the exponent is incremented by one, which is the next higher FP value
+    // to the unrounded bfloat16 value. When the bfloat16 value is subnormal
+    // with an exponent of 0x00 and a mantissa of 0x7f, it may be rounded up
+    // to a normal value with an exponent of 0x01 and a mantissa of 0x00.
+    // When the bfloat16 value has an exponent of 0xFE and a mantissa of 0x7F,
+    // incrementing it causes it to become an exponent of 0xFF and a mantissa
+    // of 0x00, which is Inf, the next higher value to the unrounded value.
+    bool flag0 = ~u.int32 & 0x7f800000;
+
+    // When all of the exponent bits are 1, the value is Inf or NaN.
+    // Inf is indicated by a zero mantissa. NaN is indicated by any nonzero
+    // mantissa bit. Quiet NaN is indicated by the most significant mantissa
+    // bit being 1. Signaling NaN is indicated by the most significant
+    // mantissa bit being 0 but some other bit(s) being 1. If any of the
+    // lower 16 bits of the mantissa are 1, we set the least significant bit
+    // of the bfloat16 mantissa, in order to preserve signaling NaN in case
+    // the bfloat16's mantissa bits are all 0.
+    bool flag1 = !flag0 && (u.int32 & 0xffff);
+
+    u.int32 += flag0 ? 0x7fff + ((u.int32 >> 16) & 1) : 0; // Round to nearest, round to even
+    u.int32 |= flag1 ? 0x10000 : 0x0;                      // Preserve signaling NaN
+
+    return uint16_t(u.int32 >> 16);
+}
+
+// convert fp16 to bfp16 via fp32 with RTN if higher precision is needed
+template <>
+inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(half_t x)
+{
+    float x_fp32 = static_cast<float>(x);
+
+    return bf16_convert_rtn<bhalf_t>(x_fp32);
+}
+
+// Declare a template function for fp8 conversion using SR
+template <typename Y, typename X>
+__host__ __device__ constexpr Y f8_convert_sr(X x);
+
+// convert fp32 to fp8 with stochastic rounding
+template <>
+inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    constexpr int seed               = 42;
+    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+// convert fp16 to fp8 with stochastic rounding
+template <>
+inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    constexpr int seed               = 42;
+    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+        x, rng);
+}
+
+} // namespace ck
diff --git a/library/include/ck/library/utility/host_tensor.hpp b/library/include/ck/library/utility/host_tensor.hpp
index 91293d29f7..816d834130 100644
--- a/library/include/ck/library/utility/host_tensor.hpp
+++ b/library/include/ck/library/utility/host_tensor.hpp
@@ -13,6 +13,7 @@
 
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/span.hpp"
+#include "ck/utility/type_convert.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
 #include "ck/library/utility/ranges.hpp"
diff --git a/test/data_type/CMakeLists.txt b/test/data_type/CMakeLists.txt
index 088fbfec71..2b63727f19 100644
--- a/test/data_type/CMakeLists.txt
+++ b/test/data_type/CMakeLists.txt
@@ -2,3 +2,6 @@ if (USE_BITINT_EXTENSION_INT4)
   add_gtest_executable(test_int4 int4.cpp)
   target_link_libraries(test_int4 PRIVATE utility)
 endif()
+
+add_gtest_executable(test_fp8 fp8.cpp)
+target_link_libraries(test_fp8 PRIVATE utility)
diff --git a/test/data_type/fp8.cpp b/test/data_type/fp8.cpp
new file mode 100644
index 0000000000..5004fe9527
--- /dev/null
+++ b/test/data_type/fp8.cpp
@@ -0,0 +1,123 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/type_convert.hpp"
+
+using ck::f8_convert_sr;
+using ck::f8_t;
+using ck::half_t;
+using ck::type_convert;
+
+TEST(FP8, NumericLimits)
+{
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Min(), 0x08);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Max(), 0x77);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Lowest(), 0xF7);
+    EXPECT_EQ(ck::NumericLimits<f8_t>::QuietNaN(), 0x80);
+}
+
+TEST(FP8, ConvertFP32Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to fp8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(type_convert<f8_t>(0.0f)), abs_tol);
+    // convert minimal float to fp8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal f8_t to float and check if equal to 240.0
+    ASSERT_NEAR(240.0f, type_convert<float>(type_convert<f8_t>(240.0f)), abs_tol);
+    // convert maximal float to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(240.0f,
+                type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to f8_t and check if it is qNan
+    ASSERT_NEAR(0x80, type_convert<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
+    // positive float value to fp8 and back, check if holds
+    float pos_float = 0.0078125f;
+    ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
+    // negative float value to fp8 and back, check if holds
+    float neg_float = -0.0156250f;
+    ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
+}
+
+TEST(FP8, ConvertFP32Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to fp8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(f8_convert_sr<f8_t>(0.0f)), abs_tol);
+    // convert minimal float to fp8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal f8_t to float and check if equal to 240.0
+    ASSERT_NEAR(240.0f, type_convert<float>(f8_convert_sr<f8_t>(240.0f)), abs_tol);
+    // convert maximal float to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(240.0f,
+                type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to f8_t and check if it is qNan
+    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
+    // positive float value to fp8 and back, check if holds
+    float pos_float = 0.0078125f;
+    ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<f8_t>(pos_float)), abs_tol);
+    // negative float value to fp8 and back, check if holds
+    float neg_float = -0.0156250f;
+    ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<f8_t>(neg_float)), abs_tol);
+}
+
+TEST(FP8, ConvertFP16Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(type_convert<f8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal f8_t to fp16 and check if equal to 240.0
+    ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(type_convert<f8_t>(half_t{240.0})), abs_tol);
+    // convert maximal fp16 to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(half_t{240.0},
+                type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
+    ASSERT_NEAR(0x80, type_convert<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
+    // positive fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
+    // negative fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.0156250};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
+}
+
+TEST(FP8, ConvertFP16Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(f8_convert_sr<f8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to fp8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal f8_t to fp16 and check if equal to 240.0
+    ASSERT_NEAR(half_t{240.0}, type_convert<half_t>(f8_convert_sr<f8_t>(half_t{240.0})), abs_tol);
+    // convert maximal fp16 to fp8 and back, check if clipped to 240.0
+    ASSERT_NEAR(half_t{240.0},
+                type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
+    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
+    // positive fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<f8_t>(pos_half)), abs_tol);
+    // negative fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.0156250};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<f8_t>(neg_half)), abs_tol);
+}

From 645eb2f2a08d1116b98710ec1b22ef4ae25b36b2 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 19 Jun 2023 16:55:03 -0700
Subject: [PATCH 043/194] do not build gemm-gemm and conv-conv examples for
 gfx94* (#761)

* do not build gemm-gemm and conv-conv examples for gfx94*

* do not build gemm-gemm and conv-conv examples on navi
---
 example/31_batched_gemm_gemm/CMakeLists.txt     | 8 ++------
 example/41_grouped_conv_conv_fwd/CMakeLists.txt | 8 ++------
 2 files changed, 4 insertions(+), 12 deletions(-)

diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index f8d139275d..83989bcd97 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -14,10 +14,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
  endif()
 endforeach()
 
-set(target 0)
-foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
    add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
-   set(target 1)
- endif()
-endforeach()
\ No newline at end of file
+endif()
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index 0c9df707b3..ae251e88d2 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -13,10 +13,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
  endif()
 endforeach()
 
-set(target 0)
-foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
    add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
-   set(target 1)
- endif()
-endforeach()
+endif()

From 05ea6452b602c0e3d218042c63713a056aaeb6fe Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Mon, 19 Jun 2023 19:24:18 -0700
Subject: [PATCH 044/194] changed pipeline v1 (#763)

---
 .../gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp        | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 3801993412..6c07a5c56c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -76,7 +76,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     // TODO: should be exposed as Tparams.
     static constexpr index_t NumGemmKPrefetchStage = 1;
     static constexpr LoopScheduler LoopSched       = make_default_loop_scheduler();
-    static constexpr PipelineVersion PipelineVer   = PipelineVersion::v2;
+    static constexpr PipelineVersion PipelineVer   = PipelineVersion::v1;
 
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,

From 32d2f52bf70a4ef80e8399951150c4b3b848feeb Mon Sep 17 00:00:00 2001
From: ltqin <letao.qin@amd.com>
Date: Tue, 20 Jun 2023 10:25:08 +0800
Subject: [PATCH 045/194] remove useless comments (#760)

---
 .../device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp   | 2 --
 1 file changed, 2 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index ca72bcdd23..73b12bfb47 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -786,12 +786,10 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
             if(arg.d0s_nl_ns_lengths_strides_[i][1] == 1 &&
                arg.d0s_nl_ns_lengths_strides_[i][0] % D0sTransferSrcScalarPerVector != 0)
             {
-                std::cout << "first" << std::endl;
                 return false;
             }
             if(arg.d0s_nl_ns_lengths_strides_[i][1] != 1 && D0sTransferSrcScalarPerVector != 1)
             {
-                std::cout << "second" << std::endl;
                 return false;
             }
         }

From 63388e84ab71a08d6aa5b1476c0ba5d04be201e8 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 21 Jun 2023 15:20:31 +0200
Subject: [PATCH 046/194] Support bf16/f32/f16 and NHWGC conv2d_bwd_data (#757)

* Support bf16/f32/f16 and NHWGC conv2d_bwd_data

* Add interface test

* clang format

* Comment fixes

* Add more friendly error message
---
 include/ck/ck.hpp                             |   4 +
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  15 +-
 .../transform_conv_bwd_data_to_gemm_v1.hpp    |  66 ++++-
 .../gpu/grouped_convolution_backward_data.hpp |  99 +++++++
 .../grouped_conv2d_bwd_data/CMakeLists.txt    |   5 +
 ...ta_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |  47 ++++
 ...ata_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  86 ++----
 ...ata_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |  47 ++++
 ...e_grouped_conv2d_bwd_data_xdl_instance.hpp | 141 ++++++++++
 ...ta_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |  47 ++++
 ...ata_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  47 ++++
 ...ata_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |  47 ++++
 profiler/README.md                            |  38 +++
 .../profile_grouped_conv_bwd_data_impl.hpp    | 257 ++++++++++++++++++
 profiler/src/CMakeLists.txt                   |   2 +
 .../src/profile_grouped_conv_bwd_data.cpp     | 157 +++++++++++
 test/CMakeLists.txt                           |   1 +
 test/grouped_convnd_bwd_data/CMakeLists.txt   |   6 +
 .../test_grouped_convnd_bwd_data.cpp          |  78 ++++++
 ...test_grouped_convnd_bwd_data_interface.cpp | 178 ++++++++++++
 20 files changed, 1285 insertions(+), 83 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
 create mode 100644 profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp
 create mode 100644 profiler/src/profile_grouped_conv_bwd_data.cpp
 create mode 100644 test/grouped_convnd_bwd_data/CMakeLists.txt
 create mode 100644 test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
 create mode 100644 test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index d7ce449bb6..161c3261ad 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -173,6 +173,10 @@
 
 // workaround: compiler issue on gfx908
 #define CK_WORKAROUND_SWDEV_388832 1
+
+// workaround: Grouped Conv2d_bwd_data fails for already implemented instance
+#define CK_WORKAROUND_SWDEV_3318619 0
+
 // flag to enable (1) or disable (0) the debugging output in some kernels
 #define DEBUG_LOG 0
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 81a4d69275..95af932519 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -459,7 +459,6 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
               p_ds_grid_{},
               p_e_grid_{static_cast<EDataType*>(p_e)},
               num_group_{a_g_n_k_wos_lengths[0]},
-              num_gemm_{},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
               cde_element_op_{cde_element_op},
@@ -508,9 +507,6 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
             const auto YTilde = ConvStrideH / GcdStrideDilationH;
             const auto XTilde = ConvStrideW / GcdStrideDilationW;
 
-            // number of GEMM
-            num_gemm_ = YTilde * XTilde;
-
             for(index_t i_ytilde = 0; i_ytilde < YTilde; ++i_ytilde)
             {
                 for(index_t i_xtilde = 0; i_xtilde < XTilde; ++i_xtilde)
@@ -626,7 +622,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
         void Print() const
         {
-            for(index_t i = 0; i < num_gemm_; i++)
+            for(std::size_t i = 0; i < a_grid_desc_ak0_m_ak1_container_.size(); i++)
             {
                 std::cout << "a_grid_desc_ak0_m_ak1_container_"
                           << a_grid_desc_ak0_m_ak1_container_[i] << std::endl;
@@ -654,7 +650,6 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
         // tensor descriptor for problem definition
         index_t num_group_;
-        index_t num_gemm_;
         std::vector<AGridDesc_M_K> a_grid_desc_m_k_container_;
         std::vector<BGridDesc_N_K> b_grid_desc_n_k_container_;
         std::vector<DsGridDesc_M_N> ds_grid_desc_m_n_container_;
@@ -708,7 +703,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
             float ave_time = 0;
 
-            for(index_t i = 0; i < arg.num_gemm_; i++)
+            for(std::size_t i = 0; i < arg.a_grid_desc_ak0_m_ak1_container_.size(); i++)
             {
                 if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_container_[i],
                                                 arg.b_grid_desc_n_k_container_[i],
@@ -807,7 +802,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
         }
 
         // vector load for A matrix from global memory to LDS
-        if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK>)
+        if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::NHWGK>)
         {
             if(!(ABlockTransferSrcVectorDim == 2 && ConvK % ABlockTransferSrcScalarPerVector == 0))
             {
@@ -862,7 +858,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
         }
 
         // vector store for E
-        if constexpr(is_same_v<ELayout, tensor_layout::convolution::GNHWC>)
+        if constexpr(is_same_v<ELayout, tensor_layout::convolution::GNHWC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::NHWGC>)
         {
             // vector store C matrix into global memory
             if(!(ConvC % CDEBlockTransferScalarPerVector_NPerBlock == 0))
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
index 505ed33d50..59f64bde77 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
@@ -13,6 +13,61 @@
 namespace ck {
 namespace tensor_operation {
 
+namespace {
+template <
+    index_t NDimSpatial,
+    typename ALayout,
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization ConvBwdDataSpecialization>
+constexpr auto
+make_out_n_ho_wo_k_grid_desc(const index_t N,
+                             const index_t Ho,
+                             const index_t Wo,
+                             const index_t K,
+                             const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_strides)
+{
+
+    if constexpr(is_same_v<ALayout, tensor_layout::convolution::NHWGK>)
+    {
+        const index_t NStride  = out_g_n_k_wos_strides[1];
+        const index_t HiStride = out_g_n_k_wos_strides[3];
+        const index_t WiStride = out_g_n_k_wos_strides[4];
+        const auto CStride     = Number<1>{};
+        if constexpr(ConvBwdDataSpecialization ==
+                     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
+                         Filter1x1Stride1Pad0)
+        {
+
+            return make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, K),
+                                                make_tuple(WiStride, CStride));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor(make_tuple(N, Ho, Wo, K),
+                                                make_tuple(NStride, HiStride, WiStride, CStride));
+        }
+    }
+    else if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK>)
+    {
+        // assume packed
+        if constexpr(ConvBwdDataSpecialization ==
+                     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
+                         Filter1x1Stride1Pad0)
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N, Ho, Wo, K));
+        }
+    }
+    else
+    {
+        throw std::runtime_error("wrong! unsupported layout: " + ALayout::name());
+    }
+}
+
+} // namespace
+
 template <
     index_t NDimSpatial,
     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization ConvBwdDataSpecialization,
@@ -29,11 +84,12 @@ struct TransformConvBwdDataToGemm_v1
 
     template <typename ALayout,
               typename std::enable_if<NDimSpatial == 2 &&
-                                          is_same_v<ALayout, tensor_layout::convolution::GNHWK>,
+                                          (is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
+                                           is_same_v<ALayout, tensor_layout::convolution::NHWGK>),
                                       bool>::type = false>
     static auto MakeADescriptor_AK0_M_AK1(
         const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_lengths,
-        const std::array<index_t, NDimSpatial + 3>& /* out_g_n_k_wos_strides */,
+        const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_strides,
         const std::array<index_t, NDimSpatial + 3>& wei_g_k_c_xs_lengths,
         const std::array<index_t, NDimSpatial + 3>& /* wei_g_k_c_xs_strides */,
         const std::array<index_t, NDimSpatial + 3>& in_g_n_c_wis_lengths,
@@ -70,9 +126,9 @@ struct TransformConvBwdDataToGemm_v1
 
         const index_t AK0 = K / AK1;
 
-        // assume packed
         const auto out_n_ho_wo_k_grid_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(N, Ho, Wo, K));
+            make_out_n_ho_wo_k_grid_desc<NDimSpatial, ALayout, ConvBwdDataSpecialization>(
+                N, Ho, Wo, K, out_g_n_k_wos_strides);
 
         if constexpr(ConvBwdDataSpecialization ==
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
@@ -80,7 +136,7 @@ struct TransformConvBwdDataToGemm_v1
         {
             // A: output tensor
             const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
-                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K)),
+                out_n_ho_wo_k_grid_desc,
                 make_tuple(make_pass_through_transform(N * Ho * Wo),
                            make_unmerge_transform(make_tuple(AK0, AK1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index fadfd19951..0b20e19a2a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -30,6 +30,76 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -78,6 +148,35 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
             }
+            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                              is_same_v<OutDataType, F32>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                              is_same_v<OutDataType, BF16>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
+            }
+        }
+        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
+        {
+            if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                         is_same_v<OutDataType, F16>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                              is_same_v<OutDataType, F32>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                              is_same_v<OutDataType, BF16>)
+            {
+                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
+            }
         }
 
         return op_ptrs;
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
index 3b2968d48e..85ec0f55aa 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
@@ -1,3 +1,8 @@
 add_instance_library(device_grouped_conv2d_bwd_data_instance
    device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
new file mode 100644
index 0000000000..01cd23b2e3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<GNHWK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          GNHWC,
+                                                          ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<GNHWK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          GNHWC,
+                                                          ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 11babea28e..bb0fd36e44 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,80 +1,14 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdDataDefault =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
-
-static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
-
-using device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances = std::tuple<
-    // clang-format off
-        // 1. Default
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-
-        // 2. Filter1x1Stride1Pad0
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise|              ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|               DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                                 |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                                 |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataFilter1x1Stride1Pad0,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
-    // clang-format on
-    >;
-
+// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -89,8 +23,22 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
-        instances, device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances{});
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
new file mode 100644
index 0000000000..b671017f51
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<GNHWK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         GNHWC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
new file mode 100644
index 0000000000..bfe6e47d19
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
+
+static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
+    ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+// f16_f16_f32_f16
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv2d_bwd_data_xdl_f16_instances = std::tuple<
+    // clang-format off
+        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
+        
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#endif
+    // clang-format on
+    >;
+
+// bf16_bf16_f32_bf16
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv2d_bwd_data_xdl_bf16_instances = std::tuple<
+    // clang-format off
+        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
+        
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
+#endif
+    // clang-format on
+    >;
+
+// f32_f32_f32_f32
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv2d_bwd_data_xdl_f32_instances = std::tuple<
+    // clang-format off
+        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
+
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#endif
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..0b1fa63ddb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<NHWGK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          NHWGC,
+                                                          ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_bf16_instances<NHWGK,
+                                                          GKYXC,
+                                                          Empty_Tuple,
+                                                          NHWGC,
+                                                          ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..1c8fc57e1c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f16_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..fcdfe8d42b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
+                                                         GKYXC,
+                                                         Empty_Tuple,
+                                                         NHWGC,
+                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/README.md b/profiler/README.md
index 7a1fb29113..3ee039cb3f 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -102,4 +102,42 @@ arg.b_grid_desc_k0_n0_n1_k1_{2048, 4096, 2}
 arg.e_grid_desc_m_n_{ 4096, 4096}
 ....
 Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
+## Profile grouped convolution backward data kernels
+```bash
+# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
+# arg2: data type (0: Output fp32, Weight fp32, Input fp32
+#                  1: Output fp16, Weight fp16, Input fp16
+#                  2: Output bf16, Weight bf16, Input bf16
+# arg3: tensor layout (0: Output[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Input[G, N, Ho, Wo, K]
+#                      1: Output[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Input[N, Ho, Wo, G, K])
+# arg4: verification (0: no, 1: yes)
+# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
+# arg6: print tensor value (0: no; 1: yes)
+# arg7: time kernel (0: no, 1: yes)
+# Following arguments (depending on number of spatial dims):
+#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  G, N, K, C, 
+#  <filter spatial dimensions>, (ie Y, X for 2D)
+#  <input image spatial dimensions>, (ie Hi, Wi for 2D)
+#  <strides>, (ie Sy, Sx for 2D)
+#  <dilations>, (ie Dy, Dx for 2D)
+#  <left padding>, (ie LeftPy, LeftPx for 2D)
+#  <right padding>, (ie RightPy, RightPx for 2D)
+
+ ################                   op   datatype  layout  verify  init  log  time  Ndims  G  N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx
+./bin/ckProfiler grouped_conv_bwd_data          1       0       1     1    0     1      2 32  4 192 192  3  3  28  28   1   1   1   1       1       1        1        1
+
+ ```
+
+Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
+```
+out: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
+wei: dim 5, lengths {32, 192, 192, 3, 3}, strides {331776, 1728, 1, 576, 192}
+in: dim 5, lengths {32, 4, 192, 28, 28}, strides {602112, 150528, 1, 5376, 192}
+....
+Best configuration parameters:
+name: DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<256, 128, 256, 32, 8, 2, Default, 32, 32, 2, 4, 8, 4, 1, 1>
+avg_time: 0.768321
+tflops: 86.6679
+GB/s: 127.947
 ```
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp
new file mode 100644
index 0000000000..93d3430bba
--- /dev/null
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_data_impl.hpp
@@ -0,0 +1,257 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <ck::index_t NDimSpatial,
+          typename OutLayout,
+          typename WeiLayout,
+          typename InLayout,
+          typename OutDataType,
+          typename WeiDataType,
+          typename InDataType>
+bool profile_grouped_conv_bwd_data_impl(int do_verification,
+                                        int init_method,
+                                        bool do_log,
+                                        bool time_kernel,
+                                        const ck::utils::conv::ConvParam& conv_param)
+{
+    using OutElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+
+    const auto out_element_op = OutElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+    const auto in_element_op  = InElementOp{};
+
+    const auto out_g_n_k_wos_desc =
+        ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(conv_param);
+
+    const auto wei_g_k_c_xs_desc =
+        ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(conv_param);
+
+    const auto in_g_n_c_wis_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_param);
+
+    Tensor<OutDataType> out(out_g_n_k_wos_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<InDataType> in_host(in_g_n_c_wis_desc);
+    Tensor<InDataType> in_device(in_g_n_c_wis_desc);
+
+    std::cout << "out: " << out.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "in: " << in_host.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
+        break;
+    case 2:
+        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
+        break;
+    default:
+        out.GenerateTensorValue(GeneratorTensor_1<OutDataType>{1});
+        wei.GenerateTensorValue(GeneratorTensor_1<WeiDataType>{1});
+    }
+
+    DeviceMem out_device_buf(sizeof(OutDataType) * out.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem in_device_buf(sizeof(InDataType) * in_device.mDesc.GetElementSpaceSize());
+
+    out_device_buf.ToDevice(out.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+
+    // reset input to zero
+    in_device_buf.SetZero();
+
+    if(do_verification)
+    {
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvBwdData<NDimSpatial,
+                                                                         InDataType,
+                                                                         WeiDataType,
+                                                                         OutDataType,
+                                                                         InElementOp,
+                                                                         WeiElementOp,
+                                                                         OutElementOp>();
+
+        auto ref_invoker = ref_conv.MakeInvoker();
+
+        in_host.SetZero();
+
+        auto ref_argument = ref_conv.MakeArgument(in_host,
+                                                  wei,
+                                                  out,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  out_element_op,
+                                                  wei_element_op,
+                                                  in_element_op);
+
+        ref_invoker.Run(ref_argument);
+    }
+
+    std::string best_op_name;
+    float best_avg_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    bool pass = true;
+
+    auto run_impl = [&](auto& op_ptr, auto& argument_ptr) {
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init output to zero before profiling next kernel
+            in_device_buf.SetZero();
+
+            std::string op_name = op_ptr->GetTypeString();
+
+            auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop      = conv_param.GetFlops();
+            std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+            float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
+
+            float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                in_device_buf.FromDevice(in_device.mData.data());
+
+                pass = pass & ck::utils::check_err(in_device, in_host);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "output : ", out.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "weight: ", wei.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "in_host  : ", in_host.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "in_device: ", in_device.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    };
+
+    // do GEMM
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
+                                                                                     OutLayout,
+                                                                                     WeiLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     InLayout,
+                                                                                     OutDataType,
+                                                                                     WeiDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     InDataType,
+                                                                                     OutElementOp,
+                                                                                     WeiElementOp,
+                                                                                     InElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+    copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+                                        static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
+                                        {},
+                                        static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+                                        out_lengths,
+                                        out_strides,
+                                        wei_lengths,
+                                        wei_strides,
+                                        {},
+                                        {},
+                                        in_lengths,
+                                        in_strides,
+                                        conv_filter_strides,
+                                        conv_filter_dilations,
+                                        input_left_pads,
+                                        input_right_pads,
+                                        out_element_op,
+                                        wei_element_op,
+                                        in_element_op);
+
+        run_impl(op_ptr, argument_ptr);
+    }
+
+    std::cout << "Best configuration parameters:"
+              << "\nname: " << best_op_name << "\navg_time: " << best_avg_time
+              << "\ntflops: " << best_tflops << "\nGB/s: " << best_gb_per_sec << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 6f768e0ae1..3f19096f48 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -35,6 +35,7 @@ set(PROFILER_SOURCES
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
     profile_batched_gemm_multi_d.cpp
+    profile_grouped_conv_bwd_data.cpp
 )
 
 set(PROFILER_EXECUTABLE ckProfiler)
@@ -79,4 +80,5 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_grouped_conv_bwd_data.cpp b/profiler/src/profile_grouped_conv_bwd_data.cpp
new file mode 100644
index 0000000000..351bb72c81
--- /dev/null
+++ b/profiler/src/profile_grouped_conv_bwd_data.cpp
@@ -0,0 +1,157 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+namespace {
+
+enum struct ConvLayout
+{
+    GNHWC_GKYXC_GNHWK, // 0
+    NHWGC_GKYXC_NHWGK, // 1
+};
+
+enum struct ConvDataType
+{
+    F32_F32_F32,    // 0
+    F16_F16_F16,    // 1
+    BF16_BF16_BF16, // 2
+};
+
+#define OP_NAME "grouped_conv_bwd_data"
+#define OP_DESC "Grouped Convolution Backward Data"
+
+static void print_helper_msg()
+{
+    std::cout
+        // clang-format off
+        << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+        << "arg2: data type (0: Output fp32, Weight fp32, Input fp32\n"
+        << "                 1: Output fp16, Weight fp16, Input fp16\n"
+        << "                 2: Output bf16, Weight bf16, Input bf16\n"
+        << "arg3: tensor layout (0: Output[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Input[G, N, Ho, Wo, K]\n"
+        << "                     1: Output[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Input[N, Ho, Wo, G, K])\n"
+        << "arg4: verification (0: no, 1: yes)\n"
+        << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
+        << "arg6: print tensor value (0: no; 1: yes)\n"
+        << "arg7: time kernel (0: no, 1: yes)\n"
+        << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+    // clang-format on
+}
+
+} // namespace
+
+int profile_grouped_conv_bwd_data(int argc, char* argv[])
+{
+    // 8 for control, 1 for num_dim_spatial
+    if(argc < 9)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto data_type       = static_cast<ConvDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<ConvLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+    const int num_dim_spatial  = std::stoi(argv[8]);
+
+    // 8 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
+    if(argc != 8 + 1 + 4 + 6 * num_dim_spatial)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto params = ck::utils::conv::parse_conv_param(num_dim_spatial, 9, argv);
+
+    using F32  = float;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+
+    using GNHWC = ck::tensor_layout::convolution::GNHWC;
+    using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+    using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+    using GNHWK = ck::tensor_layout::convolution::GNHWK;
+    using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+    constexpr auto I2 = ck::Number<2>{};
+
+    auto profile = [&](auto num_dim_spatial_tmp,
+                       auto out_layout,
+                       auto wei_layout,
+                       auto in_layout,
+                       auto wei_type,
+                       auto out_type,
+                       auto in_type) {
+        constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
+
+        using OutLayout = decltype(out_layout);
+        using WeiLayout = decltype(wei_layout);
+        using InLayout  = decltype(in_layout);
+
+        using OutDataType = decltype(out_type);
+        using WeiDataType = decltype(wei_type);
+        using InDataType  = decltype(in_type);
+
+        bool pass = ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
+                                                                     OutLayout,
+                                                                     WeiLayout,
+                                                                     InLayout,
+                                                                     OutDataType,
+                                                                     WeiDataType,
+                                                                     InDataType>(
+            do_verification, init_method, do_log, time_kernel, params);
+
+        return pass ? 0 : 1;
+    };
+
+    // GNHWC_GKYXC_GNHWK
+    if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_BF16_BF16)
+        {
+            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, BF16{}, BF16{}, BF16{});
+        }
+    }
+    // NHWGC_GKYXC_NHWGK
+    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_BF16_BF16)
+        {
+            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, BF16{}, BF16{}, BF16{});
+        }
+    }
+
+    std::cout << "this data_type & layout is not implemented" << std::endl;
+
+    return 1;
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_grouped_conv_bwd_data);
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index e3385b9dd4..ad08e94702 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -59,6 +59,7 @@ add_subdirectory(batchnorm)
 add_subdirectory(contraction)
 add_subdirectory(pool_fwd)
 add_subdirectory(batched_gemm_multi_d)
+add_subdirectory(grouped_convnd_bwd_data)
 if(GPU_TARGETS MATCHES "gfx1100")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/grouped_convnd_bwd_data/CMakeLists.txt b/test/grouped_convnd_bwd_data/CMakeLists.txt
new file mode 100644
index 0000000000..d7c9f46554
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/CMakeLists.txt
@@ -0,0 +1,6 @@
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
+    add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+    add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface.cpp)
+    target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+endif()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
new file mode 100644
index 0000000000..94808669a3
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "profiler/profile_grouped_conv_bwd_data_impl.hpp"
+
+template <typename Tuple>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    using DataType  = std::tuple_element_t<0, Tuple>;
+    using OutLayout = std::tuple_element_t<1, Tuple>;
+    using WeiLayout = std::tuple_element_t<2, Tuple>;
+    using InLayout  = std::tuple_element_t<3, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+
+    template <ck::index_t NDimSpatial>
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_grouped_conv_bwd_data_impl<NDimSpatial,
+                                                                            OutLayout,
+                                                                            WeiLayout,
+                                                                            InLayout,
+                                                                            DataType,
+                                                                            DataType,
+                                                                            DataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes = ::testing::Types<std::tuple<float, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<ck::bhalf_t, GNHWK, GKYXC, GNHWC>,
+                                     std::tuple<float, NHWGK, GKYXC, NHWGC>,
+                                     std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
+                                     std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>>;
+TYPED_TEST_SUITE(TestGroupedConvndBwdData, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdData, Test2D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back(
+        {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->template Run<2>();
+}
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp
new file mode 100644
index 0000000000..bc592ba665
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using DataType    = ck::half_t;
+using AccDataType = float;
+using Pass        = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvBackwardDataSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization;
+
+static constexpr auto ConvBwdDataDefault   = ConvBackwardDataSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 = ConvBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvBackwardDataSpecialization ConvSpec>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 2;
+
+    using OutLayout = std::tuple_element_t<0, Tuple>;
+    using WeiLayout = std::tuple_element_t<1, Tuple>;
+    using InLayout  = std::tuple_element_t<2, Tuple>;
+
+    // clang-format off
+    using GroupedConvBwdDataDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
+    // ######| NDimSpatial|   ALayout|   BLayout|   DsLayout|  ELayout|       AData|       BData|     AccData|         CShuffle|       DsData|      EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+    // ######|            |          |          |           |         |        Type|        Type|        Type|         DataType|         Type|       Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+    // ######|            |          |          |           |         |            |            |            |                 |             |           |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+    // ######|            |          |          |           |         |            |            |            |                 |             |           |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+    < NDimSpatial, OutLayout, WeiLayout, ck::Tuple<>, InLayout, DataType, DataType, AccDataType, DataType,   ck::Tuple<>, DataType, Pass, Pass,    Pass,  ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    template <ck::index_t NDimSpatial>
+    bool Run()
+    {
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+        copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+        copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+        copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+        copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+        copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+        copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+        copy(conv_param.conv_filter_strides_, conv_filter_strides);
+        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+        copy(conv_param.input_left_pads_, input_left_pads);
+        copy(conv_param.input_right_pads_, input_right_pads);
+
+        auto conv = GroupedConvBwdDataDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          std::array<const void*, 0>{},
+                                          nullptr,
+                                          out_lengths,
+                                          out_strides,
+                                          wei_lengths,
+                                          wei_strides,
+                                          {},
+                                          {},
+                                          in_lengths,
+                                          in_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          Pass{},
+                                          Pass{},
+                                          Pass{});
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes =
+    ::testing::Types<std::tuple<GNHWK, GKYXC, GNHWC>, std::tuple<NHWGK, GKYXC, NHWGC>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataDefault : public TestGroupedConvndBwdData<Tuple, ConvBwdDataDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataFilter1x1
+    : public TestGroupedConvndBwdData<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataDefault, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataFilter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdDataFilter1x1, SpecializationCheck)
+{
+    // Check filter 3,3 instead of 1,1
+    this->conv_param  = {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2,2 instead of 1,1
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdDataDefault, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param  = {2, 2, 128, 129, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {2, 2, 128, 128, 257, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}

From 3b18f1e38cd7dc5a621d4fb4e1ac82304583db28 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 21 Jun 2023 10:47:35 -0700
Subject: [PATCH 047/194] do not build gfx941/942 targets during CI (#766)

---
 Jenkinsfile | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index ad2baa00e0..fbff349fc3 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -696,7 +696,7 @@ pipeline {
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
                         setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')

From d3adc665819dc0535a9f710ee7b8190ce85b3245 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 28 Jun 2023 06:58:28 -0700
Subject: [PATCH 048/194] Add rocm5.6 RC4 and rocm5.7 to docker build options.
 (#770)

* upgrade to rocm5.6 rc4

* add rocm5.7 docker
---
 Dockerfile  | 12 ++++++++----
 Jenkinsfile |  2 +-
 2 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index bc47768060..cef2ae245e 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -12,7 +12,7 @@ RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 RUN chmod 1777 /tmp
 RUN apt-get update
 RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
-RUN if [ "$ROCMVERSION" != "5.6" ]; then \
+RUN if [ "$ROCMVERSION" != "5.6" ] && [ "$ROCMVERSION" != "5.7" ]; then \
         wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
         sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"; \
     elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "" ]; then \
@@ -20,11 +20,15 @@ RUN if [ "$ROCMVERSION" != "5.6" ]; then \
          apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
          amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914 && \
          amdgpu-install -y --usecase=rocm --no-dkms; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc3" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
+    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc4" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
          sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.6-20.04-1_all.deb" && \
          apt update && apt-get install -y ./amdgpu-install-internal_5.6-20.04-1_all.deb && \
-         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-45  > /etc/apt/sources.list.d/rocm-build.list' && \
-         amdgpu-repo --amdgpu-build=1602498 && amdgpu-install -y --usecase=rocm --no-dkms; \
+         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-62  > /etc/apt/sources.list.d/rocm-build.list' && \
+         amdgpu-repo --amdgpu-build=1609671 && amdgpu-install -y --usecase=rocm --no-dkms; \
+    elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
+         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
+         apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
+         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354 && amdgpu-install -y --usecase=rocm --no-dkms; \
     fi
 
 RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
diff --git a/Jenkinsfile b/Jenkinsfile
index fbff349fc3..21e4626f6a 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -19,7 +19,7 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.ROCMVERSION != "5.6"){
+    if (params.ROCMVERSION != "5.6" && params.ROCMVERSION != "5.7"){
        if (params.COMPILER_VERSION == "") {
            img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
        }

From 7797bd3d2b047ab6ccd7b3618bf40f79bf234723 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 30 Jun 2023 08:06:54 -0700
Subject: [PATCH 049/194] Upgrade default docker to ROCM5.6 release. (#778)

* upgrade default compiler to rocm5.6 release

* do daily runs with rocm5.6 instead of 5.5
---
 Dockerfile  | 29 ++++++++++++++---------------
 Jenkinsfile |  4 ++--
 2 files changed, 16 insertions(+), 17 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index cef2ae245e..3ff29c9a98 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -12,28 +12,27 @@ RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
 RUN chmod 1777 /tmp
 RUN apt-get update
 RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
-RUN if [ "$ROCMVERSION" != "5.6" ] && [ "$ROCMVERSION" != "5.7" ]; then \
+
+ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
+RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+
+RUN wget https://repo.radeon.com/amdgpu-install/5.6/ubuntu/focal/amdgpu-install_5.6.50600-1_all.deb  --no-check-certificate
+RUN apt-get update && \
+DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
+    ./amdgpu-install_5.6.50600-1_all.deb
+
+RUN if [ "$ROCMVERSION" != "5.7" ]; then \
         wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
-        sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1567752 --rocm-build=compute-rocm-dkms-no-npi-hipclang/11914 && \
-         amdgpu-install -y --usecase=rocm --no-dkms; \
-    elif [ "$ROCMVERSION" = "5.6" ] && [ "$compiler_version" = "rc4" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.6-20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amdgpu-install-internal_5.6-20.04-1_all.deb && \
-         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.6 rel-62  > /etc/apt/sources.list.d/rocm-build.list' && \
-         amdgpu-repo --amdgpu-build=1609671 && amdgpu-install -y --usecase=rocm --no-dkms; \
+        sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
+        sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
     elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
          sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
          apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354 && amdgpu-install -y --usecase=rocm --no-dkms; \
+         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
     fi
 
-RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
-RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
+RUN amdgpu-install -y --usecase=rocm --no-dkms
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
diff --git a/Jenkinsfile b/Jenkinsfile
index 21e4626f6a..808fb9d1bf 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -19,7 +19,7 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.ROCMVERSION != "5.6" && params.ROCMVERSION != "5.7"){
+    if (params.ROCMVERSION != "5.7"){
        if (params.COMPILER_VERSION == "") {
            img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
        }
@@ -598,7 +598,7 @@ def process_results(Map conf=[:]){
 
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
 CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true
-                                              0 21 * * * % ROCMVERSION=5.5;COMPILER_VERSION=release;COMPILER_COMMIT=
+                                              0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
                                               0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {

From 1cf5003179970b3ce9f8252395962810adb50f76 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Tue, 4 Jul 2023 21:38:49 -0500
Subject: [PATCH 050/194] Add fp8 GEMM and an example for it (#767)

* Add fp8 xdl gemm

* Add example

* Use int8 intrinsics for buffer load/store

* Format

* Update cmakelists
---
 example/01_gemm/CMakeLists.txt                |  4 ++
 example/01_gemm/gemm_xdl_f8.cpp               | 38 +++++++++++
 .../tensor_operation/gpu/warp/xdlops_gemm.hpp | 62 +++++++++++++++++-
 include/ck/utility/amd_buffer_addressing.hpp  | 57 ++++++++++++++---
 include/ck/utility/amd_xdlops.hpp             | 63 +++++++++++++++++++
 5 files changed, 212 insertions(+), 12 deletions(-)
 create mode 100644 example/01_gemm/gemm_xdl_f8.cpp

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index c5a8295188..66afe73c2a 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -44,3 +44,7 @@ if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS
   add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
 endif()
 
+if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
+endif()
diff --git a/example/01_gemm/gemm_xdl_f8.cpp b/example/01_gemm/gemm_xdl_f8.cpp
new file mode 100644
index 0000000000..1015926777
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_f8.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::f8_t;
+using CDataType        = ck::f8_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::f8_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index faaa2c5a95..814969ef42 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -29,7 +29,9 @@ enum struct MfmaInstr
     mfma_i32_16x16x16i8,
     mfma_i32_32x32x16i8,
     mfma_i32_16x16x32i8,
-    mfma_f64_16x16x4f64
+    mfma_f64_16x16x4f64,
+    mfma_f32_32x32x16f8f8,
+    mfma_f32_16x16x32f8f8
 };
 
 template <MfmaInstr instr>
@@ -454,6 +456,50 @@ struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
     }
 };
 
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8f8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_32x32x16f8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8f8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_16x16x32f8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
 template <typename base_type, index_t MPerXdlops, index_t NPerXdlops>
 struct MfmaSelector
 {
@@ -594,6 +640,18 @@ struct MfmaSelector
     }
 #endif
 
+    template <>
+    static constexpr auto GetMfma<f8_t, 32, 32>()
+    {
+        return MfmaInstr::mfma_f32_32x32x16f8f8;
+    }
+
+    template <>
+    static constexpr auto GetMfma<f8_t, 16, 16>()
+    {
+        return MfmaInstr::mfma_f32_16x16x32f8f8;
+    }
+
     static constexpr auto selected_mfma = mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops>()>{};
 
     __host__ __device__ constexpr MfmaSelector()
@@ -794,7 +852,7 @@ struct XdlopsGemm
     {
         static_assert(is_same<base_type, double>::value || is_same<base_type, float>::value ||
                           is_same<base_type, half_t>::value || is_same<base_type, bhalf_t>::value ||
-                          is_same<base_type, int8_t>::value,
+                          is_same<base_type, int8_t>::value || is_same<base_type, f8_t>::value,
                       "base base_type must be double, float, half, bfloat16, and int8_t!");
 
         static_for<0, KPack / mfma_instr.k_per_blk, 1>{}([&](auto k) {
diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index 38ee76d883..ea231154ea 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -1114,13 +1114,30 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 #if CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK
     uint32_t src_addr_shift = src_thread_element_valid ? 0 : 0x80000000;
 
-    return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
-        src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+    if constexpr(is_same<scalar_t, f8_t>::value)
+    {
+        auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+        return bit_cast<vector_t>(tmp);
+    }
+    else
+    {
+        return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+    }
 #else
-    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
-        src_wave_buffer_resource, src_thread_addr_offset, 0);
-
-    return src_thread_element_valid ? tmp : vector_t(0);
+    if constexpr(is_same<scalar_t, f8_t>::value)
+    {
+        auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_thread_addr_offset, 0);
+        return src_thread_element_valid ? bit_cast<vector_t>(tmp) : vector_t(0);
+    }
+    else
+    {
+        vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+            src_wave_buffer_resource, src_thread_addr_offset, 0);
+        return src_thread_element_valid ? tmp : vector_t(0);
+    }
 #endif
 }
 
@@ -1179,13 +1196,33 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
 #if CK_EXPERIMENTAL_USE_BUFFER_STORE_OOB_CHECK_OFFSET_TRICK
     uint32_t dst_addr_shift = dst_thread_element_valid ? 0 : 0x80000000;
 
-    amd_buffer_store_impl<scalar_t, vector_size, coherence>(
-        src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+    if constexpr(is_same<scalar_t, f8_t>::value)
+    {
+        auto tmp =
+            bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(src_thread_data);
+        amd_buffer_store_impl<int8_t, vector_size, coherence>(
+            tmp, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+    }
+    else
+    {
+        amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+            src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+    }
 #else
     if(dst_thread_element_valid)
     {
-        amd_buffer_store_impl<scalar_t, vector_size, coherence>(
-            src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+        if constexpr(is_same<scalar_t, f8_t>::value)
+        {
+            auto tmp = bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
+                src_thread_data);
+            amd_buffer_store_impl<int8_t, vector_size, coherence>(
+                tmp, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+        }
+        else
+        {
+            amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+                src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+        }
     }
 #endif
 }
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index d00b7cd078..ca38077cf5 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -354,5 +354,68 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
 #endif
     }
 };
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_32x32x16f8f8;
+
+template <>
+struct intrin_mfma_f32_32x32x16f8f8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const f8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__)
+        reg_c.template AsType<float16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8(
+                bit_cast<long>(reg_a),
+                bit_cast<long>(reg_b),
+                reg_c.template AsType<float16_t>()[Number<0>{}],
+                0,
+                0,
+                0);
+#else
+        vector_type<f8_t, 8> reg_a_v(reg_a);
+        vector_type<f8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<f8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<f8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_32x32x2f32<32, 32>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_16x16x32f8f8;
+
+template <>
+struct intrin_mfma_f32_16x16x32f8f8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const f8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f32_16x16x32_fp8_fp8(
+            bit_cast<long>(reg_a),
+            bit_cast<long>(reg_b),
+            reg_c.template AsType<float4_t>()[Number<0>{}],
+            0,
+            0,
+            0);
+#else
+        vector_type<f8_t, 8> reg_a_v(reg_a);
+        vector_type<f8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<f8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<f8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_16x16x4f32<16, 16>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
 } // namespace ck
 #endif

From 61dc9aa93277561bde816098ba8487d985b1b439 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Wed, 5 Jul 2023 18:29:56 -0500
Subject: [PATCH 051/194] Add the missing archs (#785)

---
 include/ck/utility/amd_xdlops.hpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index ca38077cf5..ea7755036f 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -364,7 +364,7 @@ struct intrin_mfma_f32_32x32x16f8f8<32, 32>
     template <class FloatC>
     __device__ static void Run(const f8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
     {
-#if defined(__gfx940__)
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
         reg_c.template AsType<float16_t>()(Number<0>{}) =
             __builtin_amdgcn_mfma_f32_32x32x16_fp8_fp8(
                 bit_cast<long>(reg_a),
@@ -396,7 +396,7 @@ struct intrin_mfma_f32_16x16x32f8f8<16, 16>
     template <class FloatC>
     __device__ static void Run(const f8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
     {
-#if defined(__gfx940__)
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
         reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f32_16x16x32_fp8_fp8(
             bit_cast<long>(reg_a),
             bit_cast<long>(reg_b),

From 2b0b6d9f465f241196e19d35fec4ed062c5bbd29 Mon Sep 17 00:00:00 2001
From: Bartlomiej Kocot <barkocot@amd.com>
Date: Wed, 5 Jul 2023 11:16:17 +0000
Subject: [PATCH 052/194] Fix copyrights for DeviceBatchedGemmMultipleD_Dl

---
 .../tensor_operation_instance/gpu/batched_gemm_multi_d.hpp      | 2 +-
 ...batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp | 2 +-
 ...mm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp | 2 +-
 ...batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp | 2 +-
 ...mm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp | 2 +-
 ...batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp | 2 +-
 ...mm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp | 2 +-
 ...batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp | 2 +-
 ...mm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp | 2 +-
 ...ce_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp | 2 +-
 ..._gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp | 2 +-
 ...ce_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp | 2 +-
 ..._gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp | 2 +-
 ...ce_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp | 2 +-
 ..._gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp | 2 +-
 ...ce_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp | 2 +-
 ..._gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp | 2 +-
 test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp         | 2 +-
 18 files changed, 18 insertions(+), 18 deletions(-)

diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
index ae12f4c7ac..3d50aae092 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
index 3fe9f78b2a..b20e8a86b1 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
index 4ab22bb03a..e98fd24eeb 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
index 80c890cdb3..75042280f0 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
index 647c583036..c8e1acfd4f 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
index 3ce582f1f9..f1ed7a6942 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
index 34c29d0380..6e25517ed8 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
index e0da8e3d5a..5a9fd16c83 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
index 3edd3f78c6..cf5502926b 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
index 33234edc17..43cdcc5f14 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
index 16107e1bde..617877499d 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
index 3e4bdb0172..1f58aecf50 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
index 8107717162..e0ead0ec49 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
index 0e943c88c3..55fa86f37f 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
index ea5e7c562d..b8af3e61a3 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
index 000a4b0130..29d20e58dd 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
index 24fb67619a..dd0f8749c0 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <cstdlib>
 
diff --git a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
index 4a82654034..6cf6145cd7 100644
--- a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
+++ b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <gtest/gtest.h>

From f4dfc060b79987580da9afc481dad746d5b3178d Mon Sep 17 00:00:00 2001
From: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Date: Thu, 6 Jul 2023 16:15:51 +0200
Subject: [PATCH 053/194] Move Device Ops implementations into impl directory.
 (#777)

Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 example/01_gemm/gemm_xdl_wavelet_fp16.cpp                       | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp                    | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp                    | 2 +-
 example/09_convnd_fwd/convnd_fwd_dl_int8.cpp                    | 2 +-
 ..._gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp | 2 +-
 .../grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp        | 2 +-
 .../conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp    | 2 +-
 .../conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp      | 2 +-
 .../conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp    | 2 +-
 .../conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp      | 2 +-
 .../conv2d_fwd_dl_perchannel_quantization_int8.cpp              | 2 +-
 .../conv2d_fwd_dl_perlayer_quantization_int8.cpp                | 2 +-
 .../splitk_gemm_bias_e_permute_xdl_fp16.cpp                     | 2 +-
 .../splitk_gemm_bias_e_permute_xdl_fp32.cpp                     | 2 +-
 .../device/{ => impl}/device_gemm_xdl_waveletmodel_cshuffle.hpp | 0
 .../device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp    | 0
 .../{ => impl}/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp    | 0
 .../device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp   | 0
 .../device_splitk_contraction_multiple_d_xdl_cshuffle.hpp       | 0
 .../device_grouped_conv2d_fwd_dl_instance.hpp                   | 2 +-
 .../quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp  | 2 +-
 test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp             | 2 +-
 22 files changed, 17 insertions(+), 17 deletions(-)
 rename include/ck/tensor_operation/gpu/device/{ => impl}/device_gemm_xdl_waveletmodel_cshuffle.hpp (100%)
 rename include/ck/tensor_operation/gpu/device/{ => impl}/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp (100%)
 rename include/ck/tensor_operation/gpu/device/{ => impl}/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp (100%)
 rename include/ck/tensor_operation/gpu/device/{ => impl}/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp (100%)
 rename include/ck/tensor_operation/gpu/device/{ => impl}/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp (100%)

diff --git a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
index d7176f75da..b0f963fee5 100644
--- a/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_wavelet_fp16.cpp
@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
 
 using ADataType        = ck::half_t;
 using BDataType        = ck::half_t;
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp b/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
index 7b6f18f46a..564fadcbf8 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_fp16.cpp
@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp b/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
index 551655b177..0bd90c9992 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_fp32.cpp
@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
diff --git a/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp b/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
index 27a3f2e2a9..9b0c8b31f1 100644
--- a/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
+++ b/example/09_convnd_fwd/convnd_fwd_dl_int8.cpp
@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_dl_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
index d0eb8fcc34..5794924294 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
diff --git a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
index 1d97474d20..97caec6053 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
+++ b/example/32_batched_gemm_scale_softmax_gemm/grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp
@@ -17,7 +17,7 @@ Gemm + Softmax + Gemm fused operation. Computes C_g_m_o = Softmax(A_g_m_k * B0_g
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
index 40b33852b5..4573c68658 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
index fc081ddc54..005f6263fd 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType   = int8_t;
 using WeiDataType  = int8_t;
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
index c390f016a9..62e5e583de 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
index 10b131a527..ef98fe7e4f 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType   = int8_t;
 using WeiDataType  = int8_t;
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
index e59d0d0759..e524ddb2b2 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perchannel_quantization_int8.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType           = int8_t;
 using WeiDataType          = int8_t;
diff --git a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
index aee5fe9e60..d29a3143c0 100644
--- a/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
+++ b/example/40_conv2d_fwd_quantization/conv2d_fwd_dl_perlayer_quantization_int8.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 using InDataType  = int8_t;
 using WeiDataType = int8_t;
diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
index b6d9b29a5b..ebba88cf41 100644
--- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
+++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp16.cpp
@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
diff --git a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
index 60a0e01fe3..4ab26293cc 100644
--- a/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
+++ b/example/43_splitk_gemm_bias_e_permute/splitk_gemm_bias_e_permute_xdl_fp32.cpp
@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/device/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
diff --git a/include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
similarity index 100%
rename from include/ck/tensor_operation/gpu/device/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
index 3d3f9b1793..631c2ce093 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 #include "device_grouped_conv2d_fwd_common.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
index bb7a570cda..2ec37c8413 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
+++ b/library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_dl_int8_instance.hpp
@@ -4,7 +4,7 @@
 #pragma once
 
 #include "conv2d_quantization_common.hpp"
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
 
 namespace ck {
 namespace tensor_operation {
diff --git a/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp b/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
index 983af7ecdd..23ff89da60 100644
--- a/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
+++ b/test/gemm/instance/gemm_wavelet_f16_tn_instance.cpp
@@ -5,7 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_waveletmodel_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"

From 8f5cafaf0453172fc16432c7d0d4561e7ecf5bdf Mon Sep 17 00:00:00 2001
From: Qianfeng <qianfeng.zhang@amd.com>
Date: Thu, 6 Jul 2023 23:58:55 +0800
Subject: [PATCH 054/194] Batchnorm splitk single kernel (#771)

* Use dim 0 as faster dim for writing mean/var/count workspace in batchnorm multiblock method [performance]

* Add CountDataType as template parameter in blockwise_welford

* Add utility/get_shift.hpp

* Add BatchNorm multiblock single-kernel implementation

* Add smem inline assembly based implementation of gms_init/gms_barrier/gms_reset for gfx90a

* Renaming in device_batchnorm_forward_impl.hpp

* Tiny fix in the batchnorm_fwd profiler

* Revert "Add smem inline assembly based implementation of gms_init/gms_barrier/gms_reset for gfx90a"

This reverts commit d16d00919c43f10759e7b4e4d112125221ed9064.

* Use the old two-kernel batchnorm multiblock method for gfx1030

* Use the old two-kernel batchnorm multiblock method for gfx908

* use the single-kernel batchnorm multiblock method only for gfx90a

* Remove get_wave_id() from utility/get_id.hpp since it is not used

* Set true for testing running mean/variance and saving mean/invvariance in the examples

* Fix to copy-right words

* Remove un-needed including in utility/get_id.hpp

* Add comments to workgroup_synchronization.hpp

* Remove un-used codes in gridwise_multiblock_batchnorm_forward.hpp

* Renaming in the kernels

* Remove un-used kernel file
---
 example/34_batchnorm/CMakeLists.txt           |   1 +
 .../batchnorm_forward_training_nhwc.cpp       |  17 +-
 ...tchnorm_forward_training_nhwc_obsolete.cpp | 598 +++++++++++++++
 .../gpu/block/blockwise_welford.hpp           |  24 +-
 .../block/reduction_functions_blockwise.hpp   |   2 +-
 .../impl/device_batchnorm_forward_impl.hpp    | 258 +++++--
 ...device_batchnorm_forward_impl_obsolete.hpp | 714 ++++++++++++++++++
 .../gridwise_multiblock_batchnorm_forward.hpp | 704 +++++++++++++++++
 ...gridwise_multiblock_welford_first_half.hpp |   4 +-
 ...half_batchnorm_forward_final_obsolete.hpp} |  19 +-
 include/ck/utility/get_shift.hpp              |  20 +
 include/ck/utility/reduction_common.hpp       |  12 -
 .../ck/utility/workgroup_synchronization.hpp  |  74 ++
 profiler/src/profile_batchnorm_fwd.cpp        |   2 +-
 14 files changed, 2330 insertions(+), 119 deletions(-)
 create mode 100644 example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
 rename include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/{gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp => gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp} (98%)
 create mode 100644 include/ck/utility/get_shift.hpp
 create mode 100644 include/ck/utility/workgroup_synchronization.hpp

diff --git a/example/34_batchnorm/CMakeLists.txt b/example/34_batchnorm/CMakeLists.txt
index d964f40d87..60824c5f4d 100644
--- a/example/34_batchnorm/CMakeLists.txt
+++ b/example/34_batchnorm/CMakeLists.txt
@@ -1,3 +1,4 @@
 add_example_executable(example_batchnorm_forward_training batchnorm_forward_training_nhwc.cpp)
+add_example_executable(example_batchnorm_forward_training_obsolete batchnorm_forward_training_nhwc_obsolete.cpp)
 add_example_executable(example_batchnorm_forward_inferring batchnorm_forward_inferring_nhwc.cpp)
 add_example_executable(example_batchnorm_backward batchnorm_backward_nhwc.cpp)
diff --git a/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp b/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
index d680818157..b27358fd9d 100644
--- a/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
+++ b/example/34_batchnorm/batchnorm_forward_training_nhwc.cpp
@@ -414,7 +414,7 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
         (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
 
         y_dev.FromDevice(y.mData.data());
-        pass = pass && ck::utils::check_err(y, y_ref);
+        pass = pass && ck::utils::check_err(y, y_ref, "Incorrect normalized output values");
 
         if(updateMovingAverage)
         {
@@ -424,8 +424,12 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
             resultRunningMean_dev.FromDevice(resultRunningMean.mData.data());
             resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data());
 
-            pass = pass && ck::utils::check_err(resultRunningMean, resultRunningMean_ref);
-            pass = pass && ck::utils::check_err(resultRunningVariance, resultRunningVariance_ref);
+            pass = pass && ck::utils::check_err(resultRunningMean,
+                                                resultRunningMean_ref,
+                                                "Incorrect running mean values");
+            pass = pass && ck::utils::check_err(resultRunningVariance,
+                                                resultRunningVariance_ref,
+                                                "Incorrect running variance values");
         };
 
         if(saveMeanAndInvVariance)
@@ -438,8 +442,11 @@ bool bnorm_fwd_nhwc_test(bool do_verification,
             resultSaveMean_dev.FromDevice(resultSaveMean.mData.data());
             resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data());
 
-            pass = pass && ck::utils::check_err(resultSaveMean, resultSaveMean_ref);
-            pass = pass && ck::utils::check_err(resultSaveInvVariance, resultSaveInvVariance_ref);
+            pass = pass && ck::utils::check_err(
+                               resultSaveMean, resultSaveMean_ref, "Incorrect saved mean values");
+            pass = pass && ck::utils::check_err(resultSaveInvVariance,
+                                                resultSaveInvVariance_ref,
+                                                "Incorrect saved invvariance values");
         };
     };
 
diff --git a/example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp b/example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp
new file mode 100644
index 0000000000..ffb9f4b584
--- /dev/null
+++ b/example/34_batchnorm/batchnorm_forward_training_nhwc_obsolete.cpp
@@ -0,0 +1,598 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <limits>
+#include <iostream>
+#include <vector>
+#include <array>
+#include <algorithm>
+#include <getopt.h>
+
+#include "ck/ck.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_batchnorm_forward.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp"
+#include "ck/library/utility/host_common_util.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+static struct option long_options[] = {{"inOutLengths", required_argument, nullptr, 'D'},
+                                       {"verify", required_argument, nullptr, 'v'},
+                                       {"help", no_argument, nullptr, '?'},
+                                       {nullptr, 0, nullptr, 0}};
+
+class BatchNormFwdArg
+{
+    private:
+    int option_index = 0;
+
+    public:
+    std::vector<size_t> inOutLengths;
+
+    bool do_verification = false;
+
+    bool updateMovingAverage;
+    bool saveMeanAndInvVariance;
+
+    int data_type               = 0;
+    int init_method             = 2;
+    bool time_kernel            = false;
+    bool use_multiblock_welford = false;
+
+    public:
+    void show_usage(const char* cmd)
+    {
+        std::cout << "Usage of " << cmd << std::endl;
+        std::cout << "--inOutLengths or -D, comma separated list of input tensor dimension "
+                     "lengths, must have 4 integers for nhwc"
+                  << std::endl;
+        std::cout << "--verify or -v, 1/0 to indicate whether to verify the batch-normalization "
+                     "result by "
+                     "comparing with the host-based batch-normalization"
+                  << std::endl;
+        std::cout << "Arg1: data type (0: fp16, 1: fp32, 3: int8, 5: bp16, 6: fp64)" << std::endl;
+        std::cout << "Arg2: 1/0 to indicate whether to update the moving average and variance "
+                     "(0=no, 1=yes)"
+                  << std::endl;
+        std::cout << "Arg3: 1/0 to indicate whether to save the calculated mean and invVariance "
+                     "(0=no, 1=yes)"
+                  << std::endl;
+        std::cout << "Arg4: init method used for bnScale and bnBias (0=no init, 1=single integer "
+                     "value, 2=scope integer "
+                     "value, 3=decimal value)"
+                  << std::endl;
+        std::cout << "Arg5: time kernel (0=no, 1=yes)" << std::endl;
+        std::cout << "Arg6: use multi-block welford (0=n0, 1=yes)" << std::endl;
+    };
+
+    int processArgs(int argc, char* argv[])
+    {
+        using ck::host_common::getTypeValuesFromString;
+
+        int ch;
+
+        while(1)
+        {
+            ch = getopt_long(argc, argv, "D:v:", long_options, &option_index);
+            if(ch == -1)
+                break;
+            switch(ch)
+            {
+            case 'D':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                inOutLengths = getTypeValuesFromString<size_t>(optarg);
+
+                if(inOutLengths.size() != 4)
+                    throw std::runtime_error(
+                        "NHWC tensor layout should have 4 length values specified!");
+                break;
+            case 'v':
+                if(!optarg)
+                    throw std::runtime_error("Invalid option format!");
+
+                do_verification = static_cast<bool>(std::atoi(optarg));
+                break;
+            case '?':
+                if(std::string(long_options[option_index].name) == "help")
+                {
+                    show_usage(argv[0]);
+                    return (-1);
+                };
+                break;
+            default: show_usage(argv[0]); return (-1);
+            };
+        };
+
+        if(optind + 6 > argc)
+            throw std::runtime_error("Invalid cmd-line arguments, more argumetns are needed!");
+
+        data_type              = std::atoi(argv[optind++]);
+        updateMovingAverage    = std::atoi(argv[optind++]);
+        saveMeanAndInvVariance = std::atoi(argv[optind++]);
+        init_method            = std::atoi(argv[optind++]);
+        time_kernel            = static_cast<bool>(std::atoi(argv[optind++]));
+        use_multiblock_welford = static_cast<bool>(std::atoi(argv[optind]));
+
+        if(data_type != 0 && data_type != 1 && data_type != 3 && data_type != 5 && data_type != 6)
+            return (-1);
+
+        return (0);
+    };
+};
+
+using namespace ck;
+
+template <typename InOutDataType, typename AccDataType, bool UseMultiblockInK>
+bool bnorm_fwd_nhwc_test(bool do_verification,
+                         int init_method,
+                         bool time_kernel,
+                         const std::vector<size_t> inOutLengths,
+                         bool updateMovingAverage,
+                         bool saveMeanAndInvVariance,
+                         double averageFactor,
+                         double epsilon)
+{
+    // for NHWC BatchNorm calculation of mean and meansquare
+    constexpr int Rank         = 4;
+    constexpr int NumReduceDim = 3;
+
+    // when using lengths[] to create a tensor, lengths[0] is the length of highest dimension
+    // eg. N of NHWC, so lengths[3] is the dimension C length of NHWC
+    const std::vector<size_t> scaleBiasMeanVarLengths = {inOutLengths[3]};
+
+    // input data of the batchnorm forward algorithm
+    Tensor<InOutDataType> x(inOutLengths);
+    Tensor<AccDataType> bnScale(scaleBiasMeanVarLengths);
+    Tensor<AccDataType> bnBias(scaleBiasMeanVarLengths);
+
+    // output data of the batchnorm forward algorithm
+    Tensor<InOutDataType> y_ref(inOutLengths);
+    Tensor<InOutDataType> y(inOutLengths);
+
+    Tensor<AccDataType> resultSaveMean_ref(scaleBiasMeanVarLengths);
+    Tensor<AccDataType> resultSaveInvVariance_ref(scaleBiasMeanVarLengths);
+
+    Tensor<AccDataType> resultRunningMean_ref(scaleBiasMeanVarLengths);
+    Tensor<AccDataType> resultRunningVariance_ref(scaleBiasMeanVarLengths);
+
+    auto inOutStrides            = x.mDesc.GetStrides();
+    auto scaleBiasMeanVarStrides = bnScale.mDesc.GetStrides();
+
+    std::size_t num_thread = std::thread::hardware_concurrency();
+
+    if(updateMovingAverage)
+    {
+        if constexpr(std::is_same<InOutDataType, int8_t>::value)
+        {
+            x.GenerateTensorValue(GeneratorTensor_2<InOutDataType>{-5, 5}, num_thread);
+
+            const float x_mean       = 0.0f;
+            const float x_stddev     = 2.5f;
+            const float noise_stddev = 0.04f;
+
+            resultRunningMean_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_mean, noise_stddev}, num_thread);
+
+            resultRunningVariance_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_stddev * x_stddev, noise_stddev}, num_thread);
+        }
+        else
+        {
+            const float x_mean       = 0.0f;
+            const float x_stddev     = 1.0f;
+            const float noise_stddev = 0.04f;
+
+            // input data in normal distribution
+            x.GenerateTensorValue(GeneratorTensor_4<InOutDataType>{x_mean, x_stddev}, num_thread);
+
+            // initialize the runningMean to be values with tiny variation to the mean of the x
+            // values
+            resultRunningMean_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_mean, noise_stddev}, num_thread);
+
+            // initialize the runningVariance to be values with tiny variation to the variance of
+            // the x values
+            resultRunningVariance_ref.GenerateTensorValue(
+                GeneratorTensor_4<AccDataType>{x_stddev * x_stddev, noise_stddev}, num_thread);
+        };
+    }
+    else
+    {
+        if constexpr(std::is_same<InOutDataType, int8_t>::value)
+            x.GenerateTensorValue(GeneratorTensor_2<InOutDataType>{-5, 5}, num_thread);
+        else
+            x.GenerateTensorValue(GeneratorTensor_3<InOutDataType>{-5.0f, 5.0f}, num_thread);
+    };
+
+    if(do_verification)
+    {
+        switch(init_method)
+        {
+        case 0:
+            bnScale.GenerateTensorValue(GeneratorTensor_0<AccDataType>{}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_0<AccDataType>{}, num_thread);
+            break;
+        case 1:
+            bnScale.GenerateTensorValue(GeneratorTensor_1<AccDataType>{1}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_1<AccDataType>{0}, num_thread);
+            break;
+        case 2:
+            bnScale.GenerateTensorValue(GeneratorTensor_2<AccDataType>{-5, 5}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_2<AccDataType>{-5, 5}, num_thread);
+            break;
+        default:
+            bnScale.GenerateTensorValue(GeneratorTensor_3<AccDataType>{-5.0f, 5.0f}, num_thread);
+            bnBias.GenerateTensorValue(GeneratorTensor_3<AccDataType>{-5.0f, 5.0f}, num_thread);
+        }
+    };
+
+    // these buffers are usually provided by the user application
+    DeviceMem x_dev(sizeof(InOutDataType) * x.mDesc.GetElementSpaceSize());
+    DeviceMem y_dev(sizeof(InOutDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem bnScale_dev(sizeof(AccDataType) * bnScale.mDesc.GetElementSpaceSize());
+    DeviceMem bnBias_dev(sizeof(AccDataType) * bnBias.mDesc.GetElementSpaceSize());
+
+    // mean_dev or resultSaveMean_dev
+    DeviceMem resultSaveMean_dev(sizeof(AccDataType) *
+                                 resultSaveMean_ref.mDesc.GetElementSpaceSize());
+    // meansquare_dev or resultSaveInvVariance_dev
+    DeviceMem resultSaveInvVariance_dev(sizeof(AccDataType) *
+                                        resultSaveInvVariance_ref.mDesc.GetElementSpaceSize());
+    // resultRunningMean_dev
+    DeviceMem resultRunningMean_dev(sizeof(AccDataType) *
+                                    resultRunningMean_ref.mDesc.GetElementSpaceSize());
+    // resultRunningVariance_dev
+    DeviceMem resultRunningVariance_dev(sizeof(AccDataType) *
+                                        resultRunningVariance_ref.mDesc.GetElementSpaceSize());
+
+    x_dev.ToDevice(x.mData.data());
+    bnScale_dev.ToDevice(bnScale.mData.data());
+    bnBias_dev.ToDevice(bnBias.mData.data());
+
+    if(updateMovingAverage)
+    {
+        resultRunningMean_dev.ToDevice(resultRunningMean_ref.mData.data());
+        resultRunningVariance_dev.ToDevice(resultRunningVariance_ref.mData.data());
+    };
+
+    std::array<index_t, Rank> i_inOutLengths;
+    std::array<index_t, Rank> i_inOutStrides;
+    std::array<index_t, Rank - NumReduceDim> i_scaleBiasMeanVarLengths;
+    std::array<index_t, Rank - NumReduceDim> i_scaleBiasMeanVarStrides;
+
+    ck::ranges::copy(inOutLengths, i_inOutLengths.begin());
+    ck::ranges::copy(inOutStrides, i_inOutStrides.begin());
+    ck::ranges::copy(scaleBiasMeanVarLengths, i_scaleBiasMeanVarLengths.begin());
+    ck::ranges::copy(scaleBiasMeanVarStrides, i_scaleBiasMeanVarStrides.begin());
+
+    using PassThroughOp = ck::tensor_operation::element_wise::PassThrough;
+
+    using DeviceBatchNormFwdInstance =
+        ck::tensor_operation::device::DeviceBatchNormFwdImpl<InOutDataType,
+                                                             InOutDataType,
+                                                             AccDataType,
+                                                             AccDataType,   // ScaleDataType
+                                                             AccDataType,   // BiasDataType
+                                                             AccDataType,   // MeanVarDataType
+                                                             PassThroughOp, // YElementwiseOp
+                                                             Rank,
+                                                             NumReduceDim,
+                                                             UseMultiblockInK,
+                                                             256,
+                                                             16,
+                                                             16,
+                                                             1,
+                                                             2,
+                                                             0,
+                                                             1,
+                                                             1,
+                                                             1,
+                                                             1,
+                                                             1>;
+
+    auto batchnorm_fwd = DeviceBatchNormFwdInstance{};
+
+    auto argument_ptr = batchnorm_fwd.MakeArgumentPointer(
+        i_inOutLengths,
+        i_inOutStrides,
+        i_inOutStrides,
+        {0, 1, 2}, // indicates physical indices of reduce dimensions in lengths[] and strides[]
+        i_scaleBiasMeanVarLengths,
+        i_scaleBiasMeanVarStrides,
+        i_scaleBiasMeanVarStrides,
+        i_scaleBiasMeanVarStrides,
+        x_dev.GetDeviceBuffer(),
+        bnScale_dev.GetDeviceBuffer(),
+        bnBias_dev.GetDeviceBuffer(),
+        epsilon,
+        PassThroughOp{},
+        y_dev.GetDeviceBuffer(),
+        saveMeanAndInvVariance ? resultSaveMean_dev.GetDeviceBuffer() : nullptr,
+        saveMeanAndInvVariance ? resultSaveInvVariance_dev.GetDeviceBuffer() : nullptr,
+        averageFactor,
+        updateMovingAverage ? resultRunningMean_dev.GetDeviceBuffer() : nullptr,
+        updateMovingAverage ? resultRunningVariance_dev.GetDeviceBuffer() : nullptr);
+
+    if(!batchnorm_fwd.IsSupportedArgument(argument_ptr.get()))
+    {
+        std::cout << "The runtime parameters seems not supported by the BatchNorm device instance, "
+                     "exiting!"
+                  << std::endl;
+        return (false);
+    };
+
+    size_t workspace_sz = batchnorm_fwd.GetWorkSpaceSize(argument_ptr.get());
+
+    DeviceMem workspace_dev(workspace_sz);
+
+    batchnorm_fwd.SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
+    auto invoker_ptr = batchnorm_fwd.MakeInvokerPointer();
+
+    if(time_kernel)
+    {
+        float avg_time   = 0.0f;
+        size_t num_bytes = 0;
+
+        size_t total_length = inOutLengths[0] * inOutLengths[1] * inOutLengths[2] * inOutLengths[3];
+        size_t invariant_length = inOutLengths[3];
+
+        avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        // inputing of x, scale, bias, outputing of y
+        num_bytes +=
+            total_length * sizeof(InOutDataType) * 2 + invariant_length * sizeof(AccDataType) * 2;
+
+        // outputing of mean, inv-variance
+        num_bytes += saveMeanAndInvVariance ? invariant_length * sizeof(AccDataType) * 2 : 0;
+
+        // updating of moving mean, variance
+        num_bytes += updateMovingAverage ? invariant_length * sizeof(AccDataType) * 4 : 0;
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        std::cout << "Perf: " << avg_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
+    }
+    else
+        (void)invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+
+        using ReferenceBatchNormFwdInstance =
+            ck::tensor_operation::host::ReferenceBatchNormFwd<InOutDataType,
+                                                              InOutDataType,
+                                                              AccDataType,
+                                                              AccDataType,
+                                                              AccDataType,
+                                                              AccDataType,
+                                                              PassThroughOp,
+                                                              Rank,
+                                                              NumReduceDim>;
+
+        auto batchNormFwd_ref = ReferenceBatchNormFwdInstance{};
+
+        auto argument_ptr_ref = batchNormFwd_ref.MakeArgumentPointer(
+            i_inOutLengths,
+            i_inOutStrides,
+            i_inOutStrides,
+            {0, 1, 2}, // indicates physical indices of reduce dimensions in lengths[] and strides[]
+            i_scaleBiasMeanVarLengths,
+            i_scaleBiasMeanVarStrides,
+            i_scaleBiasMeanVarStrides,
+            i_scaleBiasMeanVarStrides,
+            x.mData.data(),
+            bnScale.mData.data(),
+            bnBias.mData.data(),
+            epsilon,
+            PassThroughOp{},
+            y_ref.mData.data(),
+            saveMeanAndInvVariance ? resultSaveMean_ref.mData.data() : nullptr,
+            saveMeanAndInvVariance ? resultSaveInvVariance_ref.mData.data() : nullptr,
+            averageFactor,
+            updateMovingAverage ? resultRunningMean_ref.mData.data() : nullptr,
+            updateMovingAverage ? resultRunningVariance_ref.mData.data() : nullptr);
+
+        if(!batchNormFwd_ref.IsSupportedArgument(argument_ptr_ref.get()))
+        {
+            std::cout << "The runtime parameters seems not supported by the BatchNorm reference "
+                         "instance, exiting!"
+                      << std::endl;
+            return (false);
+        };
+
+        auto invoker_ptr_ref = batchNormFwd_ref.MakeInvokerPointer();
+
+        (void)invoker_ptr_ref->Run(argument_ptr_ref.get());
+
+        y_dev.FromDevice(y.mData.data());
+        pass = pass && ck::utils::check_err(y, y_ref, "Incorrect normalized output values");
+
+        if(updateMovingAverage)
+        {
+            Tensor<AccDataType> resultRunningMean(scaleBiasMeanVarLengths);
+            Tensor<AccDataType> resultRunningVariance(scaleBiasMeanVarLengths);
+
+            resultRunningMean_dev.FromDevice(resultRunningMean.mData.data());
+            resultRunningVariance_dev.FromDevice(resultRunningVariance.mData.data());
+
+            pass = pass && ck::utils::check_err(resultRunningMean,
+                                                resultRunningMean_ref,
+                                                "Incorrect running mean values");
+            pass = pass && ck::utils::check_err(resultRunningVariance,
+                                                resultRunningVariance_ref,
+                                                "Incorrect running variance values");
+        };
+
+        if(saveMeanAndInvVariance)
+        {
+            using ck::host_common::dumpBufferToFile;
+
+            Tensor<AccDataType> resultSaveMean(scaleBiasMeanVarLengths);
+            Tensor<AccDataType> resultSaveInvVariance(scaleBiasMeanVarLengths);
+
+            resultSaveMean_dev.FromDevice(resultSaveMean.mData.data());
+            resultSaveInvVariance_dev.FromDevice(resultSaveInvVariance.mData.data());
+
+            pass = pass && ck::utils::check_err(
+                               resultSaveMean, resultSaveMean_ref, "Incorrect saved mean values");
+            pass = pass && ck::utils::check_err(resultSaveInvVariance,
+                                                resultSaveInvVariance_ref,
+                                                "Incorrect saved invvariance values");
+        };
+    };
+
+    return (pass);
+};
+
+const double epsilon              = std::numeric_limits<float>::epsilon();
+static const double averageFactor = 0.1;
+
+int main(int argc, char* argv[])
+{
+    bool pass = true;
+
+    if(argc > 1)
+    {
+        BatchNormFwdArg arg;
+
+        if(arg.processArgs(argc, argv) < 0)
+            return (-1);
+
+        if(arg.data_type == 0)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<ck::half_t, float, true>(arg.do_verification,
+                                                                    arg.init_method,
+                                                                    arg.time_kernel,
+                                                                    arg.inOutLengths,
+                                                                    arg.updateMovingAverage,
+                                                                    arg.saveMeanAndInvVariance,
+                                                                    averageFactor,
+                                                                    epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<ck::half_t, float, false>(arg.do_verification,
+                                                                     arg.init_method,
+                                                                     arg.time_kernel,
+                                                                     arg.inOutLengths,
+                                                                     arg.updateMovingAverage,
+                                                                     arg.saveMeanAndInvVariance,
+                                                                     averageFactor,
+                                                                     epsilon);
+        }
+        else if(arg.data_type == 1)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<float, float, true>(arg.do_verification,
+                                                               arg.init_method,
+                                                               arg.time_kernel,
+                                                               arg.inOutLengths,
+                                                               arg.updateMovingAverage,
+                                                               arg.saveMeanAndInvVariance,
+                                                               averageFactor,
+                                                               epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<float, float, false>(arg.do_verification,
+                                                                arg.init_method,
+                                                                arg.time_kernel,
+                                                                arg.inOutLengths,
+                                                                arg.updateMovingAverage,
+                                                                arg.saveMeanAndInvVariance,
+                                                                averageFactor,
+                                                                epsilon);
+        }
+        else if(arg.data_type == 3)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<int8_t, float, true>(arg.do_verification,
+                                                                arg.init_method,
+                                                                arg.time_kernel,
+                                                                arg.inOutLengths,
+                                                                arg.updateMovingAverage,
+                                                                arg.saveMeanAndInvVariance,
+                                                                averageFactor,
+                                                                epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<int8_t, float, false>(arg.do_verification,
+                                                                 arg.init_method,
+                                                                 arg.time_kernel,
+                                                                 arg.inOutLengths,
+                                                                 arg.updateMovingAverage,
+                                                                 arg.saveMeanAndInvVariance,
+                                                                 averageFactor,
+                                                                 epsilon);
+        }
+        else if(arg.data_type == 5)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<ck::bhalf_t, float, true>(arg.do_verification,
+                                                                     arg.init_method,
+                                                                     arg.time_kernel,
+                                                                     arg.inOutLengths,
+                                                                     arg.updateMovingAverage,
+                                                                     arg.saveMeanAndInvVariance,
+                                                                     averageFactor,
+                                                                     epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<ck::bhalf_t, float, false>(arg.do_verification,
+                                                                      arg.init_method,
+                                                                      arg.time_kernel,
+                                                                      arg.inOutLengths,
+                                                                      arg.updateMovingAverage,
+                                                                      arg.saveMeanAndInvVariance,
+                                                                      averageFactor,
+                                                                      epsilon);
+        }
+        else if(arg.data_type == 6)
+        {
+            if(arg.use_multiblock_welford)
+                pass = bnorm_fwd_nhwc_test<double, double, true>(arg.do_verification,
+                                                                 arg.init_method,
+                                                                 arg.time_kernel,
+                                                                 arg.inOutLengths,
+                                                                 arg.updateMovingAverage,
+                                                                 arg.saveMeanAndInvVariance,
+                                                                 averageFactor,
+                                                                 epsilon);
+            else
+                pass = bnorm_fwd_nhwc_test<double, double, false>(arg.do_verification,
+                                                                  arg.init_method,
+                                                                  arg.time_kernel,
+                                                                  arg.inOutLengths,
+                                                                  arg.updateMovingAverage,
+                                                                  arg.saveMeanAndInvVariance,
+                                                                  averageFactor,
+                                                                  epsilon);
+        }
+    }
+    else
+    {
+        pass = bnorm_fwd_nhwc_test<ck::half_t, float, true>(true,
+                                                            2,
+                                                            false, // don't time kernel
+                                                            {128, 16, 6, 512},
+                                                            true,
+                                                            true,
+                                                            averageFactor,
+                                                            epsilon);
+
+        pass = pass && bnorm_fwd_nhwc_test<ck::half_t, float, false>(true,
+                                                                     2,
+                                                                     false, // don't time kernel
+                                                                     {128, 16, 3, 1024},
+                                                                     true,
+                                                                     true,
+                                                                     averageFactor,
+                                                                     epsilon);
+    };
+
+    return (pass ? 0 : 1);
+}
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp b/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
index a3813ea248..820a08fc48 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_welford.hpp
@@ -4,7 +4,7 @@
 #pragma once
 
 #include "ck/tensor_description/cluster_descriptor.hpp"
-#include "ck/utility/reduction_common.hpp"
+#include "ck/utility/get_shift.hpp"
 
 namespace ck {
 
@@ -35,10 +35,11 @@ struct BlockwiseWelford
     static constexpr auto thread_cluster_desc =
         make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
 
+    template <typename CountDataType>
     __device__ static inline void
-    Merge(T& mean_a, T& var_a, int& count_a, T mean_b, T var_b, int count_b)
+    Merge(T& mean_a, T& var_a, CountDataType& count_a, T mean_b, T var_b, CountDataType count_b)
     {
-        int count            = count_a + count_b;
+        CountDataType count  = count_a + count_b;
         T count_b_over_count = count == 0 ? type_convert<T>(0) : type_convert<T>(count_b) / count;
         T delta              = mean_b - mean_a;
         mean_a += delta * count_b_over_count;
@@ -46,11 +47,12 @@ struct BlockwiseWelford
         count_a = count;
     }
 
-    __device__ static void Run(T& mean_value, T& var_value, int& count)
+    template <typename CountDataType>
+    __device__ static void Run(T& mean_value, T& var_value, CountDataType& count)
     {
         __shared__ T mean_block_buf[BlockSize];
         __shared__ T var_block_buf[BlockSize];
-        __shared__ int count_block_buf[BlockSize];
+        __shared__ CountDataType count_block_buf[BlockSize];
 
         constexpr auto cluster_len_shift = get_shift<BufferLength_K>();
 
@@ -76,13 +78,13 @@ struct BlockwiseWelford
                 index_t offset2 = block_buf_desc_m_k.CalculateOffset(thread_cluster_idx +
                                                                      make_tuple(0, indOffset));
 
-                T mean1    = mean_block_buf[offset1];
-                T var1     = var_block_buf[offset1];
-                int count1 = count_block_buf[offset1];
+                T mean1              = mean_block_buf[offset1];
+                T var1               = var_block_buf[offset1];
+                CountDataType count1 = count_block_buf[offset1];
 
-                T mean2    = mean_block_buf[offset2];
-                T var2     = var_block_buf[offset2];
-                int count2 = count_block_buf[offset2];
+                T mean2              = mean_block_buf[offset2];
+                T var2               = var_block_buf[offset2];
+                CountDataType count2 = count_block_buf[offset2];
 
                 Merge(mean1, var1, count1, mean2, var2, count2);
 
diff --git a/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp b/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
index 6c13513cfb..82667e2352 100644
--- a/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
+++ b/include/ck/tensor_operation/gpu/block/reduction_functions_blockwise.hpp
@@ -4,7 +4,7 @@
 #pragma once
 
 #include "ck/tensor_description/cluster_descriptor.hpp"
-#include "ck/utility/reduction_common.hpp"
+#include "ck/utility/get_shift.hpp"
 #include "ck/utility/reduction_functions_accumulate.hpp"
 
 namespace ck {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
index ce0320b2d7..ad8e795603 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl.hpp
@@ -10,12 +10,14 @@
 #include "ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/device/welford_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp"
 #include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp"
-#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -114,8 +116,8 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
 
     static auto MakeMeanVarCountOutputMG2dDescriptor(int invariantLength, int blkGroupSize)
     {
-        const auto grid_desc_m_g =
-            make_naive_tensor_descriptor_packed(make_tuple(invariantLength, blkGroupSize));
+        const auto grid_desc_m_g = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, blkGroupSize), make_tuple(1, invariantLength));
 
         const auto mPad =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
@@ -132,9 +134,9 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
 
     static auto MakeMeanVarCountInputMK2dDescriptor(int invariantLength, int blkGroupSize)
     {
-        const auto reduceLength = blkGroupSize;
-        const auto grid_desc_m_k =
-            make_naive_tensor_descriptor_packed(make_tuple(invariantLength, reduceLength));
+        const auto reduceLength  = blkGroupSize;
+        const auto grid_desc_m_k = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, reduceLength), make_tuple(1, invariantLength));
 
         const auto mPad =
             math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
@@ -244,8 +246,8 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
                     int testBlkGroupSize = (reduce_length_ + (K_BlockTileSize * iterations) - 1) /
                                            (K_BlockTileSize * iterations);
 
-                    // we want the blkGroupSize be not more than 128
-                    if(testBlkGroupSize <= 128)
+                    // we want the blkGroupSize be not more than 16
+                    if(testBlkGroupSize <= 16)
                         break;
 
                     iterations++;
@@ -319,6 +321,8 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
         void* workspace_mean_;
         void* workspace_variance_;
         void* workspace_count_;
+
+        void* control_;
     };
 
     size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
@@ -340,6 +344,11 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
             // workspace for welford intermediate count
             workspace_size +=
                 pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(int32_t) + 64;
+
+            // workspace for barrier objects, each barrier object consists of two integers
+            // TODO: allocate barrier object memory globally to reuse it by other operators
+            workspace_size += (pArg_->invariant_length_ + M_BlockTileSize - 1) / M_BlockTileSize *
+                              sizeof(int) * 2;
         }
 
         return (workspace_size);
@@ -353,7 +362,6 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
 
         if(UseMultiblockInK && pArg_->blkGroupSize_ > 1)
         {
-
             // setup buffer used for intermediate welford mean
             pArg_->workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
 
@@ -374,6 +382,18 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
             // setup buffer used for intermediate welfor count
             pArg_->workspace_count_ =
                 reinterpret_cast<char*>(pArg_->workspace_variance_) + variance_space_sz;
+
+            index_t count_space_sz =
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(int32_t);
+
+            count_space_sz = math::integer_least_multiple(count_space_sz, 64);
+
+            pArg_->control_ = reinterpret_cast<char*>(pArg_->workspace_count_) + count_space_sz;
+
+            index_t control_space_sz = (pArg_->invariant_length_ + M_BlockTileSize - 1) /
+                                       M_BlockTileSize * sizeof(int) * 2;
+
+            hip_check_error(hipMemset(pArg_->control_, 0, control_space_sz));
         };
     };
 
@@ -402,6 +422,32 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
                 using MeanVarCountGridDesc_M_G = decltype(mean_var_count_grid_desc_m_g);
                 using MeanVarCountGridDesc_M_K = decltype(mean_var_count_grid_desc_m_k);
 
+                using GridwiseMultiblockBatchNormForward_ =
+                    GridwiseMultiblockBatchNormForward<XDataType,
+                                                       YDataType,
+                                                       AccDataType,
+                                                       ScaleDataType,
+                                                       BiasDataType,
+                                                       MeanVarDataType,
+                                                       YElementwiseOp,
+                                                       XYGridDesc_M_K,
+                                                       MeanVarCountGridDesc_M_G,
+                                                       MeanVarCountGridDesc_M_K,
+                                                       ScaleBiasMeanVarGridDesc_M,
+                                                       ScaleBiasMeanVarGridDesc_M,
+                                                       GetReduceCountPerThreadFunctor,
+                                                       BlockSize,
+                                                       MThreadClusterSize,
+                                                       KThreadClusterSize,
+                                                       MThreadSliceSize,
+                                                       KThreadSliceSize,
+                                                       XSrcYDstVectorDim,
+                                                       XSrcVectorSize,
+                                                       YDstVectorSize,
+                                                       ScaleSrcVectorSize,
+                                                       BiasSrcVectorSize,
+                                                       MeanVarSrcDstVectorSize>;
+
                 using GridwiseMultiblockWelfordFirstHalf_ =
                     GridwiseMultiblockWelfordFirstHalf<XDataType,
                                                        AccDataType,
@@ -441,78 +487,136 @@ struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
                                                                    BiasSrcVectorSize,
                                                                    MeanVarSrcDstVectorSize>;
 
-                index_t numMeanVarCountBlockTileIteration =
-                    (arg.blkGroupSize_ + KThreadClusterSize - 1) / KThreadClusterSize;
+                // It is found that:
+                // 1) gfx1030 does not support the GLC enabled vector load/store, so using the
+                //    two-kernel method for gfx1030
+                // 2) Profiler on gfx908 could hang even though it works when running examples
+                // 3) Single-kernel method works on gfx1100, but the performance it not better
+                //    than two-kernel method (due to more warps participating the barrier)
+                if(ck::get_device_name() == "gfx90a")
+                {
+                    const auto kern_multiblock_batchnorm_fwd_ =
+                        kernel_multiblock_batchnorm_forward<GridwiseMultiblockBatchNormForward_,
+                                                            XDataType,
+                                                            YDataType,
+                                                            AccDataType,
+                                                            ScaleDataType,
+                                                            BiasDataType,
+                                                            MeanVarDataType,
+                                                            YElementwiseOp,
+                                                            XYGridDesc_M_K,
+                                                            MeanVarCountGridDesc_M_G,
+                                                            MeanVarCountGridDesc_M_K,
+                                                            ScaleBiasMeanVarGridDesc_M,
+                                                            ScaleBiasMeanVarGridDesc_M,
+                                                            GetReduceCountPerThreadFunctor>;
 
-                const auto kern_multiblock_welford_first_half =
-                    kernel_multiblock_welford_first_half<GridwiseMultiblockWelfordFirstHalf_,
-                                                         XDataType,
-                                                         MeanVarDataType,
-                                                         XYGridDesc_M_K,
-                                                         MeanVarCountGridDesc_M_G,
-                                                         GetReduceCountPerThreadFunctor>;
+                    avg_time += launch_and_time_kernel(
+                        stream_config,
+                        kern_multiblock_batchnorm_fwd_,
+                        dim3(arg.gridSize_),
+                        dim3(BlockSize),
+                        0,
+                        arg.x_grid_desc_m_k_,
+                        arg.y_grid_desc_m_k_,
+                        mean_var_count_grid_desc_m_g, // for writing to mean/variance/count
+                                                      // workspace by multiple workgroups
+                        mean_var_count_grid_desc_m_k, // for reading from mean/variance/count
+                                                      // workspace by each workgroup
+                        arg.scale_grid_desc_m_,
+                        arg.bias_grid_desc_m_,
+                        arg.mean_var_grid_desc_m_,
+                        get_reduce_count_per_thread,
+                        arg.numBlockTileIteration_,
+                        arg.epsilon_,
+                        arg.p_x_,
+                        static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                        static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                        static_cast<int32_t*>(arg.workspace_count_),
+                        static_cast<int*>(arg.control_),
+                        arg.p_scale_,
+                        arg.p_bias_,
+                        arg.y_elementwise_op_,
+                        arg.p_y_,
+                        arg.updateMovingAverage_, // true or false
+                        arg.averageFactor_,
+                        arg.resultRunningMean_,
+                        arg.resultRunningVariance_,
+                        arg.saveMeanInvVariance_, // true or false
+                        arg.resultSaveMean_,
+                        arg.resultSaveInvVariance_);
+                }
+                else
+                {
+                    const auto kern_multiblock_welford_first_half =
+                        kernel_multiblock_welford_first_half<GridwiseMultiblockWelfordFirstHalf_,
+                                                             XDataType,
+                                                             MeanVarDataType,
+                                                             XYGridDesc_M_K,
+                                                             MeanVarCountGridDesc_M_G,
+                                                             GetReduceCountPerThreadFunctor>;
 
-                const auto kern_welford_second_half_batchnorm_forward_final =
-                    kernel_welford_second_half_batchnorm_forward_final<
-                        GridwiseWelfordSecondHalfBatchNormForwardFinal_,
-                        XDataType,
-                        YDataType,
-                        AccDataType,
-                        ScaleDataType,
-                        BiasDataType,
-                        MeanVarDataType,
-                        YElementwiseOp,
-                        XYGridDesc_M_K,
-                        MeanVarCountGridDesc_M_K,
-                        ScaleBiasMeanVarGridDesc_M,
-                        ScaleBiasMeanVarGridDesc_M>;
+                    const auto kern_welford_second_half_batchnorm_forward_final =
+                        kernel_welford_second_half_batchnorm_forward_final<
+                            GridwiseWelfordSecondHalfBatchNormForwardFinal_,
+                            XDataType,
+                            YDataType,
+                            AccDataType,
+                            ScaleDataType,
+                            BiasDataType,
+                            MeanVarDataType,
+                            YElementwiseOp,
+                            XYGridDesc_M_K,
+                            MeanVarCountGridDesc_M_K,
+                            ScaleBiasMeanVarGridDesc_M,
+                            ScaleBiasMeanVarGridDesc_M>;
 
-                avg_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kern_multiblock_welford_first_half,
-                                           dim3(arg.gridSize_),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.x_grid_desc_m_k_,
-                                           mean_var_count_grid_desc_m_g,
-                                           get_reduce_count_per_thread,
-                                           arg.numBlockTileIteration_,
-                                           arg.p_x_,
-                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
-                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
-                                           static_cast<int32_t*>(arg.workspace_count_));
+                    avg_time += launch_and_time_kernel(
+                        stream_config,
+                        kern_multiblock_welford_first_half,
+                        dim3(arg.gridSize_),
+                        dim3(BlockSize),
+                        0,
+                        arg.x_grid_desc_m_k_,
+                        mean_var_count_grid_desc_m_g,
+                        get_reduce_count_per_thread,
+                        arg.numBlockTileIteration_,
+                        arg.p_x_,
+                        static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                        static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                        static_cast<int32_t*>(arg.workspace_count_));
 
-                avg_time +=
-                    launch_and_time_kernel(stream_config,
-                                           kern_welford_second_half_batchnorm_forward_final,
-                                           dim3(arg.gridSize_),
-                                           dim3(BlockSize),
-                                           0,
-                                           arg.x_grid_desc_m_k_,
-                                           arg.y_grid_desc_m_k_,
-                                           mean_var_count_grid_desc_m_k,
-                                           arg.scale_grid_desc_m_,
-                                           arg.bias_grid_desc_m_,
-                                           arg.mean_var_grid_desc_m_,
-                                           arg.blkGroupSize_,
-                                           arg.numBlockTileIteration_,
-                                           numMeanVarCountBlockTileIteration,
-                                           arg.epsilon_,
-                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
-                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
-                                           static_cast<int32_t*>(arg.workspace_count_),
-                                           arg.p_x_,
-                                           arg.p_scale_,
-                                           arg.p_bias_,
-                                           arg.y_elementwise_op_,
-                                           arg.p_y_,
-                                           arg.updateMovingAverage_,
-                                           arg.averageFactor_,
-                                           arg.resultRunningMean_,
-                                           arg.resultRunningVariance_,
-                                           arg.saveMeanInvVariance_,
-                                           arg.resultSaveMean_,
-                                           arg.resultSaveInvVariance_);
+                    avg_time += launch_and_time_kernel(
+                        stream_config,
+                        kern_welford_second_half_batchnorm_forward_final,
+                        dim3(arg.gridSize_),
+                        dim3(BlockSize),
+                        0,
+                        arg.x_grid_desc_m_k_,
+                        arg.y_grid_desc_m_k_,
+                        mean_var_count_grid_desc_m_k,
+                        arg.scale_grid_desc_m_,
+                        arg.bias_grid_desc_m_,
+                        arg.mean_var_grid_desc_m_,
+                        arg.blkGroupSize_,
+                        arg.numBlockTileIteration_,
+                        arg.epsilon_,
+                        static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                        static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                        static_cast<int32_t*>(arg.workspace_count_),
+                        arg.p_x_,
+                        arg.p_scale_,
+                        arg.p_bias_,
+                        arg.y_elementwise_op_,
+                        arg.p_y_,
+                        arg.updateMovingAverage_,
+                        arg.averageFactor_,
+                        arg.resultRunningMean_,
+                        arg.resultRunningVariance_,
+                        arg.saveMeanInvVariance_,
+                        arg.resultSaveMean_,
+                        arg.resultSaveInvVariance_);
+                };
             }
             else
             {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp
new file mode 100644
index 0000000000..b826793c27
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batchnorm_forward_impl_obsolete.hpp
@@ -0,0 +1,714 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/reduction_operator.hpp"
+#include "ck/tensor_operation/gpu/device/device_batchnorm_forward.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
+#include "ck/tensor_operation/gpu/device/welford_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp"
+#include "ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_batchnorm_forward_blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          typename YElementwiseOp,
+          index_t Rank,
+          index_t NumBatchNormReduceDim,
+          bool UseMultiblockInK,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcYDstVectorDim,
+          index_t XSrcVectorSize,
+          index_t YDstVectorSize,
+          index_t ScaleSrcVectorSize,
+          index_t BiasSrcVectorSize,
+          index_t MeanVarSrcDstVectorSize>
+struct DeviceBatchNormFwdImpl : public DeviceBatchNormFwd<XDataType,
+                                                          YDataType,
+                                                          AccDataType,
+                                                          ScaleDataType,
+                                                          BiasDataType,
+                                                          MeanVarDataType,
+                                                          YElementwiseOp,
+                                                          Rank,
+                                                          NumBatchNormReduceDim>
+{
+    static_assert(Rank <= 6, "Bigger Rank size is not supported!");
+    static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize,
+                  "Invalid thread cluster size assignments!");
+
+    static_assert((XSrcYDstVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcYDstVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static constexpr index_t NumInvariantDim = Rank - NumBatchNormReduceDim;
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto MakeXY2dDescriptor(const std::array<index_t, Rank>& xyLengths,
+                                   const std::array<index_t, Rank>& xyStrides,
+                                   int blkGroupSize,
+                                   int numBlockTileIteration)
+    {
+        const auto tupleXYLengths =
+            generate_tuple([&](auto I) { return xyLengths[I]; }, Number<Rank>{});
+        const auto tupleXYStrides =
+            generate_tuple([&](auto I) { return xyStrides[I]; }, Number<Rank>{});
+
+        const auto raw_grid_desc = make_naive_tensor_descriptor(tupleXYLengths, tupleXYStrides);
+
+        const auto grid_desc_m_k = [&]() {
+            using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+            using ReduceDims    = typename arithmetic_sequence_gen<NumInvariantDim, Rank, 1>::type;
+
+            const auto reduceDimLengths =
+                generate_tuple([&](auto I) { return xyLengths[NumInvariantDim + I]; },
+                               Number<NumBatchNormReduceDim>{});
+            const auto invariantDimLengths =
+                generate_tuple([&](auto I) { return xyLengths[I]; }, Number<NumInvariantDim>{});
+
+            return transform_tensor_descriptor(raw_grid_desc,
+                                               make_tuple(make_merge_transform(invariantDimLengths),
+                                                          make_merge_transform(reduceDimLengths)),
+                                               make_tuple(InvariantDims{}, ReduceDims{}),
+                                               make_tuple(Sequence<0>{}, Sequence<1>{}));
+        }();
+
+        const auto invariantLength = grid_desc_m_k.GetLength(Number<0>{});
+        const auto reduceLength    = grid_desc_m_k.GetLength(Number<1>{});
+
+        const int workSizePerBlock = K_BlockTileSize * numBlockTileIteration;
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto kPad = workSizePerBlock * blkGroupSize - reduceLength;
+
+        auto grid_desc_m_k_padded =
+            transform_tensor_descriptor(grid_desc_m_k,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad),
+                                                   make_right_pad_transform(reduceLength, kPad)),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (grid_desc_m_k_padded);
+    };
+
+    static auto MakeMeanVarCountOutputMG2dDescriptor(int invariantLength, int blkGroupSize)
+    {
+        const auto grid_desc_m_g = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, blkGroupSize), make_tuple(1, invariantLength));
+
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto grid_desc_m_g_padded =
+            transform_tensor_descriptor(grid_desc_m_g,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad),
+                                                   make_pass_through_transform(blkGroupSize)),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (grid_desc_m_g_padded);
+    };
+
+    static auto MakeMeanVarCountInputMK2dDescriptor(int invariantLength, int blkGroupSize)
+    {
+        const auto reduceLength  = blkGroupSize;
+        const auto grid_desc_m_k = make_naive_tensor_descriptor(
+            make_tuple(invariantLength, reduceLength), make_tuple(1, invariantLength));
+
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+        const auto kPad =
+            math::integer_least_multiple(reduceLength, KThreadClusterSize) - reduceLength;
+
+        auto grid_desc_m_k_padded =
+            transform_tensor_descriptor(grid_desc_m_k,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad),
+                                                   make_right_pad_transform(reduceLength, kPad)),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return (grid_desc_m_k_padded);
+    };
+
+    static auto
+    MakeScaleBiasMeanVar1dDescriptor(const std::array<index_t, NumInvariantDim>& lengths,
+                                     const std::array<index_t, NumInvariantDim>& strides)
+    {
+        const auto tupleLengths =
+            generate_tuple([&](auto I) { return lengths[I]; }, Number<NumInvariantDim>{});
+        const auto tupleStrides =
+            generate_tuple([&](auto I) { return strides[I]; }, Number<NumInvariantDim>{});
+
+        auto raw_grid_desc = make_naive_tensor_descriptor(tupleLengths, tupleStrides);
+
+        auto grid_desc_m = transform_tensor_descriptor(
+            raw_grid_desc,
+            make_tuple(make_merge_transform(tupleLengths)),
+            make_tuple(typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type{}),
+            make_tuple(Sequence<0>{}));
+
+        const auto invariantLength = grid_desc_m.GetLength(Number<0>{});
+
+        const auto mPad =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto grid_desc_m_padded =
+            transform_tensor_descriptor(grid_desc_m,
+                                        make_tuple(make_right_pad_transform(invariantLength, mPad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+        return (grid_desc_m_padded);
+    };
+
+    using XYGridDesc_M_K             = decltype(MakeXY2dDescriptor({1}, {1}, 1, 1));
+    using ScaleBiasMeanVarGridDesc_M = decltype(MakeScaleBiasMeanVar1dDescriptor({1}, {1}));
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const std::array<index_t, Rank> xyLengths,
+                 const std::array<index_t, Rank> xStrides,
+                 const std::array<index_t, Rank> yStrides,
+                 const std::array<int, NumBatchNormReduceDim> reduceDims,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleBiasMeanVarLengths,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleStrides,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnBiasStrides,
+                 const std::array<index_t, Rank - NumBatchNormReduceDim> bnMeanVarStrides,
+                 const XDataType* p_x,
+                 const ScaleDataType* p_scale,
+                 const BiasDataType* p_bias,
+                 const YElementwiseOp y_elementwise_op,
+                 double epsilon,
+                 YDataType* p_y,
+                 MeanVarDataType* resultSaveMean,
+                 MeanVarDataType* resultSaveInvVariance,
+                 double averageFactor,
+                 MeanVarDataType* resultRunningMean,
+                 MeanVarDataType* resultRunningVariance)
+            : bnScaleBiasMeanVarLengths_(bnScaleBiasMeanVarLengths),
+              bnScaleStrides_(bnScaleStrides),
+              bnBiasStrides_(bnBiasStrides),
+              bnMeanVarStrides_(bnMeanVarStrides),
+              p_x_(p_x),
+              p_scale_(p_scale),
+              p_bias_(p_bias),
+              y_elementwise_op_(y_elementwise_op),
+              p_y_(p_y),
+              resultSaveMean_(resultSaveMean),
+              resultSaveInvVariance_(resultSaveInvVariance),
+              resultRunningMean_(resultRunningMean),
+              resultRunningVariance_(resultRunningVariance)
+        {
+            xyLengths_ =
+                shuffle_tensor_dimensions<Rank, NumBatchNormReduceDim>(xyLengths, reduceDims);
+            xStrides_ =
+                shuffle_tensor_dimensions<Rank, NumBatchNormReduceDim>(xStrides, reduceDims);
+            yStrides_ =
+                shuffle_tensor_dimensions<Rank, NumBatchNormReduceDim>(yStrides, reduceDims);
+
+            std::tie(invariant_length_, reduce_length_) =
+                get_2d_lengths<Rank, NumBatchNormReduceDim>(xyLengths_);
+
+            epsilon_       = type_convert<AccDataType>(epsilon);
+            averageFactor_ = type_convert<AccDataType>(averageFactor);
+
+            updateMovingAverage_ =
+                (resultRunningMean != nullptr && resultRunningVariance != nullptr);
+            saveMeanInvVariance_ = (resultSaveMean != nullptr && resultSaveInvVariance_ != nullptr);
+
+            if(UseMultiblockInK)
+            {
+                int iterations = 1;
+                while(true)
+                {
+                    int testBlkGroupSize = (reduce_length_ + (K_BlockTileSize * iterations) - 1) /
+                                           (K_BlockTileSize * iterations);
+
+                    // we want the blkGroupSize be not more than 16
+                    if(testBlkGroupSize <= 16)
+                        break;
+
+                    iterations++;
+                };
+
+                blkGroupSize_ = (reduce_length_ + (K_BlockTileSize * iterations) - 1) /
+                                (K_BlockTileSize * iterations);
+
+                numBlockTileIteration_ = iterations;
+            }
+            else
+            {
+                blkGroupSize_          = 1;
+                numBlockTileIteration_ = (reduce_length_ + K_BlockTileSize - 1) / K_BlockTileSize;
+            };
+
+            gridSize_ = (invariant_length_ + M_BlockTileSize - 1) / M_BlockTileSize * blkGroupSize_;
+
+            x_grid_desc_m_k_ =
+                MakeXY2dDescriptor(xyLengths_, xStrides_, blkGroupSize_, numBlockTileIteration_);
+            y_grid_desc_m_k_ =
+                MakeXY2dDescriptor(xyLengths_, yStrides_, blkGroupSize_, numBlockTileIteration_);
+            scale_grid_desc_m_ =
+                MakeScaleBiasMeanVar1dDescriptor(bnScaleBiasMeanVarLengths, bnScaleStrides_);
+            bias_grid_desc_m_ =
+                MakeScaleBiasMeanVar1dDescriptor(bnScaleBiasMeanVarLengths, bnBiasStrides_);
+            mean_var_grid_desc_m_ =
+                MakeScaleBiasMeanVar1dDescriptor(bnScaleBiasMeanVarLengths, bnMeanVarStrides_);
+        }
+
+        AccDataType epsilon_;
+        AccDataType averageFactor_;
+
+        bool updateMovingAverage_;
+        bool saveMeanInvVariance_;
+
+        std::array<index_t, Rank> xyLengths_;
+        std::array<index_t, Rank> xStrides_;
+        std::array<index_t, Rank> yStrides_;
+
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleBiasMeanVarLengths_;
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleStrides_;
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnBiasStrides_;
+        std::array<index_t, Rank - NumBatchNormReduceDim> bnMeanVarStrides_;
+
+        const XDataType* p_x_;
+        const ScaleDataType* p_scale_;
+        const BiasDataType* p_bias_;
+        const YElementwiseOp y_elementwise_op_;
+        YDataType* p_y_;
+
+        MeanVarDataType* resultSaveMean_;
+        MeanVarDataType* resultSaveInvVariance_;
+
+        MeanVarDataType* resultRunningMean_;
+        MeanVarDataType* resultRunningVariance_;
+
+        long_index_t invariant_length_;
+        long_index_t reduce_length_;
+
+        int blkGroupSize_;
+        int numBlockTileIteration_;
+        size_t gridSize_;
+
+        XYGridDesc_M_K x_grid_desc_m_k_;
+        XYGridDesc_M_K y_grid_desc_m_k_;
+        ScaleBiasMeanVarGridDesc_M scale_grid_desc_m_;
+        ScaleBiasMeanVarGridDesc_M bias_grid_desc_m_;
+        ScaleBiasMeanVarGridDesc_M mean_var_grid_desc_m_;
+
+        void* workspace_mean_;
+        void* workspace_variance_;
+        void* workspace_count_;
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        size_t workspace_size = 0;
+
+        if(UseMultiblockInK && pArg_->blkGroupSize_ > 1)
+        {
+            // workspace for welford intermediate mean
+            workspace_size +=
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType) + 64;
+
+            // workspace for welford intermediate variance
+            workspace_size +=
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType) + 64;
+
+            // workspace for welford intermediate count
+            workspace_size +=
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(int32_t) + 64;
+        }
+
+        return (workspace_size);
+    };
+
+    void SetWorkSpacePointer(BaseArgument* pArg, void* p_workspace) const override
+    {
+        Argument* pArg_ = dynamic_cast<Argument*>(pArg);
+
+        pArg_->p_workspace_ = p_workspace;
+
+        if(UseMultiblockInK && pArg_->blkGroupSize_ > 1)
+        {
+
+            // setup buffer used for intermediate welford mean
+            pArg_->workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
+
+            index_t mean_space_sz =
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType);
+
+            mean_space_sz = math::integer_least_multiple(mean_space_sz, 64);
+
+            // setup buffer used for intermediate welford varirance
+            pArg_->workspace_variance_ =
+                reinterpret_cast<char*>(pArg_->workspace_mean_) + mean_space_sz;
+
+            index_t variance_space_sz =
+                pArg_->invariant_length_ * pArg_->blkGroupSize_ * sizeof(MeanVarDataType);
+
+            variance_space_sz = math::integer_least_multiple(variance_space_sz, 64);
+
+            // setup buffer used for intermediate welfor count
+            pArg_->workspace_count_ =
+                reinterpret_cast<char*>(pArg_->workspace_variance_) + variance_space_sz;
+        };
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            float avg_time = 0;
+
+            if(UseMultiblockInK && arg.blkGroupSize_ > 1)
+            {
+                using GetReduceCountPerThreadFunctor =
+                    GetReduceCountPerThreadForMultiblockWelford<K_BlockTileSize, KThreadSliceSize>;
+
+                GetReduceCountPerThreadFunctor get_reduce_count_per_thread(
+                    arg.blkGroupSize_, arg.numBlockTileIteration_, arg.reduce_length_);
+
+                const auto mean_var_count_grid_desc_m_g =
+                    DeviceBatchNormFwdImpl::MakeMeanVarCountOutputMG2dDescriptor(
+                        arg.invariant_length_, arg.blkGroupSize_);
+
+                const auto mean_var_count_grid_desc_m_k =
+                    DeviceBatchNormFwdImpl::MakeMeanVarCountInputMK2dDescriptor(
+                        arg.invariant_length_, arg.blkGroupSize_);
+
+                using MeanVarCountGridDesc_M_G = decltype(mean_var_count_grid_desc_m_g);
+                using MeanVarCountGridDesc_M_K = decltype(mean_var_count_grid_desc_m_k);
+
+                using GridwiseMultiblockWelfordFirstHalf_ =
+                    GridwiseMultiblockWelfordFirstHalf<XDataType,
+                                                       AccDataType,
+                                                       MeanVarDataType,
+                                                       XYGridDesc_M_K,
+                                                       MeanVarCountGridDesc_M_G,
+                                                       GetReduceCountPerThreadFunctor,
+                                                       BlockSize,
+                                                       MThreadClusterSize,
+                                                       KThreadClusterSize,
+                                                       MThreadSliceSize,
+                                                       KThreadSliceSize,
+                                                       XSrcYDstVectorDim,
+                                                       XSrcVectorSize>;
+
+                using GridwiseWelfordSecondHalfBatchNormForwardFinal_ =
+                    GridwiseWelfordSecondHalfBatchNormForwardFinal<XDataType,
+                                                                   YDataType,
+                                                                   AccDataType,
+                                                                   ScaleDataType,
+                                                                   BiasDataType,
+                                                                   MeanVarDataType,
+                                                                   YElementwiseOp,
+                                                                   XYGridDesc_M_K,
+                                                                   MeanVarCountGridDesc_M_K,
+                                                                   ScaleBiasMeanVarGridDesc_M,
+                                                                   ScaleBiasMeanVarGridDesc_M,
+                                                                   BlockSize,
+                                                                   MThreadClusterSize,
+                                                                   KThreadClusterSize,
+                                                                   MThreadSliceSize,
+                                                                   KThreadSliceSize,
+                                                                   XSrcYDstVectorDim,
+                                                                   XSrcVectorSize,
+                                                                   YDstVectorSize,
+                                                                   ScaleSrcVectorSize,
+                                                                   BiasSrcVectorSize,
+                                                                   MeanVarSrcDstVectorSize>;
+
+                const auto kern_multiblock_welford_first_half =
+                    kernel_multiblock_welford_first_half<GridwiseMultiblockWelfordFirstHalf_,
+                                                         XDataType,
+                                                         MeanVarDataType,
+                                                         XYGridDesc_M_K,
+                                                         MeanVarCountGridDesc_M_G,
+                                                         GetReduceCountPerThreadFunctor>;
+
+                const auto kern_welford_second_half_batchnorm_forward_final =
+                    kernel_welford_second_half_batchnorm_forward_final<
+                        GridwiseWelfordSecondHalfBatchNormForwardFinal_,
+                        XDataType,
+                        YDataType,
+                        AccDataType,
+                        ScaleDataType,
+                        BiasDataType,
+                        MeanVarDataType,
+                        YElementwiseOp,
+                        XYGridDesc_M_K,
+                        MeanVarCountGridDesc_M_K,
+                        ScaleBiasMeanVarGridDesc_M,
+                        ScaleBiasMeanVarGridDesc_M>;
+
+                avg_time +=
+                    launch_and_time_kernel(stream_config,
+                                           kern_multiblock_welford_first_half,
+                                           dim3(arg.gridSize_),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.x_grid_desc_m_k_,
+                                           mean_var_count_grid_desc_m_g,
+                                           get_reduce_count_per_thread,
+                                           arg.numBlockTileIteration_,
+                                           arg.p_x_,
+                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                                           static_cast<int32_t*>(arg.workspace_count_));
+
+                avg_time +=
+                    launch_and_time_kernel(stream_config,
+                                           kern_welford_second_half_batchnorm_forward_final,
+                                           dim3(arg.gridSize_),
+                                           dim3(BlockSize),
+                                           0,
+                                           arg.x_grid_desc_m_k_,
+                                           arg.y_grid_desc_m_k_,
+                                           mean_var_count_grid_desc_m_k,
+                                           arg.scale_grid_desc_m_,
+                                           arg.bias_grid_desc_m_,
+                                           arg.mean_var_grid_desc_m_,
+                                           arg.blkGroupSize_,
+                                           arg.numBlockTileIteration_,
+                                           arg.epsilon_,
+                                           static_cast<MeanVarDataType*>(arg.workspace_mean_),
+                                           static_cast<MeanVarDataType*>(arg.workspace_variance_),
+                                           static_cast<int32_t*>(arg.workspace_count_),
+                                           arg.p_x_,
+                                           arg.p_scale_,
+                                           arg.p_bias_,
+                                           arg.y_elementwise_op_,
+                                           arg.p_y_,
+                                           arg.updateMovingAverage_,
+                                           arg.averageFactor_,
+                                           arg.resultRunningMean_,
+                                           arg.resultRunningVariance_,
+                                           arg.saveMeanInvVariance_,
+                                           arg.resultSaveMean_,
+                                           arg.resultSaveInvVariance_);
+            }
+            else
+            {
+                using GetReduceCountPerThreadFunctor =
+                    GetReduceCountPerThreadForBlockwiseWelford<K_BlockTileSize, KThreadSliceSize>;
+
+                GetReduceCountPerThreadFunctor get_reduce_count_per_thread(
+                    arg.numBlockTileIteration_, arg.reduce_length_);
+
+                using GridwiseBatchNormForwardWithBlockwiseWelford_ =
+                    GridwiseBatchNormForwardWithBlockwiseWelford<XDataType,
+                                                                 YDataType,
+                                                                 AccDataType,
+                                                                 ScaleDataType,
+                                                                 BiasDataType,
+                                                                 MeanVarDataType,
+                                                                 YElementwiseOp,
+                                                                 XYGridDesc_M_K,
+                                                                 ScaleBiasMeanVarGridDesc_M,
+                                                                 ScaleBiasMeanVarGridDesc_M,
+                                                                 GetReduceCountPerThreadFunctor,
+                                                                 BlockSize,
+                                                                 MThreadClusterSize,
+                                                                 KThreadClusterSize,
+                                                                 MThreadSliceSize,
+                                                                 KThreadSliceSize,
+                                                                 XSrcYDstVectorDim,
+                                                                 XSrcVectorSize,
+                                                                 YDstVectorSize,
+                                                                 ScaleSrcVectorSize,
+                                                                 BiasSrcVectorSize,
+                                                                 MeanVarSrcDstVectorSize>;
+
+                const auto kern_batchnorm_fwd = kernel_batchnorm_forward_with_blockwise_welford<
+                    GridwiseBatchNormForwardWithBlockwiseWelford_,
+                    XDataType,
+                    YDataType,
+                    AccDataType,
+                    ScaleDataType,
+                    BiasDataType,
+                    MeanVarDataType,
+                    YElementwiseOp,
+                    XYGridDesc_M_K,
+                    ScaleBiasMeanVarGridDesc_M,
+                    ScaleBiasMeanVarGridDesc_M,
+                    GetReduceCountPerThreadFunctor>;
+
+                avg_time += launch_and_time_kernel(stream_config,
+                                                   kern_batchnorm_fwd,
+                                                   dim3(arg.gridSize_),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.x_grid_desc_m_k_,
+                                                   arg.y_grid_desc_m_k_,
+                                                   arg.scale_grid_desc_m_,
+                                                   arg.bias_grid_desc_m_,
+                                                   arg.mean_var_grid_desc_m_,
+                                                   get_reduce_count_per_thread,
+                                                   arg.numBlockTileIteration_,
+                                                   arg.epsilon_,
+                                                   arg.p_x_,
+                                                   arg.p_scale_,
+                                                   arg.p_bias_,
+                                                   arg.y_elementwise_op_,
+                                                   arg.p_y_,
+                                                   arg.updateMovingAverage_, // true or false
+                                                   arg.averageFactor_,
+                                                   arg.resultRunningMean_,
+                                                   arg.resultRunningVariance_,
+                                                   arg.saveMeanInvVariance_, // true or false
+                                                   arg.resultSaveMean_,
+                                                   arg.resultSaveInvVariance_);
+            };
+
+            return (avg_time);
+        };
+
+        float Run(const BaseArgument* pArg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(pArg), stream_config);
+        };
+    };
+
+    bool IsSupportedArgument(const BaseArgument* pArg) override
+    {
+        const Argument* pArg_ = dynamic_cast<const Argument*>(pArg);
+
+        if constexpr(XSrcYDstVectorDim == 0)
+        {
+            if(pArg_->xStrides_[NumInvariantDim - 1] != 1 ||
+               pArg_->yStrides_[NumInvariantDim - 1] != 1)
+                return false;
+
+            if(pArg_->xyLengths_[NumInvariantDim - 1] % XSrcVectorSize != 0 ||
+               pArg_->xyLengths_[NumInvariantDim - 1] % YDstVectorSize != 0)
+                return false;
+        }
+        else
+        {
+            if(pArg_->xStrides_[Rank - 1] != 1 || pArg_->yStrides_[Rank - 1] != 1)
+                return false;
+
+            if(pArg_->xyLengths_[Rank - 1] % XSrcVectorSize != 0 ||
+               pArg_->xyLengths_[Rank - 1] % YDstVectorSize != 0)
+                return false;
+        };
+
+        if(pArg_->bnScaleStrides_[NumInvariantDim - 1] != 1 && ScaleSrcVectorSize != 1)
+            return false;
+        if(pArg_->bnBiasStrides_[NumInvariantDim - 1] != 1 && BiasSrcVectorSize != 1)
+            return false;
+
+        if(pArg_->bnScaleBiasMeanVarLengths_[NumInvariantDim - 1] % ScaleSrcVectorSize != 0)
+            return false;
+        if(pArg_->bnScaleBiasMeanVarLengths_[NumInvariantDim - 1] % BiasSrcVectorSize != 0)
+            return false;
+
+        if(pArg_->bnMeanVarStrides_[NumInvariantDim - 1] != 1 && MeanVarSrcDstVectorSize != 1)
+            return false;
+
+        if(pArg_->bnScaleBiasMeanVarLengths_[NumInvariantDim - 1] % MeanVarSrcDstVectorSize != 0)
+            return false;
+
+        bool is_valid = true;
+
+        static_for<0, NumInvariantDim, 1>{}([&](auto I) {
+            if(pArg_->xyLengths_[I] != pArg_->bnScaleBiasMeanVarLengths_[I])
+                is_valid = false;
+        });
+
+        if(!is_valid)
+            return false;
+
+        return true;
+    };
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(
+        const std::array<index_t, Rank> xyLengths,
+        const std::array<index_t, Rank> xStrides,
+        const std::array<index_t, Rank> yStrides,
+        const std::array<int, NumBatchNormReduceDim> reduceDims,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleBiasMeanVarLengths,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnScaleStrides,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnBiasStrides,
+        const std::array<index_t, Rank - NumBatchNormReduceDim> bnMeanVarStrides,
+        const void* p_x,
+        const void* p_scale,
+        const void* p_bias,
+        double epsilon,
+        const YElementwiseOp y_elementwise_op,
+        void* p_y,
+        void* resultSaveMean,
+        void* resultSaveInvVariance,
+        double averageFactor,
+        void* resultRunningMean,
+        void* resultRunningVariance) override
+    {
+        return std::make_unique<Argument>(xyLengths,
+                                          xStrides,
+                                          yStrides,
+                                          reduceDims,
+                                          bnScaleBiasMeanVarLengths,
+                                          bnScaleStrides,
+                                          bnBiasStrides,
+                                          bnMeanVarStrides,
+                                          static_cast<const XDataType*>(p_x),
+                                          static_cast<const ScaleDataType*>(p_scale),
+                                          static_cast<const BiasDataType*>(p_bias),
+                                          y_elementwise_op,
+                                          epsilon,
+                                          static_cast<YDataType*>(p_y),
+                                          static_cast<MeanVarDataType*>(resultSaveMean),
+                                          static_cast<MeanVarDataType*>(resultSaveInvVariance),
+                                          averageFactor,
+                                          static_cast<MeanVarDataType*>(resultRunningMean),
+                                          static_cast<MeanVarDataType*>(resultRunningVariance));
+    };
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>();
+    };
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceBatchNormFwdImpl<" << BlockSize << ",";
+        str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
+        str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
+        str << "XSrcYDstVectorDim_" << XSrcYDstVectorDim  << ",";
+        str << "VectorSize_X" << XSrcVectorSize << "_scale_" << ScaleSrcVectorSize << "_bias_" << BiasSrcVectorSize << "_mean_var_" << MeanVarSrcDstVectorSize << "_Y" << YDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
new file mode 100644
index 0000000000..f73fe282db
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
@@ -0,0 +1,704 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/math_v2.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_welford.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/utility/workgroup_synchronization.hpp"
+
+namespace ck {
+
+template <typename GridwiseMultiblockBatchNormForward_,
+          typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          typename YElementwiseOp,
+          typename XYGridDesc_M_K,
+          typename MeanVarCountGridDesc_M_G,
+          typename MeanVarCountGridDesc_M_K,
+          typename ScaleBiasGridDesc_M,
+          typename MeanVarGridDesc_M,
+          typename GetReduceCountPerThreadFunctor>
+__global__ void kernel_multiblock_batchnorm_forward(
+    const XYGridDesc_M_K x_grid_desc_m_k,
+    const XYGridDesc_M_K y_grid_desc_m_k,
+    const MeanVarCountGridDesc_M_G mean_var_count_grid_desc_m_g,
+    const MeanVarCountGridDesc_M_K mean_var_count_grid_desc_m_k,
+    const ScaleBiasGridDesc_M scale_grid_desc_m,
+    const ScaleBiasGridDesc_M bias_grid_desc_m,
+    const MeanVarGridDesc_M mean_var_grid_desc_m,
+    const GetReduceCountPerThreadFunctor get_reduce_count_per_thread,
+    index_t num_k_block_tile_iteration,
+    AccDataType epsilon,
+    const XDataType* const __restrict__ p_x,
+    MeanVarDataType* const __restrict__ p_welford_mean,
+    MeanVarDataType* const __restrict__ p_welford_variance,
+    int32_t* const __restrict__ p_welford_count,
+    int32_t* const __restrict__ p_control,
+    const ScaleDataType* const __restrict__ p_scale,
+    const BiasDataType* const __restrict__ p_bias,
+    const YElementwiseOp y_elementwise_op,
+    YDataType* const __restrict__ p_y,
+    bool updateMovingAverage,
+    AccDataType averageFactor,
+    MeanVarDataType* const __restrict__ resultRunningMean,
+    MeanVarDataType* const __restrict__ resultRunningVariance,
+    bool saveMeanInvVariance,
+    MeanVarDataType* const __restrict__ resultSaveMean,
+    MeanVarDataType* const __restrict__ resultSaveInvVariance)
+{
+    GridwiseMultiblockBatchNormForward_::Run(x_grid_desc_m_k,
+                                             y_grid_desc_m_k,
+                                             mean_var_count_grid_desc_m_g,
+                                             mean_var_count_grid_desc_m_k,
+                                             scale_grid_desc_m,
+                                             bias_grid_desc_m,
+                                             mean_var_grid_desc_m,
+                                             get_reduce_count_per_thread,
+                                             num_k_block_tile_iteration,
+                                             epsilon,
+                                             p_x,
+                                             p_welford_mean,
+                                             p_welford_variance,
+                                             p_welford_count,
+                                             p_control,
+                                             p_scale,
+                                             p_bias,
+                                             y_elementwise_op,
+                                             p_y,
+                                             updateMovingAverage,
+                                             averageFactor,
+                                             resultRunningMean,
+                                             resultRunningVariance,
+                                             saveMeanInvVariance,
+                                             resultSaveMean,
+                                             resultSaveInvVariance);
+};
+
+template <typename XDataType,
+          typename YDataType,
+          typename AccDataType,
+          typename ScaleDataType,
+          typename BiasDataType,
+          typename MeanVarDataType,
+          typename YElementwiseOp,
+          typename XYGridDesc_M_K,
+          typename MeanVarCountGridDesc_M_G,
+          typename MeanVarCountGridDesc_M_K,
+          typename ScaleBiasGridDesc_M,
+          typename MeanVarGridDesc_M,
+          typename GetReduceCountPerThreadFunctor,
+          index_t BlockSize,
+          index_t MThreadClusterSize,
+          index_t KThreadClusterSize,
+          index_t MThreadSliceSize,
+          index_t KThreadSliceSize,
+          index_t XSrcYDstVectorDim,
+          index_t XSrcVectorSize,
+          index_t YDstVectorSize,
+          index_t ScaleSrcVectorSize,
+          index_t BiasSrcVectorSize,
+          index_t MeanVarSrcDstVectorSize>
+struct GridwiseMultiblockBatchNormForward
+{
+    static_assert((XSrcYDstVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
+                      (XSrcYDstVectorDim == 1 && KThreadSliceSize % XSrcVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static_assert((XSrcYDstVectorDim == 0 && MThreadSliceSize % YDstVectorSize == 0) ||
+                      (XSrcYDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
+                  "Invalid thread slice sizes and/or vector sizes configuration, please check!");
+
+    static constexpr bool reorder_thread_cluster = (XSrcYDstVectorDim == 0);
+
+    using ThreadClusterLengths_M_K = Sequence<MThreadClusterSize, KThreadClusterSize>;
+
+    using ThreadBufferDimAccessOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    using ThreadClusterArrangeOrder =
+        typename conditional<reorder_thread_cluster, Sequence<1, 0>, Sequence<0, 1>>::type;
+
+    static constexpr auto thread_cluster_desc =
+        make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
+
+    using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{})));
+    using ThreadReduceDstDesc_M =
+        decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
+
+    using ThreadReduceSrcDesc_M_1 = decltype(
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
+
+    using ThreadwiseWelford1 =
+        ThreadwiseWelford<AccDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
+
+    using ThreadwiseWelford2 =
+        ThreadwiseWelfordMerge<AccDataType, ThreadReduceSrcDesc_M_1, ThreadReduceDstDesc_M>;
+
+    using BlockwiseWelford1 = BlockwiseWelford<AccDataType,
+                                               BlockSize,
+                                               ThreadClusterLengths_M_K,
+                                               ThreadClusterArrangeOrder,
+                                               false>;
+
+    using BlockwiseWelford2 = BlockwiseWelford<AccDataType,
+                                               BlockSize,
+                                               ThreadClusterLengths_M_K,
+                                               ThreadClusterArrangeOrder,
+                                               true>;
+
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    __device__ static void Run(const XYGridDesc_M_K& x_grid_desc_m_k,
+                               const XYGridDesc_M_K& y_grid_desc_m_k,
+                               const MeanVarCountGridDesc_M_G& mean_var_count_grid_desc_m_g,
+                               const MeanVarCountGridDesc_M_K& mean_var_count_grid_desc_m_k,
+                               const ScaleBiasGridDesc_M& scale_grid_desc_m,
+                               const ScaleBiasGridDesc_M& bias_grid_desc_m,
+                               const MeanVarGridDesc_M& mean_var_grid_desc_m,
+                               const GetReduceCountPerThreadFunctor& get_reduce_count_per_thread,
+                               index_t num_k_block_tile_iteration,
+                               AccDataType epsilon,
+                               const XDataType* const __restrict__ p_x,
+                               MeanVarDataType* const __restrict__ p_welford_mean,
+                               MeanVarDataType* const __restrict__ p_welford_variance,
+                               int32_t* const __restrict__ p_welford_count,
+                               int32_t* const __restrict__ p_control,
+                               const ScaleDataType* const __restrict__ p_scale,
+                               const BiasDataType* const __restrict__ p_bias,
+                               const YElementwiseOp y_elementwise_op,
+                               YDataType* const __restrict__ p_y,
+                               bool updateMovingAverage,
+                               AccDataType averageFactor,
+                               MeanVarDataType* const __restrict__ resultRunningMean,
+                               MeanVarDataType* const __restrict__ resultRunningVariance,
+                               bool saveMeanInvVariance,
+                               MeanVarDataType* const __restrict__ resultSaveMean,
+                               MeanVarDataType* const __restrict__ resultSaveInvVariance)
+    {
+        using ck::math::sqrt;
+
+        const index_t blkgroup_size = mean_var_count_grid_desc_m_g.GetLength(I1);
+
+        const index_t thread_local_id = get_thread_local_1d_id();
+        const index_t block_global_id = get_block_1d_id();
+        const index_t blkgroup_id     = block_global_id / blkgroup_size;
+        const index_t block_local_id  = block_global_id % blkgroup_size;
+
+        if(block_local_id == 0)
+            gms_init(BlockSize / warpSize * blkgroup_size, &p_control[blkgroup_id * 2]);
+
+        const auto thread_cluster_idx =
+            thread_cluster_desc.CalculateBottomIndex(make_multi_index(thread_local_id));
+
+        const auto thread_m_cluster_id = thread_cluster_idx[I0];
+        const auto thread_k_cluster_id = thread_cluster_idx[I1];
+
+        using ThreadBufferLengths_M_K = Sequence<MThreadSliceSize, KThreadSliceSize>;
+        using ThreadBufferLengths_M   = Sequence<MThreadSliceSize>;
+        using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
+
+        constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
+            make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{}));
+        constexpr auto thread_buffer_desc_m =
+            make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}));
+        constexpr auto thread_buffer_desc_m_1 = make_naive_tensor_descriptor_packed(
+            make_tuple(Number<MThreadSliceSize>{}, Number<1>{}));
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
+            x_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true> count_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+            tmp_mean_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+            tmp_var_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true> tmp_count_thread_buf;
+
+        const index_t reduceSizePerBlock = K_BlockTileSize * num_k_block_tile_iteration;
+
+        auto threadwise_x_load = ThreadwiseTensorSliceTransfer_v2<XDataType,
+                                                                  AccDataType,
+                                                                  XYGridDesc_M_K,
+                                                                  decltype(thread_buffer_desc_m_k),
+                                                                  ThreadBufferLengths_M_K,
+                                                                  ThreadBufferDimAccessOrder,
+                                                                  XSrcYDstVectorDim,
+                                                                  XSrcVectorSize,
+                                                                  1,
+                                                                  true>(
+            x_grid_desc_m_k,
+            make_multi_index(blkgroup_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                             block_local_id * reduceSizePerBlock +
+                                 thread_k_cluster_id * KThreadSliceSize));
+
+        constexpr auto xy_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileSize);
+
+        const auto x_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_x, x_grid_desc_m_k.GetElementSpaceSize());
+
+        // Step 1: each workgroup does local welford reduction
+
+        auto threadwise_welford_1 = ThreadwiseWelford1();
+        threadwise_welford_1.max_count_ =
+            get_reduce_count_per_thread(block_local_id, thread_k_cluster_id);
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I) = type_convert<AccDataType>(0.0f);
+            var_thread_buf(I)  = type_convert<AccDataType>(0.0f);
+        });
+
+        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+        {
+            threadwise_x_load.Run(x_grid_desc_m_k,
+                                  x_global_val_buf,
+                                  thread_buffer_desc_m_k,
+                                  make_tuple(I0, I0),
+                                  x_thread_buf);
+
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, xy_copy_fwd_step_m_k);
+            threadwise_welford_1.Run(x_thread_buf, mean_thread_buf, var_thread_buf);
+        }
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            count_thread_buf(I) = threadwise_welford_1.cur_count_;
+            BlockwiseWelford1::Run(mean_thread_buf(I), var_thread_buf(I), count_thread_buf(I));
+        });
+
+        // Step 2: each workgroup writes its local welford result to workspace memory
+
+        auto mean_global_val_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                p_welford_mean, mean_var_count_grid_desc_m_g.GetElementSpaceSize());
+
+        auto var_global_val_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                p_welford_variance, mean_var_count_grid_desc_m_g.GetElementSpaceSize());
+
+        auto count_global_val_buf =
+            make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                p_welford_count, mean_var_count_grid_desc_m_g.GetElementSpaceSize());
+
+        auto threadwise_mean_var_store_m_g =
+            ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                               MeanVarDataType,
+                                               decltype(thread_buffer_desc_m_1),
+                                               MeanVarCountGridDesc_M_G,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M_1,
+                                               Sequence<0, 1>,
+                                               0,
+                                               1,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                mean_var_count_grid_desc_m_g,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_local_id),
+                PassThroughOp{});
+
+        auto threadwise_count_store_m_g =
+            ThreadwiseTensorSliceTransfer_v1r3<int32_t,
+                                               int32_t,
+                                               decltype(thread_buffer_desc_m_1),
+                                               MeanVarCountGridDesc_M_G,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M_1,
+                                               Sequence<0, 1>,
+                                               0,
+                                               1,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                mean_var_count_grid_desc_m_g,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 block_local_id),
+                PassThroughOp{});
+
+        if(thread_k_cluster_id == 0)
+        {
+            threadwise_mean_var_store_m_g.Run(thread_buffer_desc_m_1,
+                                              make_tuple(I0, I0),
+                                              mean_thread_buf,
+                                              mean_var_count_grid_desc_m_g,
+                                              mean_global_val_buf);
+
+            threadwise_mean_var_store_m_g.Run(thread_buffer_desc_m_1,
+                                              make_tuple(I0, I0),
+                                              var_thread_buf,
+                                              mean_var_count_grid_desc_m_g,
+                                              var_global_val_buf);
+
+            threadwise_count_store_m_g.Run(thread_buffer_desc_m_1,
+                                           make_tuple(I0, I0),
+                                           count_thread_buf,
+                                           mean_var_count_grid_desc_m_g,
+                                           count_global_val_buf);
+        };
+
+        gms_barrier(&p_control[blkgroup_id * 2]);
+
+        if(block_local_id == 0)
+            gms_reset(&p_control[blkgroup_id * 2]);
+
+        // Step 3: each workgroup reads welford results from workspace memory and does final welford
+        // reduction
+
+        auto threadwise_mean_var_load_m_k =
+            ThreadwiseTensorSliceTransfer_v2<MeanVarDataType,
+                                             AccDataType,
+                                             MeanVarCountGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             0,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_count_grid_desc_m_k,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * 1));
+
+        auto threadwise_count_load_m_k =
+            ThreadwiseTensorSliceTransfer_v2<int32_t,
+                                             int32_t,
+                                             MeanVarCountGridDesc_M_K,
+                                             decltype(thread_buffer_desc_m_1),
+                                             ThreadBufferLengths_M_1,
+                                             Sequence<0, 1>,
+                                             0,
+                                             1,
+                                             1,
+                                             true>(
+                mean_var_count_grid_desc_m_k,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize,
+                                 thread_k_cluster_id * 1));
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            mean_thread_buf(I)  = type_convert<AccDataType>(0.0f);
+            var_thread_buf(I)   = type_convert<AccDataType>(0.0f);
+            count_thread_buf(I) = 0;
+        });
+
+        constexpr auto mean_var_count_read_fwd_step_m_k = make_multi_index(0, KThreadClusterSize);
+
+        int32_t reducedSize = 0;
+        while(reducedSize < blkgroup_size)
+        {
+            threadwise_mean_var_load_m_k.Run(mean_var_count_grid_desc_m_k,
+                                             mean_global_val_buf,
+                                             thread_buffer_desc_m_1,
+                                             make_tuple(I0, I0),
+                                             tmp_mean_thread_buf);
+
+            threadwise_mean_var_load_m_k.Run(mean_var_count_grid_desc_m_k,
+                                             var_global_val_buf,
+                                             thread_buffer_desc_m_1,
+                                             make_tuple(I0, I0),
+                                             tmp_var_thread_buf);
+
+            threadwise_count_load_m_k.Run(mean_var_count_grid_desc_m_k,
+                                          count_global_val_buf,
+                                          thread_buffer_desc_m_1,
+                                          make_tuple(I0, I0),
+                                          tmp_count_thread_buf);
+
+            ThreadwiseWelford2::Run(tmp_mean_thread_buf,
+                                    tmp_var_thread_buf,
+                                    tmp_count_thread_buf,
+                                    mean_thread_buf,
+                                    var_thread_buf,
+                                    count_thread_buf);
+
+            reducedSize += KThreadClusterSize;
+
+            threadwise_mean_var_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
+                                                            mean_var_count_read_fwd_step_m_k);
+            threadwise_count_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
+                                                         mean_var_count_read_fwd_step_m_k);
+        };
+
+        static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+            if constexpr(I > 0)
+                block_sync_lds();
+
+            BlockwiseWelford2::Run(mean_thread_buf(I), var_thread_buf(I), count_thread_buf(I));
+        });
+
+        // Step 4: do normalization using the mean/variance
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> scale_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true> bias_thread_buf;
+
+        StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize * KThreadSliceSize, true>
+            y_thread_buf;
+
+        auto threadwise_y_store =
+            ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                               YDataType,
+                                               decltype(thread_buffer_desc_m_k),
+                                               XYGridDesc_M_K,
+                                               YElementwiseOp,
+                                               ThreadBufferLengths_M_K,
+                                               ThreadBufferDimAccessOrder,
+                                               XSrcYDstVectorDim,
+                                               YDstVectorSize,
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                y_grid_desc_m_k,
+                make_multi_index(
+                    blkgroup_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                    block_local_id * reduceSizePerBlock + thread_k_cluster_id * KThreadSliceSize),
+                y_elementwise_op);
+
+        auto threadwise_scale_load =
+            ThreadwiseTensorSliceTransfer_v2<ScaleDataType,
+                                             AccDataType,
+                                             ScaleBiasGridDesc_M,
+                                             decltype(thread_buffer_desc_m),
+                                             ThreadBufferLengths_M,
+                                             Sequence<0>,
+                                             0,
+                                             ScaleSrcVectorSize,
+                                             1,
+                                             true>(
+                scale_grid_desc_m,
+                make_multi_index(blkgroup_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize));
+
+        auto threadwise_bias_load = ThreadwiseTensorSliceTransfer_v2<BiasDataType,
+                                                                     AccDataType,
+                                                                     ScaleBiasGridDesc_M,
+                                                                     decltype(thread_buffer_desc_m),
+                                                                     ThreadBufferLengths_M,
+                                                                     Sequence<0>,
+                                                                     0,
+                                                                     BiasSrcVectorSize,
+                                                                     1,
+                                                                     true>(
+            bias_grid_desc_m,
+            make_multi_index(blkgroup_id * M_BlockTileSize +
+                             thread_m_cluster_id * MThreadSliceSize));
+
+        const auto scale_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_scale, scale_grid_desc_m.GetElementSpaceSize());
+
+        const auto bias_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_bias, bias_grid_desc_m.GetElementSpaceSize());
+
+        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_y, y_grid_desc_m_k.GetElementSpaceSize());
+
+        threadwise_scale_load.Run(scale_grid_desc_m,
+                                  scale_global_val_buf,
+                                  thread_buffer_desc_m,
+                                  make_tuple(I0),
+                                  scale_thread_buf);
+
+        threadwise_bias_load.Run(bias_grid_desc_m,
+                                 bias_global_val_buf,
+                                 thread_buffer_desc_m,
+                                 make_tuple(I0),
+                                 bias_thread_buf);
+
+        threadwise_x_load.SetSrcSliceOrigin(
+            x_grid_desc_m_k,
+            make_multi_index(blkgroup_id * M_BlockTileSize + thread_m_cluster_id * MThreadSliceSize,
+                             block_local_id * reduceSizePerBlock +
+                                 thread_k_cluster_id * KThreadSliceSize));
+
+        for(index_t reducedTiles = 0; reducedTiles < num_k_block_tile_iteration; ++reducedTiles)
+        {
+            threadwise_x_load.Run(x_grid_desc_m_k,
+                                  x_global_val_buf,
+                                  thread_buffer_desc_m_k,
+                                  make_tuple(I0, I0),
+                                  x_thread_buf);
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
+                AccDataType multiplier =
+                    scale_thread_buf[Number<iM>{}] / sqrt(var_thread_buf[iM] + epsilon);
+
+                AccDataType fused_mean_bias =
+                    bias_thread_buf[Number<iM>{}] - mean_thread_buf[iM] * multiplier;
+
+                static_for<0, KThreadSliceSize, 1>{}([&](auto iK) {
+                    constexpr auto offset =
+                        thread_buffer_desc_m_k.CalculateOffset(make_tuple(iM, iK));
+
+                    // normalize
+                    y_thread_buf(Number<offset>{}) =
+                        x_thread_buf[Number<offset>{}] * multiplier + fused_mean_bias;
+                });
+            });
+
+            threadwise_y_store.Run(thread_buffer_desc_m_k,
+                                   make_tuple(I0, I0),
+                                   y_thread_buf,
+                                   y_grid_desc_m_k,
+                                   y_global_val_buf);
+
+            threadwise_x_load.MoveSrcSliceWindow(x_grid_desc_m_k, xy_copy_fwd_step_m_k);
+            threadwise_y_store.MoveDstSliceWindow(y_grid_desc_m_k, xy_copy_fwd_step_m_k);
+        }
+
+        // Step 5: update the moving average of mean and variance (optional)
+
+        if(updateMovingAverage && block_local_id == 0 && thread_k_cluster_id == 0)
+        {
+            StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+                running_mean_thread_buf;
+            StaticBuffer<AddressSpaceEnum::Vgpr, AccDataType, MThreadSliceSize, true>
+                running_var_thread_buf;
+
+            auto running_mean_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultRunningMean, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            auto running_var_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultRunningVariance, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            auto threadwise_mean_var_load =
+                ThreadwiseTensorSliceTransfer_v2<MeanVarDataType,
+                                                 AccDataType,
+                                                 MeanVarGridDesc_M,
+                                                 decltype(thread_buffer_desc_m),
+                                                 ThreadBufferLengths_M,
+                                                 Sequence<0>,
+                                                 0,
+                                                 MeanVarSrcDstVectorSize,
+                                                 1,
+                                                 true>(
+                    mean_var_grid_desc_m,
+                    make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize));
+
+            threadwise_mean_var_load.Run(mean_var_grid_desc_m,
+                                         running_mean_global_buf,
+                                         thread_buffer_desc_m,
+                                         make_tuple(I0),
+                                         running_mean_thread_buf);
+
+            threadwise_mean_var_load.Run(mean_var_grid_desc_m,
+                                         running_var_global_buf,
+                                         thread_buffer_desc_m,
+                                         make_tuple(I0),
+                                         running_var_thread_buf);
+
+            AccDataType oneMinusAverageFactor = type_convert<AccDataType>(1.0) - averageFactor;
+
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                running_mean_thread_buf(I) = running_mean_thread_buf[I] * oneMinusAverageFactor +
+                                             mean_thread_buf[I] * averageFactor;
+                running_var_thread_buf(I) = running_var_thread_buf[I] * oneMinusAverageFactor +
+                                            var_thread_buf[I] * averageFactor;
+            });
+
+            auto threadwise_mean_var_store =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   MeanVarDataType,
+                                                   decltype(thread_buffer_desc_m),
+                                                   MeanVarGridDesc_M,
+                                                   PassThroughOp,
+                                                   ThreadBufferLengths_M,
+                                                   Sequence<0>,
+                                                   0,
+                                                   MeanVarSrcDstVectorSize,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>(
+                    mean_var_grid_desc_m,
+                    make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize),
+                    PassThroughOp{});
+
+            threadwise_mean_var_store.Run(thread_buffer_desc_m,
+                                          make_tuple(I0),
+                                          running_mean_thread_buf,
+                                          mean_var_grid_desc_m,
+                                          running_mean_global_buf);
+
+            threadwise_mean_var_store.Run(thread_buffer_desc_m,
+                                          make_tuple(I0),
+                                          running_var_thread_buf,
+                                          mean_var_grid_desc_m,
+                                          running_var_global_buf);
+        };
+
+        // Step 6: save mean and inv-variance (optional)
+
+        if(saveMeanInvVariance && block_local_id == 0 && thread_k_cluster_id == 0)
+        {
+            auto result_mean_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultSaveMean, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            auto result_inv_var_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+                resultSaveInvVariance, mean_var_grid_desc_m.GetElementSpaceSize());
+
+            // calculate inv-variance as 1/sqrt(epsilon+variance), stored in place of variance
+            static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
+                var_thread_buf(I) =
+                    type_convert<AccDataType>(1.0f) / sqrt(epsilon + var_thread_buf[I]);
+            });
+
+            auto threadwise_mean_inv_var_store =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   MeanVarDataType,
+                                                   decltype(thread_buffer_desc_m),
+                                                   MeanVarGridDesc_M,
+                                                   PassThroughOp,
+                                                   ThreadBufferLengths_M,
+                                                   Sequence<0>,
+                                                   0,
+                                                   MeanVarSrcDstVectorSize,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>(
+                    mean_var_grid_desc_m,
+                    make_multi_index(blkgroup_id * M_BlockTileSize +
+                                     thread_m_cluster_id * MThreadSliceSize),
+                    PassThroughOp{});
+
+            threadwise_mean_inv_var_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              mean_thread_buf,
+                                              mean_var_grid_desc_m,
+                                              result_mean_global_buf);
+
+            threadwise_mean_inv_var_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              var_thread_buf,
+                                              mean_var_grid_desc_m,
+                                              result_inv_var_global_buf);
+        };
+    }
+}; // namespace ck
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
index fc263138f9..a82a173500 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_first_half.hpp
@@ -161,7 +161,7 @@ struct GridwiseMultiblockWelfordFirstHalf
                                                PassThroughOp,
                                                ThreadBufferLengths_M_1,
                                                Sequence<0, 1>,
-                                               1,
+                                               0,
                                                1,
                                                InMemoryDataOperationEnum::Set,
                                                1,
@@ -180,7 +180,7 @@ struct GridwiseMultiblockWelfordFirstHalf
                                                PassThroughOp,
                                                ThreadBufferLengths_M_1,
                                                Sequence<0, 1>,
-                                               1,
+                                               0,
                                                1,
                                                InMemoryDataOperationEnum::Set,
                                                1,
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
similarity index 98%
rename from include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
rename to include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
index 1f8990e6df..dd514567d2 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
@@ -33,7 +33,6 @@ __global__ void kernel_welford_second_half_batchnorm_forward_final(
     const MeanVarGridDesc_M mean_var_grid_desc_m,
     index_t blkgroup_size,
     index_t num_xy_k_block_tile_iteration,
-    index_t num_mean_var_count_k_block_tile_iteration,
     AccDataType epsilon,
     const MeanVarDataType* const __restrict__ p_in_welford_mean,
     const MeanVarDataType* const __restrict__ p_in_welford_variance,
@@ -59,7 +58,6 @@ __global__ void kernel_welford_second_half_batchnorm_forward_final(
                                                          mean_var_grid_desc_m,
                                                          blkgroup_size,
                                                          num_xy_k_block_tile_iteration,
-                                                         num_mean_var_count_k_block_tile_iteration,
                                                          epsilon,
                                                          p_in_welford_mean,
                                                          p_in_welford_variance,
@@ -152,7 +150,6 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                const MeanVarGridDesc_M& mean_var_grid_desc_m,
                                index_t blkgroup_size,
                                index_t num_xy_k_block_tile_iteration,
-                               index_t num_mean_var_count_k_block_tile_iteration,
                                AccDataType epsilon,
                                const MeanVarDataType* const __restrict__ p_in_welford_mean,
                                const MeanVarDataType* const __restrict__ p_in_welford_variance,
@@ -223,7 +220,7 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                              decltype(thread_buffer_desc_m_1),
                                              ThreadBufferLengths_M_1,
                                              Sequence<0, 1>,
-                                             1,
+                                             0,
                                              1,
                                              1,
                                              true>(
@@ -239,7 +236,7 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                              decltype(thread_buffer_desc_m_1),
                                              ThreadBufferLengths_M_1,
                                              Sequence<0, 1>,
-                                             1,
+                                             0,
                                              1,
                                              1,
                                              true>(
@@ -257,9 +254,6 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
         const auto welford_count_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_in_welford_count, mean_var_count_grid_desc_m_k.GetElementSpaceSize());
 
-        constexpr auto mean_var_count_thread_copy_step_m_k =
-            make_multi_index(0, KThreadClusterSize * 1);
-
         // Step 1: do final welford reduction to get mean and variance
 
         static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
@@ -268,8 +262,11 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
             welford_count_thread_buf(I) = 0;
         });
 
-        for(index_t reducedTiles = 0; reducedTiles < num_mean_var_count_k_block_tile_iteration;
-            ++reducedTiles)
+        constexpr auto mean_var_count_thread_copy_step_m_k =
+            make_multi_index(0, KThreadClusterSize);
+
+        int32_t reducedSize = 0;
+        while(reducedSize < blkgroup_size)
         {
             threadwise_mean_var_load_m_k.Run(mean_var_count_grid_desc_m_k,
                                              welford_mean_global_val_buf,
@@ -296,6 +293,8 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
                                    welford_var_thread_buf,
                                    welford_count_thread_buf);
 
+            reducedSize += KThreadClusterSize;
+
             threadwise_mean_var_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
                                                             mean_var_count_thread_copy_step_m_k);
             threadwise_count_load_m_k.MoveSrcSliceWindow(mean_var_count_grid_desc_m_k,
diff --git a/include/ck/utility/get_shift.hpp b/include/ck/utility/get_shift.hpp
new file mode 100644
index 0000000000..0a93081cfd
--- /dev/null
+++ b/include/ck/utility/get_shift.hpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+
+template <index_t N>
+static constexpr __device__ index_t get_shift()
+{
+    return (get_shift<N / 2>() + 1);
+};
+
+template <>
+constexpr __device__ index_t get_shift<1>()
+{
+    return (0);
+}
+
+} // namespace ck
diff --git a/include/ck/utility/reduction_common.hpp b/include/ck/utility/reduction_common.hpp
index 3777d297c8..75fdd85825 100644
--- a/include/ck/utility/reduction_common.hpp
+++ b/include/ck/utility/reduction_common.hpp
@@ -25,16 +25,4 @@ struct float_equal_zero
     };
 };
 
-template <index_t N>
-static constexpr __device__ index_t get_shift()
-{
-    return (get_shift<N / 2>() + 1);
-};
-
-template <>
-constexpr __device__ index_t get_shift<1>()
-{
-    return (0);
-}
-
 } // namespace ck
diff --git a/include/ck/utility/workgroup_synchronization.hpp b/include/ck/utility/workgroup_synchronization.hpp
new file mode 100644
index 0000000000..24858fdbdc
--- /dev/null
+++ b/include/ck/utility/workgroup_synchronization.hpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+#include "ck/host_utility/hip_check_error.hpp"
+
+namespace ck {
+
+// Initialization flag of Barrier object, can be any value except for zero
+static constexpr int BarrierInitFlag = 0x7856;
+
+// 1) only the first thread-block in the synchronizaton group is supposed to call this function. It
+// is the responsibility of the user to ensure the two integer values in p_control_bits are zeros
+// before calling gms_init().
+// 2) Aftercalling gms_reset(), the two integer values in p_control_bits will be zeros, so no
+// repetitious initialization of p_control_bits buffer is required
+static __device__ void gms_init(int NumWarps, int* p_control_bits)
+{
+    union
+    {
+        int two32[2];
+        unsigned long one64;
+    } regs;
+
+    regs.two32[0] = BarrierInitFlag;
+    regs.two32[1] = NumWarps;
+
+    if(threadIdx.x == 0)
+        atomicCAS(reinterpret_cast<unsigned long*>(p_control_bits), 0, regs.one64);
+};
+
+// all the workgroups in the synchronization group is supposed to call this function
+static __device__ void gms_barrier(int* p_control_bits)
+{
+    constexpr int mask = warpSize - 1;
+
+    if((threadIdx.x & mask) == 0)
+    {
+        // ensure the barrier object is initialized
+        do
+        {
+            const int r0 = __atomic_load_n(&p_control_bits[0], __ATOMIC_RELAXED);
+
+            if(r0 == BarrierInitFlag)
+                break;
+
+        } while(true);
+
+        // go ahead toward the barrier line
+        atomicSub(&p_control_bits[1], 1);
+
+        // wait until all warps have arrived
+        do
+        {
+            const int r1 = __atomic_load_n(&p_control_bits[1], __ATOMIC_RELAXED);
+
+            if(r1 == 0)
+                break;
+
+        } while(true);
+    };
+};
+
+// 1) Only the first thread-block in the synchronizaton group is supposed to call this function.
+// 2) Aftercalling gms_reset(), the two integer values in p_control_bits will be zeros, so no
+// repetitious initialization of p_control_bits buffer is required
+static __device__ void gms_reset(int* p_control_bits)
+{
+    // reset the barrier object
+    if(threadIdx.x == 0)
+        (void)atomicCAS(&p_control_bits[0], BarrierInitFlag, 0);
+};
+
+} // namespace ck
diff --git a/profiler/src/profile_batchnorm_fwd.cpp b/profiler/src/profile_batchnorm_fwd.cpp
index 2b3e4eea41..507fb4b450 100644
--- a/profiler/src/profile_batchnorm_fwd.cpp
+++ b/profiler/src/profile_batchnorm_fwd.cpp
@@ -148,7 +148,7 @@ int profile_batchnorm_forward(int argc, char* argv[])
     {
         if(arg_parser.inLengths.size() == 4 && arg_parser.reduceDims.size() == 3)
         {
-            profile_batchnorm_forward_impl<F16, F16, F32, F16, F16, F16, 4, 3>(
+            profile_batchnorm_forward_impl<F16, F16, F32, F16, F16, F32, 4, 3>(
                 arg_parser.do_verification,
                 arg_parser.init_method,
                 arg_parser.do_dumpout,

From 850144a0d3237ea1f0e0f64422358e8b49cc8810 Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Thu, 6 Jul 2023 23:59:35 +0800
Subject: [PATCH 055/194] Split GEMM instance library & enable pipeline v2
 optimization (#783)

* Move source file into sub-directories

* Add missing include directive

* Split DeviceGemmXdl<> fp16 instances

* Fix format

* Remove unnecessary CMakeLists.txt

* Add macros to toggle new features

* Remove debug message

* Turn off GEMM v2 pipeline optimization by default

* Fix format

* Extract duplicated string as list

* Enlarge indent in CMakeLists.txt
---
 include/ck/ck.hpp                             |  19 +++
 .../grid/gridwise_gemm_pipeline_selector.hpp  |   2 +
 .../gpu/grid/gridwise_gemm_pipeline_v2.hpp    |   4 +
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |   6 +
 .../gpu/gemm/CMakeLists.txt                   | 146 +++++++++++++-----
 .../device_gemm_xdl_f16_f16_f16/common.hpp    |  43 ++++++
 .../km_kn_mn_add_instance.cpp                 |  38 +++++
 .../km_kn_mn_default_pipeline_v1_instance.cpp |  40 +++++
 .../km_kn_mn_default_pipeline_v2_instance.cpp |  42 +++++
 ...kn_mn_default_pipeline_v2_opt_instance.cpp |  35 +++++
 ...m_kn_mn_interwave_pipeline_v1_instance.cpp |  42 +++++
 ...irregular_default_pipeline_v1_instance.cpp |  32 ++++
 ...irregular_default_pipeline_v2_instance.cpp |  34 ++++
 ...regular_interwave_pipeline_v1_instance.cpp |  34 ++++
 .../km_nk_mn_add_instance.cpp                 |  38 +++++
 .../km_nk_mn_default_pipeline_v1_instance.cpp |  40 +++++
 .../km_nk_mn_default_pipeline_v2_instance.cpp |  42 +++++
 ...nk_mn_default_pipeline_v2_opt_instance.cpp |  35 +++++
 ...m_nk_mn_interwave_pipeline_v1_instance.cpp |  42 +++++
 ...irregular_default_pipeline_v1_instance.cpp |  32 ++++
 ...irregular_default_pipeline_v2_instance.cpp |  34 ++++
 ...regular_interwave_pipeline_v1_instance.cpp |  34 ++++
 .../mk_kn_mn_add_instance.cpp                 |  38 +++++
 .../mk_kn_mn_default_pipeline_v1_instance.cpp |  49 ++++++
 .../mk_kn_mn_default_pipeline_v2_instance.cpp |  51 ++++++
 ...kn_mn_default_pipeline_v2_opt_instance.cpp |  35 +++++
 ...k_kn_mn_interwave_pipeline_v1_instance.cpp |  51 ++++++
 ...irregular_default_pipeline_v1_instance.cpp |  32 ++++
 ...irregular_default_pipeline_v2_instance.cpp |  34 ++++
 ...regular_interwave_pipeline_v1_instance.cpp |  34 ++++
 .../mk_nk_mn_add_instance.cpp                 |  38 +++++
 .../mk_nk_mn_default_pipeline_v1_instance.cpp |  44 ++++++
 .../mk_nk_mn_default_pipeline_v2_instance.cpp |  46 ++++++
 ...nk_mn_default_pipeline_v2_opt_instance.cpp |  34 ++++
 ...k_nk_mn_interwave_pipeline_v1_instance.cpp |  46 ++++++
 ...irregular_default_pipeline_v1_instance.cpp |  34 ++++
 ...irregular_default_pipeline_v2_instance.cpp |  36 +++++
 ...regular_interwave_pipeline_v1_instance.cpp |  36 +++++
 ...gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp | 110 -------------
 ...gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp | 110 -------------
 ...gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp | 137 ----------------
 ...gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp | 130 ----------------
 42 files changed, 1411 insertions(+), 528 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 161c3261ad..6a1ac09021 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -27,6 +27,21 @@
 #define CK_WAVELET_MIN_BLOCK_PER_CU 2
 #endif
 
+// kernel attribute: amdgpu_waves_per_eu()
+#ifdef CK_USE_WAVES_PER_EU
+// for 1-wave kernels, control arguments of amdgpu_waves_per_eu() attribute
+#ifndef CK_MIN_WAVES_PER_EU
+#define CK_MIN_WAVES_PER_EU 0
+#endif
+
+#ifndef CK_MAX_WAVES_PER_EU
+#define CK_MAX_WAVES_PER_EU 0
+#endif
+
+#else
+#define CK_USE_WAVES_PER_EU 0
+#endif
+
 // buffer resource
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_BUFFER_RESOURCE_3RD_DWORD -1
@@ -148,6 +163,10 @@
 #define CK_EXPERIMENTAL_INTER_WAVE_INSTANCES 1
 // experimental feature: add instances using pipeline v2
 #define CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES 1
+// experimental feature: optimize pipeline v2 by IGLP strategy (value=ID of strategy)
+#ifndef CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT
+#define CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT 0
+#endif
 
 // hack: have underlying assumption that need to be satsified, otherwise it's a bug
 // hack for forcing register to keep idx_diff_low_const in SGPR. idx_diff_low_const must be
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
index 9090c16dcd..f760feb2ed 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include <iostream>
+
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp"
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
index f54345b04f..d3d7d5af85 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
@@ -79,6 +79,10 @@ struct GridwiseGemmPipeline_v2
 
             do
             {
+#if CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT
+                __builtin_amdgcn_iglp_opt(CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT);
+#endif
+
                 block_sync_lds();
 
                 // GEMM i
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index 7b8bbd301e..3d1b13f998 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -27,6 +27,9 @@ template <typename GridwiseGemm,
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+#if CK_USE_WAVES_PER_EU
+        __attribute__((amdgpu_waves_per_eu(CK_MIN_WAVES_PER_EU, CK_MAX_WAVES_PER_EU)))
 #endif
         kernel_gemm_xdlops_v2r3(const FloatAB* __restrict__ p_a_grid,
                                 const FloatAB* __restrict__ p_b_grid,
@@ -60,6 +63,9 @@ template <typename GridwiseGemm, bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+#if CK_USE_WAVES_PER_EU
+        __attribute__((amdgpu_waves_per_eu(CK_MIN_WAVES_PER_EU, CK_MAX_WAVES_PER_EU)))
 #endif
         kernel_gemm_xdlops_v2r3(const typename GridwiseGemm::Argument karg)
 {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index d66010af73..11227d01f5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -1,51 +1,115 @@
 add_instance_library(device_gemm_instance
-   device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
    device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
    device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
    device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+   device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
+   device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
+   device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
+   device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
    device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
    device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
    device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
    device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
+   device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
+   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
+
+   device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
+   device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
+   device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
+   device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
+   device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
 )
+
+set(ENABLE_PIPELINE_V2_OPT OFF)
+
+if (ENABLE_PIPELINE_V2_OPT)
+set(MAX_ILP_OPTS
+   -mllvm
+   -amdgpu-enable-max-ilp-scheduling-strategy
+)
+set(WAVES_PER_EU_DEFS
+   CK_USE_WAVES_PER_EU=1
+   CK_MIN_WAVES_PER_EU=1
+   CK_MAX_WAVES_PER_EU=1
+)
+set(IGLP_OPT_DEFS
+   CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT=1
+)
+
+# layout=NT
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS ";;"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+# layout=NN
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+# layout=TT
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS}")
+# layout=TN
+set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+endif(ENABLE_PIPELINE_V2_OPT)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp
new file mode 100644
index 0000000000..de96a22630
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/common.hpp
@@ -0,0 +1,43 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/utility/data_type.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+using InstanceNT = DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+using InstanceNN = DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+using InstanceTT = DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+using InstanceTN = DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>;
+
+template <typename Instance>
+using OwnerList = std::vector<std::unique_ptr<Instance>>;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
new file mode 100644
index 0000000000..d887b16e60
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceNT;
+using Instances = OwnerList<InstanceNT>;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..81fedd50f0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+        // clang-format off
+        // pipeline v1, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..5a0c52c2df
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000..59ffb80bd4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|           LoopScheduler|             Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                        |                     |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                        |                     |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                        |                     |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..a64424e8ac
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..e313a790e1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1, LoopScheduler::Default,          PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..050ff6e231
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1, LoopScheduler::Default,          PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..0a0406baec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|            LoopScheduler|                   Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                         |                           |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                         |                           |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                         |                           |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1, LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
new file mode 100644
index 0000000000..7fb1a4845d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceNN;
+using Instances = OwnerList<Instance>;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..a0dd60c0f5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+        // clang-format off
+        // pipeline v1, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..122fff4960
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000..7fc55b017e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|              Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                      |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..3671bea7a3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,42 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..d7e7a05505
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..15e191e5a8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..95fc8ecb46
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceNN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
new file mode 100644
index 0000000000..e3a7ccefb4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceTT;
+using Instances = OwnerList<Instance>;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..98db8bad1c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+        // clang-format off
+        // pipeline v1, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..532c348b7e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000..5314c7fc20
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|              Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                      |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                      |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..fa53a3bf0f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,51 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using Instances =
+    std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..bc4b3e801b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..0c8e8b2c41
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..c9d1913aec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTT>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
new file mode 100644
index 0000000000..70bfd0a80c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
@@ -0,0 +1,38 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Instance  = InstanceTN;
+using Instances = OwnerList<Instance>;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_opt_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_interwave_pipeline_v1_instances(Instances&);
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v2_instances(Instances&);
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(Instances& instances)
+{
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_opt_instances(instances);
+
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_interwave_pipeline_v1_instances(instances);
+    add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v2_instances(instances);
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..0f4ccceb50
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
@@ -0,0 +1,44 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..134755299d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
new file mode 100644
index 0000000000..1aba1fa8d9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|           LoopScheduler|             Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                        |                     |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                        |                     |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                        |                     |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_default_pipeline_v2_opt_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..0410eabb70
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..e6e73cdcca
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+    // clang-format off
+        // pipeline v1, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
new file mode 100644
index 0000000000..78f18fd582
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
+        // pipeline v2, 1 wave
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_default_pipeline_v2_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
new file mode 100644
index 0000000000..a41919aab7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
@@ -0,0 +1,36 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// irregular tile size
+using Instances = std::tuple<
+// clang-format off
+#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
+        // pipeline v1, 2 waves
+        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
+        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
+        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
+#endif
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_interwave_pipeline_v1_instances(
+    OwnerList<InstanceTN>& instances)
+{
+    add_device_operation_instances(instances, Instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
deleted file mode 100644
index 8a81b77891..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_kn_mn_instance.cpp
+++ /dev/null
@@ -1,110 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_km_kn_mn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler| Pipeline|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |         |
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |         |
-        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |         |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-        // clang-format on
-        >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   64,     16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_km_kn_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_km_kn_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
deleted file mode 100644
index e1983add04..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_km_nk_mn_instance.cpp
+++ /dev/null
@@ -1,110 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_km_nk_mn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-        // clang-format on
-        >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmMNPadding,   64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_km_nk_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_km_nk_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
deleted file mode 100644
index 47a180e127..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ /dev/null
@@ -1,137 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances =
-    std::tuple<
-        // clang-format off
-        //##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   256,     4,  8,   32,   32,    1,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    32,     4,  8,   32,   32,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   256,     4,  8,   16,   16,    1,    8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,   128,     4,  8,   16,   16,    1,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    64,     4,  8,   16,   16,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    16,    32,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-        // clang-format on
-        >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
deleted file mode 100644
index b8e994e91a..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
+++ /dev/null
@@ -1,130 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault   = ck::tensor_operation::device::GemmSpecialization::Default;
-static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
-
-// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
-using device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-// irregular tile size
-using device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
-    // clang-format off
-        //###########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumPrefetch|          LoopScheduler|                     Pipeline|
-        //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                             |
-        //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                             |
-        // pipeline v1, 1 wave
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v1>
-#if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
-        // pipeline v1, 2 waves
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Interwave,        PipelineVersion::v1>
-#endif
-#if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
-        // pipeline v2, 1 wave
-        ,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     8,  8,   16,   16,    2,    9,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8,  8, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   144,     4,  8,   16,   16,    2,    9,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 4>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               7,               1,           1,  LoopScheduler::Default,        PipelineVersion::v2>
-#endif
-    // clang-format on
-    >;
-
-void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances{});
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck

From 237f9cd3aa80184daabaf28104c114f605c0b381 Mon Sep 17 00:00:00 2001
From: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Date: Thu, 6 Jul 2023 18:01:06 +0200
Subject: [PATCH 056/194] Add basic setup for precommit (#749) (#764)

* Add basic setup for precommit

* Update README.md with instructions on installing precommit hooks

---------

Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
---
 .pre-commit-config.yaml        | 14 ++++++++++++++
 README.md                      | 18 ++++++++++++++++++
 script/check_copyright_year.sh | 17 +++++++++++++++++
 script/install_precommit.sh    | 20 ++++++++++++++++++++
 script/uninstall_precommit.sh  |  1 +
 5 files changed, 70 insertions(+)
 create mode 100644 .pre-commit-config.yaml
 create mode 100755 script/check_copyright_year.sh
 create mode 100755 script/install_precommit.sh
 create mode 100755 script/uninstall_precommit.sh

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
new file mode 100644
index 0000000000..3e1e8c0169
--- /dev/null
+++ b/.pre-commit-config.yaml
@@ -0,0 +1,14 @@
+repos:
+-   repo: local
+    hooks:
+    -   id: clang-format
+        name: clang-format
+        entry: clang-format-10 -i --style=file
+        language: system
+        types_or: [c++, inc]
+    -   id: copyright-year-checker
+        name: copyright-year-checker
+        entry: script/check_copyright_year.sh
+        verbose: false
+        language: script
+        types: [c++]
diff --git a/README.md b/README.md
index a45f61a37d..fd6f7e37c2 100644
--- a/README.md
+++ b/README.md
@@ -109,6 +109,24 @@ make install
 
 Instructions for using CK as a pre-built kernel library are under [client_example](/client_example)
 
+## Contributing
+
+When you contribute to Composable Kernel, make sure to run `clang-format` on all the changed files. We highly recommend using git hooks that are managed by the `pre-commit` framework. To install hooks, run:
+
+```bash
+sudo script/install_precommit.sh
+```
+
+This way, `pre-commit` will add the appropriate hooks to your local repository and automatically run `clang-format` (and possibly additional checks) before any commit is created.
+
+If you need to uninstall hooks from the repository, you can do so by running the following command:
+
+```bash
+script/uninstall_precommit.sh
+```
+
+If for any reason, you need to temporarily disable precommit hooks, you can add the `--no-verify` option to the `git commit` command.
+
 ## Caveat
 ### Kernel Timing and Verification
 
diff --git a/script/check_copyright_year.sh b/script/check_copyright_year.sh
new file mode 100755
index 0000000000..f7709472ef
--- /dev/null
+++ b/script/check_copyright_year.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+
+current_year=$(date +%Y)
+exit_code=0
+
+for file in $@; do
+    if grep -q "Copyright (c)" $file
+    then
+        if ! grep -q "Copyright (c).*$current_year" $file
+        then
+            echo "ERROR: File $file has a copyright notice without the current year ($current_year)."
+            exit_code=1
+        fi
+    fi
+done
+
+exit $exit_code
diff --git a/script/install_precommit.sh b/script/install_precommit.sh
new file mode 100755
index 0000000000..daa5cb29ae
--- /dev/null
+++ b/script/install_precommit.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+
+run_and_check() {
+    "$@"
+    status=$?
+    if [ $status -ne 0 ]; then
+        echo "Error with \"$@\": Exited with status $status"
+        exit $status
+    fi
+    return $status
+}
+
+echo "I: Installing tools required for pre-commit checks..."
+run_and_check apt install clang-format-10
+
+echo "I: Installing pre-commit itself..."
+run_and_check pip3 install pre-commit
+run_and_check pre-commit install
+
+echo "I: Installation successful."
diff --git a/script/uninstall_precommit.sh b/script/uninstall_precommit.sh
new file mode 100755
index 0000000000..b0d4d15166
--- /dev/null
+++ b/script/uninstall_precommit.sh
@@ -0,0 +1 @@
+pre-commit uninstall

From 87f2bbcf5c36a85cce4407aecf98fe3fc4b94f0a Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 6 Jul 2023 18:17:25 -0700
Subject: [PATCH 057/194] change the build thread usage in CI (#787)

---
 Jenkinsfile | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 808fb9d1bf..df9d13167f 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -11,6 +11,20 @@ def show_node_info() {
     """
 }
 
+def nthreads() {
+    def nproc = sh(returnStdout: true, script: 'nproc')
+    echo "Number of cores: ${nproc}"
+    def n = nproc.toInteger()
+    if (n > 32){
+        n /= 2
+    }
+    if (n > 64){
+        n = 64
+    }
+    echo "Number of threads used for building: ${n}"
+    return n
+}
+
 def runShell(String command){
     def responseCode = sh returnStatus: true, script: "${command} > tmp.txt"
     def output = readFile(file: "tmp.txt")
@@ -219,7 +233,8 @@ def cmake_build(Map conf=[:]){
         """
     def setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
     // reduce parallelism when compiling, clang uses too much memory
-    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j\$(( \$(nproc) / 2 )) ${config_targets}")
+    def nt = nthreads()
+    def build_cmd = conf.get("build_cmd", "${build_envs} dumb-init make  -j${nt} ${config_targets}")
     def execute_cmd = conf.get("execute_cmd", "")
 
     def cmd = conf.get("cmd", """
@@ -461,7 +476,7 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
-                        if ( runShell('grep -n "gfx1030" clinfo.log') ){
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
                             navi_node = 1
                         }
                     }
@@ -482,7 +497,7 @@ def Build_CK(Map conf=[:]){
                         else{
                             echo "GPU is OK"
                         }
-                        if ( runShell('grep -n "gfx1030" clinfo.log') ){
+                        if ( runShell('grep -n "gfx1030" clinfo.log') || runShell('grep -n "gfx1101" clinfo.log') ){
                             navi_node = 1
                         }
                     }
@@ -493,8 +508,9 @@ def Build_CK(Map conf=[:]){
                 {
                     cmake_build(conf)
                     dir("build"){
-                        //run tests and examples 	
-                        sh 'make -j\$(( \$(nproc) / 2 )) check'
+                        //run tests and examples
+                        def nt = nthreads()	
+                        sh 'make -j${nt} check'
                         if (navi_node == 0 ){
                             //we only need the ckProfiler to run the performance tests, so we pack and stash it
                             //do not stash profiler on Navi nodes
@@ -717,7 +733,7 @@ pipeline {
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
                     }
                 }
-                stage("Build CK and run Tests on Navi")
+                stage("Build CK and run Tests on Navi21")
                 {
                     when {
                         beforeAgent true

From 1ee99dcaa644dd520a04f70bf9be1a9db78c5d70 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 12 Jul 2023 15:25:02 +0200
Subject: [PATCH 058/194] Support NHWGC conv2d_bwd_weight (#769)

* Support NHWGC conv2d_bwd_weight

* Fix client example

* Fix client example

* Fix comments

* Redesign grouped_conv_bwd_weight instances

* Clang format fix

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../11_grouped_conv_bwd_weight/common.hpp     |  14 +-
 .../grouped_conv1d_bwd_weight_fp16.cpp        |  23 +-
 .../grouped_conv2d_bwd_weight_fp16.cpp        |  26 +-
 .../grouped_conv3d_bwd_weight_fp16.cpp        |  27 +-
 .../grouped_conv3d_bwd_weight_fp32.cpp        |  36 +-
 .../grouped_conv_bwd_weight_xdl_bf16.cpp      |  18 +-
 .../grouped_conv_bwd_weight_xdl_fp16.cpp      |  18 +-
 .../run_grouped_conv_bwd_weight_example.inc   |   6 +
 .../device/device_grouped_conv_bwd_weight.hpp |  24 +-
 ...uped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp | 164 +++---
 ..._grouped_conv_bwd_weight_xdl_cshuffle.hpp} | 538 +++++++++++++-----
 ...e_grouped_conv_bwd_weight_xdl_instance.hpp | 119 ++++
 .../grouped_convolution_backward_weight.hpp   | 169 ++++--
 ...eight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp |  86 +--
 ...weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp |  84 +--
 ...weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp |  83 +--
 .../grouped_conv2d_bwd_weight/CMakeLists.txt  |   3 +
 ...ht_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |  93 +--
 ...ght_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  92 +--
 ...ght_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |  91 +--
 ...ht_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |  47 ++
 ...ght_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  47 ++
 ...ght_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |  47 ++
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |  86 +--
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |  86 +--
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |  85 +--
 profiler/README.md                            |  43 ++
 .../profile_grouped_conv_bwd_weight_impl.hpp  |   6 +
 .../src/profile_grouped_conv_bwd_weight.cpp   |  21 +
 test/grouped_convnd_bwd_weight/CMakeLists.txt |   4 +-
 .../grouped_convnd_bwd_weight.cpp             |  91 ---
 .../test_grouped_convnd_bwd_weight.cpp        | 125 ++++
 ...st_grouped_convnd_bwd_weight_interface.cpp | 180 ++++++
 33 files changed, 1547 insertions(+), 1035 deletions(-)
 rename include/ck/tensor_operation/gpu/device/impl/{device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp => device_grouped_conv_bwd_weight_xdl_cshuffle.hpp} (70%)
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
 delete mode 100644 test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
 create mode 100644 test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
 create mode 100644 test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp

diff --git a/client_example/11_grouped_conv_bwd_weight/common.hpp b/client_example/11_grouped_conv_bwd_weight/common.hpp
index 62eb7bcf55..f63e5f2157 100644
--- a/client_example/11_grouped_conv_bwd_weight/common.hpp
+++ b/client_example/11_grouped_conv_bwd_weight/common.hpp
@@ -101,13 +101,15 @@ template <ck::index_t NumDimSpatial,
           typename WeiLayout,
           typename OutLayout>
 bool run_grouped_conv_bwd_weight(
-    ck::index_t G,
-    ck::index_t N,
-    ck::index_t K,
-    ck::index_t C,
+    const ck::index_t G,
+    const ck::index_t N,
+    const ck::index_t K,
+    const ck::index_t C,
     const std::array<ck::index_t, NumDimSpatial>& input_spatial_lengths,
     const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths,
     const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths,
+    const std::array<ck::index_t, NumDimSpatial + 3>& input_strides,
+    const std::array<ck::index_t, NumDimSpatial + 3>& output_strides,
     const std::array<ck::index_t, NumDimSpatial>& conv_filter_strides,
     const std::array<ck::index_t, NumDimSpatial>& conv_filter_dilations,
     const std::array<ck::index_t, NumDimSpatial>& input_left_pads,
@@ -157,6 +159,8 @@ bool run_grouped_conv_bwd_weight(
                                                         input_spatial_lengths,
                                                         filter_spatial_lengths,
                                                         output_spatial_lengths,
+                                                        input_strides,
+                                                        output_strides,
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
                                                         input_left_pads,
@@ -224,6 +228,8 @@ bool run_grouped_conv_bwd_weight(
                                                         input_spatial_lengths,
                                                         filter_spatial_lengths,
                                                         output_spatial_lengths,
+                                                        input_strides,
+                                                        output_strides,
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
                                                         input_left_pads,
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
index 788d50ddef..1c6f485da2 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
@@ -22,6 +22,15 @@ static constexpr ck::index_t C             = 192;
 static constexpr ck::index_t X             = 3;
 static constexpr ck::index_t Wi            = 28;
 static constexpr ck::index_t Wo            = 28;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{N * Wi * C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{N * Wo * K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1};
 
 int main()
 {
@@ -31,7 +40,19 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G, N, K, C, {Wi}, {X}, {Wo}, {1}, {1}, {1}, {1})
+                                       OutLayout>(G,
+                                                  N,
+                                                  K,
+                                                  C,
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  output_spatial_lengths,
+                                                  input_strides,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
index 1903bd95b6..25e82f3896 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
@@ -25,6 +25,17 @@ static constexpr ck::index_t Hi            = 28;
 static constexpr ck::index_t Wi            = 28;
 static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 28;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
+    N * Hi * Wi * C, Hi* Wi* C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
+    N * Ho * Wo * K, Ho* Wo* K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
 
 int main()
 {
@@ -34,8 +45,19 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(
-               G, N, K, C, {Hi, Wi}, {Y, X}, {Ho, Wo}, {1, 1}, {1, 1}, {1, 1}, {1, 1})
+                                       OutLayout>(G,
+                                                  N,
+                                                  K,
+                                                  C,
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  output_spatial_lengths,
+                                                  input_strides,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
index 2f2b5d4e21..a5f5e628ff 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
@@ -28,6 +28,17 @@ static constexpr ck::index_t Wi            = 3;
 static constexpr ck::index_t Do            = 28;
 static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 3;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Di, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Z, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Do, Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
+    N * Di * Hi * Wi * C, Di* Hi* Wi* C, Hi* Wi* C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
+    N * Do * Ho * Wo * K, Do* Ho* Wo* K, Ho* Wo* K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
 
 int main()
 {
@@ -41,13 +52,15 @@ int main()
                                                   N,
                                                   K,
                                                   C,
-                                                  {Di, Hi, Wi},
-                                                  {Z, Y, X},
-                                                  {Do, Ho, Wo},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1})
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  output_spatial_lengths,
+                                                  input_strides,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
index 796311d231..d95e8a205e 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
@@ -28,6 +28,17 @@ static constexpr ck::index_t Wi            = 3;
 static constexpr ck::index_t Do            = 28;
 static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 3;
+static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Di, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Z, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Do, Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
+    N * Di * Hi * Wi * C, Di* Hi* Wi* C, Hi* Wi* C, Wi* C, C, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
+    N * Do * Ho * Wo * K, Do* Ho* Wo* K, Ho* Wo* K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
 
 int main()
 {
@@ -37,17 +48,20 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G,
-                                                  N,
-                                                  K,
-                                                  C,
-                                                  {Di, Hi, Wi},
-                                                  {Z, Y, X},
-                                                  {Do, Ho, Wo},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1},
-                                                  {1, 1, 1})
+                                       OutLayout>(
+               G,
+               N,
+               K,
+               C,
+               {Di, Hi, Wi},
+               {Z, Y, X},
+               {Do, Ho, Wo},
+               {N * Di * Hi * Wi * C, Di * Hi * Wi * C, Hi * Wi * C, Wi * C, C, 1},
+               {N * Do * Ho * Wo * K, Do * Ho * Wo * K, Ho * Wo * K, Wo * K, K, 1},
+               {1, 1, 1},
+               {1, 1, 1},
+               {1, 1, 1},
+               {1, 1, 1})
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
index 3cd70d0f36..31e277b5c7 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 
 using InDataType = BF16;
 // bf16 kernel use fp32 atomic add to accumulate Weight tensor into global memory
@@ -17,8 +17,20 @@ using OutElementOp = PassThrough;
 
 template <ck::index_t NDimSpatial>
 using DeviceConvBwdWeightInstance =
-    ck::tensor_operation::device::DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<
-        NDimSpatial,          // NDimSpatial
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle<
+        NDimSpatial,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                      ck::tensor_layout::convolution::GNHWC,
+                                      ck::tensor_layout::convolution::GNDHWC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                      ck::tensor_layout::convolution::GKYXC,
+                                      ck::tensor_layout::convolution::GKZYXC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                      ck::tensor_layout::convolution::GNHWK,
+                                      ck::tensor_layout::convolution::GNDHWK>>,
         InDataType,           // InDataType
         WeiDataType,          // WeiDataType
         OutDataType,          // OutDataType
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
index 966b586767..69c831cc54 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 
 using InDataType  = F16;
 using WeiDataType = F16;
@@ -16,8 +16,20 @@ using OutElementOp = PassThrough;
 
 template <ck::index_t NDimSpatial>
 using DeviceConvBwdWeightInstance =
-    ck::tensor_operation::device::DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<
-        NDimSpatial,          // NDimSpatial
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle<
+        NDimSpatial,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                      ck::tensor_layout::convolution::GNHWC,
+                                      ck::tensor_layout::convolution::GNDHWC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                      ck::tensor_layout::convolution::GKYXC,
+                                      ck::tensor_layout::convolution::GKZYXC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                      ck::tensor_layout::convolution::GNHWK,
+                                      ck::tensor_layout::convolution::GNDHWK>>,
         InDataType,           // InDataType
         WeiDataType,          // WeiDataType
         OutDataType,          // OutDataType
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 39b9100bf8..49bd9fc7f0 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -75,6 +75,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> output_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -85,6 +87,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
     range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
     range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+    range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+    range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
     range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
     range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
     range_copy(conv_param.input_left_pads_, begin(input_left_pads));
@@ -103,6 +107,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
                                       input_spatial_lengths,
                                       filter_spatial_lengths,
                                       output_spatial_lengths,
+                                      input_strides,
+                                      output_strides,
                                       conv_filter_strides,
                                       conv_filter_dilations,
                                       input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
index de54f9be20..a103bdff35 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
@@ -27,17 +27,19 @@ struct DeviceGroupedConvBwdWeight : public BaseOperator
     MakeArgumentPointer(const void* p_in,
                         void* p_wei,
                         const void* p_out,
-                        ck::index_t G,
-                        ck::index_t N,
-                        ck::index_t K,
-                        ck::index_t C,
-                        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                        std::array<ck::index_t, NDimSpatial> input_left_pads,
-                        std::array<ck::index_t, NDimSpatial> input_right_pads,
+                        const ck::index_t G,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t C,
+                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                         InElementwiseOperation in_element_op,
                         WeiElementwiseOperation wei_element_op,
                         OutElementwiseOperation out_element_op,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
index 0473eaf762..003a508f07 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
@@ -195,17 +195,17 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -347,17 +347,17 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     } // function end
     template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -515,17 +515,17 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -784,17 +784,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         Argument(const InDataType* p_in_grid,
                  WeiDataType* p_wei_grid,
                  const OutDataType* p_out_grid,
-                 ck::index_t G,
-                 ck::index_t N,
-                 ck::index_t K,
-                 ck::index_t C,
-                 std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                 std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                 std::array<ck::index_t, NDimSpatial> input_left_pads,
-                 std::array<ck::index_t, NDimSpatial> input_right_pads,
+                 const ck::index_t G,
+                 const ck::index_t N,
+                 const ck::index_t K,
+                 const ck::index_t C,
+                 const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial + 3>& /*input_strides*/,
+                 const std::array<ck::index_t, NDimSpatial + 3>& /*output_strides*/,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                 const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                  InElementwiseOperation in_element_op,
                  WeiElementwiseOperation wei_element_op,
                  OutElementwiseOperation out_element_op,
@@ -897,18 +899,18 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         InElementwiseOperation c_element_op_;
 
         // for checking IsSupportedArgument()
-        index_t Conv_G_;
-        index_t Conv_N_;
-        index_t Conv_K_;
-        index_t Conv_C_;
+        const index_t Conv_G_;
+        const index_t Conv_N_;
+        const index_t Conv_K_;
+        const index_t Conv_C_;
 
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides_;
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations_;
-        std::array<ck::index_t, NDimSpatial> input_left_pads_;
-        std::array<ck::index_t, NDimSpatial> input_right_pads_;
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations_;
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads_;
         index_t k_batch_;
     };
 
@@ -1111,17 +1113,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     static auto MakeArgument(const InDataType* p_in_grid,
                              WeiDataType* p_wei_grid,
                              const OutDataType* p_out_grid,
-                             ck::index_t G,
-                             ck::index_t N,
-                             ck::index_t K,
-                             ck::index_t C,
-                             std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                             std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                             std::array<ck::index_t, NDimSpatial> input_left_pads,
-                             std::array<ck::index_t, NDimSpatial> input_right_pads,
+                             const ck::index_t G,
+                             const ck::index_t N,
+                             const ck::index_t K,
+                             const ck::index_t C,
+                             const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                             const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                             const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                             const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                              InElementwiseOperation in_element_op,
                              WeiElementwiseOperation wei_element_op,
                              OutElementwiseOperation out_element_op,
@@ -1137,6 +1141,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                         input_spatial_lengths,
                         filter_spatial_lengths,
                         output_spatial_lengths,
+                        input_strides,
+                        output_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -1153,17 +1159,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     MakeArgumentPointer(const void* p_in_grid,
                         void* p_wei_grid,
                         const void* p_out_grid,
-                        ck::index_t G,
-                        ck::index_t N,
-                        ck::index_t K,
-                        ck::index_t C,
-                        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                        std::array<ck::index_t, NDimSpatial> input_left_pads,
-                        std::array<ck::index_t, NDimSpatial> input_right_pads,
+                        const ck::index_t G,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t C,
+                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                         InElementwiseOperation in_element_op,
                         WeiElementwiseOperation wei_element_op,
                         OutElementwiseOperation out_element_op,
@@ -1179,6 +1187,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                                           input_spatial_lengths,
                                           filter_spatial_lengths,
                                           output_spatial_lengths,
+                                          input_strides,
+                                          output_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
similarity index 70%
rename from include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
index 4f4e0d576a..7475dc414d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -126,6 +126,9 @@ __global__ void
 
 // out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
 template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout,
           typename InDataType,
           typename WeiDataType,
           typename OutDataType,
@@ -161,29 +164,19 @@ template <ck::index_t NDimSpatial,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CBlockTransferScalarPerVector_NWaveNPerXdl>
-struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
-    : public DeviceGroupedConvBwdWeight<
-          NDimSpatial,
-          ck::tuple_element_t<NDimSpatial - 1,
-                              ck::Tuple<ck::tensor_layout::convolution::GNWC,
-                                        ck::tensor_layout::convolution::GNHWC,
-                                        ck::tensor_layout::convolution::GNDHWC>>,
-          ck::tuple_element_t<NDimSpatial - 1,
-                              ck::Tuple<ck::tensor_layout::convolution::GKXC,
-                                        ck::tensor_layout::convolution::GKYXC,
-                                        ck::tensor_layout::convolution::GKZYXC>>,
-          ck::tuple_element_t<NDimSpatial - 1,
-                              ck::Tuple<ck::tensor_layout::convolution::GNWK,
-                                        ck::tensor_layout::convolution::GNHWK,
-                                        ck::tensor_layout::convolution::GNDHWK>>,
-          InDataType,
-          WeiDataType,
-          OutDataType,
-          InElementwiseOperation,
-          WeiElementwiseOperation,
-          OutElementwiseOperation>
+struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
+    : public DeviceGroupedConvBwdWeight<NDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        OutDataType,
+                                        InElementwiseOperation,
+                                        WeiElementwiseOperation,
+                                        OutElementwiseOperation>
 {
-    using DeviceOp = DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle;
+    using DeviceOp = DeviceGroupedConvBwdWeight_Xdl_CShuffle;
 
     using ADataType = OutDataType;
     using BDataType = InDataType;
@@ -222,17 +215,19 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -282,14 +277,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
             const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
                 in_gemmktotal_gemmn_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
+                           make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0>{}, Sequence<1>{}));
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
                 in_gemmkpad_gemmn_grid_desc,
                 make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                           make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
@@ -372,19 +367,25 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         }
     }
 
-    template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
+    template <ck::index_t NDim,
+              typename ck::enable_if<NDim == 2 &&
+                                         is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
+                                         is_same_v<OutLayout, tensor_layout::convolution::GNHWK>,
+                                     bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -447,14 +448,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
             const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
                 in_gemmktotal_gemmn_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
+                           make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0>{}, Sequence<1>{}));
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
                 in_gemmkpad_gemmn_grid_desc,
                 make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                           make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
@@ -539,19 +540,202 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         }
     }
 
+    template <ck::index_t NDim,
+              typename ck::enable_if<NDim == 2 &&
+                                         is_same_v<InLayout, tensor_layout::convolution::NHWGC> &&
+                                         is_same_v<OutLayout, tensor_layout::convolution::NHWGK>,
+                                     bool>::type = false>
+    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
+    {
+        using namespace ck;
+
+        const index_t Hi = input_spatial_lengths[0];
+        const index_t Wi = input_spatial_lengths[1];
+
+        const index_t Ho = output_spatial_lengths[0];
+        const index_t Wo = output_spatial_lengths[1];
+
+        const index_t Y = filter_spatial_lengths[0];
+        const index_t X = filter_spatial_lengths[1];
+
+        const index_t ConvStrideH = conv_filter_strides[0];
+        const index_t ConvStrideW = conv_filter_strides[1];
+
+        const index_t ConvDilationH = conv_filter_dilations[0];
+        const index_t ConvDilationW = conv_filter_dilations[1];
+
+        const index_t InLeftPadH = input_left_pads[0];
+        const index_t InLeftPadW = input_left_pads[1];
+
+        const index_t InRightPadH = input_right_pads[0];
+        const index_t InRightPadW = input_right_pads[1];
+
+        const index_t GemmKTotal = N * Ho * Wo;
+        const index_t GemmM      = K;
+        const index_t GemmN      = C * X * Y;
+
+        const index_t NStride  = input_strides[1];
+        const index_t HiStride = input_strides[3];
+        const index_t WiStride = input_strides[4];
+        const auto CStride     = input_strides[2];
+
+        const index_t WoStride = output_strides[4];
+        const auto KStride     = Number<1>{};
+
+        const index_t GemmKBatch = batch_k;
+        const index_t GemmK0 =
+            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
+            K0PerBlock;
+        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // A: output tensor
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Ho * Wo, K), make_tuple(WoStride, KStride));
+
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_gemmktotal_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Hi * Wi, C), make_tuple(WiStride, CStride));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+        else
+        {
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Ho * Wo, K), make_tuple(WoStride, KStride));
+            const auto in_n_hi_wi_c_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N, Hi, Wi, C), make_tuple(NStride, HiStride, WiStride, CStride));
+
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_gemmktotal_gemmm_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // B: input tensor
+            const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                in_n_hi_wi_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                in_n_hip_wip_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
+                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
+
+            const auto in_gemmktotal_gemmn_grid_desc =
+                transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
+                                            make_tuple(make_merge_transform(make_tuple(Y, X, C)),
+                                                       make_merge_transform(make_tuple(N, Ho, Wo))),
+                                            make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
+                                            make_tuple(Sequence<1>{}, Sequence<0>{}));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+
+            // C: weight tensor
+            const auto wei_gemmm_gemmn_grid_desc =
+                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_gemmm_gemmn_grid_desc);
+        }
+    }
+
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        ck::index_t N,
-        ck::index_t K,
-        ck::index_t C,
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-        std::array<ck::index_t, NDimSpatial> input_left_pads,
-        std::array<ck::index_t, NDimSpatial> input_right_pads,
-        ck::index_t batch_k)
+        const ck::index_t N,
+        const ck::index_t K,
+        const ck::index_t C,
+        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+        const ck::index_t batch_k)
     {
         using namespace ck;
 
@@ -621,14 +805,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
             const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
                 in_gemmktotal_gemmn_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
+                           make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0>{}, Sequence<1>{}));
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
                 in_gemmkpad_gemmn_grid_desc,
                 make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                           make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
@@ -725,31 +909,70 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto GetABCGridDesc()
     {
-        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<1>(
-            1, 1, 1, {1}, {1}, {1}, {1}, {1}, {1}, {1}, 1);
+        const ck::index_t dim   = 1;
+        const ck::index_t batch = 1;
+        const std::array<ck::index_t, NDimSpatial> lengths{1};
+        const std::array<ck::index_t, NDimSpatial + 3> strides{1, 1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial> params{1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<1>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  batch);
     }
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
     static auto GetABCGridDesc()
     {
-        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<2>(
-            1, 1, 1, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, 1);
+        const ck::index_t dim   = 1;
+        const ck::index_t batch = 1;
+        const std::array<ck::index_t, NDimSpatial> lengths{1, 1};
+        const std::array<ck::index_t, NDimSpatial + 3> strides{1, 1, 1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial> params{1, 1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<2>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  batch);
     }
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     static auto GetABCGridDesc()
     {
-        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<3>(1,
-                                                                  1,
-                                                                  1,
-                                                                  {1, 1, 1},
-                                                                  {1, 1, 1},
-                                                                  {1, 1, 1},
-                                                                  {1, 1, 1},
-                                                                  {1, 1, 1},
-                                                                  {1, 1, 1},
-                                                                  {1, 1, 1},
-                                                                  1);
+        const ck::index_t dim   = 1;
+        const ck::index_t batch = 1;
+        const std::array<ck::index_t, NDimSpatial> lengths{1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial + 3> strides{1, 1, 1, 1, 1, 1};
+        const std::array<ck::index_t, NDimSpatial> params{1, 1, 1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<3>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  batch);
     }
 
     // type convert descs
@@ -863,19 +1086,21 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         Argument(const InDataType* p_in_grid,
                  WeiDataType* p_wei_grid,
                  const OutDataType* p_out_grid,
-                 ck::index_t G,
-                 ck::index_t N,
-                 ck::index_t K,
-                 ck::index_t C,
-                 std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                 std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                 std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                 std::array<ck::index_t, NDimSpatial> input_left_pads,
-                 std::array<ck::index_t, NDimSpatial> input_right_pads,
-                 ck::index_t M01,
-                 ck::index_t N01,
+                 const ck::index_t G,
+                 const ck::index_t N,
+                 const ck::index_t K,
+                 const ck::index_t C,
+                 const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                 const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                 const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                 const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+                 const ck::index_t M01,
+                 const ck::index_t N01,
                  InElementwiseOperation in_element_op,
                  WeiElementwiseOperation wei_element_op,
                  OutElementwiseOperation out_element_op,
@@ -913,6 +1138,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                     input_spatial_lengths,
                     filter_spatial_lengths,
                     output_spatial_lengths,
+                    input_strides,
+                    output_strides,
                     conv_filter_strides,
                     conv_filter_dilations,
                     input_left_pads,
@@ -927,18 +1154,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                 GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
 
             // A/B/C Batch Stride
-            compute_ptr_offset_of_batch_.BatchStrideA_ =
-                N * K *
-                std::accumulate(begin(output_spatial_lengths),
-                                end(output_spatial_lengths),
-                                index_t{1},
-                                std::multiplies<>{});
-            compute_ptr_offset_of_batch_.BatchStrideB_ =
-                N * C *
-                std::accumulate(begin(input_spatial_lengths),
-                                end(input_spatial_lengths),
-                                index_t{1},
-                                std::multiplies<>{});
+            compute_ptr_offset_of_batch_.BatchStrideA_ = output_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = input_strides[0];
             compute_ptr_offset_of_batch_.BatchStrideC_ =
                 K * C *
                 std::accumulate(begin(filter_spatial_lengths),
@@ -977,16 +1194,16 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         WeiElementwiseOperation c_element_op_;
 
         // for checking IsSupportedArgument()
-        index_t Conv_G_;
-        index_t Conv_N_;
-        index_t Conv_K_;
-        index_t Conv_C_;
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides_;
-        std::array<ck::index_t, NDimSpatial> input_left_pads_;
-        std::array<ck::index_t, NDimSpatial> input_right_pads_;
-        index_t k_batch_;
+        const index_t Conv_G_;
+        const index_t Conv_N_;
+        const index_t Conv_K_;
+        const index_t Conv_C_;
+        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths_;
+        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
+        const std::array<ck::index_t, NDimSpatial>& input_right_pads_;
+        const index_t k_batch_;
     };
 
     // Invoker
@@ -1091,6 +1308,45 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if constexpr(NDimSpatial == 1)
+        {
+            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNWC> &&
+                           is_same_v<WeiLayout, tensor_layout::convolution::GKXC> &&
+                           is_same_v<OutLayout, tensor_layout::convolution::GNWK>))
+            {
+                return false;
+            }
+        }
+        else if constexpr(NDimSpatial == 2)
+        {
+            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
+                           is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
+                           is_same_v<OutLayout,
+                                     tensor_layout::convolution::
+                                         GNHWK>)&&!(is_same_v<InLayout,
+                                                              tensor_layout::convolution::NHWGC> &&
+                                                    is_same_v<WeiLayout,
+                                                              tensor_layout::convolution::GKYXC> &&
+                                                    is_same_v<OutLayout,
+                                                              tensor_layout::convolution::NHWGK>))
+            {
+                return false;
+            }
+        }
+        else if constexpr(NDimSpatial == 3)
+        {
+            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNDHWC> &&
+                           is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+                           is_same_v<OutLayout, tensor_layout::convolution::GNDHWK>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
@@ -1134,21 +1390,23 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
     static auto MakeArgument(const InDataType* p_in_grid,
                              WeiDataType* p_wei_grid,
                              const OutDataType* p_out_grid,
-                             ck::index_t G,
-                             ck::index_t N,
-                             ck::index_t K,
-                             ck::index_t C,
-                             std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                             std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                             std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                             std::array<ck::index_t, NDimSpatial> input_left_pads,
-                             std::array<ck::index_t, NDimSpatial> input_right_pads,
+                             const ck::index_t G,
+                             const ck::index_t N,
+                             const ck::index_t K,
+                             const ck::index_t C,
+                             const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                             const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                             const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                             const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                             const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                             const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                              InElementwiseOperation in_element_op,
                              WeiElementwiseOperation wei_element_op,
                              OutElementwiseOperation out_element_op,
-                             ck::index_t split_k)
+                             const ck::index_t split_k)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
@@ -1160,6 +1418,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                         input_spatial_lengths,
                         filter_spatial_lengths,
                         output_spatial_lengths,
+                        input_strides,
+                        output_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -1178,21 +1438,23 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
     MakeArgumentPointer(const void* p_in_grid,
                         void* p_wei_grid,
                         const void* p_out_grid,
-                        ck::index_t G,
-                        ck::index_t N,
-                        ck::index_t K,
-                        ck::index_t C,
-                        std::array<ck::index_t, NDimSpatial> input_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> output_spatial_lengths,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_strides,
-                        std::array<ck::index_t, NDimSpatial> conv_filter_dilations,
-                        std::array<ck::index_t, NDimSpatial> input_left_pads,
-                        std::array<ck::index_t, NDimSpatial> input_right_pads,
+                        const ck::index_t G,
+                        const ck::index_t N,
+                        const ck::index_t K,
+                        const ck::index_t C,
+                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
                         InElementwiseOperation in_element_op,
                         WeiElementwiseOperation wei_element_op,
                         OutElementwiseOperation out_element_op,
-                        ck::index_t split_k) override
+                        const ck::index_t split_k) override
     {
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
                                           static_cast<WeiDataType*>(p_wei_grid),
@@ -1204,6 +1466,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
                                           input_spatial_lengths,
                                           filter_spatial_lengths,
                                           output_spatial_lengths,
+                                          input_strides,
+                                          output_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
@@ -1226,7 +1490,7 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle"
+        str << "DeviceGroupedConvBwdWeight_Xdl_CShuffle"
             << "<"
             << BlockSize << ", "
             << MPerBlock << ", "
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
new file mode 100644
index 0000000000..0f44952810
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdWeightDefault =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
+
+static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances = std::tuple<
+    // clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances = std::tuple<
+    // clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances = std::tuple<
+    // clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index 377ce083c7..d0c3338f24 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -91,6 +91,42 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
 // conv3d backward weight
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
@@ -162,66 +198,103 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
-        if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC> &&
-                     is_same_v<WeiLayout, GKXC> && is_same_v<OutLayout, GNWK>)
+        if constexpr(NumDimSpatial == 1)
         {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+            if constexpr(is_same_v<InLayout, GNWC> && is_same_v<WeiLayout, GKXC> &&
+                         is_same_v<OutLayout, GNWK>)
             {
-                add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
-            {
-                add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
-            {
-                add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
-                    op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+        else if constexpr(NumDimSpatial == 2)
         {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+            if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
+                         is_same_v<OutLayout, GNHWK>)
             {
-                add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+            else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
+                              is_same_v<OutLayout, NHWGK>)
             {
-                add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
-            {
-                add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
-                    op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
-                          is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
+        else if constexpr(NumDimSpatial == 3)
         {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
-                         is_same_v<OutDataType, float>)
+            if(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
+               is_same_v<OutLayout, GNDHWK>)
             {
-                add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
-                    op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
-            {
-                add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
-                    op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
-            {
-                add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
-                    op_ptrs);
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index ccbfaeaf4f..25ea4f48cf 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -5,81 +5,17 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = bhalf_t;
-using F32  = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
-using device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -92,12 +28,22 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_insta
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_bf16_f32_bf16_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_bf16_f32_bf16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<1,
+                                                                    GNWC,
+                                                                    GKXC,
+                                                                    GNWK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       1,
+                                       GNWC,
+                                       GKXC,
+                                       GNWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index e10de67f94..8065012f12 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -5,80 +5,17 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
-using device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f16_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-using device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f16_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -91,11 +28,22 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f16_default_instances{});
-    add_device_operation_instances(
-        instances, device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<1,
+                                                                   GNWC,
+                                                                   GKXC,
+                                                                   GNWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       1,
+                                       GNWC,
+                                       GKXC,
+                                       GNWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index 8b47e82f6d..c70a54c2dc 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -5,79 +5,17 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, wi, c] * wei[k, x, c] = out[n, wo, k]
-using device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f32_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f32_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       1,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -90,11 +28,22 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv1d_bwd_weight_xdl_c_shuffle_gnwc_gkxc_gnwk_f32_default_instances{});
-    add_device_operation_instances(
-        instances, device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_1x1_s1_p0_f32_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<1,
+                                                                   GNWC,
+                                                                   GKXC,
+                                                                   GNWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       1,
+                                       GNWC,
+                                       GKXC,
+                                       GNWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
index 4009121e7f..b7b9fc92d1 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
@@ -2,5 +2,8 @@ add_instance_library(device_grouped_conv2d_bwd_weight_instance
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
 )
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index c2c0fc553c..cf39c86010 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -1,85 +1,16 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = bhalf_t;
-using F32  = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
-using device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -92,12 +23,22 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<2,
+                                                                    GNHWC,
+                                                                    GKYXC,
+                                                                    GNHWK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 5be7443eca..d52f0b4d1c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -1,84 +1,16 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
-using device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f16_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-using device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f16_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -91,12 +23,22 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f16_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<2,
+                                                                   GNHWC,
+                                                                   GKYXC,
+                                                                   GNHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 2828b432ff..62547a5014 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -1,83 +1,16 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, hi, wi, c] * wei[k, y, x, c] = out[n, ho, wo, k]
-using device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f32_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f32_instances = std::tuple<
-    // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       2,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-    // clang-format on
-    >;
-
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -90,12 +23,22 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_weight_xdl_c_shuffle_gnhwc_gkyxc_gnhwk_f32_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_1x1_s1_p0_f32_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<2,
+                                                                   GNHWC,
+                                                                   GKYXC,
+                                                                   GNHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       GNHWC,
+                                       GKYXC,
+                                       GNHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..1cb9991a86
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<2,
+                                                                    NHWGC,
+                                                                    GKYXC,
+                                                                    NHWGK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..e64d55c3bb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<2,
+                                                                   NHWGC,
+                                                                   GKYXC,
+                                                                   NHWGK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..5fa4c9ba3c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<2,
+                                                                   NHWGC,
+                                                                   GKYXC,
+                                                                   NHWGK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index 7ae87eed5d..c8f456db83 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -5,81 +5,17 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = bhalf_t;
-using F32  = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
-using device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,               ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_bf16_f32_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                       ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|   BBlockTransfer|  BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                             Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                     Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          |  Lengths_K0_N_K1|    ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                   |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                 |                |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,  ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -92,12 +28,22 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_bf16_f32_bf16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<3,
+                                                                    GNDHWC,
+                                                                    GKZYXC,
+                                                                    GNDHWK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       GNDHWC,
+                                       GKZYXC,
+                                       GNDHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index ab07341b59..099123ecbc 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -5,81 +5,17 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
-using device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f16_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-using device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f16_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8>,   
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -92,12 +28,22 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f16_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f16_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<3,
+                                                                   GNDHWC,
+                                                                   GKZYXC,
+                                                                   GNDHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       GNDHWC,
+                                       GKZYXC,
+                                       GNDHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index 15045bedd8..0eda980b30 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -5,80 +5,17 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdWeightDefault =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
-
-static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
-    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
-
-// Compilation parameters for in[n, di, hi, wi, c] * wei[k, z, y, x, c] = out[n, do, ho, wo, k]
-using device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f32_default_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,              ConvBwdWeightDefault,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
-using device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f32_instances =
-    std::tuple<
-        // clang-format off
-        //#########################################|     Num| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    64,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,   128,    64,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   256,    64,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              1,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,   128,    32,   128,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              1,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    64,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
-        DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Xdl_CShuffle<       3,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough, ConvBwdWeightFilter1x1Stride1Pad0,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -91,12 +28,22 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances)
 {
+    // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv3d_bwd_weight_xdl_c_shuffle_gndhwc_gkzyxc_gndhwk_f32_default_instances{});
-    add_device_operation_instances(
-        instances,
-        device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_1x1_s1_p0_f32_instances{});
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<3,
+                                                                   GNDHWC,
+                                                                   GKZYXC,
+                                                                   GNDHWK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       GNDHWC,
+                                       GKZYXC,
+                                       GNDHWK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/profiler/README.md b/profiler/README.md
index 3ee039cb3f..a18ebc2265 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -141,3 +141,46 @@ avg_time: 0.768321
 tflops: 86.6679
 GB/s: 127.947
 ```
+
+## Profile grouped convolution backward weight kernels
+```bash
+# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
+# arg2: data type (0: Input fp32, Weight fp32, Output fp32
+#                  1: Input fp16, Weight fp16, Output fp16
+#                  2: Input bf16, Weight fp32, Output bf16)
+# arg3: tensor layout (0: Input[G, N, C, Hi, Wi], Weight[G, K, C, Y, X], Output[G, N, K, Ho, Wo]
+#                      1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, N, Ho, Wo, K]
+#                      2: Input[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Output[N, Ho, Wo, G, K]
+# arg4: verification (0: no, 1: yes)
+# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
+# arg6: print tensor value (0: no; 1: yes)
+# arg7: time kernel (0: no, 1: yes)
+# Following arguments (depending on number of spatial dims):
+#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  G, N, K, C, 
+#  <filter spatial dimensions>, (ie Y, X for 2D)
+#  <input image spatial dimensions>, (ie Hi, Wi for 2D)
+#  <strides>, (ie Sy, Sx for 2D)
+#  <dilations>, (ie Dy, Dx for 2D)
+#  <left padding>, (ie LeftPy, LeftPx for 2D)
+#  <right padding>, (ie RightPy, RightPx for 2D)
+# SplitK
+
+ ################                   op   datatype  layout  verify  init  log  time  Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx  SplitK
+./bin/ckProfiler grouped_conv_bwd_data          1       0       1     1    0     1      2 32 256 256 512  3  3  28  28   1   1   1   1       1       0        0        0       1
+
+ ```
+
+Result (MI100, FP16, GNHWC_GKYXC_GNHWK)
+```
+input: dim 5, lengths {32, 512, 1024, 28, 28}, strides {411041792, 802816, 1, 28672, 1024}
+weight: dim 5, lengths {32, 512, 1024, 3, 3}, strides {4718592, 9216, 1, 3072, 1024}
+output: dim 5, lengths {32, 512, 512, 26, 26}, strides {177209344, 346112, 1, 13312, 512}
+....
+Best configuration parameters:
+name: DeviceGroupedConvBwdWeight_Xdl_CShuffle<256, 256, 128, 4, Default, 8, 4, 2, 8, 4, 8, 2, 1, 1, 8>
+avg_time: 68.5216
+tflops: 95.337
+GB/s: 69.2301
+```
+Note: This kernel use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
index dc6739773f..3a4295eeb3 100644
--- a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
@@ -139,6 +139,8 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> output_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -149,6 +151,8 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
     range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
     range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
     range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+    range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+    range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
     range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
     range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
     range_copy(conv_param.input_left_pads_, begin(input_left_pads));
@@ -167,6 +171,8 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
                                         input_spatial_lengths,
                                         filter_spatial_lengths,
                                         output_spatial_lengths,
+                                        input_strides,
+                                        output_strides,
                                         conv_filter_strides,
                                         conv_filter_dilations,
                                         input_left_pads,
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index 7a062ed519..d8edcf0f65 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -15,6 +15,7 @@ enum struct ConvLayout
 {
     GNCHW_GKCYX_GNKHW, // 0
     GNHWC_GKYXC_GNHWK, // 1
+    NHWGC_GKYXC_NHWGK, // 2
 };
 
 enum struct ConvDataType
@@ -37,6 +38,8 @@ static void print_helper_msg()
                  "N, K, Ho, Wo]\n"
               << "                     1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, "
                  "N, Ho, Wo, K]\n"
+              << "                     2: Input[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Output[N, "
+                 "Ho, Wo, G, K]\n"
               << "arg4: verification (0: no, 1: yes)\n"
               << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
               << "arg6: print tensor value (0: no; 1: yes)\n"
@@ -82,6 +85,7 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
 
     using GNWC   = ck::tensor_layout::convolution::GNWC;
     using GNHWC  = ck::tensor_layout::convolution::GNHWC;
+    using NHWGC  = ck::tensor_layout::convolution::NHWGC;
     using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
 
     using GKXC   = ck::tensor_layout::convolution::GKXC;
@@ -90,6 +94,7 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
 
     using GNWK   = ck::tensor_layout::convolution::GNWK;
     using GNHWK  = ck::tensor_layout::convolution::GNHWK;
+    using NHWGK  = ck::tensor_layout::convolution::NHWGK;
     using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
 
     constexpr auto I1 = ck::Number<1>{};
@@ -157,6 +162,22 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
             return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, BF16{}, F32{}, BF16{});
         }
     }
+    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_F32_BF16)
+        {
+            // fp32 atomic add is used for weight tensor in bf16 kernel
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, BF16{}, F32{}, BF16{});
+        }
+    }
     else if(num_dim_spatial == 3 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
diff --git a/test/grouped_convnd_bwd_weight/CMakeLists.txt b/test/grouped_convnd_bwd_weight/CMakeLists.txt
index 8be872bc69..f1a3f04b77 100644
--- a/test/grouped_convnd_bwd_weight/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_weight/CMakeLists.txt
@@ -2,8 +2,10 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_gtest_executable(test_grouped_convnd_bwd_weight grouped_convnd_bwd_weight.cpp)
+   add_gtest_executable(test_grouped_convnd_bwd_weight test_grouped_convnd_bwd_weight.cpp)
    target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+   add_gtest_executable(test_grouped_convnd_bwd_weight_interface test_grouped_convnd_bwd_weight_interface.cpp)
+   target_link_libraries(test_grouped_convnd_bwd_weight_interface PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
deleted file mode 100644
index 207cdab7c7..0000000000
--- a/test/grouped_convnd_bwd_weight/grouped_convnd_bwd_weight.cpp
+++ /dev/null
@@ -1,91 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-#include <iostream>
-#include <initializer_list>
-#include <tuple>
-#include <vector>
-
-#include <gtest/gtest.h>
-
-#include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
-
-template <typename Tuple>
-class TestGroupedConvndBwdWeight : public ::testing::Test
-{
-    protected:
-    using DataType = std::tuple_element_t<0, Tuple>;
-    std::vector<ck::utils::conv::ConvParam> conv_params;
-    ck::index_t split_k{2};
-
-    template <ck::index_t NDimSpatial>
-    void Run()
-    {
-        for(auto& param : conv_params)
-        {
-            bool pass;
-            EXPECT_FALSE(conv_params.empty());
-            pass = ck::profiler::profile_grouped_conv_bwd_weight_impl<
-                NDimSpatial,
-                ck::tuple_element_t<NDimSpatial - 1,
-                                    ck::Tuple<ck::tensor_layout::convolution::GNWC,
-                                              ck::tensor_layout::convolution::GNHWC,
-                                              ck::tensor_layout::convolution::GNDHWC>>,
-                ck::tuple_element_t<NDimSpatial - 1,
-                                    ck::Tuple<ck::tensor_layout::convolution::GKXC,
-                                              ck::tensor_layout::convolution::GKYXC,
-                                              ck::tensor_layout::convolution::GKZYXC>>,
-                ck::tuple_element_t<NDimSpatial - 1,
-                                    ck::Tuple<ck::tensor_layout::convolution::GNWK,
-                                              ck::tensor_layout::convolution::GNHWK,
-                                              ck::tensor_layout::convolution::GNDHWK>>,
-                DataType,
-                DataType,
-                DataType>(true,  // do_verification
-                          1,     // init_method: integer value
-                          false, // do_log
-                          false, // time_kernel
-                          param,
-                          split_k);
-            EXPECT_TRUE(pass);
-        }
-    }
-};
-
-using KernelTypes =
-    ::testing::Types<std::tuple<float>, std::tuple<ck::half_t>, std::tuple<ck::bhalf_t>>;
-TYPED_TEST_SUITE(TestGroupedConvndBwdWeight, KernelTypes);
-
-TYPED_TEST(TestGroupedConvndBwdWeight, Test1D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
-    this->conv_params.push_back({1, 2, 32, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
-    this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
-    this->template Run<1>();
-}
-
-TYPED_TEST(TestGroupedConvndBwdWeight, Test2D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back(
-        {2, 2, 64, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
-    this->conv_params.push_back(
-        {2, 2, 4, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    this->conv_params.push_back(
-        {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
-    this->template Run<2>();
-}
-
-TYPED_TEST(TestGroupedConvndBwdWeight, Test3D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back(
-        {3, 2, 16, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    this->conv_params.push_back(
-        {3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-    this->conv_params.push_back(
-        {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    this->template Run<3>();
-}
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
new file mode 100644
index 0000000000..b6c79fa80b
--- /dev/null
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -0,0 +1,125 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+#include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight : public ::testing::Test
+{
+    protected:
+    using InDataType  = std::tuple_element_t<0, Tuple>;
+    using WeiDataType = std::tuple_element_t<1, Tuple>;
+    using OutDataType = std::tuple_element_t<2, Tuple>;
+    using InLayout    = std::tuple_element_t<3, Tuple>;
+    using WeiLayout   = std::tuple_element_t<4, Tuple>;
+    using OutLayout   = std::tuple_element_t<5, Tuple>;
+    using NDimSpatial = std::tuple_element_t<6, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+    ck::index_t split_k{2};
+
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial{},
+                                                                              InLayout,
+                                                                              WeiLayout,
+                                                                              OutLayout,
+                                                                              InDataType,
+                                                                              WeiDataType,
+                                                                              OutDataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param,
+                               split_k);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight1d : public TestGroupedConvndBwdWeight<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight2d : public TestGroupedConvndBwdWeight<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeight3d : public TestGroupedConvndBwdWeight<Tuple>
+{
+};
+
+using namespace ck::tensor_layout::convolution;
+
+using KernelTypes1d = ::testing::Types<
+    std::tuple<float, float, float, GNWC, GKXC, GNWK, ck::Number<1>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, GNWC, GKXC, GNWK, ck::Number<1>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNWC, GKXC, GNWK, ck::Number<1>>>;
+using KernelTypes2d = ::testing::Types<
+    std::tuple<float, float, float, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
+    std::tuple<float, float, float, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, NHWGC, GKYXC, NHWGK, ck::Number<2>>>;
+using KernelTypes3d = ::testing::Types<
+    std::tuple<float, float, float, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>>;
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeight1d, KernelTypes1d);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeight2d, KernelTypes2d);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeight3d, KernelTypes3d);
+
+TYPED_TEST(TestGroupedConvndBwdWeight1d, Test1D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 2, 32, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
+    this->Run();
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeight2d, Test2D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {2, 2, 64, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 2, 4, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->Run();
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeight3d, Test3D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {3, 2, 16, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->Run();
+}
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
new file mode 100644
index 0000000000..57d4e4186c
--- /dev/null
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
@@ -0,0 +1,180 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using F16         = ck::half_t;
+using F32         = float;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvolutionBackwardWeightSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization;
+
+static constexpr auto ConvBwdWeightDefault = ConvolutionBackwardWeightSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 =
+    ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvolutionBackwardWeightSpecialization ConvSpec>
+class TestGroupedConvndBwdWeight : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 2;
+
+    using InLayout  = std::tuple_element_t<2, Tuple>;
+    using WeiLayout = std::tuple_element_t<1, Tuple>;
+    using OutLayout = std::tuple_element_t<0, Tuple>;
+
+    // clang-format off
+    using GroupedConvBwdWeightDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle
+        //##########|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //##########|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //##########| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //##########|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        <         NDimSpatial,  InLayout, WeiLayout,OutLayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+    ck::index_t split_k{2};
+
+    template <ck::index_t NDimSpatial>
+    bool Run()
+    {
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> output_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
+
+        range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
+        range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
+        range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+        range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+        range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
+        range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
+        range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
+        range_copy(conv_param.input_left_pads_, begin(input_left_pads));
+        range_copy(conv_param.input_right_pads_, begin(input_right_pads));
+
+        auto conv = GroupedConvBwdWeightDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          nullptr,
+                                          conv_param.G_,
+                                          conv_param.N_,
+                                          conv_param.K_,
+                                          conv_param.C_,
+                                          input_spatial_lengths,
+                                          filter_spatial_lengths,
+                                          output_spatial_lengths,
+                                          input_strides,
+                                          output_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          PassThrough{},
+                                          PassThrough{},
+                                          PassThrough{},
+                                          split_k);
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes =
+    ::testing::Types<std::tuple<GNHWK, GKYXC, GNHWC>, std::tuple<NHWGK, GKYXC, NHWGC>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightDefault
+    : public TestGroupedConvndBwdWeight<Tuple, ConvBwdWeightDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightFilter1x1
+    : public TestGroupedConvndBwdWeight<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightDefault, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightFilter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdWeightFilter1x1, SpecializationCheck)
+{
+    // Check filter 3,3 instead of 1,1
+    this->conv_param  = {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2,2 instead of 1,1
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeightDefault, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param  = {2, 2, 128, 129, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {2, 2, 128, 128, 257, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}

From 03d3395b3cb70e6a47d0c1774dbeb258e6b29aad Mon Sep 17 00:00:00 2001
From: arvindcheru <90783369+arvindcheru@users.noreply.github.com>
Date: Fri, 14 Jul 2023 20:00:20 -0400
Subject: [PATCH 059/194] Disable Werror to ignore xnack+ warnings (#794)

* Disable Werror to ignore xnack+ warnings
---
 cmake/EnableCompilerWarnings.cmake | 1 +
 1 file changed, 1 insertion(+)

diff --git a/cmake/EnableCompilerWarnings.cmake b/cmake/EnableCompilerWarnings.cmake
index 369cd0b54c..66139cc710 100644
--- a/cmake/EnableCompilerWarnings.cmake
+++ b/cmake/EnableCompilerWarnings.cmake
@@ -67,6 +67,7 @@ else()
             -Wunused
             -Wno-reserved-identifier
             -Werror
+            -Wno-option-ignored
             -Wsign-compare
             -Wno-extra-semi-stmt
         )

From 4867db429067bb0822b2b4a8d286b2f7a24a7e86 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 17 Jul 2023 09:44:40 -0700
Subject: [PATCH 060/194] Add check for compiler GPU target support. (#800)

* check if gpu_targets are supported by compiler

* set default list of targets and filter for them
---
 CMakeLists.txt | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c9fb6b4552..33c25b499e 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -16,11 +16,24 @@ include(ROCMSetupVersion)
 include(ROCMInstallSymlinks)
 include(ROCMCreatePackage)
 include(CheckCXXCompilerFlag)
-
+include(ROCMCheckTargetIds)
 rocm_setup_version(VERSION 0.2.0)
 include(TargetFlags)
 list(APPEND CMAKE_PREFIX_PATH ${CMAKE_INSTALL_PREFIX} ${CMAKE_INSTALL_PREFIX}/llvm ${CMAKE_INSTALL_PREFIX}/hip /opt/rocm /opt/rocm/llvm /opt/rocm/hip)
 
+message("GPU_TARGETS= ${GPU_TARGETS}")
+
+message("checking which targets are supported")
+#This is the list of targets to be used in case GPU_TARGETS is not set on command line
+#These targets will be filtered and only supported ones will be used
+#Setting GPU_TARGETS on command line will override this list
+rocm_check_target_ids(DEFAULT_GPU_TARGETS
+    TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102"
+)
+message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}")
+set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " ")
+find_package(hip)
+
 option(USE_BITINT_EXTENSION_INT4, "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
 option(USE_OPT_NAVI3X, "Whether to enable LDS cumode and Wavefront32 mode for NAVI3X silicons." OFF)
 

From 189ea3b9aa1cf311869b8dd8d058e4531bf54bd4 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 17 Jul 2023 18:02:42 -0700
Subject: [PATCH 061/194] Add mechanism to build CK for select data types, add
 Navi3x CI. (#790)

* allow building CK for specific data types

* add CI build and test stage on Naiv3x without some int8 instances

* add missing gemm fp16 instances

* add the changes to the missed cmake file

* add empty lines at end of source files

* Do not build quantization client example on navi3 in CI

* disable batched_gemm_multi_d_int8 instances with DTYPES

* disable device_conv2d_bwd_data_instance with DTYPES

* fix ckprofiler for conv_bwd_data for int8

* properly isolate the conv_bwd_data int8 instances

* remove empty line
---
 CMakeLists.txt                                |  25 +++
 Jenkinsfile                                   |  16 ++
 client_example/09_quantization/CMakeLists.txt |   2 +
 client_example/CMakeLists.txt                 |  25 +++
 example/01_gemm/CMakeLists.txt                |  14 +-
 example/14_gemm_quantization/CMakeLists.txt   |   4 +-
 .../40_conv2d_fwd_quantization/CMakeLists.txt |   4 +-
 .../gpu/convolution_backward_data.hpp         |  21 ++-
 .../tensor_operation_instance/gpu/gemm.hpp    |  46 ++---
 .../gpu/CMakeLists.txt                        |  39 +++-
 .../gpu/batched_gemm_multi_d/CMakeLists.txt   |  40 +++--
 .../gpu/conv2d_bwd_data/CMakeLists.txt        |  27 +--
 ..._data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp |   3 +-
 .../gpu/gemm/CMakeLists.txt                   | 169 +++++++++---------
 ...ice_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp |   3 +-
 ...l_i8_i8_i8_km_kn_mn_irregular_instance.cpp |   3 +-
 ...ice_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp |   3 +-
 ...l_i8_i8_i8_km_nk_mn_irregular_instance.cpp |   3 +-
 ...ice_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp |   3 +-
 ...l_i8_i8_i8_mk_kn_mn_irregular_instance.cpp |   3 +-
 ...ice_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp |   3 +-
 ...l_i8_i8_i8_mk_nk_mn_irregular_instance.cpp |   3 +-
 ...l_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp |   3 +-
 ...l_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp |   3 +-
 ...l_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp |   3 +-
 ...l_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp |   3 +-
 .../gpu/quantization/CMakeLists.txt           |   2 +
 profiler/src/profile_batched_gemm_multi_d.cpp |   6 +-
 profiler/src/profile_conv_bwd_data.cpp        |   8 +
 profiler/src/profile_gemm.cpp                 |  14 +-
 .../test_batched_gemm_multi_d.cpp             |   6 +-
 test/gemm/CMakeLists.txt                      |  24 +--
 32 files changed, 356 insertions(+), 175 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 33c25b499e..3a3b98d627 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -5,6 +5,31 @@ project(composable_kernel)
 
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake")
 
+if (DTYPES)
+        add_definitions(-DDTYPES)
+        if (DTYPES MATCHES "int8")
+                add_definitions(-D__int8__)
+        endif()
+        if (DTYPES MATCHES "fp8")
+                add_definitions(-D__fp8__)
+        endif()
+        if (DTYPES MATCHES "fp16")
+                add_definitions(-D__fp16__)
+        endif()
+        if (DTYPES MATCHES "fp32")
+                add_definitions(-D__fp32__)
+        endif()
+        if (DTYPES MATCHES "fp64")
+                add_definitions(-D__fp64__)
+        endif()
+        if (DTYPES MATCHES "bf16")
+                add_definitions(-D__bf16__)
+        endif()
+        message("DTYPES macro set to ${DTYPES}")
+else()
+        add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
+endif()
+
 enable_testing()
 
 set(ROCM_SYMLINK_LIBS OFF)
diff --git a/Jenkinsfile b/Jenkinsfile
index df9d13167f..764dc71220 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -749,6 +749,22 @@ pipeline {
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
                     }
                 }
+                stage("Build CK and run Tests on Navi32")
+                {
+                    when {
+                        beforeAgent true
+                        expression { !params.RUN_FULL_QA.toBoolean() }
+                    }
+                    agent{ label rocmnode("navi32") }
+                    environment{
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DDTYPES="fp16;fp32;bf16" -DGPU_TARGETS="gfx1101" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -DDTYPES="fp16;fp32;bf16" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
+
+                    }
+                    steps{
+                        Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                    }
+                }
             }
         }
 
diff --git a/client_example/09_quantization/CMakeLists.txt b/client_example/09_quantization/CMakeLists.txt
index 2b7d6fc806..ac11aad45d 100644
--- a/client_example/09_quantization/CMakeLists.txt
+++ b/client_example/09_quantization/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 add_executable(client_conv2d_fwd_bias_tanh_perchannel_quantization conv2d_fwd_bias_tanh_perchannel_quantization.cpp)
 target_link_libraries(client_conv2d_fwd_bias_tanh_perchannel_quantization PRIVATE composable_kernel::device_operations)
 
@@ -18,3 +19,4 @@ target_link_libraries(client_conv2d_fwd_perlayer_quantization PRIVATE composable
 
 add_executable(client_gemm_quantization gemm_quantization.cpp)
 target_link_libraries(client_gemm_quantization PRIVATE composable_kernel::device_operations)
+endif()
diff --git a/client_example/CMakeLists.txt b/client_example/CMakeLists.txt
index 14c066e4a2..526abf5e77 100644
--- a/client_example/CMakeLists.txt
+++ b/client_example/CMakeLists.txt
@@ -2,6 +2,31 @@ cmake_minimum_required(VERSION 3.15)
 project(ck_app)
 add_compile_options(-std=c++17)
 
+if (DTYPES)
+        add_definitions(-DDTYPES)
+        if (DTYPES MATCHES "int8")
+                add_definitions(-D__int8__)
+        endif()
+        if (DTYPES MATCHES "fp8")
+                add_definitions(-D__fp8__)
+        endif()
+        if (DTYPES MATCHES "fp16")
+                add_definitions(-D__fp16__)
+        endif()
+        if (DTYPES MATCHES "fp32")
+                add_definitions(-D__fp32__)
+        endif()
+        if (DTYPES MATCHES "fp64")
+                add_definitions(-D__fp64__)
+        endif()
+        if (DTYPES MATCHES "bf16")
+                add_definitions(-D__bf16__)
+        endif()
+        message("DTYPES macro set to ${DTYPES}")
+else()
+        add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
+endif()
+
 find_package(composable_kernel 1.0.0 COMPONENTS device_operations)
 find_package(hip REQUIRED PATHS /opt/rocm)
 message(STATUS "Build with HIP ${hip_VERSION}")
diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 66afe73c2a..ec4061213e 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -2,11 +2,14 @@ add_custom_target(example_gemm_dl)
 
 add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
 add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
-add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
 
 add_dependencies(example_gemm_dl example_gemm_dl_fp32)
 add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-add_dependencies(example_gemm_dl example_gemm_dl_int8)
+
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+   add_dependencies(example_gemm_dl example_gemm_dl_int8)
+endif()
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
@@ -19,13 +22,16 @@ add_custom_target(example_gemm_xdl)
 add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
 add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
 add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
-add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
 
 add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
 add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
-add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
 add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
 
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+  add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
+endif()
+
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
diff --git a/example/14_gemm_quantization/CMakeLists.txt b/example/14_gemm_quantization/CMakeLists.txt
index 72bdff5ab4..ca06265e31 100644
--- a/example/14_gemm_quantization/CMakeLists.txt
+++ b/example/14_gemm_quantization/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 # dlops
 add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
 
@@ -10,4 +11,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
    add_example_executable(example_gemm_xdl_quantization_int8 gemm_xdl_quantization_int8.cpp)
    set(target 1)
  endif()
-endforeach()
\ No newline at end of file
+endforeach()
+endif()
\ No newline at end of file
diff --git a/example/40_conv2d_fwd_quantization/CMakeLists.txt b/example/40_conv2d_fwd_quantization/CMakeLists.txt
index b82a013b58..1962b6013b 100644
--- a/example/40_conv2d_fwd_quantization/CMakeLists.txt
+++ b/example/40_conv2d_fwd_quantization/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -25,4 +26,5 @@ add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_i
 add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
 
 # Conv + bias + tanh perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
\ No newline at end of file
+add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
+endif()
\ No newline at end of file
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index 1efe073669..628fb5e479 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -39,7 +39,7 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#ifdef __int8__
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
                                                   NWC,
@@ -51,7 +51,7 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
 // conv2d backward data
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -88,7 +88,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#ifdef __int8__
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -100,7 +100,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
 // conv2d dl
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -125,7 +125,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#ifdef __int8__
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -137,6 +137,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
+#endif
 // conv3d backward data
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
@@ -173,7 +174,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#ifdef __int8__
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -185,7 +186,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -239,11 +240,13 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(op_ptrs);
             }
+#ifdef __int8__
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWC> &&
                           is_same_v<WeiLayout, KYXC> && is_same_v<OutLayout, NHWK>)
@@ -266,12 +269,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
+#ifdef __int8__
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(op_ptrs);
                 add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWC> &&
                           is_same_v<WeiLayout, KZYXC> && is_same_v<OutLayout, NDHWK>)
@@ -292,11 +297,13 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(op_ptrs);
             }
+#ifdef __int8__
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(op_ptrs);
             }
+#endif
         }
 
         return op_ptrs;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index adac7d0dcf..ad298009ae 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -77,7 +77,7 @@ void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#ifdef __int8__
 void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -118,6 +118,27 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+#endif
+
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -183,26 +204,6 @@ void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
 void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -388,6 +389,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(op_ptrs);
             }
         }
+#ifdef __int8__
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<CDataType, int8_t>)
         {
@@ -420,7 +422,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
index c206c4dc04..597aeaed8b 100644
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
@@ -12,9 +12,42 @@ set(CK_DEVICE_INSTANCES)
 FOREACH(subdir_path ${dir_list})
 set(target_dir)
 IF(IS_DIRECTORY "${subdir_path}")
-    get_filename_component(target_dir ${subdir_path} NAME)
-    add_subdirectory(${target_dir})
-    list(APPEND CK_DEVICE_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
+    set(cmake_instance)
+    file(READ "${subdir_path}/CMakeLists.txt" cmake_instance)
+    set(add_inst 0)
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8")
+            #message("fp8 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
+            #message("fp16 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32")
+            #message("fp32 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64")
+            #message("fp64 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16")
+            #message("bf16 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8")
+            #message("int8 instance found!")
+            set(add_inst 1)
+    endif()
+    if(NOT "${cmake_instance}" MATCHES "DTYPES")
+            #message("instance should be built for all types!")
+            set(add_inst 1)
+    endif()
+    if(add_inst EQUAL 1 OR NOT DEFINED DTYPES)
+      get_filename_component(target_dir ${subdir_path} NAME)
+      add_subdirectory(${target_dir})
+      list(APPEND CK_DEVICE_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
+    endif()
 ENDIF()
 ENDFOREACH()
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
index fda55a9303..8f4e3b6948 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
@@ -1,18 +1,22 @@
-add_instance_library(device_batched_gemm_multi_d_instance
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp
-   device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp
-)
+set(BATCHED_GEMM_MULTID_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp)
+   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp)
+endif()
+add_instance_library(device_batched_gemm_multi_d_instance ${BATCHED_GEMM_MULTID_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
index 281453b586..582eab3064 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
@@ -1,10 +1,17 @@
-add_instance_library(device_conv2d_bwd_data_instance
-   device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
-   device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
-   device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
-   device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
-
-   device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
-   device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
-   device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
-) 
+set(CONV2D_BWD_DATA_INSTANCES)
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp)
+   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp)
+endif()
+add_instance_library(device_conv2d_bwd_data_instance ${CONV2D_BWD_DATA_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index bb9b696864..4bd714980a 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -151,3 +151,4 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 11227d01f5..62cdae7cdf 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -1,84 +1,93 @@
-add_instance_library(device_gemm_instance
-   device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
-   device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
-   device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
-   device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
-   device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
-   device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
-   device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+set(GEMM_INSTANCES)
+if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp)
+    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp)
+    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp)
+    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp)
+   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp)
+endif()
 
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
-   device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
-
-   device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
-   device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
-)
+add_instance_library(device_gemm_instance ${GEMM_INSTANCES})
 
 set(ENABLE_PIPELINE_V2_OPT OFF)
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
index 3d9a265c29..8377b13ba6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -80,3 +80,4 @@ void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
index 0439201511..2976c3ad1f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -78,3 +78,4 @@ void add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
index 240384d19c..db4ca19a8f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -80,3 +80,4 @@ void add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
index 350834f7e5..e6643ff085 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -78,3 +78,4 @@ void add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
index e96905247d..872eb28caf 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -80,3 +80,4 @@ void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
index 27397527bc..1c19cdd15c 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -78,3 +78,4 @@ void add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
index 124b818b28..75aac31a26 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -80,3 +80,4 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
index b99f3f2b63..87fda6cbe0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -78,3 +78,4 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
index bf24bc76b1..2c8b2e3acb 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -66,3 +66,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
index 023f987127..e5156c4ce8 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -66,3 +66,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
index ffb199e58f..3aa4ea90f6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -66,3 +66,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
index 90e979d891..a5de884004 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -63,3 +63,4 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
index db650dbfbd..3b4c8935a4 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 set(CONV2D_PERLAYER_QUANT_SRC
     conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
     conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
@@ -36,3 +37,4 @@ add_instance_library(device_quantization_instance
     ${CONV2D_BIAS_PERCHANNEL_QUANT_SRC}
     ${GEMM_QUANT_SRC}
 )
+endif()
\ No newline at end of file
diff --git a/profiler/src/profile_batched_gemm_multi_d.cpp b/profiler/src/profile_batched_gemm_multi_d.cpp
index 98b462d950..8c0c75c1bc 100644
--- a/profiler/src/profile_batched_gemm_multi_d.cpp
+++ b/profiler/src/profile_batched_gemm_multi_d.cpp
@@ -70,8 +70,10 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
 
     const int BatchCount = std::stoi(argv[17]);
 
-    using F16  = ck::half_t;
+    using F16 = ck::half_t;
+#ifdef __int8__
     using INT8 = int8_t;
+#endif
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -163,6 +165,7 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
     {
         return profile(F16{}, F16{}, F16{}, Col{}, Col{}, Row{});
     }
+#ifdef __int8__
     else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(INT8{}, INT8{}, INT8{}, Row{}, Row{}, Row{});
@@ -179,6 +182,7 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
     {
         return profile(INT8{}, INT8{}, INT8{}, Col{}, Col{}, Row{});
     }
+#endif
     else
     {
         std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/profiler/src/profile_conv_bwd_data.cpp b/profiler/src/profile_conv_bwd_data.cpp
index 465abacc4a..f8314b5169 100644
--- a/profiler/src/profile_conv_bwd_data.cpp
+++ b/profiler/src/profile_conv_bwd_data.cpp
@@ -77,7 +77,9 @@ int profile_conv_bwd_data(int argc, char* argv[])
     using F32  = float;
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
+#ifdef __int8__
     using INT8 = int8_t;
+#endif
 
     using NWC   = ck::tensor_layout::convolution::NWC;
     using NHWC  = ck::tensor_layout::convolution::NHWC;
@@ -138,10 +140,12 @@ int profile_conv_bwd_data(int argc, char* argv[])
         {
             return profile(I1, NWC{}, KXC{}, NWK{}, BF16{}, BF16{}, BF16{});
         }
+#ifdef __int8__
         else if(data_type == ConvDataType::INT8_INT8_INT8)
         {
             return profile(I1, NWC{}, KXC{}, NWK{}, INT8{}, INT8{}, INT8{});
         }
+#endif
     }
     else if(num_dim_spatial == 2 && layout == ConvLayout::NHWC_KYXC_NHWK)
     {
@@ -157,10 +161,12 @@ int profile_conv_bwd_data(int argc, char* argv[])
         {
             return profile(I2, NHWC{}, KYXC{}, NHWK{}, BF16{}, BF16{}, BF16{});
         }
+#ifdef __int8__
         else if(data_type == ConvDataType::INT8_INT8_INT8)
         {
             return profile(I2, NHWC{}, KYXC{}, NHWK{}, INT8{}, INT8{}, INT8{});
         }
+#endif
     }
     else if(num_dim_spatial == 3 && layout == ConvLayout::NHWC_KYXC_NHWK)
     {
@@ -176,10 +182,12 @@ int profile_conv_bwd_data(int argc, char* argv[])
         {
             return profile(I3, NDHWC{}, KZYXC{}, NDHWK{}, BF16{}, BF16{}, BF16{});
         }
+#ifdef __int8__
         else if(data_type == ConvDataType::INT8_INT8_INT8)
         {
             return profile(I3, NDHWC{}, KZYXC{}, NDHWK{}, INT8{}, INT8{}, INT8{});
         }
+#endif
     }
 
     std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/profiler/src/profile_gemm.cpp b/profiler/src/profile_gemm.cpp
index b3587ea98d..fbef123860 100644
--- a/profiler/src/profile_gemm.cpp
+++ b/profiler/src/profile_gemm.cpp
@@ -67,11 +67,15 @@ int profile_gemm(int argc, char* argv[])
     const int StrideB = std::stoi(argv[12]);
     const int StrideC = std::stoi(argv[13]);
 
-    using F32   = float;
-    using F16   = ck::half_t;
-    using BF16  = ck::bhalf_t;
+    using F32 = float;
+    using F16 = ck::half_t;
+#ifdef __bf16__
+    using BF16 = ck::bhalf_t;
+#endif
+#ifdef __int8__
     using INT8  = int8_t;
     using INT32 = int32_t;
+#endif
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -149,6 +153,7 @@ int profile_gemm(int argc, char* argv[])
     {
         return profile(Col{}, Col{}, Row{}, F16{}, F16{}, F32{}, F16{});
     }
+#ifdef __bf16__
     else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, BF16{}, BF16{}, F32{}, BF16{});
@@ -165,6 +170,8 @@ int profile_gemm(int argc, char* argv[])
     {
         return profile(Col{}, Col{}, Row{}, BF16{}, BF16{}, F32{}, BF16{});
     }
+#endif
+#ifdef __int8__
     else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, INT8{}, INT8{}, INT32{}, INT8{});
@@ -181,6 +188,7 @@ int profile_gemm(int argc, char* argv[])
     {
         return profile(Col{}, Col{}, Row{}, INT8{}, INT8{}, INT32{}, INT8{});
     }
+#endif
     else
     {
         std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
index 6cf6145cd7..5ccae83805 100644
--- a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
+++ b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
@@ -68,7 +68,9 @@ using KernelTypes = ::testing::Types<std::tuple<Row, Row, Row>,
 } // namespace
 
 TYPED_TEST_SUITE(TestBatchedGemmMultiD, KernelTypes);
-
+#ifdef __fp16
 TYPED_TEST(TestBatchedGemmMultiD, f16) { this->template Run<F16>(); }
-
+#endif
+#ifdef __int8__
 TYPED_TEST(TestBatchedGemmMultiD, int8) { this->template Run<int8_t>(); }
+#endif
diff --git a/test/gemm/CMakeLists.txt b/test/gemm/CMakeLists.txt
index 186b06d911..7a8836bfef 100644
--- a/test/gemm/CMakeLists.txt
+++ b/test/gemm/CMakeLists.txt
@@ -1,19 +1,12 @@
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
 add_test_executable(test_gemm_fp32 gemm_fp32.cpp)
 target_link_libraries(test_gemm_fp32 PRIVATE utility)
 target_link_libraries(test_gemm_fp32 PRIVATE device_gemm_instance)
-
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_test_executable(test_gemm_fp16 gemm_fp16.cpp)
 target_link_libraries(test_gemm_fp16 PRIVATE utility)
 target_link_libraries(test_gemm_fp16 PRIVATE device_gemm_instance)
-
-add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
-target_link_libraries(test_gemm_bf16 PRIVATE utility)
-target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance)
-
-add_test_executable(test_gemm_int8 gemm_int8.cpp)
-target_link_libraries(test_gemm_int8 PRIVATE utility)
-target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance)
-
 add_library(gemm_standalone_xdl_fp16_instances STATIC
     instance/gemm_f16_nn_instance.cpp
     instance/gemm_f16_nt_instance.cpp
@@ -24,3 +17,14 @@ add_library(gemm_standalone_xdl_fp16_instances STATIC
 add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp)
 target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility)
 target_include_directories(test_gemm_standalone_xdl_fp16 PRIVATE instance/)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
+target_link_libraries(test_gemm_bf16 PRIVATE utility)
+target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+add_test_executable(test_gemm_int8 gemm_int8.cpp)
+target_link_libraries(test_gemm_int8 PRIVATE utility)
+target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance)
+endif()
\ No newline at end of file

From f82bd5938902142ee9daf4bc6aac753a6dccfaeb Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Tue, 18 Jul 2023 09:44:51 -0500
Subject: [PATCH 062/194] Remove type_convert bf16 to int32 and back (#802)

---
 include/ck/utility/type_convert.hpp | 18 ------------------
 1 file changed, 18 deletions(-)

diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index ed83964936..65d8940377 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -62,24 +62,6 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, half_t>(half_
     return type_convert<bhalf_t>(x_fp32);
 }
 
-// convert bfp16 to int32 via fp32
-template <>
-inline __host__ __device__ constexpr int32_t type_convert<int32_t, bhalf_t>(bhalf_t x)
-{
-    float x_fp32 = type_convert<float>(x);
-
-    return static_cast<int32_t>(x_fp32);
-}
-
-// convert int32 to bfp16 via fp32
-template <>
-inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int32_t>(int32_t x)
-{
-    float x_fp32 = static_cast<float>(x);
-
-    return type_convert<bhalf_t>(x_fp32);
-}
-
 // convert bfp16 to int8 via fp32
 template <>
 inline __host__ __device__ constexpr int8_t type_convert<int8_t, bhalf_t>(bhalf_t x)

From 49180fd60b7ce0e1fa5961413865f6b219346a75 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 18 Jul 2023 18:01:33 +0200
Subject: [PATCH 063/194] Grouped 3d conv backward data support (#799)

* Grouped 3d conv backward data support

* Fix comments
---
 ...nvolution_backward_data_specialization.hpp |   3 +-
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp | 242 +++--
 .../transform_conv_bwd_data_to_gemm_v1.hpp    | 947 +++++++++++++-----
 ...ice_grouped_conv_bwd_data_xdl_instance.hpp | 141 +++
 .../gpu/grouped_convolution_backward_data.hpp | 204 +++-
 ...ta_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |  26 +-
 ...ata_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  26 +-
 ...ata_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |  26 +-
 ...e_grouped_conv2d_bwd_data_xdl_instance.hpp | 141 ---
 ...ta_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |  26 +-
 ...ata_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  26 +-
 ...ata_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |  26 +-
 .../grouped_conv3d_bwd_data/CMakeLists.txt    |   8 +
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |  50 +
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |  50 +
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |  50 +
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |  50 +
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |  50 +
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |  50 +
 profiler/src/CMakeLists.txt                   |   1 +
 .../src/profile_grouped_conv_bwd_data.cpp     |  83 +-
 test/grouped_convnd_bwd_data/CMakeLists.txt   |   2 +-
 .../test_grouped_convnd_bwd_data.cpp          |  51 +-
 23 files changed, 1654 insertions(+), 625 deletions(-)
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp

diff --git a/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp b/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
index f3263c7216..cab5e21365 100644
--- a/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
+++ b/include/ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp
@@ -19,8 +19,7 @@ getConvBackwardDataSpecializationString(const ConvolutionBackwardDataSpecializat
     switch(s)
     {
     case ConvolutionBackwardDataSpecialization::Default: return "Default";
-    case ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0:
-        return "FFilter1x1Stride1Pad0";
+    case ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0: return "Filter1x1Stride1Pad0";
     default: return "Unrecognized specialization!";
     }
 }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 95af932519..d932fd4069 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -258,7 +258,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
                                                CDEElementwiseOp>
 {
     // FIXME
-    static_assert(NDimSpatial == 2, "wrong! only implemented for 2D now");
+    static_assert(NDimSpatial == 2 || NDimSpatial == 3,
+                  "wrong! only implemented for 2D and 3D now");
 
     using DeviceOp = DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1;
 
@@ -491,130 +492,172 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
                 compute_ptr_offset_of_batch_.BatchStrideDs_(i) = ds_g_n_c_wis_strides[i][0];
             });
 
+            static constexpr auto NonSpatialDimsNum = Number<3>{};
+
+            static constexpr auto DIdx = Number<NonSpatialDimsNum>{};
+            static constexpr auto HIdx =
+                NDimSpatial == 2 ? Number<NonSpatialDimsNum>{} : Number<NonSpatialDimsNum + 1>{};
+            static constexpr auto WIdx = NDimSpatial == 2 ? Number<NonSpatialDimsNum + 1>{}
+                                                          : Number<NonSpatialDimsNum + 2>{};
+
+            static constexpr auto ZIdx = Number<NonSpatialDimsNum>{};
+            static constexpr auto YIdx =
+                NDimSpatial == 2 ? Number<NonSpatialDimsNum>{} : Number<NonSpatialDimsNum + 1>{};
+            static constexpr auto XIdx = NDimSpatial == 2 ? Number<NonSpatialDimsNum + 1>{}
+                                                          : Number<NonSpatialDimsNum + 2>{};
+
             // problem definition
-            const index_t Y = b_g_k_c_xs_lengths[3];
-            const index_t X = b_g_k_c_xs_lengths[4];
+            const index_t Z = b_g_k_c_xs_lengths[ZIdx];
+            const index_t Y = b_g_k_c_xs_lengths[YIdx];
+            const index_t X = b_g_k_c_xs_lengths[XIdx];
 
-            const index_t ConvStrideH = conv_filter_strides_[0];
-            const index_t ConvStrideW = conv_filter_strides_[1];
+            const index_t ConvStrideD = conv_filter_strides[DIdx - NonSpatialDimsNum];
+            const index_t ConvStrideH = conv_filter_strides[HIdx - NonSpatialDimsNum];
+            const index_t ConvStrideW = conv_filter_strides[WIdx - NonSpatialDimsNum];
 
-            const index_t ConvDilationH = conv_filter_dilations_[0];
-            const index_t ConvDilationW = conv_filter_dilations_[1];
+            const index_t ConvDilationD = conv_filter_dilations[DIdx - NonSpatialDimsNum];
+            const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
+            const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
 
+            const auto GcdStrideDilationD = math::gcd(ConvStrideD, ConvDilationD);
             const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
             const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
 
+            const auto ZTilde = NDimSpatial == 3 ? ConvStrideD / GcdStrideDilationD : 1;
             const auto YTilde = ConvStrideH / GcdStrideDilationH;
             const auto XTilde = ConvStrideW / GcdStrideDilationW;
 
-            for(index_t i_ytilde = 0; i_ytilde < YTilde; ++i_ytilde)
+            for(index_t i_ztilde = 0; i_ztilde < ZTilde; ++i_ztilde)
             {
-                for(index_t i_xtilde = 0; i_xtilde < XTilde; ++i_xtilde)
+
+                for(index_t i_ytilde = 0; i_ytilde < YTilde; ++i_ytilde)
                 {
-                    // check slice is valid
-                    const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
-                    const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
-
-                    if(YDotSlice * XDotSlice <= 0)
+                    for(index_t i_xtilde = 0; i_xtilde < XTilde; ++i_xtilde)
                     {
-                        continue;
-                    }
+                        // check slice is valid
+                        const auto ZDotSlice =
+                            NDimSpatial == 3 ? math::integer_divide_ceil(Z - i_ztilde, ZTilde) : 1;
+                        const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
+                        const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
-                    const auto a_grid_desc_ak0_m_ak1 =
-                        transform_conv_to_gemm.template MakeADescriptor_AK0_M_AK1<ALayout>(
-                            a_g_n_k_wos_lengths,
-                            a_g_n_k_wos_strides,
-                            b_g_k_c_xs_lengths,
-                            b_g_k_c_xs_strides,
-                            e_g_n_c_wis_lengths,
-                            e_g_n_c_wis_strides,
-                            conv_filter_strides,
-                            conv_filter_dilations,
-                            input_left_pads,
-                            input_right_pads,
-                            {i_ytilde, i_xtilde});
+                        if(YDotSlice * XDotSlice * ZDotSlice <= 0)
+                        {
+                            continue;
+                        }
 
-                    const auto b_grid_desc_bk0_n_bk1 =
-                        transform_conv_to_gemm.template MakeBDescriptor_BK0_N_BK1<BLayout>(
-                            a_g_n_k_wos_lengths,
-                            a_g_n_k_wos_strides,
-                            b_g_k_c_xs_lengths,
-                            b_g_k_c_xs_strides,
-                            e_g_n_c_wis_lengths,
-                            e_g_n_c_wis_strides,
-                            conv_filter_strides,
-                            conv_filter_dilations,
-                            input_left_pads,
-                            input_right_pads,
-                            {i_ytilde, i_xtilde});
+                        std::array<index_t, NDimSpatial> tildes;
+                        if constexpr(NDimSpatial == 2)
+                        {
+                            tildes = {i_ytilde, i_xtilde};
+                        }
+                        else if constexpr(NDimSpatial == 3)
+                        {
+                            tildes = {i_ztilde, i_ytilde, i_xtilde};
+                        }
+                        else
+                        {
+                            throw std::runtime_error("wrong! only implemented for 2D and 3D now");
+                        }
 
-                    DsGridDesc_M_N ds_grid_desc_m_n;
-
-                    // populate Ds desc
-                    static_for<0, NumDTensor, 1>{}([&](auto i) {
-                        using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
-
-                        ds_grid_desc_m_n(i) =
-                            transform_conv_to_gemm.template MakeCDescriptor_M_N<DLayout>(
+                        const auto a_grid_desc_ak0_m_ak1 =
+                            transform_conv_to_gemm.template MakeADescriptor_AK0_M_AK1<ALayout>(
                                 a_g_n_k_wos_lengths,
                                 a_g_n_k_wos_strides,
                                 b_g_k_c_xs_lengths,
                                 b_g_k_c_xs_strides,
-                                ds_g_n_c_wis_lengths[i],
-                                ds_g_n_c_wis_strides[i],
+                                e_g_n_c_wis_lengths,
+                                e_g_n_c_wis_strides,
                                 conv_filter_strides,
                                 conv_filter_dilations,
                                 input_left_pads,
                                 input_right_pads,
-                                {i_ytilde, i_xtilde});
-                    });
+                                tildes);
 
-                    const auto e_grid_desc_m_n =
-                        transform_conv_to_gemm.template MakeCDescriptor_M_N<ELayout>(
-                            a_g_n_k_wos_lengths,
-                            a_g_n_k_wos_strides,
-                            b_g_k_c_xs_lengths,
-                            b_g_k_c_xs_strides,
-                            e_g_n_c_wis_lengths,
-                            e_g_n_c_wis_strides,
-                            conv_filter_strides,
-                            conv_filter_dilations,
-                            input_left_pads,
-                            input_right_pads,
-                            {i_ytilde, i_xtilde});
+                        const auto b_grid_desc_bk0_n_bk1 =
+                            transform_conv_to_gemm.template MakeBDescriptor_BK0_N_BK1<BLayout>(
+                                a_g_n_k_wos_lengths,
+                                a_g_n_k_wos_strides,
+                                b_g_k_c_xs_lengths,
+                                b_g_k_c_xs_strides,
+                                e_g_n_c_wis_lengths,
+                                e_g_n_c_wis_strides,
+                                conv_filter_strides,
+                                conv_filter_dilations,
+                                input_left_pads,
+                                input_right_pads,
+                                tildes);
 
-                    // desc for problem definition
-                    const auto a_grid_desc_m_k = transform_k0_m_k1_to_m_k(a_grid_desc_ak0_m_ak1);
-                    const auto b_grid_desc_n_k = transform_k0_m_k1_to_m_k(b_grid_desc_bk0_n_bk1);
+                        DsGridDesc_M_N ds_grid_desc_m_n;
 
-                    a_grid_desc_m_k_container_.push_back(a_grid_desc_m_k);
-                    b_grid_desc_n_k_container_.push_back(b_grid_desc_n_k);
-                    ds_grid_desc_m_n_container_.push_back(ds_grid_desc_m_n);
-                    e_grid_desc_m_n_container_.push_back(e_grid_desc_m_n);
+                        // populate Ds desc
+                        static_for<0, NumDTensor, 1>{}([&](auto i) {
+                            using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
 
-                    // desc for blockwise copy
-                    a_grid_desc_ak0_m_ak1_container_.push_back(a_grid_desc_ak0_m_ak1);
-                    b_grid_desc_bk0_n_bk1_container_.push_back(b_grid_desc_bk0_n_bk1);
+                            ds_grid_desc_m_n(i) =
+                                transform_conv_to_gemm.template MakeCDescriptor_M_N<DLayout>(
+                                    a_g_n_k_wos_lengths,
+                                    a_g_n_k_wos_strides,
+                                    b_g_k_c_xs_lengths,
+                                    b_g_k_c_xs_strides,
+                                    ds_g_n_c_wis_lengths[i],
+                                    ds_g_n_c_wis_strides[i],
+                                    conv_filter_strides,
+                                    conv_filter_dilations,
+                                    input_left_pads,
+                                    input_right_pads,
+                                    tildes);
+                        });
 
-                    // block-to-e-tile-map
-                    auto block_2_etile_map =
-                        GridwiseGemm::MakeDefaultBlock2ETileMap(e_grid_desc_m_n);
+                        const auto e_grid_desc_m_n =
+                            transform_conv_to_gemm.template MakeCDescriptor_M_N<ELayout>(
+                                a_g_n_k_wos_lengths,
+                                a_g_n_k_wos_strides,
+                                b_g_k_c_xs_lengths,
+                                b_g_k_c_xs_strides,
+                                e_g_n_c_wis_lengths,
+                                e_g_n_c_wis_strides,
+                                conv_filter_strides,
+                                conv_filter_dilations,
+                                input_left_pads,
+                                input_right_pads,
+                                tildes);
 
-                    block_2_etile_map_container_.push_back(block_2_etile_map);
+                        // desc for problem definition
+                        const auto a_grid_desc_m_k =
+                            transform_k0_m_k1_to_m_k(a_grid_desc_ak0_m_ak1);
+                        const auto b_grid_desc_n_k =
+                            transform_k0_m_k1_to_m_k(b_grid_desc_bk0_n_bk1);
 
-                    if(GridwiseGemm::CheckValidity(a_grid_desc_m_k,
-                                                   b_grid_desc_n_k,
-                                                   ds_grid_desc_m_n,
-                                                   e_grid_desc_m_n,
-                                                   block_2_etile_map))
-                    {
-                        ds_grid_desc_mblock_mperblock_nblock_nperblock_container_.push_back(
-                            GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                                ds_grid_desc_m_n));
+                        a_grid_desc_m_k_container_.push_back(a_grid_desc_m_k);
+                        b_grid_desc_n_k_container_.push_back(b_grid_desc_n_k);
+                        ds_grid_desc_m_n_container_.push_back(ds_grid_desc_m_n);
+                        e_grid_desc_m_n_container_.push_back(e_grid_desc_m_n);
 
-                        e_grid_desc_mblock_mperblock_nblock_nperblock_container_.push_back(
-                            GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                                e_grid_desc_m_n));
+                        // desc for blockwise copy
+                        a_grid_desc_ak0_m_ak1_container_.push_back(a_grid_desc_ak0_m_ak1);
+                        b_grid_desc_bk0_n_bk1_container_.push_back(b_grid_desc_bk0_n_bk1);
+
+                        // block-to-e-tile-map
+                        auto block_2_etile_map =
+                            GridwiseGemm::MakeDefaultBlock2ETileMap(e_grid_desc_m_n);
+
+                        block_2_etile_map_container_.push_back(block_2_etile_map);
+
+                        if(GridwiseGemm::CheckValidity(a_grid_desc_m_k,
+                                                       b_grid_desc_n_k,
+                                                       ds_grid_desc_m_n,
+                                                       e_grid_desc_m_n,
+                                                       block_2_etile_map))
+                        {
+                            ds_grid_desc_mblock_mperblock_nblock_nperblock_container_.push_back(
+                                GridwiseGemm::
+                                    MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                                        ds_grid_desc_m_n));
+
+                            e_grid_desc_mblock_mperblock_nblock_nperblock_container_.push_back(
+                                GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                                    e_grid_desc_m_n));
+                        }
                     }
                 }
             }
@@ -803,7 +846,9 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
         // vector load for A matrix from global memory to LDS
         if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
-                     is_same_v<ALayout, tensor_layout::convolution::NHWGK>)
+                     is_same_v<ALayout, tensor_layout::convolution::GNDHWK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::NHWGK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::NDHWGK>)
         {
             if(!(ABlockTransferSrcVectorDim == 2 && ConvK % ABlockTransferSrcScalarPerVector == 0))
             {
@@ -816,7 +861,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
         }
 
         // vector load for B matrix from global memory to LDS
-        if constexpr(is_same_v<BLayout, tensor_layout::convolution::GKYXC>)
+        if constexpr(is_same_v<BLayout, tensor_layout::convolution::GKYXC> ||
+                     is_same_v<BLayout, tensor_layout::convolution::GKZYXC>)
         {
             if(!(BBlockTransferSrcVectorDim == 1 && ConvC % BBlockTransferSrcScalarPerVector == 0))
             {
@@ -835,7 +881,9 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
             using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
 
             if constexpr(is_same_v<DLayout, tensor_layout::convolution::GNHWC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::GNDHWC> ||
                          is_same_v<DLayout, tensor_layout::convolution::NHWGC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::NDHWGC> ||
                          is_same_v<DLayout, tensor_layout::convolution::G_NHW_C> ||
                          is_same_v<DLayout, tensor_layout::convolution::GC> ||
                          is_same_v<DLayout, tensor_layout::convolution::G_C>)
@@ -859,7 +907,9 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
         // vector store for E
         if constexpr(is_same_v<ELayout, tensor_layout::convolution::GNHWC> ||
-                     is_same_v<ELayout, tensor_layout::convolution::NHWGC>)
+                     is_same_v<ELayout, tensor_layout::convolution::GNDHWC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::NHWGC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::NDHWGC>)
         {
             // vector store C matrix into global memory
             if(!(ConvC % CDEBlockTransferScalarPerVector_NPerBlock == 0))
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
index 59f64bde77..98a46ef450 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
@@ -18,32 +18,53 @@ template <
     index_t NDimSpatial,
     typename ALayout,
     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization ConvBwdDataSpecialization>
-constexpr auto
-make_out_n_ho_wo_k_grid_desc(const index_t N,
-                             const index_t Ho,
-                             const index_t Wo,
-                             const index_t K,
-                             const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_strides)
+constexpr auto make_out_grid_desc(const index_t N,
+                                  const index_t Do,
+                                  const index_t Ho,
+                                  const index_t Wo,
+                                  const index_t K,
+                                  const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_strides)
 {
+    const auto KStride = Number<1>{};
 
     if constexpr(is_same_v<ALayout, tensor_layout::convolution::NHWGK>)
     {
         const index_t NStride  = out_g_n_k_wos_strides[1];
         const index_t HiStride = out_g_n_k_wos_strides[3];
         const index_t WiStride = out_g_n_k_wos_strides[4];
-        const auto CStride     = Number<1>{};
         if constexpr(ConvBwdDataSpecialization ==
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
                          Filter1x1Stride1Pad0)
         {
 
             return make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, K),
-                                                make_tuple(WiStride, CStride));
+                                                make_tuple(WiStride, KStride));
         }
         else
         {
             return make_naive_tensor_descriptor(make_tuple(N, Ho, Wo, K),
-                                                make_tuple(NStride, HiStride, WiStride, CStride));
+                                                make_tuple(NStride, HiStride, WiStride, KStride));
+        }
+    }
+    else if constexpr(is_same_v<ALayout, tensor_layout::convolution::NDHWGK>)
+    {
+        const index_t NStride  = out_g_n_k_wos_strides[1];
+        const index_t DoStride = out_g_n_k_wos_strides[3];
+        const index_t HoStride = out_g_n_k_wos_strides[4];
+        const index_t WoStride = out_g_n_k_wos_strides[5];
+        if constexpr(ConvBwdDataSpecialization ==
+                     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
+                         Filter1x1Stride1Pad0)
+        {
+
+            return make_naive_tensor_descriptor(make_tuple(N * Do * Ho * Wo, K),
+                                                make_tuple(WoStride, KStride));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor(
+                make_tuple(N, Do, Ho, Wo, K),
+                make_tuple(NStride, DoStride, HoStride, WoStride, KStride));
         }
     }
     else if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK>)
@@ -60,12 +81,80 @@ make_out_n_ho_wo_k_grid_desc(const index_t N,
             return make_naive_tensor_descriptor_packed(make_tuple(N, Ho, Wo, K));
         }
     }
+    else if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNDHWK>)
+    {
+        // assume packed
+        if constexpr(ConvBwdDataSpecialization ==
+                     ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
+                         Filter1x1Stride1Pad0)
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N, Do, Ho, Wo, K));
+        }
+    }
     else
     {
         throw std::runtime_error("wrong! unsupported layout: " + ALayout::name());
     }
 }
 
+template <typename BLayout>
+constexpr auto make_wei_grid_desc(
+    const index_t K, const index_t Z, const index_t Y, const index_t X, const index_t C)
+{
+
+    if constexpr(is_same_v<BLayout, tensor_layout::convolution::GKYXC>)
+    {
+        return make_naive_tensor_descriptor_packed(make_tuple(K, Y, X, C));
+    }
+    else if constexpr(is_same_v<BLayout, tensor_layout::convolution::GKZYXC>)
+    {
+        return make_naive_tensor_descriptor_packed(make_tuple(K, Z, Y, X, C));
+    }
+    else
+    {
+        throw std::runtime_error("wrong! unsupported layout: " + BLayout::name());
+    }
+}
+
+template <index_t NDimSpatial, typename CLayout>
+constexpr auto make_in_grid_desc(const index_t N,
+                                 const index_t Di,
+                                 const index_t Hi,
+                                 const index_t Wi,
+                                 const index_t C,
+                                 const std::array<index_t, NDimSpatial + 3>& in_g_n_c_wis_strides)
+{
+
+    if constexpr(is_same_v<CLayout, tensor_layout::convolution::GNHWC> ||
+                 is_same_v<CLayout, tensor_layout::convolution::NHWGC> ||
+                 is_same_v<CLayout, tensor_layout::convolution::G_NHW_C>)
+    {
+        return make_naive_tensor_descriptor(make_tuple(N, Hi, Wi, C),
+                                            make_tuple(in_g_n_c_wis_strides[1],
+                                                       in_g_n_c_wis_strides[3],
+                                                       in_g_n_c_wis_strides[4],
+                                                       in_g_n_c_wis_strides[2]));
+    }
+    else if constexpr(is_same_v<CLayout, tensor_layout::convolution::GNDHWC> ||
+                      is_same_v<CLayout, tensor_layout::convolution::NDHWGC>)
+    {
+        return make_naive_tensor_descriptor(make_tuple(N, Di, Hi, Wi, C),
+                                            make_tuple(in_g_n_c_wis_strides[1],
+                                                       in_g_n_c_wis_strides[3],
+                                                       in_g_n_c_wis_strides[4],
+                                                       in_g_n_c_wis_strides[5],
+                                                       in_g_n_c_wis_strides[2]));
+    }
+    else
+    {
+        throw std::runtime_error("wrong! unsupported layout: " + CLayout::name());
+    }
+}
+
 } // namespace
 
 template <
@@ -82,10 +171,26 @@ struct TransformConvBwdDataToGemm_v1
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
 
+    static constexpr auto NonSpatialDimsNum = Number<3>{};
+
+    static constexpr auto DIdx = Number<NonSpatialDimsNum>{};
+    static constexpr auto HIdx =
+        NDimSpatial == 2 ? Number<NonSpatialDimsNum>{} : Number<NonSpatialDimsNum + 1>{};
+    static constexpr auto WIdx =
+        NDimSpatial == 2 ? Number<NonSpatialDimsNum + 1>{} : Number<NonSpatialDimsNum + 2>{};
+
+    static constexpr auto ZIdx = Number<NonSpatialDimsNum>{};
+    static constexpr auto YIdx =
+        NDimSpatial == 2 ? Number<NonSpatialDimsNum>{} : Number<NonSpatialDimsNum + 1>{};
+    static constexpr auto XIdx =
+        NDimSpatial == 2 ? Number<NonSpatialDimsNum + 1>{} : Number<NonSpatialDimsNum + 2>{};
+
     template <typename ALayout,
-              typename std::enable_if<NDimSpatial == 2 &&
+              typename std::enable_if<(NDimSpatial == 2 || NDimSpatial == 3) &&
                                           (is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
-                                           is_same_v<ALayout, tensor_layout::convolution::NHWGK>),
+                                           is_same_v<ALayout, tensor_layout::convolution::GNDHWK> ||
+                                           is_same_v<ALayout, tensor_layout::convolution::NHWGK> ||
+                                           is_same_v<ALayout, tensor_layout::convolution::NDHWGK>),
                                       bool>::type = false>
     static auto MakeADescriptor_AK0_M_AK1(
         const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_lengths,
@@ -100,35 +205,43 @@ struct TransformConvBwdDataToGemm_v1
         const std::array<index_t, NDimSpatial>& /* input_right_pads */,
         const std::array<index_t, NDimSpatial>& tildes)
     {
-        index_t i_ytilde = tildes[0];
-        index_t i_xtilde = tildes[1];
+        index_t i_ztilde = tildes[ZIdx - NonSpatialDimsNum];
+        index_t i_ytilde = tildes[YIdx - NonSpatialDimsNum];
+        index_t i_xtilde = tildes[XIdx - NonSpatialDimsNum];
 
         const index_t N = in_g_n_c_wis_lengths[1];
         const index_t K = wei_g_k_c_xs_lengths[1];
 
-        const index_t Hi = in_g_n_c_wis_lengths[3];
-        const index_t Wi = in_g_n_c_wis_lengths[4];
+        const index_t Di = NDimSpatial == 3 ? in_g_n_c_wis_lengths[DIdx] : 1;
+        const index_t Hi = in_g_n_c_wis_lengths[HIdx];
+        const index_t Wi = in_g_n_c_wis_lengths[WIdx];
 
-        const index_t Ho = out_g_n_k_wos_lengths[3];
-        const index_t Wo = out_g_n_k_wos_lengths[4];
+        const index_t Do = NDimSpatial == 3 ? out_g_n_k_wos_lengths[DIdx] : 1;
+        const index_t Ho = out_g_n_k_wos_lengths[HIdx];
+        const index_t Wo = out_g_n_k_wos_lengths[WIdx];
 
-        const index_t Y = wei_g_k_c_xs_lengths[3];
-        const index_t X = wei_g_k_c_xs_lengths[4];
+        const index_t Z = NDimSpatial == 3 ? wei_g_k_c_xs_lengths[ZIdx] : 1;
+        const index_t Y = wei_g_k_c_xs_lengths[YIdx];
+        const index_t X = wei_g_k_c_xs_lengths[XIdx];
 
-        const index_t InLeftPadH = input_left_pads[0];
-        const index_t InLeftPadW = input_left_pads[1];
+        const index_t InLeftPadD = input_left_pads[DIdx - NonSpatialDimsNum];
+        const index_t InLeftPadH = input_left_pads[HIdx - NonSpatialDimsNum];
+        const index_t InLeftPadW = input_left_pads[WIdx - NonSpatialDimsNum];
 
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
+        const index_t ConvStrideD = conv_filter_strides[DIdx - NonSpatialDimsNum];
+        const index_t ConvStrideH = conv_filter_strides[HIdx - NonSpatialDimsNum];
+        const index_t ConvStrideW = conv_filter_strides[WIdx - NonSpatialDimsNum];
 
-        const index_t ConvDilationH = conv_filter_dilations[0];
-        const index_t ConvDilationW = conv_filter_dilations[1];
+        const index_t ConvDilationD = conv_filter_dilations[DIdx - NonSpatialDimsNum];
+        const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
+        const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
 
         const index_t AK0 = K / AK1;
 
-        const auto out_n_ho_wo_k_grid_desc =
-            make_out_n_ho_wo_k_grid_desc<NDimSpatial, ALayout, ConvBwdDataSpecialization>(
-                N, Ho, Wo, K, out_g_n_k_wos_strides);
+        // n_do_ho_wo_k for 3d or n_ho_wo_k for 2d
+        const auto out_grid_desc =
+            make_out_grid_desc<NDimSpatial, ALayout, ConvBwdDataSpecialization>(
+                N, Do, Ho, Wo, K, out_g_n_k_wos_strides);
 
         if constexpr(ConvBwdDataSpecialization ==
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
@@ -136,8 +249,8 @@ struct TransformConvBwdDataToGemm_v1
         {
             // A: output tensor
             const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
-                out_n_ho_wo_k_grid_desc,
-                make_tuple(make_pass_through_transform(N * Ho * Wo),
+                out_grid_desc,
+                make_tuple(make_pass_through_transform(N * Do * Ho * Wo),
                            make_unmerge_transform(make_tuple(AK0, AK1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<1>{}, Sequence<0, 2>{}));
@@ -152,103 +265,208 @@ struct TransformConvBwdDataToGemm_v1
         }
         else
         {
+            const auto GcdStrideDilationD = math::gcd(ConvStrideD, ConvDilationD);
             const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
             const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
 
+            const auto ZTilde = ConvStrideD / GcdStrideDilationD;
             const auto YTilde = ConvStrideH / GcdStrideDilationH;
             const auto XTilde = ConvStrideW / GcdStrideDilationW;
 
+            const auto ZDot = math::integer_divide_ceil(Z, ZTilde);
             const auto YDot = math::integer_divide_ceil(Y, YTilde);
             const auto XDot = math::integer_divide_ceil(X, XTilde);
 
+            const auto DTilde =
+                Do + math::integer_divide_ceil(ConvDilationD * (Z - I1), ConvStrideD);
             const auto HTilde =
                 Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
             const auto WTilde =
                 Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
 
             // only work on HTilde and WTilde that contribute to non-padding area of input tensor
+            const auto IDTildeSliceBegin = math::integer_divide_floor(
+                math::max(I0, InLeftPadD - ConvDilationD * (ZTilde - I1)), ConvStrideD);
             const auto IHTildeSliceBegin = math::integer_divide_floor(
                 math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
             const auto IWTildeSliceBegin = math::integer_divide_floor(
                 math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
 
+            const auto IDTildeSliceEnd = math::min(
+                DTilde, math::integer_divide_ceil(InLeftPadD + Di - I1, ConvStrideD) + I1);
             const auto IHTildeSliceEnd = math::min(
                 HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
             const auto IWTildeSliceEnd = math::min(
                 WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
 
+            const auto DTildeSlice = IDTildeSliceEnd - IDTildeSliceBegin;
             const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
             const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
 
             // GemmK is different for each GEMM
+            const auto ZDotSlice = math::integer_divide_ceil(Z - i_ztilde, ZTilde);
             const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
             const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
-            // A: output tensor
-            const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
-                out_n_ho_wo_k_grid_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_pad_transform(Ho, I0, I0),
-                           make_pad_transform(Wo, I0, I0),
-                           make_pass_through_transform(K)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-            const auto out_n_ydot_htilde_xdot_wtilde_k_grid_desc = transform_tensor_descriptor(
-                out_n_hop_wop_k_grid_desc,
-                make_tuple(
-                    make_pass_through_transform(N),
-                    make_embed_transform(make_tuple(YDot, HTilde),
-                                         make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
-                    make_embed_transform(make_tuple(XDot, WTilde),
-                                         make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
-                    make_pass_through_transform(K)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-            const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc =
-                transform_tensor_descriptor(
-                    out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
+            if constexpr(NDimSpatial == 2)
+            {
+                // A: output tensor
+                const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
+                    out_grid_desc,
                     make_tuple(make_pass_through_transform(N),
-                               make_slice_transform(YDot, I0, YDotSlice),
-                               make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                               make_slice_transform(XDot, I0, XDotSlice),
-                               make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                               make_unmerge_transform(make_tuple(AK0, AK1))),
-                    make_tuple(Sequence<0>{},
-                               Sequence<1>{},
-                               Sequence<2>{},
-                               Sequence<3>{},
-                               Sequence<4>{},
-                               Sequence<5>{}),
-                    make_tuple(Sequence<0>{},
-                               Sequence<1>{},
-                               Sequence<2>{},
-                               Sequence<3>{},
-                               Sequence<4>{},
-                               Sequence<5, 6>{}));
+                               make_pad_transform(Ho, I0, I0),
+                               make_pad_transform(Wo, I0, I0),
+                               make_pass_through_transform(K)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
 
-            const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
-                out_n_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc,
-                make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, AK0)),
-                           make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                           make_pass_through_transform(AK1)),
-                make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+                const auto out_n_ydot_htilde_xdot_wtilde_k_grid_desc = transform_tensor_descriptor(
+                    out_n_hop_wop_k_grid_desc,
+                    make_tuple(
+                        make_pass_through_transform(N),
+                        make_embed_transform(make_tuple(YDot, HTilde),
+                                             make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
+                        make_embed_transform(make_tuple(XDot, WTilde),
+                                             make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
+                        make_pass_through_transform(K)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
 
-            const auto out_gemmak0_gemmm_gemmak1_grid_desc =
-                ck::tensor_operation::device::PadTensorDescriptor(
-                    out_gemmak0_gemmmraw_gemmak1_grid_desc,
-                    make_tuple(AK0, GemmMPerBlock, AK1),
-                    Sequence<false, DoPadGemmM, false>{});
+                const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc =
+                    transform_tensor_descriptor(
+                        out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
+                        make_tuple(make_pass_through_transform(N),
+                                   make_slice_transform(YDot, I0, YDotSlice),
+                                   make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
+                                   make_slice_transform(XDot, I0, XDotSlice),
+                                   make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
+                                   make_unmerge_transform(make_tuple(AK0, AK1))),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{},
+                                   Sequence<5>{}),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{},
+                                   Sequence<5, 6>{}));
 
-            return out_gemmak0_gemmm_gemmak1_grid_desc;
+                const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
+                    out_n_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc,
+                    make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, AK0)),
+                               make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
+                               make_pass_through_transform(AK1)),
+                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+                const auto out_gemmak0_gemmm_gemmak1_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        out_gemmak0_gemmmraw_gemmak1_grid_desc,
+                        make_tuple(AK0, GemmMPerBlock, AK1),
+                        Sequence<false, DoPadGemmM, false>{});
+
+                return out_gemmak0_gemmm_gemmak1_grid_desc;
+            }
+            else if constexpr(NDimSpatial == 3)
+            {
+                // A: output tensor
+                const auto out_n_hop_wop_k_grid_desc = transform_tensor_descriptor(
+                    out_grid_desc,
+                    make_tuple(make_pass_through_transform(N),
+                               make_pad_transform(Do, I0, I0),
+                               make_pad_transform(Ho, I0, I0),
+                               make_pad_transform(Wo, I0, I0),
+                               make_pass_through_transform(K)),
+                    make_tuple(
+                        Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                    make_tuple(
+                        Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+                const auto out_n_zdot_dtilde_ydot_htilde_xdot_wtilde_k_grid_desc =
+                    transform_tensor_descriptor(
+                        out_n_hop_wop_k_grid_desc,
+                        make_tuple(make_pass_through_transform(N),
+                                   make_embed_transform(
+                                       make_tuple(ZDot, DTilde),
+                                       make_tuple(-ConvDilationD / GcdStrideDilationD, I1)),
+                                   make_embed_transform(
+                                       make_tuple(YDot, HTilde),
+                                       make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
+                                   make_embed_transform(
+                                       make_tuple(XDot, WTilde),
+                                       make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
+                                   make_pass_through_transform(K)),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{}),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1, 2>{},
+                                   Sequence<3, 4>{},
+                                   Sequence<5, 6>{},
+                                   Sequence<7>{}));
+
+                const auto
+                    out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc =
+                        transform_tensor_descriptor(
+                            out_n_zdot_dtilde_ydot_htilde_xdot_wtilde_k_grid_desc,
+                            make_tuple(make_pass_through_transform(N),
+                                       make_slice_transform(ZDot, I0, ZDotSlice),
+                                       make_slice_transform(DTilde, IDTildeSliceBegin, DTildeSlice),
+                                       make_slice_transform(YDot, I0, YDotSlice),
+                                       make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
+                                       make_slice_transform(XDot, I0, XDotSlice),
+                                       make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
+                                       make_unmerge_transform(make_tuple(AK0, AK1))),
+                            make_tuple(Sequence<0>{},
+                                       Sequence<1>{},
+                                       Sequence<2>{},
+                                       Sequence<3>{},
+                                       Sequence<4>{},
+                                       Sequence<5>{},
+                                       Sequence<6>{},
+                                       Sequence<7>{}),
+                            make_tuple(Sequence<0>{},
+                                       Sequence<1>{},
+                                       Sequence<2>{},
+                                       Sequence<3>{},
+                                       Sequence<4>{},
+                                       Sequence<5>{},
+                                       Sequence<6>{},
+                                       Sequence<7, 8>{}));
+
+                const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
+                    out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc,
+                    make_tuple(
+                        make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice, AK0)),
+                        make_merge_transform(make_tuple(N, DTildeSlice, HTildeSlice, WTildeSlice)),
+                        make_pass_through_transform(AK1)),
+                    make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}, Sequence<8>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+                const auto out_gemmak0_gemmm_gemmak1_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        out_gemmak0_gemmmraw_gemmak1_grid_desc,
+                        make_tuple(AK0, GemmMPerBlock, AK1),
+                        Sequence<false, DoPadGemmM, false>{});
+
+                return out_gemmak0_gemmm_gemmak1_grid_desc;
+            }
+            else
+            {
+                throw std::runtime_error("wrong! only implemented for 2D and 3D now");
+            }
         }
     }
 
     template <typename BLayout,
-              typename std::enable_if<NDimSpatial == 2 &&
-                                          is_same_v<BLayout, tensor_layout::convolution::GKYXC>,
+              typename std::enable_if<(NDimSpatial == 2 || NDimSpatial == 3) &&
+                                          (is_same_v<BLayout, tensor_layout::convolution::GKYXC> ||
+                                           is_same_v<BLayout, tensor_layout::convolution::GKZYXC>),
                                       bool>::type = false>
     static auto MakeBDescriptor_BK0_N_BK1(
         const std::array<index_t, NDimSpatial + 3>& out_g_n_k_wos_lengths,
@@ -263,30 +481,35 @@ struct TransformConvBwdDataToGemm_v1
         const std::array<index_t, NDimSpatial>& /* input_right_pads */,
         const std::array<index_t, NDimSpatial>& tildes)
     {
-        index_t i_ytilde = tildes[0];
-        index_t i_xtilde = tildes[1];
+        index_t i_ztilde = tildes[ZIdx - NonSpatialDimsNum];
+        index_t i_ytilde = tildes[YIdx - NonSpatialDimsNum];
+        index_t i_xtilde = tildes[XIdx - NonSpatialDimsNum];
 
         const index_t N = in_g_n_c_wis_lengths[1];
         const index_t K = wei_g_k_c_xs_lengths[1];
         const index_t C = wei_g_k_c_xs_lengths[2];
 
-        const index_t Ho = out_g_n_k_wos_lengths[3];
-        const index_t Wo = out_g_n_k_wos_lengths[4];
+        const index_t Do = NDimSpatial == 3 ? out_g_n_k_wos_lengths[DIdx] : 1;
+        const index_t Ho = out_g_n_k_wos_lengths[HIdx];
+        const index_t Wo = out_g_n_k_wos_lengths[WIdx];
 
-        const index_t Y = wei_g_k_c_xs_lengths[3];
-        const index_t X = wei_g_k_c_xs_lengths[4];
+        const index_t Z = NDimSpatial == 3 ? wei_g_k_c_xs_lengths[ZIdx] : 1;
+        const index_t Y = wei_g_k_c_xs_lengths[YIdx];
+        const index_t X = wei_g_k_c_xs_lengths[XIdx];
 
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
+        const index_t ConvStrideD = conv_filter_strides[DIdx - NonSpatialDimsNum];
+        const index_t ConvStrideH = conv_filter_strides[HIdx - NonSpatialDimsNum];
+        const index_t ConvStrideW = conv_filter_strides[WIdx - NonSpatialDimsNum];
 
-        const index_t ConvDilationH = conv_filter_dilations[0];
-        const index_t ConvDilationW = conv_filter_dilations[1];
+        const index_t ConvDilationD = conv_filter_dilations[DIdx - NonSpatialDimsNum];
+        const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
+        const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
 
         const index_t BK0 = K / BK1;
 
         // assume packed
-        const auto wei_k_y_x_c_grid_desc =
-            make_naive_tensor_descriptor_packed(make_tuple(K, Y, X, C));
+        // k_y_x_c for 2d or k_z_y_x_c for 3d
+        const auto wei_grid_desc = make_wei_grid_desc<BLayout>(K, Z, Y, X, C);
 
         if constexpr(ConvBwdDataSpecialization ==
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
@@ -299,7 +522,7 @@ struct TransformConvBwdDataToGemm_v1
                                                        make_pass_through_transform(C)),
                                             make_tuple(Sequence<0>{}, Sequence<1>{}),
                                             make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
-            make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, C), make_tuple(I0, I1));
+            make_naive_tensor_descriptor(make_tuple(N * Do * Ho * Wo, C), make_tuple(I0, I1));
 
             const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc =
                 ck::tensor_operation::device::PadTensorDescriptor(
@@ -311,75 +534,163 @@ struct TransformConvBwdDataToGemm_v1
         }
         else
         {
+            const auto GcdStrideDilationD = math::gcd(ConvStrideD, ConvDilationD);
             const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
             const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
 
+            const auto ZTilde = ConvStrideD / GcdStrideDilationD;
             const auto YTilde = ConvStrideH / GcdStrideDilationH;
             const auto XTilde = ConvStrideW / GcdStrideDilationW;
 
+            const auto ZDot = math::integer_divide_ceil(Z, ZTilde);
             const auto YDot = math::integer_divide_ceil(Y, YTilde);
             const auto XDot = math::integer_divide_ceil(X, XTilde);
 
             // GemmK is different for each GEMM
+            const auto ZDotSlice = math::integer_divide_ceil(Z - i_ztilde, ZTilde);
             const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
             const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
             // B weight tensor
-            const auto wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc = transform_tensor_descriptor(
-                wei_k_y_x_c_grid_desc,
-                make_tuple(make_pass_through_transform(K),
-                           make_embed_transform(make_tuple(YDot, YTilde),
-                                                make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
-                           make_embed_transform(make_tuple(XDot, XTilde),
-                                                make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-            const auto wei_bk0_bk1_ydotslice_xdotslice_c_grid_desc =
-                transform_tensor_descriptor(wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                                            make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                                       make_slice_transform(YDot, I0, YDotSlice),
-                                                       make_slice_transform(XDot, I0, XDotSlice),
-                                                       make_freeze_transform(i_ytilde),
-                                                       make_freeze_transform(i_xtilde),
-                                                       make_pass_through_transform(C)),
-                                            make_tuple(Sequence<0>{},
-                                                       Sequence<1>{},
-                                                       Sequence<3>{},
-                                                       Sequence<2>{},
-                                                       Sequence<4>{},
-                                                       Sequence<5>{}),
-                                            make_tuple(Sequence<0, 1>{},
-                                                       Sequence<2>{},
-                                                       Sequence<3>{},
-                                                       Sequence<>{},
-                                                       Sequence<>{},
-                                                       Sequence<4>{}));
-
-            const auto wei_gemmbk0_gemmnraw_gemmbk1_grid_desc = transform_tensor_descriptor(
-                wei_bk0_bk1_ydotslice_xdotslice_c_grid_desc,
-                make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, BK0)),
-                           make_pass_through_transform(C),
-                           make_pass_through_transform(BK1)),
-                make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-
-            const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc =
-                ck::tensor_operation::device::PadTensorDescriptor(
-                    wei_gemmbk0_gemmnraw_gemmbk1_grid_desc,
+            if constexpr(NDimSpatial == 2)
+            {
+                const auto wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc = transform_tensor_descriptor(
+                    wei_grid_desc,
                     make_tuple(
-                        wei_gemmbk0_gemmnraw_gemmbk1_grid_desc.GetLength(I0), GemmNPerBlock, BK1),
-                    Sequence<false, DoPadGemmN, false>{});
+                        make_pass_through_transform(K),
+                        make_embed_transform(make_tuple(YDot, YTilde),
+                                             make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
+                        make_embed_transform(make_tuple(XDot, XTilde),
+                                             make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
+                        make_pass_through_transform(C)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
 
-            return wei_gemmbk0_gemmn_gemmbk1_grid_desc;
+                const auto wei_bk0_bk1_ydotslice_xdotslice_c_grid_desc =
+                    transform_tensor_descriptor(
+                        wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
+                        make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                                   make_slice_transform(YDot, I0, YDotSlice),
+                                   make_slice_transform(XDot, I0, XDotSlice),
+                                   make_freeze_transform(i_ytilde),
+                                   make_freeze_transform(i_xtilde),
+                                   make_pass_through_transform(C)),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<3>{},
+                                   Sequence<2>{},
+                                   Sequence<4>{},
+                                   Sequence<5>{}),
+                        make_tuple(Sequence<0, 1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<>{},
+                                   Sequence<>{},
+                                   Sequence<4>{}));
+
+                const auto wei_gemmbk0_gemmnraw_gemmbk1_grid_desc = transform_tensor_descriptor(
+                    wei_bk0_bk1_ydotslice_xdotslice_c_grid_desc,
+                    make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, BK0)),
+                               make_pass_through_transform(C),
+                               make_pass_through_transform(BK1)),
+                    make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+                const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        wei_gemmbk0_gemmnraw_gemmbk1_grid_desc,
+                        make_tuple(wei_gemmbk0_gemmnraw_gemmbk1_grid_desc.GetLength(I0),
+                                   GemmNPerBlock,
+                                   BK1),
+                        Sequence<false, DoPadGemmN, false>{});
+
+                return wei_gemmbk0_gemmn_gemmbk1_grid_desc;
+            }
+            else if constexpr(NDimSpatial == 3)
+            {
+                const auto wei_k_zdot_ztilde_ydot_ytilde_xdot_xtilde_c_grid_desc =
+                    transform_tensor_descriptor(
+                        wei_grid_desc,
+                        make_tuple(
+                            make_pass_through_transform(K),
+                            make_embed_transform(make_tuple(ZDot, ZTilde),
+                                                 make_tuple(ConvStrideD / GcdStrideDilationD, I1)),
+                            make_embed_transform(make_tuple(YDot, YTilde),
+                                                 make_tuple(ConvStrideH / GcdStrideDilationH, I1)),
+                            make_embed_transform(make_tuple(XDot, XTilde),
+                                                 make_tuple(ConvStrideW / GcdStrideDilationW, I1)),
+                            make_pass_through_transform(C)),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{}),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1, 2>{},
+                                   Sequence<3, 4>{},
+                                   Sequence<5, 6>{},
+                                   Sequence<7>{}));
+
+                const auto wei_bk0_bk1_zdotslice_ydotslice_xdotslice_c_grid_desc =
+                    transform_tensor_descriptor(
+                        wei_k_zdot_ztilde_ydot_ytilde_xdot_xtilde_c_grid_desc,
+                        make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                                   make_slice_transform(ZDot, I0, ZDotSlice),
+                                   make_slice_transform(YDot, I0, YDotSlice),
+                                   make_slice_transform(XDot, I0, XDotSlice),
+                                   make_freeze_transform(i_ztilde),
+                                   make_freeze_transform(i_ytilde),
+                                   make_freeze_transform(i_xtilde),
+                                   make_pass_through_transform(C)),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<3>{},
+                                   Sequence<5>{},
+                                   Sequence<2>{},
+                                   Sequence<4>{},
+                                   Sequence<6>{},
+                                   Sequence<7>{}),
+                        make_tuple(Sequence<0, 1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{},
+                                   Sequence<>{},
+                                   Sequence<>{},
+                                   Sequence<>{},
+                                   Sequence<5>{}));
+
+                const auto wei_gemmbk0_gemmnraw_gemmbk1_grid_desc = transform_tensor_descriptor(
+                    wei_bk0_bk1_zdotslice_ydotslice_xdotslice_c_grid_desc,
+                    make_tuple(
+                        make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice, BK0)),
+                        make_pass_through_transform(C),
+                        make_pass_through_transform(BK1)),
+                    make_tuple(Sequence<2, 3, 4, 0>{}, Sequence<5>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+                const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        wei_gemmbk0_gemmnraw_gemmbk1_grid_desc,
+                        make_tuple(wei_gemmbk0_gemmnraw_gemmbk1_grid_desc.GetLength(I0),
+                                   GemmNPerBlock,
+                                   BK1),
+                        Sequence<false, DoPadGemmN, false>{});
+
+                return wei_gemmbk0_gemmn_gemmbk1_grid_desc;
+            }
+            else
+            {
+                throw std::runtime_error("wrong! only implemented for 2D and 3D now");
+            }
         }
     }
 
     template <typename CLayout,
-              typename std::enable_if<NDimSpatial == 2 &&
+              typename std::enable_if<(NDimSpatial == 2 || NDimSpatial == 3) &&
                                           (is_same_v<CLayout, tensor_layout::convolution::GNHWC> ||
+                                           is_same_v<CLayout, tensor_layout::convolution::GNDHWC> ||
                                            is_same_v<CLayout, tensor_layout::convolution::NHWGC> ||
+                                           is_same_v<CLayout, tensor_layout::convolution::NDHWGC> ||
                                            is_same_v<CLayout, tensor_layout::convolution::G_NHW_C>),
                                       bool>::type = false>
     static auto
@@ -395,153 +706,309 @@ struct TransformConvBwdDataToGemm_v1
                         const std::array<index_t, NDimSpatial>& input_right_pads,
                         const std::array<index_t, NDimSpatial>& tildes)
     {
-        index_t i_ytilde = tildes[0];
-        index_t i_xtilde = tildes[1];
+        index_t i_ztilde = tildes[ZIdx - NonSpatialDimsNum];
+        index_t i_ytilde = tildes[YIdx - NonSpatialDimsNum];
+        index_t i_xtilde = tildes[XIdx - NonSpatialDimsNum];
 
         const index_t N = in_g_n_c_wis_lengths[1];
         const index_t C = wei_g_k_c_xs_lengths[2];
 
-        const index_t Hi = in_g_n_c_wis_lengths[3];
-        const index_t Wi = in_g_n_c_wis_lengths[4];
+        const index_t Di = NDimSpatial == 3 ? in_g_n_c_wis_lengths[DIdx] : 1;
+        const index_t Hi = in_g_n_c_wis_lengths[HIdx];
+        const index_t Wi = in_g_n_c_wis_lengths[WIdx];
 
-        const index_t Ho = out_g_n_k_wos_lengths[3];
-        const index_t Wo = out_g_n_k_wos_lengths[4];
+        const index_t Do = NDimSpatial == 3 ? out_g_n_k_wos_lengths[DIdx] : 1;
+        const index_t Ho = out_g_n_k_wos_lengths[HIdx];
+        const index_t Wo = out_g_n_k_wos_lengths[WIdx];
 
-        const index_t Y = wei_g_k_c_xs_lengths[3];
-        const index_t X = wei_g_k_c_xs_lengths[4];
+        const index_t Z = NDimSpatial == 3 ? wei_g_k_c_xs_lengths[ZIdx] : 1;
+        const index_t Y = wei_g_k_c_xs_lengths[YIdx];
+        const index_t X = wei_g_k_c_xs_lengths[XIdx];
 
-        const index_t InLeftPadH = input_left_pads[0];
-        const index_t InLeftPadW = input_left_pads[1];
+        const index_t InLeftPadD = input_left_pads[DIdx - NonSpatialDimsNum];
+        const index_t InLeftPadH = input_left_pads[HIdx - NonSpatialDimsNum];
+        const index_t InLeftPadW = input_left_pads[WIdx - NonSpatialDimsNum];
 
-        const index_t InRightPadH = input_right_pads[0];
-        const index_t InRightPadW = input_right_pads[1];
+        const index_t InRightPadD = input_right_pads[DIdx - NonSpatialDimsNum];
+        const index_t InRightPadH = input_right_pads[HIdx - NonSpatialDimsNum];
+        const index_t InRightPadW = input_right_pads[WIdx - NonSpatialDimsNum];
 
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
+        const index_t ConvStrideD = conv_filter_strides[DIdx - NonSpatialDimsNum];
+        const index_t ConvStrideH = conv_filter_strides[HIdx - NonSpatialDimsNum];
+        const index_t ConvStrideW = conv_filter_strides[WIdx - NonSpatialDimsNum];
 
-        const index_t ConvDilationH = conv_filter_dilations[0];
-        const index_t ConvDilationW = conv_filter_dilations[1];
+        const index_t ConvDilationD = conv_filter_dilations[DIdx - NonSpatialDimsNum];
+        const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
+        const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
 
         // assume strided
-        const auto in_n_hi_wi_c_grid_desc =
-            make_naive_tensor_descriptor(make_tuple(N, Hi, Wi, C),
-                                         make_tuple(in_g_n_c_wis_strides[1],
-                                                    in_g_n_c_wis_strides[3],
-                                                    in_g_n_c_wis_strides[4],
-                                                    in_g_n_c_wis_strides[2]));
+        // n_hi_wi_c for 2d n_di_hi_wi_c for 3d
+        const auto in_grid_desc =
+            make_in_grid_desc<NDimSpatial, CLayout>(N, Di, Hi, Wi, C, in_g_n_c_wis_strides);
 
         if constexpr(ConvBwdDataSpecialization ==
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
                          Filter1x1Stride1Pad0)
         {
             // C: input tensor
-            const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-                in_n_hi_wi_c_grid_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_embed_transform(make_tuple(I1, Ho), make_tuple(I1, ConvStrideH)),
-                           make_embed_transform(make_tuple(I1, Wo), make_tuple(I1, ConvStrideW)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
+            if constexpr(NDimSpatial == 2)
+            {
+                const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                    in_grid_desc,
+                    make_tuple(
+                        make_pass_through_transform(N),
+                        make_embed_transform(make_tuple(I1, Ho), make_tuple(I1, ConvStrideH)),
+                        make_embed_transform(make_tuple(I1, Wo), make_tuple(I1, ConvStrideW)),
+                        make_pass_through_transform(C)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
 
-            const auto in_gemmmraw_gemmnraw_grid_desc = transform_tensor_descriptor(
-                in_n_y_ho_x_wo_c_grid_desc,
-                make_tuple(make_freeze_transform(I0),
-                           make_freeze_transform(I0),
-                           make_merge_transform(make_tuple(N, Ho, Wo)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<1>{}, Sequence<3>{}, Sequence<0, 2, 4>{}, Sequence<5>{}),
-                make_tuple(Sequence<>{}, Sequence<>{}, Sequence<0>{}, Sequence<1>{}));
+                const auto in_gemmmraw_gemmnraw_grid_desc = transform_tensor_descriptor(
+                    in_n_y_ho_x_wo_c_grid_desc,
+                    make_tuple(make_freeze_transform(I0),
+                               make_freeze_transform(I0),
+                               make_merge_transform(make_tuple(N, Ho, Wo)),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<1>{}, Sequence<3>{}, Sequence<0, 2, 4>{}, Sequence<5>{}),
+                    make_tuple(Sequence<>{}, Sequence<>{}, Sequence<0>{}, Sequence<1>{}));
 
-            const auto in_gemmm_gemmn_grid_desc = ck::tensor_operation::device::PadTensorDescriptor(
-                in_gemmmraw_gemmnraw_grid_desc,
-                make_tuple(GemmMPerBlock, GemmNPerBlock),
-                Sequence<DoPadGemmM, DoPadGemmN>{});
+                const auto in_gemmm_gemmn_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        in_gemmmraw_gemmnraw_grid_desc,
+                        make_tuple(GemmMPerBlock, GemmNPerBlock),
+                        Sequence<DoPadGemmM, DoPadGemmN>{});
 
-            return in_gemmm_gemmn_grid_desc;
+                return in_gemmm_gemmn_grid_desc;
+            }
+            else if constexpr(NDimSpatial == 3)
+            {
+
+                // C: input tensor
+                const auto in_n_x_do_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                    in_grid_desc,
+                    make_tuple(
+                        make_pass_through_transform(N),
+                        make_embed_transform(make_tuple(I1, Do), make_tuple(I1, ConvStrideD)),
+                        make_embed_transform(make_tuple(I1, Ho), make_tuple(I1, ConvStrideH)),
+                        make_embed_transform(make_tuple(I1, Wo), make_tuple(I1, ConvStrideW)),
+                        make_pass_through_transform(C)),
+                    make_tuple(
+                        Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                    make_tuple(Sequence<0>{},
+                               Sequence<1, 2>{},
+                               Sequence<3, 4>{},
+                               Sequence<5, 6>{},
+                               Sequence<7>{}));
+
+                const auto in_gemmmraw_gemmnraw_grid_desc = transform_tensor_descriptor(
+                    in_n_x_do_y_ho_x_wo_c_grid_desc,
+                    make_tuple(make_freeze_transform(I0),
+                               make_freeze_transform(I0),
+                               make_freeze_transform(I0),
+                               make_merge_transform(make_tuple(N, Do, Ho, Wo)),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<1>{},
+                               Sequence<3>{},
+                               Sequence<5>{},
+                               Sequence<0, 2, 4, 6>{},
+                               Sequence<7>{}),
+                    make_tuple(
+                        Sequence<>{}, Sequence<>{}, Sequence<>{}, Sequence<0>{}, Sequence<1>{}));
+
+                const auto in_gemmm_gemmn_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        in_gemmmraw_gemmnraw_grid_desc,
+                        make_tuple(GemmMPerBlock, GemmNPerBlock),
+                        Sequence<DoPadGemmM, DoPadGemmN>{});
+
+                return in_gemmm_gemmn_grid_desc;
+            }
+            else
+            {
+                throw std::runtime_error("wrong! only implemented for 2D and 3D now");
+            }
         }
         else
         {
+            const auto GcdStrideDilationD = math::gcd(ConvStrideD, ConvDilationD);
             const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
             const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
 
+            const auto ZTilde = ConvStrideD / GcdStrideDilationD;
             const auto YTilde = ConvStrideH / GcdStrideDilationH;
             const auto XTilde = ConvStrideW / GcdStrideDilationW;
 
+            const auto DTilde =
+                Do + math::integer_divide_ceil(ConvDilationD * (Z - I1), ConvStrideD);
             const auto HTilde =
                 Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
             const auto WTilde =
                 Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
 
-            // only work on HTilde and WTilde that contribute to non-padding area of input tensor
+            // only work on DTilde, HTilde and WTilde that contribute to
+            // non-padding area of input tensor
+            const auto IDTildeSliceBegin = math::integer_divide_floor(
+                math::max(I0, InLeftPadD - ConvDilationD * (ZTilde - I1)), ConvStrideD);
             const auto IHTildeSliceBegin = math::integer_divide_floor(
                 math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
             const auto IWTildeSliceBegin = math::integer_divide_floor(
                 math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
 
+            const auto IDTildeSliceEnd = math::min(
+                DTilde, math::integer_divide_ceil(InLeftPadD + Di - I1, ConvStrideD) + I1);
             const auto IHTildeSliceEnd = math::min(
                 HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
             const auto IWTildeSliceEnd = math::min(
                 WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
 
+            const auto DTildeSlice = IDTildeSliceEnd - IDTildeSliceBegin;
             const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
             const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
 
             // C: input tensor
-            const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-                in_n_hi_wi_c_grid_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+            if constexpr(NDimSpatial == 2)
+            {
+                const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                    in_grid_desc,
+                    make_tuple(make_pass_through_transform(N),
+                               make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                               make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
 
-            const auto in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc = transform_tensor_descriptor(
-                in_n_hip_wip_c_grid_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_embed_transform(make_tuple(YTilde, HTilde),
-                                                make_tuple(ConvDilationH, ConvStrideH)),
-                           make_embed_transform(make_tuple(XTilde, WTilde),
-                                                make_tuple(ConvDilationW, ConvStrideW)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
+                const auto in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc =
+                    transform_tensor_descriptor(
+                        in_n_hip_wip_c_grid_desc,
+                        make_tuple(make_pass_through_transform(N),
+                                   make_embed_transform(make_tuple(YTilde, HTilde),
+                                                        make_tuple(ConvDilationH, ConvStrideH)),
+                                   make_embed_transform(make_tuple(XTilde, WTilde),
+                                                        make_tuple(ConvDilationW, ConvStrideW)),
+                                   make_pass_through_transform(C)),
+                        make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                        make_tuple(
+                            Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
 
-            const auto in_n_htildeslice_wtildeslice_c_grid_desc = transform_tensor_descriptor(
-                in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_freeze_transform(i_ytilde),
-                           make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
-                           make_freeze_transform(i_xtilde),
-                           make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{},
-                           Sequence<1>{},
-                           Sequence<2>{},
-                           Sequence<3>{},
-                           Sequence<4>{},
-                           Sequence<5>{}),
-                make_tuple(Sequence<0>{},
-                           Sequence<>{},
-                           Sequence<1>{},
-                           Sequence<>{},
-                           Sequence<2>{},
-                           Sequence<3>{}));
+                const auto in_n_htildeslice_wtildeslice_c_grid_desc = transform_tensor_descriptor(
+                    in_n_ytilde_htilde_xtilde_wtilde_c_grid_desc,
+                    make_tuple(make_pass_through_transform(N),
+                               make_freeze_transform(i_ytilde),
+                               make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
+                               make_freeze_transform(i_xtilde),
+                               make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<0>{},
+                               Sequence<1>{},
+                               Sequence<2>{},
+                               Sequence<3>{},
+                               Sequence<4>{},
+                               Sequence<5>{}),
+                    make_tuple(Sequence<0>{},
+                               Sequence<>{},
+                               Sequence<1>{},
+                               Sequence<>{},
+                               Sequence<2>{},
+                               Sequence<3>{}));
 
-            const auto in_gemmmraw_gemmnraw_grid_desc = transform_tensor_descriptor(
-                in_n_htildeslice_wtildeslice_c_grid_desc,
-                make_tuple(make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0, 1, 2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+                const auto in_gemmmraw_gemmnraw_grid_desc = transform_tensor_descriptor(
+                    in_n_htildeslice_wtildeslice_c_grid_desc,
+                    make_tuple(make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<0, 1, 2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-            const auto in_gemmm_gemmn_grid_desc = ck::tensor_operation::device::PadTensorDescriptor(
-                in_gemmmraw_gemmnraw_grid_desc,
-                make_tuple(GemmMPerBlock, GemmNPerBlock),
-                Sequence<DoPadGemmM, DoPadGemmN>{});
+                const auto in_gemmm_gemmn_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        in_gemmmraw_gemmnraw_grid_desc,
+                        make_tuple(GemmMPerBlock, GemmNPerBlock),
+                        Sequence<DoPadGemmM, DoPadGemmN>{});
 
-            return in_gemmm_gemmn_grid_desc;
+                return in_gemmm_gemmn_grid_desc;
+            }
+            else if(NDimSpatial == 3)
+            {
+                const auto in_n_dip_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                    in_grid_desc,
+                    make_tuple(make_pass_through_transform(N),
+                               make_pad_transform(Di, InLeftPadD, InRightPadD),
+                               make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                               make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                               make_pass_through_transform(C)),
+                    make_tuple(
+                        Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                    make_tuple(
+                        Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+                const auto in_n_ztilde_dtilde_ytilde_htilde_xtilde_wtilde_c_grid_desc =
+                    transform_tensor_descriptor(
+                        in_n_dip_hip_wip_c_grid_desc,
+                        make_tuple(make_pass_through_transform(N),
+                                   make_embed_transform(make_tuple(ZTilde, DTilde),
+                                                        make_tuple(ConvDilationD, ConvStrideD)),
+                                   make_embed_transform(make_tuple(YTilde, HTilde),
+                                                        make_tuple(ConvDilationH, ConvStrideH)),
+                                   make_embed_transform(make_tuple(XTilde, WTilde),
+                                                        make_tuple(ConvDilationW, ConvStrideW)),
+                                   make_pass_through_transform(C)),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{}),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1, 2>{},
+                                   Sequence<3, 4>{},
+                                   Sequence<5, 6>{},
+                                   Sequence<7>{}));
+
+                const auto in_n_dtildeslice_htildeslice_wtildeslice_c_grid_desc =
+                    transform_tensor_descriptor(
+                        in_n_ztilde_dtilde_ytilde_htilde_xtilde_wtilde_c_grid_desc,
+                        make_tuple(make_pass_through_transform(N),
+                                   make_freeze_transform(i_ztilde),
+                                   make_slice_transform(DTilde, IDTildeSliceBegin, DTildeSlice),
+                                   make_freeze_transform(i_ytilde),
+                                   make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
+                                   make_freeze_transform(i_xtilde),
+                                   make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
+                                   make_pass_through_transform(C)),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
+                                   Sequence<2>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{},
+                                   Sequence<5>{},
+                                   Sequence<6>{},
+                                   Sequence<7>{}),
+                        make_tuple(Sequence<0>{},
+                                   Sequence<>{},
+                                   Sequence<1>{},
+                                   Sequence<>{},
+                                   Sequence<2>{},
+                                   Sequence<>{},
+                                   Sequence<3>{},
+                                   Sequence<4>{}));
+
+                const auto in_gemmmraw_gemmnraw_grid_desc = transform_tensor_descriptor(
+                    in_n_dtildeslice_htildeslice_wtildeslice_c_grid_desc,
+                    make_tuple(
+                        make_merge_transform(make_tuple(N, DTildeSlice, HTildeSlice, WTildeSlice)),
+                        make_pass_through_transform(C)),
+                    make_tuple(Sequence<0, 1, 2, 3>{}, Sequence<4>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+                const auto in_gemmm_gemmn_grid_desc =
+                    ck::tensor_operation::device::PadTensorDescriptor(
+                        in_gemmmraw_gemmnraw_grid_desc,
+                        make_tuple(GemmMPerBlock, GemmNPerBlock),
+                        Sequence<DoPadGemmM, DoPadGemmN>{});
+                return in_gemmm_gemmn_grid_desc;
+            }
+            else
+            {
+                throw std::runtime_error("wrong! only implemented for 2D and 3D now");
+            }
         }
     }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
new file mode 100644
index 0000000000..c6c1697f0b
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -0,0 +1,141 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
+
+static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
+    ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+// f16_f16_f32_f16
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_f16_instances =
+    std::tuple<
+        // clang-format off
+        // ##############################################|       NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
+        
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#endif
+        // clang-format on
+        >;
+
+// bf16_bf16_f32_bf16
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_bf16_instances =
+    std::tuple<
+        // clang-format off
+        // ##############################################|       NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
+        
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
+#endif
+        // clang-format on
+        >;
+
+// f32_f32_f32_f32
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_f32_instances =
+    std::tuple<
+        // clang-format off
+         // ##############################################|       NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+         // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
+
+#ifdef CK_WORKAROUND_SWDEV_3318619
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#endif
+        // clang-format on
+        >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index 0b20e19a2a..58c9064535 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -16,7 +16,7 @@ namespace device {
 namespace instance {
 
 // conv2d backward data
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -30,7 +30,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -44,7 +44,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -58,7 +58,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -72,7 +72,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -86,7 +86,7 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -100,6 +100,91 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 
+// conv3d backward data
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -139,43 +224,96 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
-        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+        if constexpr(NumDimSpatial == 2)
         {
-            if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                         is_same_v<OutDataType, F16>)
+
+            if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
+                         is_same_v<OutLayout, GNHWK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
+                        op_ptrs);
+                }
             }
-            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                              is_same_v<OutDataType, F32>)
+            else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
+                              is_same_v<OutLayout, NHWGK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                              is_same_v<OutDataType, BF16>)
-            {
-                add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
-                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
+        else if constexpr(NumDimSpatial == 3)
         {
-            if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                         is_same_v<OutDataType, F16>)
+
+            if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
+                         is_same_v<OutLayout, GNDHWK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
+                        op_ptrs);
+                }
             }
-            else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
-                              is_same_v<OutDataType, F32>)
+            else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
+                              is_same_v<OutLayout, NDHWGK>)
             {
-                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
-                              is_same_v<OutDataType, BF16>)
-            {
-                add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
+                if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                             is_same_v<OutDataType, F16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
+                                  is_same_v<OutDataType, F32>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
+                                  is_same_v<OutDataType, BF16>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
+                        op_ptrs);
+                }
             }
         }
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index 01cd23b2e3..226dca5083 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -2,14 +2,14 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 // Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -26,19 +26,21 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
     // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_bf16_instances<GNHWK,
-                                                          GKYXC,
-                                                          Empty_Tuple,
-                                                          GNHWC,
-                                                          ConvBwdDataDefault>{});
+        device_grouped_conv_bwd_data_xdl_bf16_instances<2,
+                                                        GNHWK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        GNHWC,
+                                                        ConvBwdDataDefault>{});
     // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_bf16_instances<GNHWK,
-                                                          GKYXC,
-                                                          Empty_Tuple,
-                                                          GNHWC,
-                                                          ConvBwdDataFilter1x1Stride1Pad0>{});
+        device_grouped_conv_bwd_data_xdl_bf16_instances<2,
+                                                        GNHWK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        GNHWC,
+                                                        ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index bb0fd36e44..64fbf8bbf2 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -2,14 +2,14 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 // Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -26,19 +26,21 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f16_instances<GNHWK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         GNHWC,
-                                                         ConvBwdDataDefault>{});
+        device_grouped_conv_bwd_data_xdl_f16_instances<2,
+                                                       GNHWK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       GNHWC,
+                                                       ConvBwdDataDefault>{});
     // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f16_instances<GNHWK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         GNHWC,
-                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+        device_grouped_conv_bwd_data_xdl_f16_instances<2,
+                                                       GNHWK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       GNHWC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index b671017f51..f9351d96f2 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -2,14 +2,14 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 // Compilation parameters for out[g, n, hi, wi, c] * wei[g, k, y, x, c] = in[g, n, ho, wo, k]
-void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
                                                                   GKYXC,
@@ -26,19 +26,21 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f32_instances<GNHWK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         GNHWC,
-                                                         ConvBwdDataDefault>{});
+        device_grouped_conv_bwd_data_xdl_f32_instances<2,
+                                                       GNHWK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       GNHWC,
+                                                       ConvBwdDataDefault>{});
     // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f32_instances<GNHWK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         GNHWC,
-                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+        device_grouped_conv_bwd_data_xdl_f32_instances<2,
+                                                       GNHWK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       GNHWC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
deleted file mode 100644
index bfe6e47d19..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_instance.hpp
+++ /dev/null
@@ -1,141 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using BF16 = ck::bhalf_t;
-using F16  = ck::half_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NHWGC = ck::tensor_layout::convolution::NHWGC;
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-
-using NHWGK = ck::tensor_layout::convolution::NHWGK;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvBwdDataDefault = ConvolutionBackwardDataSpecialization::Default;
-
-static constexpr auto ConvBwdDataFilter1x1Stride1Pad0 =
-    ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
-
-// f16_f16_f32_f16
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          ConvolutionBackwardDataSpecialization ConvSpec>
-using device_grouped_conv2d_bwd_data_xdl_f16_instances = std::tuple<
-    // clang-format off
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
-        
-#ifdef CK_WORKAROUND_SWDEV_3318619
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-#endif
-    // clang-format on
-    >;
-
-// bf16_bf16_f32_bf16
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          ConvolutionBackwardDataSpecialization ConvSpec>
-using device_grouped_conv2d_bwd_data_xdl_bf16_instances = std::tuple<
-    // clang-format off
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
-        
-#ifdef CK_WORKAROUND_SWDEV_3318619
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
-#endif
-    // clang-format on
-    >;
-
-// f32_f32_f32_f32
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          ConvolutionBackwardDataSpecialization ConvSpec>
-using device_grouped_conv2d_bwd_data_xdl_f32_instances = std::tuple<
-    // clang-format off
-        // ##############################################|    NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################| Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|        |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
-
-#ifdef CK_WORKAROUND_SWDEV_3318619
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<       2, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-#endif
-    // clang-format on
-    >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
index 0b1fa63ddb..5d9194798b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -2,14 +2,14 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 // Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -26,19 +26,21 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
     // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_bf16_instances<NHWGK,
-                                                          GKYXC,
-                                                          Empty_Tuple,
-                                                          NHWGC,
-                                                          ConvBwdDataDefault>{});
+        device_grouped_conv_bwd_data_xdl_bf16_instances<2,
+                                                        NHWGK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        NHWGC,
+                                                        ConvBwdDataDefault>{});
     // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_bf16_instances<NHWGK,
-                                                          GKYXC,
-                                                          Empty_Tuple,
-                                                          NHWGC,
-                                                          ConvBwdDataFilter1x1Stride1Pad0>{});
+        device_grouped_conv_bwd_data_xdl_bf16_instances<2,
+                                                        NHWGK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        NHWGC,
+                                                        ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
index 1c8fc57e1c..5269bb9652 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -2,14 +2,14 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 // Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -26,19 +26,21 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f16_instances<NHWGK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         NHWGC,
-                                                         ConvBwdDataDefault>{});
+        device_grouped_conv_bwd_data_xdl_f16_instances<2,
+                                                       NHWGK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       NHWGC,
+                                                       ConvBwdDataDefault>{});
     // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f16_instances<NHWGK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         NHWGC,
-                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+        device_grouped_conv_bwd_data_xdl_f16_instances<2,
+                                                       NHWGK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       NHWGC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
index fcdfe8d42b..adfa08c1f7 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -2,14 +2,14 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_bwd_data_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 // Compilation parameters for out[n, hi, wi, g, c] * wei[g, k, y, x, c] = in[n, ho, wo, g, k]
-void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
+void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
                                                                   GKYXC,
@@ -26,19 +26,21 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
     // 1. Default
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         NHWGC,
-                                                         ConvBwdDataDefault>{});
+        device_grouped_conv_bwd_data_xdl_f32_instances<2,
+                                                       NHWGK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       NHWGC,
+                                                       ConvBwdDataDefault>{});
     // 2. Filter1x1Stride1Pad0
     add_device_operation_instances(
         instances,
-        device_grouped_conv2d_bwd_data_xdl_f32_instances<NHWGK,
-                                                         GKYXC,
-                                                         Empty_Tuple,
-                                                         NHWGC,
-                                                         ConvBwdDataFilter1x1Stride1Pad0>{});
+        device_grouped_conv_bwd_data_xdl_f32_instances<2,
+                                                       NHWGK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       NHWGC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
new file mode 100644
index 0000000000..8383d622b2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
@@ -0,0 +1,8 @@
+add_instance_library(device_grouped_conv3d_bwd_data_instance
+   device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+   device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+   device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+   device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+   device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
new file mode 100644
index 0000000000..8331ea1fda
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_bf16_instances<3,
+                                                        GNDHWK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        GNDHWC,
+                                                        ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_bf16_instances<3,
+                                                        GNDHWK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        GNDHWC,
+                                                        ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
new file mode 100644
index 0000000000..1885d49c81
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f16_instances<3,
+                                                       GNDHWK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       GNDHWC,
+                                                       ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f16_instances<3,
+                                                       GNDHWK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       GNDHWC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
new file mode 100644
index 0000000000..77135fcc05
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = in[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f32_instances<3,
+                                                       GNDHWK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       GNDHWC,
+                                                       ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f32_instances<3,
+                                                       GNDHWK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       GNDHWC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..663d41fe0b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  BF16,
+                                                                  BF16,
+                                                                  Empty_Tuple,
+                                                                  BF16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_bf16_instances<3,
+                                                        NDHWGK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        NDHWGC,
+                                                        ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_bf16_instances<3,
+                                                        NDHWGK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        NDHWGC,
+                                                        ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..ac0ab44ce3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f16_instances<3,
+                                                       NDHWGK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       NDHWGC,
+                                                       ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f16_instances<3,
+                                                       NDHWGK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       NDHWGC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..50d5cce73d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F32,
+                                                                  F32,
+                                                                  Empty_Tuple,
+                                                                  F32,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f32_instances<3,
+                                                       NDHWGK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       NDHWGC,
+                                                       ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_f32_instances<3,
+                                                       NDHWGK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       NDHWGC,
+                                                       ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 3f19096f48..d35c743939 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -81,4 +81,5 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_in
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_grouped_conv_bwd_data.cpp b/profiler/src/profile_grouped_conv_bwd_data.cpp
index 351bb72c81..55d199317a 100644
--- a/profiler/src/profile_grouped_conv_bwd_data.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_data.cpp
@@ -77,15 +77,10 @@ int profile_grouped_conv_bwd_data(int argc, char* argv[])
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
 
-    using GNHWC = ck::tensor_layout::convolution::GNHWC;
-    using NHWGC = ck::tensor_layout::convolution::NHWGC;
-
-    using GKYXC = ck::tensor_layout::convolution::GKYXC;
-
-    using GNHWK = ck::tensor_layout::convolution::GNHWK;
-    using NHWGK = ck::tensor_layout::convolution::NHWGK;
+    using namespace ck::tensor_layout::convolution;
 
     constexpr auto I2 = ck::Number<2>{};
+    constexpr auto I3 = ck::Number<3>{};
 
     auto profile = [&](auto num_dim_spatial_tmp,
                        auto out_layout,
@@ -116,36 +111,70 @@ int profile_grouped_conv_bwd_data(int argc, char* argv[])
         return pass ? 0 : 1;
     };
 
-    // GNHWC_GKYXC_GNHWK
-    if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
+    if(num_dim_spatial == 2)
     {
-        if(data_type == ConvDataType::F32_F32_F32)
+        if(layout == ConvLayout::GNHWC_GKYXC_GNHWK)
         {
-            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F32{}, F32{}, F32{});
+            if(data_type == ConvDataType::F32_F32_F32)
+            {
+                return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F32{}, F32{}, F32{});
+            }
+            else if(data_type == ConvDataType::F16_F16_F16)
+            {
+                return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F16{}, F16{}, F16{});
+            }
+            else if(data_type == ConvDataType::BF16_BF16_BF16)
+            {
+                return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, BF16{}, BF16{}, BF16{});
+            }
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        else if(layout == ConvLayout::NHWGC_GKYXC_NHWGK)
         {
-            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, F16{}, F16{}, F16{});
-        }
-        else if(data_type == ConvDataType::BF16_BF16_BF16)
-        {
-            return profile(I2, GNHWK{}, GKYXC{}, GNHWC{}, BF16{}, BF16{}, BF16{});
+            if(data_type == ConvDataType::F32_F32_F32)
+            {
+                return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F32{}, F32{}, F32{});
+            }
+            else if(data_type == ConvDataType::F16_F16_F16)
+            {
+                return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F16{}, F16{}, F16{});
+            }
+            else if(data_type == ConvDataType::BF16_BF16_BF16)
+            {
+                return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, BF16{}, BF16{}, BF16{});
+            }
         }
     }
-    // NHWGC_GKYXC_NHWGK
-    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    else if(num_dim_spatial == 3)
     {
-        if(data_type == ConvDataType::F32_F32_F32)
+        if(layout == ConvLayout::GNHWC_GKYXC_GNHWK)
         {
-            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F32{}, F32{}, F32{});
+            if(data_type == ConvDataType::F32_F32_F32)
+            {
+                return profile(I3, GNDHWK{}, GKZYXC{}, GNDHWC{}, F32{}, F32{}, F32{});
+            }
+            else if(data_type == ConvDataType::F16_F16_F16)
+            {
+                return profile(I3, GNDHWK{}, GKZYXC{}, GNDHWC{}, F16{}, F16{}, F16{});
+            }
+            else if(data_type == ConvDataType::BF16_BF16_BF16)
+            {
+                return profile(I3, GNDHWK{}, GKZYXC{}, GNDHWC{}, BF16{}, BF16{}, BF16{});
+            }
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        else if(layout == ConvLayout::NHWGC_GKYXC_NHWGK)
         {
-            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, F16{}, F16{}, F16{});
-        }
-        else if(data_type == ConvDataType::BF16_BF16_BF16)
-        {
-            return profile(I2, NHWGK{}, GKYXC{}, NHWGC{}, BF16{}, BF16{}, BF16{});
+            if(data_type == ConvDataType::F32_F32_F32)
+            {
+                return profile(I3, NDHWGK{}, GKZYXC{}, NDHWGC{}, F32{}, F32{}, F32{});
+            }
+            else if(data_type == ConvDataType::F16_F16_F16)
+            {
+                return profile(I3, NDHWGK{}, GKZYXC{}, NDHWGC{}, F16{}, F16{}, F16{});
+            }
+            else if(data_type == ConvDataType::BF16_BF16_BF16)
+            {
+                return profile(I3, NDHWGK{}, GKZYXC{}, NDHWGC{}, BF16{}, BF16{}, BF16{});
+            }
         }
     }
 
diff --git a/test/grouped_convnd_bwd_data/CMakeLists.txt b/test/grouped_convnd_bwd_data/CMakeLists.txt
index d7c9f46554..436ab2af76 100644
--- a/test/grouped_convnd_bwd_data/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_data/CMakeLists.txt
@@ -1,6 +1,6 @@
 if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
     add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
-    target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+    target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
     add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface.cpp)
     target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
 endif()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
index 94808669a3..c2aab749d3 100644
--- a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
@@ -46,23 +46,36 @@ class TestGroupedConvndBwdData : public ::testing::Test
     }
 };
 
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-using NHWGC = ck::tensor_layout::convolution::NHWGC;
+using namespace ck::tensor_layout::convolution;
 
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
+using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWK, GKYXC, GNHWC>,
+                                       std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
+                                       std::tuple<ck::bhalf_t, GNHWK, GKYXC, GNHWC>,
+                                       std::tuple<float, NHWGK, GKYXC, NHWGC>,
+                                       std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
+                                       std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>>;
 
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-using NHWGK = ck::tensor_layout::convolution::NHWGK;
+using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWK, GKZYXC, GNDHWC>,
+                                       std::tuple<ck::half_t, GNDHWK, GKZYXC, GNDHWC>,
+                                       std::tuple<ck::bhalf_t, GNDHWK, GKZYXC, GNDHWC>,
+                                       std::tuple<float, NDHWGK, GKZYXC, NDHWGC>,
+                                       std::tuple<ck::half_t, NDHWGK, GKZYXC, NDHWGC>,
+                                       std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>>;
 
-using KernelTypes = ::testing::Types<std::tuple<float, GNHWK, GKYXC, GNHWC>,
-                                     std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
-                                     std::tuple<ck::bhalf_t, GNHWK, GKYXC, GNHWC>,
-                                     std::tuple<float, NHWGK, GKYXC, NHWGC>,
-                                     std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
-                                     std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>>;
-TYPED_TEST_SUITE(TestGroupedConvndBwdData, KernelTypes);
+template <typename Tuple>
+class TestGroupedConvndBwdData2d : public TestGroupedConvndBwdData<Tuple>
+{
+};
 
-TYPED_TEST(TestGroupedConvndBwdData, Test2D)
+template <typename Tuple>
+class TestGroupedConvndBwdData3d : public TestGroupedConvndBwdData<Tuple>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdData2d, KernelTypes2d);
+TYPED_TEST_SUITE(TestGroupedConvndBwdData3d, KernelTypes3d);
+
+TYPED_TEST(TestGroupedConvndBwdData2d, Test2D)
 {
     this->conv_params.clear();
 
@@ -76,3 +89,15 @@ TYPED_TEST(TestGroupedConvndBwdData, Test2D)
         {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
     this->template Run<2>();
 }
+
+TYPED_TEST(TestGroupedConvndBwdData3d, Test3D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {3, 2, 16, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->template Run<3>();
+}

From 10732847e73496e59f398d894c50dd9a920f1bd4 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Fri, 21 Jul 2023 19:00:55 +0200
Subject: [PATCH 064/194] Grouped conv bwd wei NDHWGC/NDHWGK (#804)

---
 ...e_grouped_conv_bwd_weight_xdl_cshuffle.hpp | 420 ++++++++----------
 .../grouped_convolution_backward_weight.hpp   |  63 ++-
 .../grouped_conv3d_bwd_weight/CMakeLists.txt  |   3 +
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |  47 ++
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |  47 ++
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |  47 ++
 profiler/README.md                            |   4 +-
 .../src/profile_grouped_conv_bwd_weight.cpp   |  30 +-
 .../test_grouped_convnd_bwd_weight.cpp        |   5 +-
 9 files changed, 414 insertions(+), 252 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
index 7475dc414d..a55da9add2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -189,6 +189,30 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
     // TODO make A/B datatype different
     using ABDataType = InDataType;
 
+    // 1d
+    static constexpr bool is_GNWK_GKXC_GNWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNWK>;
+    // 2d
+    static constexpr bool is_NHWGK_GKYXC_NHWGC =
+        is_same_v<InLayout, tensor_layout::convolution::NHWGC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::NHWGK>;
+    static constexpr bool is_GNHWK_GKYXC_GNHWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNHWK>;
+    // 3d
+    static constexpr bool is_NDHWGK_GKZYXC_NDHWGC =
+        is_same_v<InLayout, tensor_layout::convolution::NDHWGC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::NDHWGK>;
+    static constexpr bool is_GNDHWK_GKZYXC_GNDHWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNDHWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNDHWK>;
+
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
@@ -213,6 +237,148 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
     static constexpr auto BBlockLdsN0PerBlock = NPerBlock / BBlockLdsN1PerBlock;
     static constexpr auto BBlockLdsN1Padding  = 4;
 
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
+    constexpr static auto
+    make_out_grid_desc(const ck::index_t N,
+                       const ck::index_t Ho,
+                       const ck::index_t Wo,
+                       const ck::index_t K,
+                       const std::array<ck::index_t, NDimSpatial + 3>& output_strides)
+    {
+        if constexpr(is_GNHWK_GKYXC_GNHWC)
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
+        }
+        else if constexpr(is_NHWGK_GKYXC_NHWGC)
+        {
+            const index_t WoStride = output_strides[4];
+            const auto KStride     = Number<1>{};
+            return make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, K),
+                                                make_tuple(WoStride, KStride));
+        }
+        else
+        {
+            throw std::runtime_error("wrong! unsupported layout: " + OutLayout::name());
+        }
+    }
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
+    constexpr static auto
+    make_in_grid_desc(const ck::index_t N,
+                      const ck::index_t Hi,
+                      const ck::index_t Wi,
+                      const ck::index_t C,
+                      const std::array<ck::index_t, NDimSpatial + 3>& input_strides)
+    {
+        if constexpr(is_GNHWK_GKYXC_GNHWC)
+        {
+            if constexpr(ConvBackwardWeightSpecialization ==
+                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+            {
+                return make_naive_tensor_descriptor_packed(make_tuple(N * Hi * Wi, C));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
+            }
+        }
+        else if constexpr(is_NHWGK_GKYXC_NHWGC)
+        {
+            const index_t NStride  = input_strides[1];
+            const index_t HiStride = input_strides[3];
+            const index_t WiStride = input_strides[4];
+            const auto CStride     = input_strides[2];
+            if constexpr(ConvBackwardWeightSpecialization ==
+                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N * Hi * Wi, C),
+                                                    make_tuple(WiStride, CStride));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(N, Hi, Wi, C), make_tuple(NStride, HiStride, WiStride, CStride));
+            }
+        }
+        else
+        {
+            throw std::runtime_error("wrong! unsupported layout: " + InLayout::name());
+        }
+    }
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    constexpr static auto
+    make_out_grid_desc(const ck::index_t N,
+                       const ck::index_t Do,
+                       const ck::index_t Ho,
+                       const ck::index_t Wo,
+                       const ck::index_t K,
+                       const std::array<ck::index_t, NDimSpatial + 3>& output_strides)
+    {
+        if constexpr(is_GNDHWK_GKZYXC_GNDHWC)
+        {
+            return make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
+        }
+        else if constexpr(is_NDHWGK_GKZYXC_NDHWGC)
+        {
+            const index_t WoStride = output_strides[5];
+            const auto KStride     = Number<1>{};
+            return make_naive_tensor_descriptor(make_tuple(N * Do * Ho * Wo, K),
+                                                make_tuple(WoStride, KStride));
+        }
+        else
+        {
+            throw std::runtime_error("wrong! unsupported layout: " + OutLayout::name());
+        }
+    }
+
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    constexpr static auto
+    make_in_grid_desc(const ck::index_t N,
+                      const ck::index_t Di,
+                      const ck::index_t Hi,
+                      const ck::index_t Wi,
+                      const ck::index_t C,
+                      const std::array<ck::index_t, NDimSpatial + 3>& input_strides)
+    {
+        if constexpr(is_GNDHWK_GKZYXC_GNDHWC)
+        {
+            if constexpr(ConvBackwardWeightSpecialization ==
+                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+            {
+                return make_naive_tensor_descriptor_packed(make_tuple(N * Di * Hi * Wi, C));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
+            }
+        }
+        else if constexpr(is_NDHWGK_GKZYXC_NDHWGC)
+        {
+            const index_t NStride  = input_strides[1];
+            const index_t DiStride = input_strides[3];
+            const index_t HiStride = input_strides[4];
+            const index_t WiStride = input_strides[5];
+            const auto CStride     = input_strides[2];
+            if constexpr(ConvBackwardWeightSpecialization ==
+                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N * Di * Hi * Wi, C),
+                                                    make_tuple(WiStride, CStride));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(N, Di, Hi, Wi, C),
+                    make_tuple(NStride, DiStride, HiStride, WiStride, CStride));
+            }
+        }
+        else
+        {
+            throw std::runtime_error("wrong! unsupported layout: " + InLayout::name());
+        }
+    }
+
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
         const ck::index_t N,
@@ -367,184 +533,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         }
     }
 
-    template <ck::index_t NDim,
-              typename ck::enable_if<NDim == 2 &&
-                                         is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
-                                         is_same_v<OutLayout, tensor_layout::convolution::GNHWK>,
-                                     bool>::type = false>
-    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        const ck::index_t N,
-        const ck::index_t K,
-        const ck::index_t C,
-        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
-        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
-        const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
-        const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
-        const std::array<ck::index_t, NDimSpatial>& input_left_pads,
-        const std::array<ck::index_t, NDimSpatial>& input_right_pads,
-        const ck::index_t batch_k)
-    {
-        using namespace ck;
-
-        const index_t Hi = input_spatial_lengths[0];
-        const index_t Wi = input_spatial_lengths[1];
-
-        const index_t Ho = output_spatial_lengths[0];
-        const index_t Wo = output_spatial_lengths[1];
-
-        const index_t Y = filter_spatial_lengths[0];
-        const index_t X = filter_spatial_lengths[1];
-
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
-
-        const index_t ConvDilationH = conv_filter_dilations[0];
-        const index_t ConvDilationW = conv_filter_dilations[1];
-
-        const index_t InLeftPadH = input_left_pads[0];
-        const index_t InLeftPadW = input_left_pads[1];
-
-        const index_t InRightPadH = input_right_pads[0];
-        const index_t InRightPadW = input_right_pads[1];
-
-        const index_t GemmKTotal = N * Ho * Wo;
-        const index_t GemmM      = K;
-        const index_t GemmN      = C * X * Y;
-
-        const index_t GemmKBatch = batch_k;
-        const index_t GemmK0 =
-            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
-            K0PerBlock;
-        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
-
-        if constexpr(ConvBackwardWeightSpecialization ==
-                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
-        {
-            // A: output tensor
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-            // B: input tensor
-            const auto in_gemmktotal_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Hi * Wi, C));
-
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-            // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
-
-            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
-        }
-        else
-        {
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-            const auto in_n_hi_wi_c_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
-
-            // A: output tensor
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-            // B: input tensor
-            const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-                in_n_hi_wi_c_grid_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-            const auto in_n_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
-                in_n_hip_wip_c_grid_desc,
-                make_tuple(
-                    make_pass_through_transform(N),
-                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                    make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-            const auto in_gemmktotal_gemmn_grid_desc =
-                transform_tensor_descriptor(in_n_y_ho_x_wo_c_grid_desc,
-                                            make_tuple(make_merge_transform(make_tuple(Y, X, C)),
-                                                       make_merge_transform(make_tuple(N, Ho, Wo))),
-                                            make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
-                                            make_tuple(Sequence<1>{}, Sequence<0>{}));
-
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
-
-            // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
-
-            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
-        }
-    }
-
-    template <ck::index_t NDim,
-              typename ck::enable_if<NDim == 2 &&
-                                         is_same_v<InLayout, tensor_layout::convolution::NHWGC> &&
-                                         is_same_v<OutLayout, tensor_layout::convolution::NHWGK>,
-                                     bool>::type = false>
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
         const ck::index_t N,
         const ck::index_t K,
@@ -587,29 +576,21 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const index_t GemmM      = K;
         const index_t GemmN      = C * X * Y;
 
-        const index_t NStride  = input_strides[1];
-        const index_t HiStride = input_strides[3];
-        const index_t WiStride = input_strides[4];
-        const auto CStride     = input_strides[2];
-
-        const index_t WoStride = output_strides[4];
-        const auto KStride     = Number<1>{};
-
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
             K0PerBlock;
         const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
 
+        const auto out_grid_desc = make_out_grid_desc<NDim>(N, Ho, Wo, K, output_strides);
+        const auto in_grid_desc  = make_in_grid_desc<NDim>(N, Hi, Wi, C, input_strides);
+
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
             // A: output tensor
-            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
-                make_tuple(N * Ho * Wo, K), make_tuple(WoStride, KStride));
-
             const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
+                out_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
                            make_pass_through_transform(GemmM)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
@@ -623,11 +604,8 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // B: input tensor
-            const auto in_gemmktotal_gemmn_grid_desc = make_naive_tensor_descriptor(
-                make_tuple(N * Hi * Wi, C), make_tuple(WiStride, CStride));
-
             const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
+                in_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
                            make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
@@ -650,14 +628,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         }
         else
         {
-            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
-                make_tuple(N * Ho * Wo, K), make_tuple(WoStride, KStride));
-            const auto in_n_hi_wi_c_grid_desc = make_naive_tensor_descriptor(
-                make_tuple(N, Hi, Wi, C), make_tuple(NStride, HiStride, WiStride, CStride));
-
             // A: output tensor
             const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
+                out_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
                            make_pass_through_transform(GemmM)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
@@ -672,7 +645,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
             // B: input tensor
             const auto in_n_hip_wip_c_grid_desc = transform_tensor_descriptor(
-                in_n_hi_wi_c_grid_desc,
+                in_grid_desc,
                 make_tuple(make_pass_through_transform(N),
                            make_pad_transform(Hi, InLeftPadH, InRightPadH),
                            make_pad_transform(Wi, InLeftPadW, InRightPadW),
@@ -729,8 +702,8 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
-        const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -777,15 +750,15 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
             K0PerBlock;
         const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
 
+        const auto out_grid_desc = make_out_grid_desc<NDim>(N, Do, Ho, Wo, K, output_strides);
+        const auto in_grid_desc  = make_in_grid_desc<NDim>(N, Di, Hi, Wi, C, input_strides);
+
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
             // A: output tensor
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
-
             const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
+                out_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
                            make_pass_through_transform(GemmM)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
@@ -799,11 +772,8 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // B: input tensor
-            const auto in_gemmktotal_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Di * Hi * Wi, C));
-
             const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
+                in_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
                            make_pass_through_transform(GemmN)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
@@ -826,14 +796,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         }
         else
         {
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
-            const auto in_n_di_hi_wi_c_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
-
             // A: output tensor
             const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
+                out_grid_desc,
                 make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
                            make_pass_through_transform(GemmM)),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
@@ -848,7 +813,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
             // B: input tensor
             const auto in_n_dip_hip_wip_c_grid_desc = transform_tensor_descriptor(
-                in_n_di_hi_wi_c_grid_desc,
+                in_grid_desc,
                 make_tuple(make_pass_through_transform(N),
                            make_pad_transform(Di, InLeftPadD, InRightPadD),
                            make_pad_transform(Hi, InLeftPadH, InRightPadH),
@@ -1310,34 +1275,21 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
     {
         if constexpr(NDimSpatial == 1)
         {
-            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNWC> &&
-                           is_same_v<WeiLayout, tensor_layout::convolution::GKXC> &&
-                           is_same_v<OutLayout, tensor_layout::convolution::GNWK>))
+            if constexpr(!is_GNWK_GKXC_GNWC)
             {
                 return false;
             }
         }
         else if constexpr(NDimSpatial == 2)
         {
-            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
-                           is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
-                           is_same_v<OutLayout,
-                                     tensor_layout::convolution::
-                                         GNHWK>)&&!(is_same_v<InLayout,
-                                                              tensor_layout::convolution::NHWGC> &&
-                                                    is_same_v<WeiLayout,
-                                                              tensor_layout::convolution::GKYXC> &&
-                                                    is_same_v<OutLayout,
-                                                              tensor_layout::convolution::NHWGK>))
+            if constexpr(!(is_NHWGK_GKYXC_NHWGC || is_GNHWK_GKYXC_GNHWC))
             {
                 return false;
             }
         }
         else if constexpr(NDimSpatial == 3)
         {
-            if constexpr(!(is_same_v<InLayout, tensor_layout::convolution::GNDHWC> &&
-                           is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
-                           is_same_v<OutLayout, tensor_layout::convolution::GNDHWK>))
+            if constexpr(!(is_NDHWGK_GKZYXC_NDHWGC || is_GNDHWK_GKZYXC_GNDHWC))
             {
                 return false;
             }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index d0c3338f24..cad3e1ace8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -164,6 +164,42 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -273,8 +309,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         }
         else if constexpr(NumDimSpatial == 3)
         {
-            if(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
-               is_same_v<OutLayout, GNDHWK>)
+            if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
+                         is_same_v<OutLayout, GNDHWK>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
@@ -296,6 +332,29 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                         op_ptrs);
                 }
             }
+            else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
+                              is_same_v<OutLayout, NDHWGK>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
+            }
         }
 
         return op_ptrs;
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
index 04cad43e75..5118599b4f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
@@ -2,4 +2,7 @@ add_instance_library(device_grouped_conv3d_bwd_weight_instance
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..1e5c1946fb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<3,
+                                                                    NDHWGC,
+                                                                    GKZYXC,
+                                                                    NDHWGK,
+                                                                    ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..e0b442bf24
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<3,
+                                                                   NDHWGC,
+                                                                   GKZYXC,
+                                                                   NDHWGK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..4bb7948245
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -0,0 +1,47 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<3,
+                                                                   NDHWGC,
+                                                                   GKZYXC,
+                                                                   NDHWGK,
+                                                                   ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances<
+                                       3,
+                                       NDHWGC,
+                                       GKZYXC,
+                                       NDHWGK,
+                                       ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/README.md b/profiler/README.md
index a18ebc2265..e8ac3a4d2b 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -144,7 +144,7 @@ GB/s: 127.947
 
 ## Profile grouped convolution backward weight kernels
 ```bash
-# arg1: tensor operation (grouped_conv_bwd_data: Grouped Convolution Backward Data)
+# arg1: tensor operation (grouped_conv_bwd_weight: Grouped Convolution Backward Weight)
 # arg2: data type (0: Input fp32, Weight fp32, Output fp32
 #                  1: Input fp16, Weight fp16, Output fp16
 #                  2: Input bf16, Weight fp32, Output bf16)
@@ -167,7 +167,7 @@ GB/s: 127.947
 # SplitK
 
  ################                   op   datatype  layout  verify  init  log  time  Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx  SplitK
-./bin/ckProfiler grouped_conv_bwd_data          1       0       1     1    0     1      2 32 256 256 512  3  3  28  28   1   1   1   1       1       0        0        0       1
+./bin/ckProfiler grouped_conv_bwd_weight          1       0       1     1    0     1      2 32 256 256 512  3  3  28  28   1   1   1   1       1       0        0        0       1
 
  ```
 
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index d8edcf0f65..be8a3230f7 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -83,19 +83,7 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
 
-    using GNWC   = ck::tensor_layout::convolution::GNWC;
-    using GNHWC  = ck::tensor_layout::convolution::GNHWC;
-    using NHWGC  = ck::tensor_layout::convolution::NHWGC;
-    using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-
-    using GKXC   = ck::tensor_layout::convolution::GKXC;
-    using GKYXC  = ck::tensor_layout::convolution::GKYXC;
-    using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-
-    using GNWK   = ck::tensor_layout::convolution::GNWK;
-    using GNHWK  = ck::tensor_layout::convolution::GNHWK;
-    using NHWGK  = ck::tensor_layout::convolution::NHWGK;
-    using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
+    using namespace ck::tensor_layout::convolution;
 
     constexpr auto I1 = ck::Number<1>{};
     constexpr auto I2 = ck::Number<2>{};
@@ -194,6 +182,22 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
             return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, BF16{}, F32{}, BF16{});
         }
     }
+    else if(num_dim_spatial == 3 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    {
+        if(data_type == ConvDataType::F32_F32_F32)
+        {
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F32{}, F32{}, F32{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16)
+        {
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{});
+        }
+        else if(data_type == ConvDataType::BF16_F32_BF16)
+        {
+            // fp32 atomic add is used for weight tensor in bf16 kernel
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, BF16{}, F32{}, BF16{});
+        }
+    }
 
     std::cout << "this data_type & layout is not implemented" << std::endl;
 
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
index b6c79fa80b..86d99c4a53 100644
--- a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -85,7 +85,10 @@ using KernelTypes2d = ::testing::Types<
 using KernelTypes3d = ::testing::Types<
     std::tuple<float, float, float, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
     std::tuple<ck::half_t, ck::half_t, ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
-    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>>;
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
+    std::tuple<float, float, float, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>>;
 
 TYPED_TEST_SUITE(TestGroupedConvndBwdWeight1d, KernelTypes1d);
 TYPED_TEST_SUITE(TestGroupedConvndBwdWeight2d, KernelTypes2d);

From 7a29f711d48198177a960ce095d9405cdd883dba Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 21 Jul 2023 15:31:19 -0700
Subject: [PATCH 065/194] add INSTANCES_ONLY cmake macro to build only
 instances (#807)

---
 CMakeLists.txt | 65 +++++++++++++++++++++++++++++++++++++++-----------
 1 file changed, 51 insertions(+), 14 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 3a3b98d627..c0de76c580 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -296,31 +296,68 @@ file(GLOB_RECURSE INSTANCE_FILES "${PROJECT_SOURCE_DIR}/*/device_*_instance.cpp"
 file(GLOB dir_list RELATIVE ${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu ${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/*)
 set(CK_DEVICE_INSTANCES)
 FOREACH(subdir_path ${dir_list})
-    IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}")
-       list(APPEND CK_DEVICE_INSTANCES device_${subdir_path}_instance)
-    ENDIF()
+set(target_dir)
+IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}")
+    set(cmake_instance)
+    file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
+    set(add_inst 0)
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8")
+            #message("fp8 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
+            #message("fp16 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32")
+            #message("fp32 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64")
+            #message("fp64 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16")
+            #message("bf16 instance found!")
+            set(add_inst 1)
+    endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8")
+            #message("int8 instance found!")
+            set(add_inst 1)
+    endif()
+    if(NOT "${cmake_instance}" MATCHES "DTYPES")
+            #message("instance should be built for all types!")
+            set(add_inst 1)
+    endif()
+    if(add_inst EQUAL 1 OR NOT DEFINED DTYPES)
+      list(APPEND CK_DEVICE_INSTANCES device_${subdir_path}_instance)
+    endif()
+ENDIF()
 ENDFOREACH()
-add_custom_target(instances DEPENDS utility;${CK_DEVICE_INSTANCES}  SOURCES ${INSTANCE_FILES})
 
-rocm_package_setup_component(tests
+add_custom_target(instances DEPENDS utility;${CK_DEVICE_INSTANCES}  SOURCES ${INSTANCE_FILES})
+add_subdirectory(library)
+
+if(NOT DEFINED INSTANCES_ONLY)
+   rocm_package_setup_component(tests
         LIBRARY_NAME composablekernel
         PACKAGE_NAME tests # Prevent -static suffix on package name
-)
+   )
 
-rocm_package_setup_component(examples
+   rocm_package_setup_component(examples
         LIBRARY_NAME composablekernel
         PACKAGE_NAME examples
-)
+   )
 
-rocm_package_setup_component(profiler
+   rocm_package_setup_component(profiler
         LIBRARY_NAME composablekernel
         PACKAGE_NAME ckProfiler
-)
+   )
 
-add_subdirectory(library)
-add_subdirectory(example)
-add_subdirectory(test)
-add_subdirectory(profiler)
+   add_subdirectory(example)
+   add_subdirectory(test)
+   add_subdirectory(profiler)
+endif()
 
 #Create an interface target for the include only files and call it "composablekernels"
 include(CMakePackageConfigHelpers)

From 844b215d924f39c5dddb1d9cacd8e63ea4a26ef5 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 21 Jul 2023 15:33:17 -0700
Subject: [PATCH 066/194] add ninja profiling tools to the base docker (#805)

---
 Dockerfile | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/Dockerfile b/Dockerfile
index 3ff29c9a98..8d11a29853 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -48,6 +48,7 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     libpthread-stubs0-dev \
     llvm-amdgpu \
     pkg-config \
+    python \
     python3 \
     python3-dev \
     python3-pip \
@@ -63,6 +64,10 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     rm -rf /var/lib/apt/lists/*
 
 #Install latest version of cmake
+RUN wget -qO /usr/local/bin/ninja.gz https://github.com/ninja-build/ninja/releases/latest/download/ninja-linux.zip
+RUN gunzip /usr/local/bin/ninja.gz
+RUN chmod a+x /usr/local/bin/ninja
+RUN git clone https://github.com/nico/ninjatracing.git
 RUN apt purge --auto-remove -y cmake
 RUN apt update
 RUN apt install -y software-properties-common lsb-release

From 50643dd5551e988425b6297d5dbe6d200bd63735 Mon Sep 17 00:00:00 2001
From: ltqin <letao.qin@amd.com>
Date: Tue, 25 Jul 2023 09:08:15 +0800
Subject: [PATCH 067/194] Add bias scalar vectorload = 1 for gemm bias gemm
 (#791)

* first change bias load

* add bias dim and scalervector parameter

* make CDE0BlockTransferSrcVectorDim not work

* changse toinstance

* add limit for CDE0BlockTransferSrcScalarPerVector
---
 ...ed_gemm_add_add_relu_gemm_add_xdl_fp16.cpp |   4 +-
 ...ultiple_d_gemm_multiple_d_xdl_cshuffle.hpp |   4 +
 ...iple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp | 114 ++++++++----------
 ...6_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp |  30 ++---
 ...6_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp |  32 ++---
 5 files changed, 91 insertions(+), 93 deletions(-)

diff --git a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
index b3d0ab6bf7..36dcf58d70 100644
--- a/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
+++ b/example/37_batched_gemm_add_add_relu_gemm_add/batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp
@@ -173,6 +173,8 @@ using DeviceGemmInstance =
         8,
         8,
         true,
+        9,           // D0sTransferSrcVectorDim
+        4,           // D0sTransferSrcScalaerPerVector
         S<8, 32, 1>, // B1BlockTransfer
         S<0, 2, 1>,
         S<0, 2, 1>,
@@ -189,7 +191,7 @@ int main(int argc, char* argv[])
 {
     bool do_verification = true;
     int init_method      = 1;
-    bool time_kernel     = false;
+    bool time_kernel     = true;
 
     // GEMM shape
     ck::index_t M              = 1024;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index d259ebffae..d92865a9ec 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -196,6 +196,8 @@ template <typename A0Layout,
           index_t B0BlockTransferSrcScalarPerVector,
           index_t B0BlockTransferDstScalarPerVector_BK1,
           bool B0BlockLdsExtraN,
+          index_t CDE0BlockTransferSrcVectorDim,
+          index_t CDE0BlockTransferSrcScalaerPerVector,
           typename B1BlockTransferThreadClusterLengths_BK0_N_BK1,
           typename B1BlockTransferThreadClusterArrangeOrder,
           typename B1BlockTransferSrcAccessOrder,
@@ -492,6 +494,8 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
         B0BlockTransferDstScalarPerVector_BK1,
         true,
         B0BlockLdsExtraN,
+        CDE0BlockTransferSrcVectorDim,
+        CDE0BlockTransferSrcScalaerPerVector,
         B1BlockTransferThreadClusterLengths_BK0_N_BK1,
         B1BlockTransferThreadClusterArrangeOrder,
         B1BlockTransferSrcAccessOrder,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
index 59d6bad5d9..2c40defddd 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
@@ -67,6 +67,8 @@ template <typename A0B0B1DataType, // FIXME: don't assume A0/B0/B1 have same dat
           index_t B0BlockTransferDstScalarPerVector_BK1,
           bool B0ThreadTransferSrcResetCoordinateAfterRun, // ignored
           index_t B0BlockLdsExtraN,
+          index_t CDE0BlockTransferSrcVectorDim,
+          index_t CDE0BlockTransferSrcScalarPerVector,
           typename B1BlockTransferThreadClusterLengths_BK0_N_BK1,
           typename B1BlockTransferThreadClusterArrangeOrder,
           typename B1BlockTransferSrcAccessOrder,
@@ -710,13 +712,13 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
         constexpr auto d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5 =
             make_naive_tensor_descriptor_packed(make_tuple(I1,   // MBlockId
                                                            I1,   // NBlockID
-                                                           I1,   // MRepeat
-                                                           I1,   // NRepeat
-                                                           I1,   // MWaveId
-                                                           I1,   // NWaveId
-                                                           I1,   // MPerXdl
-                                                           I1,   // NGroupNum
-                                                           I1,   // NInputNum
+                                                           m0,   // MRepeat
+                                                           n0,   // NRepeat
+                                                           m1,   // MWaveId
+                                                           n1,   // NWaveId
+                                                           m2,   // MPerXdl
+                                                           n2,   // NGroupNum
+                                                           n3,   // NInputNum
                                                            n4)); // registerNum
 
         auto d0s_thread_buf = generate_tuple(
@@ -732,8 +734,9 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
         const auto wave_id     = GetGemm0WaveIdx();
         const auto wave_m_n_id = GetGemm0WaveMNIdx(wave_id[I2]); // I2: 0~63
 
-        constexpr auto acc0_thread_desc = make_naive_tensor_descriptor_packed(
-            make_tuple(Number<Gemm0MXdlPerWave>{}, Number<Gemm0NXdlPerWave>{}, n2, n4));
+        static_assert(CDE0BlockTransferSrcScalarPerVector <= n4,
+                      "vector load must be not greater than n4");
+        static_assert(n4 % CDE0BlockTransferSrcScalarPerVector == 0);
 
         auto d0s_threadwise_copy = generate_tuple(
             [&](auto i) {
@@ -742,10 +745,19 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
                     A0B0B1DataType,
                     decltype(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i]),
                     decltype(d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5),
-                    Sequence<I1, I1, I1, I1, I1, I1, I1, I1, I1, n4>,
+                    Sequence<I1, // MBlockId
+                             I1, // NBlockID
+                             m0, // MRepeat
+                             n0, // NRepeat
+                             m1, // MWaveId
+                             n1, // NWaveId
+                             m2, // MPerXdl
+                             n2, // NGroupNum
+                             n3, // NInputNum
+                             n4>,
                     Sequence<0, 1, 2, 3, 4, 5, 6, 7, 8, 9>,
-                    9,
-                    n4,
+                    9, // CDE0BlockTransferSrcVectorDim
+                    CDE0BlockTransferSrcScalarPerVector,
                     1,
                     false>(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
                            make_multi_index(block_work_idx[I0], // MBlockId
@@ -898,66 +910,42 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
                                                                     blockwise_gemm0,
                                                                     acc0_thread_buf,
                                                                     num_k_block_main_loop);
-            // bias+gelu
+            // multiple d
+            if constexpr(NumD0Tensor)
             {
-                static_for<0, Gemm0MXdlPerWave, 1>{}([&](auto mr) {
-                    static_for<0, Gemm0NXdlPerWave, 1>{}([&](auto nr) {
-                        static_for<0, n2, 1>{}([&](auto groupid) {
-                            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                                d0s_threadwise_copy(i).Run(
-                                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                    d0s_grid_buf[i],
-                                    d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
-                                    make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
-                                    d0s_thread_buf(i));
-                            });
+                static_for<0, NumD0Tensor, 1>{}([&](auto i) {
+                    d0s_threadwise_copy(i).Run(d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
+                                               d0s_grid_buf[i],
+                                               d0_thread_desc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5,
+                                               make_tuple(I0, I0, I0, I0, I0, I0, I0, I0, I0, I0),
+                                               d0s_thread_buf(i));
+                });
+                static_for<0, m0 * n0 * n2 * n4, 1>{}([&](auto i) {
+                    // get reference to src data
+                    const auto src_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iSrc) -> const auto& { return d0s_thread_buf[iSrc][i]; },
+                        Number<NumD0Tensor>{});
 
-                            static_for<0, n4, 1>{}([&](auto i) {
-                                constexpr index_t c_offset = acc0_thread_desc.CalculateOffset(
-                                    make_tuple(mr, nr, groupid, i));
+                    // get reference to dst data
+                    auto dst_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto) -> auto& { return acc0_thread_buf(i); },
+                        Number<2>{});
 
-                                // get reference to src data
-                                const auto src_data_refs = generate_tie(
-                                    // return type should be lvalue
-                                    [&](auto iSrc) -> const auto& {
-                                        return d0s_thread_buf[iSrc][i];
-                                    },
-                                    Number<NumD0Tensor>{});
-
-                                // get reference to dst data
-                                auto dst_data_refs = generate_tie(
-                                    // return type should be lvalue
-                                    [&](auto) -> auto& {
-                                        return acc0_thread_buf(Number<c_offset>{});
-                                    },
-                                    Number<2>{});
-
-                                unpack2(cde0_element_op, dst_data_refs, src_data_refs);
-                            });
-                            static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                                d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                                    d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                    make_multi_index(0, 0, 0, 0, 0, 0, 0, 1, 0, 0));
-                            });
-                        });
-                        static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                            d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                                d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                                make_multi_index(0, 0, 0, 1, 0, 0, 0, -n2.value, 0, 0));
-                        });
-                    });
-                    static_for<0, NumD0Tensor, 1>{}([&](auto i) {
-                        d0s_threadwise_copy(i).MoveSrcSliceWindow(
-                            d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                            make_multi_index(0, 0, 1, -Gemm0NXdlPerWave, 0, 0, 0, 0, 0, 0));
-                    });
+                    unpack2(cde0_element_op, dst_data_refs, src_data_refs);
                 });
                 static_for<0, NumD0Tensor, 1>{}([&](auto i) {
                     d0s_threadwise_copy(i).MoveSrcSliceWindow(
                         d0s_griddesc_m0_n0_m1_n1_m2_n2_m3_n3_n4_n5[i],
-                        make_multi_index(0, 1, -Gemm0MXdlPerWave, 0, 0, 0, 0, 0, 0, 0));
+                        make_multi_index(0, 1, 0, 0, 0, 0, 0, 0, 0, 0));
                 });
             }
+            else
+            {
+                static_for<0, acc0_thread_buf.Size(), 1>{}(
+                    [&](auto i) { cde0_element_op(acc_thread_buf(i), acc0_thread_buf[i]); });
+            }
             // gemm1
             {
                 // TODO: explore using dynamic buffer for a1 thread buffer
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
index 4db05589b1..e750f18cc7 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
@@ -31,22 +31,24 @@ using CDE1ElementOp = ck::tensor_operation::element_wise::Add;
 using device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances =
     std::tuple<
         // clang-format off
-        //##################################################| A0Layout| B0Layout|       D0Layout| B1Layout|      D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType|     D0DataType| B1Data| Acc1CData| CShuffle|        D1sData| E1Data|          A0|          B0|          CDE0|          B1|          CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K|NumGemm0K| Block|  Gemm0| Gemm0| Gemm0| Gemm1| Gemm1|A0K1|B0K1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|   C1Shuffle|   C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer|
-        //##################################################|         |         |               |         |               |         |   Type|   Type|         Type|           Type|   Type|      Type| DataType|           Type|   Type| Elementwise| Elementwise|   Elementwise| Elementwise|   Elementwise|          |          |          |          |          | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|  ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|            _MBlock_MWaveMPerXdl|   ScalarPerVector|
-        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |   Operation|   Operation|     Operation|   Operation|     Operation|          |          |          |          |          |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per|Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|            _NBlock_NWaveNPerXdl|     _NWaveNPerXdl|
-        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |            |            |              |            |              |          |          |          |          |          |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|               |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                                |                  |
+        //##################################################| A0Layout| B0Layout|       D0Layout| B1Layout|      D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType|     D0DataType| B1Data| Acc1CData| CShuffle|        D1sData| E1Data|          A0|          B0|          CDE0|          B1|          CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K|NumGemm0K| Block|  Gemm0| Gemm0| Gemm0| Gemm1| Gemm1|A0K1|B0K1| B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| CDE0BlockTransfer| CDE0BlockTransfer|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|   C1Shuffle|   C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer|
+        //##################################################|         |         |               |         |               |         |   Type|   Type|         Type|           Type|   Type|      Type| DataType|           Type|   Type| Elementwise| Elementwise|   Elementwise| Elementwise|   Elementwise|          |          |          |          |          | Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|    |    |     |  XDL|  XDL|  MXdl|  NXdl|  NXdl|  ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|      SrcVectorDim|         SrcScalar|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|            _MBlock_MWaveMPerXdl|   ScalarPerVector|
+        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |   Operation|   Operation|     Operation|   Operation|     Operation|          |          |          |          |          |    Stage|      |  Block| Block| Block| Block| Block|    |    |     |     |     |   Per|   Per|   Per|Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |                  |         PerVector|  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|            _NBlock_NWaveNPerXdl|     _NWaveNPerXdl|
+        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |            |            |              |            |              |          |          |          |          |          |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|               |               |               |               |               |               |          |                 |                |                |                |                |                |           |                  |                  |                 |                |                |                |                |                |           |            |            |                                |                  |
+        //generic
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 1,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
         // no padding
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    64,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    32,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    32,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    32,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,                  S<1, 16, 1,16>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    32,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,                  S<1, 16, 1,16>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    64,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,                 9,                 4,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    32,    64,    32,   8,   8,    2,   32,   32,     1,     4,     2,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,                 9,                 4,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,    128,   128,    32,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    32,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,                  S<1, 16, 1,16>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    32,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    64,   128,    32,   8,   8,    2,   16,   16,     1,    16,     8,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           8,                  S<1, 16, 1,16>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,        1,   256,     64,   256,    64,    64,    32,   8,   8,    2,   16,   16,     1,    16,     4,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           4,                  S<1, 32, 1, 8>,               8>,
         // Padded fallback kernel
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,        1,   256,    128,   128,    64,   128,    32,   8,   8,    2,   32,   32,     1,     4,     4,    S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,                 9,                 4,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Row, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,        1,   256,    128,    64,    32,   128,    32,   8,   8,    2,   32,   32,     1,     2,     4,    S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,                  S<1, 32, 1, 8>,               8>
         // clang-format on
         >;
 
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
index e25f903a82..37342e5f45 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
@@ -31,23 +31,25 @@ using CDE1ElementOp = ck::tensor_operation::element_wise::Add;
 using device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instances =
     std::tuple<
         // clang-format off
-        //##################################################| A0Layout| B0Layout|       D0Layout| B1Layout|      D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType|     D0DataType| B1Data| Acc1CData| CShuffle|        D1sData| E1Data|          A0|          B0|          CDE0|          B1|          CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K| NumGemm0K| Block|  Gemm0| Gemm0| Gemm0| Gemm1| Gemm1| A0K1| B0K1|B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1| A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|   C1Shuffle|   C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer|
-        //##################################################|         |         |               |         |               |         |   Type|   Type|         Type|           Type|   Type|      Type| DataType|           Type|   Type| Elementwise| Elementwise|   Elementwise| Elementwise|   Elementwise|          |          |          |          |          |  Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|     |     |    |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|            _MBlock_MWaveMPerXdl|   ScalarPerVector|
-        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |   Operation|   Operation|     Operation|   Operation|     Operation|          |          |          |          |          |     Stage|      |  Block| Block| Block| Block| Block|     |     |    |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|            _NBlock_NWaveNPerXdl|     _NWaveNPerXdl|
-        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |            |            |              |            |              |          |          |          |          |          |          |      |       |      |      |      |      |     |     |    |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                                |                  |
+        //##################################################| A0Layout| B0Layout|       D0Layout| B1Layout|      D1sLayout| E1Layout| A0Data| B0Data| Acc0DataType|     D0DataType| B1Data| Acc1CData| CShuffle|        D1sData| E1Data|          A0|          B0|          CDE0|          B1|          CDE1| PadGemm0M| PadGemm0N| PadGemm0K| PadGemm1N| PadGemm1K| NumGemm0K| Block|  Gemm0| Gemm0| Gemm0| Gemm1| Gemm1| A0K1| B0K1|B1K1| MPer| NPer| Gemm0| Gemm0| Gemm1| A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockTransfer|A0BlockLds|  B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockTransfer| B0BlockLds| CDE0BlockTransfer| CDE0BlockTransfer|  B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockTransfer| B1BlockLds|   C1Shuffle|   C1Shuffle| CDE1BlockTransferClusterLengths| CDE1BlockTransfer|
+        //##################################################|         |         |               |         |               |         |   Type|   Type|         Type|           Type|   Type|      Type| DataType|           Type|   Type| Elementwise| Elementwise|   Elementwise| Elementwise|   Elementwise|          |          |          |          |          |  Prefetch|  Size|   MPer|  NPer|  KPer|  NPer|  KPer|     |     |    |  XDL|  XDL|  MXdl|  NXdl|  NXdl|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN|      SrcVectorDim|         SrcScalar|    ThreadCluster|   ThreadCluster|  SrcAccessOrder|    SrcVectorDim|       SrcScalar|       DstScalar|  AddExtraN| MXdlPerWave| NXdlPerWave|            _MBlock_MWaveMPerXdl|   ScalarPerVector|
+        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |   Operation|   Operation|     Operation|   Operation|     Operation|          |          |          |          |          |     Stage|      |  Block| Block| Block| Block| Block|     |     |    |     |     |   Per|   Per|   Per| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          |  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |                  |         PerVector|  Lengths_K0_N_K1|    ArrangeOrder|                |                |       PerVector|    PerVector_K1|           |  PerShuffle|  PerShuffle|            _NBlock_NWaveNPerXdl|     _NWaveNPerXdl|
+        //##################################################|         |         |               |         |               |         |       |       |             |               |       |          |         |               |       |            |            |              |            |              |          |          |          |          |          |          |      |       |      |      |      |      |     |     |    |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                  |                  |                 |                |                |                |                |                |           |            |            |                                |                  |
+        //generic
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,         1,   256,    128,    64,    32,   128,    32,    8,   8,    4,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 1,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
         // no padding
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    256,   128,    32,   128,    32,    8,   8,    4,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    64,    64,    32,    8,   8,    4,   32,   32,     1,     4,     2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    32,    64,    32,    8,   8,    4,   32,   32,     1,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    64,   128,    32,    8,   8,    4,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    32,   128,    32,    8,   8,    4,   32,   32,     1,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    32,   128,    32,    8,   8,    4,   16,   16,     1,    16,     8,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           8,                  S<1, 16, 1,16>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    32,    64,    32,    8,   8,    4,   16,   16,     1,    16,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           4,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    64,   128,    32,    8,   8,    4,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           8,                  S<1, 16, 1,16>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    64,    64,    32,    8,   8,    4,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           4,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    256,   128,    32,   128,    32,    8,   8,    4,   32,   32,     2,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    64,    64,    32,    8,   8,    4,   32,   32,     1,     4,     2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    32,    64,    32,    8,   8,    4,   32,   32,     1,     4,     2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    64,   128,    32,    8,   8,    4,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,    128,   128,    32,   128,    32,    8,   8,    4,   32,   32,     1,     4,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    32,   128,    32,    8,   8,    4,   16,   16,     1,    16,     8,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           8,                  S<1, 16, 1,16>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    32,    64,    32,    8,   8,    4,   16,   16,     1,    16,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           4,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    64,   128,    32,    8,   8,    4,   16,   16,     1,    16,     8,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           8,                  S<1, 16, 1,16>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,     false,     false,     false,     false,     false,         1,   256,     64,   256,    64,    64,    32,    8,   8,    4,   16,   16,     1,    16,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           4,                  S<1, 32, 1, 8>,               8>,
         // Padded fallback kernel                                                                                                                                                                                                       
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,         1,   256,    128,   128,    64,   128,    32,    8,   8,    4,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
-        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,         1,   256,    128,    64,    32,   128,    32,    8,   8,    4,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,         1,   256,    128,   128,    64,   128,    32,    8,   8,    4,   32,   32,     1,     4,     4,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,     false,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,      false,                 9,                 4,      S<8, 32, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,           1,           2,                  S<1, 32, 1, 8>,               8>,
+        DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle<      Row,      Col, ck::Tuple<Row>,      Col, ck::Tuple<Row>,      Row,    F16,    F16,          F32, ck::Tuple<F16>,    F16,       F32,      F32, ck::Tuple<F16>,    F16, PassThrough, PassThrough, CDE0ElementOp, PassThrough, CDE1ElementOp,      true,      true,      true,      true,      true,         1,   256,    128,    64,    32,   128,    32,    8,   8,    4,   32,   32,     1,     2,     4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               8,               8,       true,                 9,                 4,      S<4, 64, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,       true,           1,           2,                  S<1, 32, 1, 8>,               8>
         // clang-format on
         >;
 

From f4ea560112dce363b2614708917c2a61c084a35b Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Wed, 26 Jul 2023 07:54:47 +0800
Subject: [PATCH 068/194] Speed-up global memory reading for GEMM instances
 (#813)

* Use better ThreadClusterLengths to speed up

* Update B tile reading pattern for layout=NN instance
---
 .../km_nk_mn_default_pipeline_v2_opt_instance.cpp               | 2 +-
 .../mk_nk_mn_default_pipeline_v2_opt_instance.cpp               | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
index 7fc55b017e..9f459aabfc 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
@@ -18,7 +18,7 @@ using Instances =
         //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                      |
         //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                      |
         //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                      |
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
 #endif
         // clang-format on
         >;
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
index 1aba1fa8d9..76d71ee97a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
@@ -17,7 +17,7 @@ using Instances = std::tuple<
         //###########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                        |                     |
         //###########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                        |                     |
         //###########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                        |                     |
-        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+        DeviceGemmXdl<   F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
 #endif
     // clang-format on
     >;

From 016bd428df2554b54007d374955982b1df1369f8 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Wed, 26 Jul 2023 12:18:56 +0800
Subject: [PATCH 069/194] Refine the dimension of host tesnor. This example
 only require 1D (#812)

---
 example/50_put_element/put_element_fp16.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/example/50_put_element/put_element_fp16.cpp b/example/50_put_element/put_element_fp16.cpp
index d4b6831bc4..791747d875 100644
--- a/example/50_put_element/put_element_fp16.cpp
+++ b/example/50_put_element/put_element_fp16.cpp
@@ -34,9 +34,9 @@ int main()
 
     int N = 1024;
 
-    Tensor<XDataType> x(HostTensorDescriptor{N, 1});
-    Tensor<IndexDataType> indices(HostTensorDescriptor{N, 1});
-    Tensor<YDataType> y(HostTensorDescriptor{N, 1});
+    Tensor<XDataType> x(HostTensorDescriptor{N});
+    Tensor<IndexDataType> indices(HostTensorDescriptor{N});
+    Tensor<YDataType> y(HostTensorDescriptor{N});
 
     x.GenerateTensorValue(GeneratorTensor_3<XDataType>{-1.0, 1.0});
     for(int i = 0; i < N; ++i)
@@ -72,7 +72,7 @@ int main()
     bool pass = true;
     if(do_verification)
     {
-        Tensor<YDataType> y_host(HostTensorDescriptor{N, 1});
+        Tensor<YDataType> y_host(HostTensorDescriptor{N});
 
         for(int i = 0; i < N; ++i)
         {

From ac6d68b3536fe9f7e563e4903733b377aaa9013f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 26 Jul 2023 16:19:55 +0200
Subject: [PATCH 070/194] Disable XDL kernels on unsupported HW Add
 ck::is_xdl_supported (#768)

* Disable XDL kernels on unsupported HW; Add ck::is_xdl_supported function (#765)

* Do not throw an error when GEMM problem is not supported.

---------

Co-authored-by: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
---
 example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp               | 6 +++---
 include/ck/host_utility/device_prop.hpp                    | 7 +++++++
 .../device_batched_contraction_multiple_d_xdl_cshuffle.hpp | 4 +---
 .../gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp  | 5 +++++
 .../device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp  | 4 +---
 .../gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp    | 4 +---
 ...atched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp | 4 +---
 .../impl/device_batched_gemm_reduce_xdl_cshuffle.hpp       | 5 +++++
 ...vice_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp | 4 +---
 .../impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp | 4 +---
 .../gpu/device/impl/device_batched_gemm_xdl.hpp            | 5 +++++
 .../gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp    | 5 +++++
 .../impl/device_contraction_multiple_d_xdl_cshuffle.hpp    | 4 +---
 ...conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp | 5 +++++
 .../impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp     | 5 +++++
 ...wd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp | 5 +++++
 ...2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp | 5 +++++
 .../device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp     | 5 +++++
 .../device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp   | 5 +++++
 .../impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp       | 5 +++++
 .../device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp | 5 +++++
 .../impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp      | 5 +++++
 .../impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp | 4 +---
 .../device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp     | 4 +---
 .../device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp    | 4 +---
 .../gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp    | 5 +++++
 .../gpu/device/impl/device_gemm_xdl_cshuffle.hpp           | 4 +---
 .../gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp | 4 +---
 .../gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp         | 5 +++++
 .../gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp   | 5 +++++
 .../device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp  | 4 +---
 .../device_grouped_contraction_multiple_d_xdl_cshuffle.hpp | 4 +---
 ...ce_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp | 5 +++++
 ...vice_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp | 4 +---
 .../gpu/device/impl/device_grouped_gemm_xdl.hpp            | 5 +++++
 .../impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp       | 5 +++++
 .../device_splitk_contraction_multiple_d_xdl_cshuffle.hpp  | 4 +---
 37 files changed, 121 insertions(+), 51 deletions(-)

diff --git a/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp b/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
index 3afd0ebdb9..4a0c23cf44 100644
--- a/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
+++ b/example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp
@@ -204,9 +204,9 @@ int main(int argc, char* argv[])
 
     if(!gemm.IsSupportedArgument(argument))
     {
-        throw std::runtime_error(
-            "wrong! device_gemm with the specified compilation parameters does "
-            "not support this GEMM problem");
+        std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
     }
 
     float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
diff --git a/include/ck/host_utility/device_prop.hpp b/include/ck/host_utility/device_prop.hpp
index bd02d5d88a..be1dbc1657 100644
--- a/include/ck/host_utility/device_prop.hpp
+++ b/include/ck/host_utility/device_prop.hpp
@@ -51,4 +51,11 @@ inline std::string get_device_name()
     return name;
 }
 
+inline bool is_xdl_supported()
+{
+    return ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+           ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+           ck::get_device_name() == "gfx942";
+}
+
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index 46e71240c1..3270325069 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -840,9 +840,7 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
index fc080df5f5..8f46e0c498 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -571,6 +571,11 @@ struct DeviceBatchedGemmEPermuteXdl : public DeviceBatchedGemmEPermute<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_m_k_,
                                            arg.b_grid_desc_n_k_,
                                            ck::Tuple<>{},
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
index 7012584aab..3dbe8c6722 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_gemm_xdl_cshuffle.hpp
@@ -589,9 +589,7 @@ struct DeviceBatchedGemmGemm_Xdl_CShuffle : public DeviceBatchedGemmGemm<ALayout
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index e00deaf000..bd87126f99 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -580,9 +580,7 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index d92865a9ec..dfc84ccf67 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -809,9 +809,7 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
index 1d52416d4d..c4567108c9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_reduce_xdl_cshuffle.hpp
@@ -801,6 +801,11 @@ struct DeviceBatchedGemmReduce_Xdl_CShuffle : public DeviceGemmReduce<0, ReduceO
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
                                            arg.b_grid_desc_bk0_n_bk1_,
                                            arg.c_grid_desc_m_n_,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index 73b12bfb47..c38b7e1c5b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -723,9 +723,7 @@ struct DeviceBatchedGemmSoftmaxGemmPermute_Xdl_CShuffle
         arg.Print();
 #endif
 
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
index 2405640ca9..f12e05fd34 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_softmax_gemm_xdl_cshuffle.hpp
@@ -613,9 +613,7 @@ struct DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index b6a0567fee..408c56d490 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -310,6 +310,11 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
 
     static bool IsSupportedArgument(const Problem& problem)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if(problem.K % K1 != 0)
         {
             return false;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
index 9ee6c6f46b..106746716d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
@@ -448,6 +448,11 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg);
     }
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index dd57c90898..db263b41d6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -582,9 +582,7 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index 2ab09ba5c8..1e50e1ae89 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -649,6 +649,11 @@ struct DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         // vector load A/B matrix from global memory
         if(!(ABlockTransferSrcVectorDim == 2 && BBlockTransferSrcVectorDim == 2 &&
              arg.Conv_K_ % ABlockTransferSrcScalarPerVector == 0 &&
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
index eb4db6f8cb..e22c5a2aa5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk.hpp
@@ -616,6 +616,11 @@ struct DeviceConv2dBwdDataXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if constexpr(ConvBackwardDataSpecialization ==
                      ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
index bb8f531613..c9e8940edc 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_add_nhwc_kyxc_nhwk.hpp
@@ -810,6 +810,11 @@ struct
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if constexpr(ConvForwardSpecialization ==
                      ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
index 1bd1e553cd..28fceb428e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_bias_activation_nhwc_kyxc_nhwk.hpp
@@ -767,6 +767,11 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Bias_Activation_Input_N_Hi_Wi_C_Weight_K_Y_X
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if constexpr(ConvForwardSpecialization ==
                      ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index de6bf27fec..ca291d3b11 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -741,6 +741,11 @@ struct DeviceConv2dFwdXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_W
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if constexpr(ConvForwardSpecialization ==
                      ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
index 88615bba31..ef94120f4e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp
@@ -524,6 +524,11 @@ struct DeviceConv2dFwdXdl_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_N_Ho_Wo_K
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if constexpr(ConvForwardSpecialization ==
                      ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
index bc6582835e..a8e586b20c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv3d_fwd_xdl_ndhwc_kzyxc_ndhwk.hpp
@@ -524,6 +524,11 @@ struct DeviceConv3dFwdXdl_Input_N_Di_Hi_Wi_C_Weight_K_Z_Y_X_C_Output_N_Do_Ho_Wo_
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
                                            arg.b_grid_desc_k0_n_k1_,
                                            arg.c_grid_desc_m_n_,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
index 77ad61d7ec..ee3f0cea1b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_convnd_bwd_data_nwc_kxc_nwk_xdl.hpp
@@ -1320,6 +1320,11 @@ struct DeviceConvNdBwdDataNwcKxcNwk_Xdl
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if constexpr(ConvBackwardDataSpecialization ==
                      ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
index 63f7fa7068..5bfef60af2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_bias_add_reduce_xdl_cshuffle.hpp
@@ -683,6 +683,11 @@ struct DeviceGemmBiasAddReduce_Xdl_CShuffle : public DeviceGemmReduce<1, ReduceO
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
                                            arg.b_grid_desc_bk0_n_bk1_,
                                            arg.c_grid_desc_m_n_,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
index 5afeaf98c4..b086151e77 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -855,9 +855,7 @@ struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index 3a0be837b7..1f3ea43c7b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -555,9 +555,7 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index d4ecea7bb7..d5b07e213a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -491,9 +491,7 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
index e60fae370f..b590fba98f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_reduce_xdl_cshuffle.hpp
@@ -645,6 +645,11 @@ struct DeviceGemmReduce_Xdl_CShuffle : public DeviceGemmReduce<0, ReduceOperatio
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_,
                                            arg.b_grid_desc_bk0_n_bk1_,
                                            arg.c_grid_desc_m_n_,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index b68ecfe7b1..4c9ddbab80 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -188,9 +188,7 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
index a2403f109c..a8cd39080c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_layernorm_cshuffle.hpp
@@ -648,9 +648,7 @@ struct DeviceGemmLayerNorm_Xdl_CShuffle : public BaseOperator
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
index ef5e267819..d54ddf433e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp
@@ -416,6 +416,11 @@ struct DeviceGemmXdlSkipBLds : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(arg.a_grid_desc_k0_m_k1_,
                                            arg.b_grid_desc_k0_n_k1_,
                                            arg.c_grid_desc_m_n_,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 6c07a5c56c..2b8f760737 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -231,6 +231,11 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
 
     static bool IsSupportedArgument(const Argument& karg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         return GridwiseGemm::CheckValidity(karg);
     }
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
index d0de101c49..a086e1a4bd 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -417,9 +417,7 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 4b1cc65ba2..39a6dd7ab2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -705,9 +705,7 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index d932fd4069..a9f69ab4e6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -826,6 +826,11 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         const index_t ConvK = arg.b_g_k_c_xs_lengths_[1];
         const index_t ConvC = arg.b_g_k_c_xs_lengths_[2];
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index ab6d2716c0..a7ff7f3d7e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -681,9 +681,7 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index bc9b3a4dcc..9d3fe77fe2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -600,6 +600,11 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if((ck::type_convert<ck::index_t>(arg.gemm_desc_kernel_arg_.size()) +
             arg.skipped_group_count_) != arg.group_count_)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 74f38b9db2..4f4413b78f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -502,6 +502,11 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
             arg.skipped_group_count_) != arg.group_count_)
         {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index f849ac799d..7ba2d96a9c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -939,9 +939,7 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
-             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
-             ck::get_device_name() == "gfx942"))
+        if(!ck::is_xdl_supported())
         {
             return false;
         }

From 9195435c77b17d554fb80086686746fa20106f76 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 26 Jul 2023 09:06:45 -0700
Subject: [PATCH 071/194] Disable DL kernels by default. (#816)

---
 CHANGELOG.md                                  |   5 +
 CMakeLists.txt                                |   8 ++
 Jenkinsfile                                   |   5 +-
 README.md                                     |  18 ++-
 .../gpu/batched_gemm_multi_d.hpp              |   2 +
 .../tensor_operation_instance/gpu/gemm.hpp    | 125 +++++++++++-------
 .../gpu/CMakeLists.txt                        |   8 +-
 .../gpu/batched_gemm_multi_d/CMakeLists.txt   |  45 ++++---
 .../gpu/gemm/CMakeLists.txt                   |  46 ++++---
 profiler/src/CMakeLists.txt                   |   8 +-
 test/batched_gemm_multi_d/CMakeLists.txt      |   4 +-
 11 files changed, 177 insertions(+), 97 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3898b5ce2d..9001c2f4e7 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -12,6 +12,11 @@ Full documentation for Composable Kernel is not yet available.
 - Improve proformance of normalization kernel
 
 ### Added
+- Added new cmake flag "DL_KERNELS" must be set to "ON" in order to build the gemm_dl and batched_gemm_multi_d_dl instances.
+- Added new cmake flag "DTYPES" which could be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build instance of select data types.
+- Added new cmake flag "INSTANCES_ONLY" which will only build CK library and instances without the tests, examples, or profiler.
+- Added new feature: if GPU_TARGETS is not set on cmake command line, CK will be built for all targets supported by compiler.
+- Added support on MI300A/MI300X.
 - Added support on NAVI3x.
 - Added user tutorial (#563).
 - Added more instances for irregular GEMM sizes (#560).
diff --git a/CMakeLists.txt b/CMakeLists.txt
index c0de76c580..514b98fde6 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -30,6 +30,14 @@ else()
         add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
 endif()
 
+if(DL_KERNELS)
+    add_definitions(-DDL_KERNELS)
+endif()
+
+if(INSTANCES_ONLY)
+    add_definitions(-DINSTANCES_ONLY)
+endif()
+
 enable_testing()
 
 set(ROCM_SYMLINK_LIBS OFF)
diff --git a/Jenkinsfile b/Jenkinsfile
index 764dc71220..6df9ea3889 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -509,8 +509,7 @@ def Build_CK(Map conf=[:]){
                     cmake_build(conf)
                     dir("build"){
                         //run tests and examples
-                        def nt = nthreads()	
-                        sh 'make -j${nt} check'
+                        sh 'make -j check'
                         if (navi_node == 0 ){
                             //we only need the ckProfiler to run the performance tests, so we pack and stash it
                             //do not stash profiler on Navi nodes
@@ -741,7 +740,7 @@ pipeline {
                     }
                     agent{ label rocmnode("navi21") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" """ 
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" -DDL_KERNELS=ON """ 
                         execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1030" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
 
                     }
diff --git a/README.md b/README.md
index fd6f7e37c2..c2b493db11 100644
--- a/README.md
+++ b/README.md
@@ -52,6 +52,8 @@ CK is released under the MIT license. [License File](/LICENSE)
 ```bash
 DOCKER_BUILDKIT=1 docker build -t ck:latest -f Dockerfile .
 ```
+Pre-built dockers are available from this public repo: 
+https://hub.docker.com/r/rocm/composable_kernel/tags
 
 ## Launch docker
 
@@ -76,12 +78,26 @@ mkdir build && cd build
 cmake                                                                                             \
 -D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
 -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
--D CMAKE_CXX_FLAGS="-O3"                                                                          \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D GPU_TARGETS="gfx908;gfx90a"                                                                    \
 ..
 ```
 
+If GPU_TARGETS is not set on the cmake command line, CK will be built for all targets supported by the 
+current compiler.
+
+
+Additional cmake flags can be used to significantly speed-up the build:
+
+INSTANCES_ONLY (by default is OFF) must be set to ON in order to build only the instances and library
+while skipping all tests, examples, and profiler. This is useful for libraries that use CK as a dependency.
+
+DTYPES (by default not set) can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build instances 
+of select data types only. Currently, building of int8 instances is taking a lot of time (the compiler fix is in the works).
+
+DL_KERNELS (by default is OFF) must be set to ON in order to build the gemm_dl and batched_gemm_multi_d_dl 
+instances. Those instances are only needed for the NAVI2x platforms.
+
 ### Build examples and tests
 
 ```bash
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
index 3d50aae092..2d7aaf8729 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
@@ -14,6 +14,7 @@
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 
+#ifdef DL_KERNELS
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -335,3 +336,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index ad298009ae..56a74f7770 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -17,6 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+#if defined(__fp16__) && defined(DL_KERNELS)
 void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -56,11 +57,11 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#if defined(__fp32__) && defined(DL_KERNELS)
 void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
-
         instances);
 
 void add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(
@@ -77,7 +78,8 @@ void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-#ifdef __int8__
+#endif
+#if defined(__int8__) && defined(DL_KERNELS)
 void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -117,7 +119,8 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __int8__
 void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -138,32 +141,12 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-
+#ifdef __fp16__
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
 void add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -184,26 +167,6 @@ void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
-void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
 void add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -224,6 +187,49 @@ void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+#endif
+#ifdef __bf16__
+void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+#endif
+#ifdef __fp32__
+void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -243,7 +249,8 @@ void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __fp64__
 void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(
     std::vector<std::unique_ptr<
 
@@ -264,7 +271,7 @@ void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
 template <typename ALayout,
           typename BLayout,
           typename CLayout,
@@ -303,28 +310,36 @@ struct DeviceOperationInstanceFactory<
                          is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f32_f32_f32_km_nk_mn_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(op_ptrs);
             }
         }
@@ -335,16 +350,20 @@ struct DeviceOperationInstanceFactory<
                          is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
             }
@@ -352,16 +371,20 @@ struct DeviceOperationInstanceFactory<
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
+#endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
@@ -397,29 +420,37 @@ struct DeviceOperationInstanceFactory<
                          is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances(op_ptrs);
+#endif
             }
             else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(op_ptrs);
+#endif
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances(op_ptrs);
+#endif
             }
             else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
                               is_same_v<CLayout, Row>)
             {
                 add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_gemm_dl_i8_i8_i8_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances(op_ptrs);
+#endif
             }
         }
 #endif
diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
index 597aeaed8b..1d54a141b7 100644
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
@@ -39,11 +39,15 @@ IF(IS_DIRECTORY "${subdir_path}")
             #message("int8 instance found!")
             set(add_inst 1)
     endif()
-    if(NOT "${cmake_instance}" MATCHES "DTYPES")
+    if(NOT "${cmake_instance}" MATCHES "DTYPES" OR NOT DEFINED DTYPES)
             #message("instance should be built for all types!")
             set(add_inst 1)
     endif()
-    if(add_inst EQUAL 1 OR NOT DEFINED DTYPES)
+    if("${cmake_instance}" MATCHES "ONLY DL_KERNELS" AND NOT DEFINED DL_KERNELS)
+	    message("Found only dl instances, but DL_KERNELS is not set. Skipping.")
+            set(add_inst 0)
+    endif()
+    if(add_inst EQUAL 1)
       get_filename_component(target_dir ${subdir_path} NAME)
       add_subdirectory(${target_dir})
       list(APPEND CK_DEVICE_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
index 8f4e3b6948..444c93b118 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
@@ -1,22 +1,25 @@
-set(BATCHED_GEMM_MULTID_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp)
+# ONLY DL_KERNELS
+if(DL_KERNELS)
+  set(BATCHED_GEMM_MULTID_INSTANCES)
+  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp)
+  endif()
+  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp)
+    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp)
+  endif()
+  add_instance_library(device_batched_gemm_multi_d_instance ${BATCHED_GEMM_MULTID_INSTANCES})
 endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp)
-   list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp)
-endif()
-add_instance_library(device_batched_gemm_multi_d_instance ${BATCHED_GEMM_MULTID_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 62cdae7cdf..b6f63f39d7 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -14,25 +14,29 @@ if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp)
+   if(DL_KERNELS)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp)
+   endif()
 endif()
 if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+   if(DL_KERNELS)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
+   endif()
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp)
@@ -67,14 +71,16 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp)
 endif()
 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp)
+   if(DL_KERNELS)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp)
+   endif()
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp)
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index d35c743939..d057dccb40 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -34,9 +34,11 @@ set(PROFILER_SOURCES
     profile_grouped_gemm_fastgelu.cpp
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
-    profile_batched_gemm_multi_d.cpp
     profile_grouped_conv_bwd_data.cpp
 )
+if(DL_KERNELS)
+  list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
+endif()
 
 set(PROFILER_EXECUTABLE ckProfiler)
 
@@ -79,7 +81,9 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgel
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
+if(DL_KERNELS)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
+endif()
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/test/batched_gemm_multi_d/CMakeLists.txt b/test/batched_gemm_multi_d/CMakeLists.txt
index 45a306551f..f8ceb545c2 100644
--- a/test/batched_gemm_multi_d/CMakeLists.txt
+++ b/test/batched_gemm_multi_d/CMakeLists.txt
@@ -1,5 +1,7 @@
 # TODO: Enable for gfx90a after complier fix
-if(NOT GPU_TARGETS MATCHES "gfx90a")
+if(DL_KERNELS)
+  if(NOT GPU_TARGETS MATCHES "gfx90a")
     add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
     target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
+  endif()
 endif()

From e7dca79d273ce63161d1ff2d3efc9d4be9283f1c Mon Sep 17 00:00:00 2001
From: carlushuang <carlus.huang@amd.com>
Date: Thu, 27 Jul 2023 03:18:15 +0800
Subject: [PATCH 072/194] initial stream-k implementation with example (#699)

* initial stream-k implementation with example

* fix unexpected change in err

* improve a little bit performance by reorganize pipeline.

* improve perf a little bit by swizzle block idx

* add profiler

* update example

* fix spelling

* shrink karg for streamk

* support dynamic buffer using memory coherence glc_slc bit from template

* control memory coherence while construct dynamic buffer

* update reduction for streamk(not ready yet)

* Add template parameter to make_dynamic_buffer to support amd_buffer coherence setting

* fix build issue

* fix several bug

* now result is correct, everything works (but has scratch)

* remove scratch by manually reset coordinate

* update device code

* fix a bug in final reduce

* fix something in example

* update async memset

* fix enum as camel case

* modify coherence enum name

* clean code and use atomic streamk by default

* remove unused var

* throw exception if have empty pointer

* fix format

* fix CI warning

* fix type in init

* modify CI error

* filter out on gfx10+

* restore changed example code

---------

Co-authored-by: Qianfeng Zhang <Qianfeng.Zhang@amd.com>
---
 example/01_gemm/CMakeLists.txt                |    2 +
 example/01_gemm/common.hpp                    |   75 +-
 example/01_gemm/gemm_xdl_streamk.cpp          |   49 +
 example/01_gemm/run_gemm_example.inc          |  169 ++-
 include/ck/host_utility/kernel_launch.hpp     |   69 +
 ...hread_group_tensor_slice_transfer_v4r1.hpp |   15 +
 ...ead_group_tensor_slice_transfer_v6r1r2.hpp |  164 +++
 .../gpu/device/device_gemm_streamk.hpp        |   64 +
 .../device/impl/device_gemm_xdl_streamk.hpp   |  357 +++++
 .../gpu/grid/block_to_ctile_map.hpp           |  404 ++++++
 .../gpu/grid/gridwise_gemm_pipeline_v3.hpp    |   89 ++
 .../gpu/grid/gridwise_gemm_xdlops_streamk.hpp | 1183 +++++++++++++++++
 ...hreadwise_tensor_slice_transfer_v6r1r2.hpp |  213 +++
 include/ck/utility/amd_buffer_addressing.hpp  |   16 +-
 include/ck/utility/magic_division.hpp         |   72 +
 include/ck/utility/math.hpp                   |   16 +
 include/ck/utility/workgroup_barrier.hpp      |   73 +
 .../gpu/gemm_streamk.hpp                      |  121 ++
 .../ck/library/utility/device_memory.hpp      |    3 +-
 .../gpu/gemm_streamk/CMakeLists.txt           |   10 +
 ..._streamk_f16_f16_f16_mk_kn_mn_instance.cpp |   71 +
 library/src/utility/device_memory.cpp         |   45 +-
 .../profiler/profile_gemm_streamk_impl.hpp    |  265 ++++
 profiler/src/CMakeLists.txt                   |    2 +
 profiler/src/profile_gemm_streamk.cpp         |  155 +++
 .../block_swizzle_test/block_swizzle_test.cpp |  406 ++++++
 test/block_swizzle_test/rebuild.sh            |    3 +
 test/block_swizzle_test/simple_args.h         |  159 +++
 28 files changed, 4234 insertions(+), 36 deletions(-)
 create mode 100644 example/01_gemm/gemm_xdl_streamk.cpp
 create mode 100644 include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1r2.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_gemm_streamk.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v3.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
 create mode 100644 include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1r2.hpp
 create mode 100644 include/ck/utility/workgroup_barrier.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 profiler/include/profiler/profile_gemm_streamk_impl.hpp
 create mode 100644 profiler/src/profile_gemm_streamk.cpp
 create mode 100644 test/block_swizzle_test/block_swizzle_test.cpp
 create mode 100644 test/block_swizzle_test/rebuild.sh
 create mode 100644 test/block_swizzle_test/simple_args.h

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index ec4061213e..42a3499385 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -50,6 +50,8 @@ if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS
   add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
 endif()
 
+add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
+
 if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
   add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
   add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
diff --git a/example/01_gemm/common.hpp b/example/01_gemm/common.hpp
index 144c9aaccd..7fd15b2833 100644
--- a/example/01_gemm/common.hpp
+++ b/example/01_gemm/common.hpp
@@ -33,6 +33,19 @@ struct ProblemSize final
     ck::index_t StrideC = 4096;
 };
 
+struct ProblemSizeStreamK final
+{
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideC = 4096;
+
+    ck::index_t NumSKBlocks = -1;
+};
+
 struct ExecutionConfig final
 {
     bool do_verification = true;
@@ -48,8 +61,17 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-inline bool
-parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfig& config)
+template <typename ProblemType>
+bool parse_cmd_args(int, char*[], ProblemType&, ExecutionConfig&)
+{
+    return false;
+}
+
+template <>
+bool parse_cmd_args<ProblemSize>(int argc,
+                                 char* argv[],
+                                 ProblemSize& problem_size,
+                                 ExecutionConfig& config)
 {
     if(argc == 1)
     {
@@ -87,3 +109,52 @@ parse_cmd_args(int argc, char* argv[], ProblemSize& problem_size, ExecutionConfi
 
     return true;
 }
+
+template <>
+bool parse_cmd_args<ProblemSizeStreamK>(int argc,
+                                        char* argv[],
+                                        ProblemSizeStreamK& problem_size,
+                                        ExecutionConfig& config)
+{
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc >= 10)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+
+        problem_size.M = std::stoi(argv[4]);
+        problem_size.N = std::stoi(argv[5]);
+        problem_size.K = std::stoi(argv[6]);
+
+        problem_size.StrideA = std::stoi(argv[7]);
+        problem_size.StrideB = std::stoi(argv[8]);
+        problem_size.StrideC = std::stoi(argv[9]);
+
+        if(argc >= 11)
+        {
+            problem_size.NumSKBlocks = std::stoi(argv[10]);
+        }
+    }
+    else
+    {
+        std::cerr << "arg1: verification (0=no, 1=yes)" << std::endl
+                  << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)"
+                  << std::endl
+                  << "arg3: time kernel (0=no, 1=yes)" << std::endl
+                  << "arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideC" << std::endl
+                  << "arg10: NumSKBlocks(optional)" << std::endl;
+        return false;
+    }
+
+    return true;
+}
diff --git a/example/01_gemm/gemm_xdl_streamk.cpp b/example/01_gemm/gemm_xdl_streamk.cpp
new file mode 100644
index 0000000000..7d433b6145
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_streamk.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp"
+
+using ADataType        = ck::half_t;
+using BDataType        = ck::half_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+using CDataType        = ck::half_t;
+
+using F16 = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Row;
+// using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+// clang-format off
+using DeviceGemmStreamK = ck::tensor_operation::device::DeviceGemmXdlStreamK
+// ######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|  Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise|   Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|       |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|          |          |          |            |        |        |        |            |            |            |       |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,      S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
+
+          // < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    256,   256,   128,     4,   8,  32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              2,         1,           1,           1,               S<1, 32, 1, 8>,              8>;
+      //  < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    128,   32,   64,     4,   8,  32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>;
+       // < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    128,   32,   128,     4,   8,  32,   32,    1,    1,     S<8, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              2,         1,           1,           1,               S<1, 32, 1, 4>,              8>;
+
+
+
+// // clang-format on
+// clang-format on
+
+using DeviceGemmInstance = DeviceGemmStreamK;
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_streamk_example(argc, argv); }
diff --git a/example/01_gemm/run_gemm_example.inc b/example/01_gemm/run_gemm_example.inc
index 38c72afc60..7be2539d90 100644
--- a/example/01_gemm/run_gemm_example.inc
+++ b/example/01_gemm/run_gemm_example.inc
@@ -3,7 +3,10 @@
 
 #pragma once
 
-bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
+#include "ck/tensor_operation/gpu/device/device_gemm_streamk.hpp"
+
+template <typename ProblemType>
+bool run_gemm(const ProblemType& problem_size, const ExecutionConfig& config)
 {
 #if defined(BUILD_INT4_EXAMPLE) && defined(CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4)
     static_assert(sizeof(ck::int4_t) == sizeof(int8_t));
@@ -11,7 +14,12 @@ bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
 
     using namespace ck::literals;
 
-    auto& [M, N, K, StrideA, StrideB, StrideC] = problem_size;
+    auto M       = problem_size.M;
+    auto N       = problem_size.N;
+    auto K       = problem_size.K;
+    auto StrideA = problem_size.StrideA;
+    auto StrideB = problem_size.StrideB;
+    auto StrideC = problem_size.StrideC;
 
     auto f_host_tensor_descriptor =
         [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
@@ -25,12 +33,37 @@ bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
             }
         };
 
+    auto f_get_default_stride =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            if(stride == 0)
+            {
+                // give a chance if stride is zero, return a default packed stride
+                if constexpr(std::is_same_v<decltype(layout), ck::tensor_layout::gemm::RowMajor>)
+                {
+                    return col;
+                }
+                else
+                {
+                    return row;
+                }
+            }
+            else
+                return stride;
+        };
+
+    StrideA = f_get_default_stride(M, K, StrideA, ALayout{});
+    StrideB = f_get_default_stride(K, N, StrideB, BLayout{});
+    StrideC = f_get_default_stride(M, N, StrideC, CLayout{});
+
     Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
     Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
 
     switch(config.init_method)
     {
-    case 0: break;
+    case 0:
+        ck::utils::FillConstant<ADataType>{static_cast<ADataType>(1.f)}(a_m_k);
+        ck::utils::FillConstant<BDataType>{static_cast<BDataType>(1.f)}(b_k_n);
+        break;
     case 1:
         ck::utils::FillUniformDistributionIntegerValue<ADataType>{-5.f, 5.f}(a_m_k);
         ck::utils::FillUniformDistributionIntegerValue<BDataType>{-5.f, 5.f}(b_k_n);
@@ -66,42 +99,114 @@ bool run_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
     a_m_k_device_buf.ToDevice(a_m_k.mData.data());
     b_k_n_device_buf.ToDevice(b_k_n.mData.data());
 #endif
+    DeviceMem workspace;
 
     auto a_element_op = AElementOp{};
     auto b_element_op = BElementOp{};
     auto c_element_op = CElementOp{};
 
+    using BaseStreamK = ck::tensor_operation::device::DeviceGemmStreamK<ALayout,
+                                                                        BLayout,
+                                                                        CLayout,
+                                                                        ADataType,
+                                                                        BDataType,
+                                                                        CDataType,
+                                                                        AElementOp,
+                                                                        BElementOp,
+                                                                        CElementOp>;
+
     // do GEMM
-    auto gemm     = DeviceGemmInstance{};
-    auto invoker  = gemm.MakeInvoker();
-    auto argument = gemm.MakeArgument(
-#ifdef BUILD_INT4_EXAMPLE
-        static_cast<KernelADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-        static_cast<KernelBDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-        static_cast<KernelCDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-#else
-        static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
-        static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
-        static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
-#endif
-        M,
-        N,
-        K,
-        StrideA,
-        StrideB,
-        StrideC,
-        a_element_op,
-        b_element_op,
-        c_element_op);
+    auto gemm      = DeviceGemmInstance{};
+    auto invoker   = gemm.MakeInvoker();
+    float ave_time = 0;
 
-    if(!gemm.IsSupportedArgument(argument))
+    if constexpr(std::is_same<ProblemType, ProblemSize>::value &&
+                 !std::is_base_of<BaseStreamK, DeviceGemmInstance>::value)
     {
-        std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+        auto argument = gemm.MakeArgument(
+#ifdef BUILD_INT4_EXAMPLE
+            static_cast<KernelADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<KernelBDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<KernelCDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+#else
+            static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+#endif
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            StrideC,
+            a_element_op,
+            b_element_op,
+            c_element_op);
 
-        return true;
+        if(!gemm.IsSupportedArgument(argument))
+        {
+            std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+            return true;
+        }
+
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
     }
+    else if constexpr(std::is_same<ProblemType, ProblemSizeStreamK>::value &&
+                      std::is_base_of<BaseStreamK, DeviceGemmInstance>::value)
+    {
+        auto argument = gemm.MakeArgument(
+#ifdef BUILD_INT4_EXAMPLE
+            static_cast<KernelADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<KernelBDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<KernelCDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+#else
+            static_cast<ADataType*>(a_m_k_device_buf.GetDeviceBuffer()),
+            static_cast<BDataType*>(b_k_n_device_buf.GetDeviceBuffer()),
+            static_cast<CDataType*>(c_m_n_device_buf.GetDeviceBuffer()),
+#endif
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            StrideC,
+            a_element_op,
+            b_element_op,
+            c_element_op,
+            problem_size.NumSKBlocks);
 
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+        if(!gemm.IsSupportedArgument(argument))
+        {
+            std::cerr << gemm.GetTypeString() << " does not support this problem" << std::endl;
+
+            return true;
+        }
+
+        std::size_t workspace_size = gemm.GetWorkSpaceSize(&argument);
+        if(workspace_size != 0)
+        {
+            workspace.Realloc(workspace_size);
+            gemm.SetWorkSpacePointer(&argument, workspace.GetDeviceBuffer());
+        }
+
+        ave_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+
+#if 0
+        // TODO!!!!!
+        if(workspace_size != 0){
+            float * ws_ptr = reinterpret_cast<float*>(malloc(workspace_size));
+            size_t ws_dwords = workspace_size / sizeof(float);
+            workspace.FromDevice(ws_ptr);
+
+            for(size_t i = 0; i < ws_dwords; i++) {
+                uint32_t rere = reinterpret_cast<uint32_t*>(ws_ptr)[i];
+                printf("%4lu : %f(0x%08x)\n", i, ws_ptr[i], rere);
+            }
+            free(ws_ptr);
+        }
+#endif
+    }
 
     std::size_t flop = 2_uz * M * N * K;
     std::size_t num_btype =
@@ -149,3 +254,11 @@ bool run_gemm_example(int argc, char* argv[])
 
     return !parse_cmd_args(argc, argv, problem_size, config) || run_gemm(problem_size, config);
 }
+
+bool run_gemm_streamk_example(int argc, char* argv[])
+{
+    ProblemSizeStreamK problem_size;
+    ExecutionConfig config;
+
+    return !parse_cmd_args(argc, argv, problem_size, config) || run_gemm(problem_size, config);
+}
diff --git a/include/ck/host_utility/kernel_launch.hpp b/include/ck/host_utility/kernel_launch.hpp
index 58740b4351..3d27103dcb 100644
--- a/include/ck/host_utility/kernel_launch.hpp
+++ b/include/ck/host_utility/kernel_launch.hpp
@@ -73,3 +73,72 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
     return 0;
 #endif
 }
+
+template <typename... Args, typename F, typename PreProcessFunc>
+float launch_and_time_kernel_with_preprocess(const StreamConfig& stream_config,
+                                             PreProcessFunc preprocess,
+                                             F kernel,
+                                             dim3 grid_dim,
+                                             dim3 block_dim,
+                                             std::size_t lds_byte,
+                                             Args... args)
+{
+#if CK_TIME_KERNEL
+    if(stream_config.time_kernel_)
+    {
+#if DEBUG_LOG
+        printf("%s: grid_dim {%d, %d, %d}, block_dim {%d, %d, %d} \n",
+               __func__,
+               grid_dim.x,
+               grid_dim.y,
+               grid_dim.z,
+               block_dim.x,
+               block_dim.y,
+               block_dim.z);
+
+        printf("Warm up 1 time\n");
+#endif
+        // warm up
+        preprocess();
+        kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+
+        const int nrepeat = 10;
+#if DEBUG_LOG
+        printf("Start running %d times...\n", nrepeat);
+#endif
+        hipEvent_t start, stop;
+
+        hip_check_error(hipEventCreate(&start));
+        hip_check_error(hipEventCreate(&stop));
+
+        hip_check_error(hipDeviceSynchronize());
+        hip_check_error(hipEventRecord(start, stream_config.stream_id_));
+
+        for(int i = 0; i < nrepeat; ++i)
+        {
+            preprocess();
+            kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+        }
+
+        hip_check_error(hipEventRecord(stop, stream_config.stream_id_));
+        hip_check_error(hipEventSynchronize(stop));
+
+        float total_time = 0;
+
+        hip_check_error(hipEventElapsedTime(&total_time, start, stop));
+
+        return total_time / nrepeat;
+    }
+    else
+    {
+        preprocess();
+        kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+
+        return 0;
+    }
+#else
+    kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+
+    return 0;
+#endif
+}
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
index c8690e5f68..2c5fbc3937 100644
--- a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp
@@ -94,6 +94,21 @@ struct ThreadGroupTensorSliceTransfer_v4r1
         }
     }
 
+    __device__ void SetSrcSliceOrigin(const SrcDesc& src_desc, const Index& src_block_slice_origin)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
+                make_multi_index(ThreadGroup::GetThreadId()));
+
+            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
+
+            threadwise_transfer_.SetSrcSliceOrigin(src_desc,
+                                                   src_block_slice_origin + thread_data_idx_begin);
+        }
+    }
+
     template <typename SrcBuffer, index_t ThreadScratchId = 0>
     __device__ void RunRead(const SrcDesc& src_desc,
                             const SrcBuffer& src_buf,
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1r2.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1r2.hpp
new file mode 100644
index 0000000000..83cb9fb5de
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1r2.hpp
@@ -0,0 +1,164 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_description/cluster_descriptor.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1r2.hpp"
+
+namespace ck {
+
+// this version does following things to avoid scratch memory issue
+// 1. Use StaticallyIndexedArray instead of C array for thread buffer
+// 2. ThreadwiseTensorSliceTransfer_v3 does not keep reference to tensor descriptor
+// 3. ThreadwiseTensorSliceTransfer_v3::Run() does not construct new tensor coordinate
+template <typename ThreadGroup,
+          typename ElementwiseOperation,
+          typename SliceLengths,
+          typename ThreadClusterLengths,
+          typename ThreadClusterArrangeOrder,
+          typename SrcData,
+          typename DstData,
+          typename SrcDesc,
+          typename DstDesc,
+          typename DimAccessOrder,
+          index_t VectorDim,
+          index_t ScalarPerVector,
+          bool ThreadTransferSrcResetCoordinateAfterRun,
+          bool ThreadTransferDstResetCoordinateAfterRun>
+struct ThreadGroupTensorSliceTransfer_v6r1r2
+{
+    static constexpr index_t nDim = remove_reference_t<SrcDesc>::GetNumOfDimension();
+
+    static constexpr auto thread_slice_lengths = SliceLengths{} / ThreadClusterLengths{};
+
+    using Index = MultiIndex<nDim>;
+
+    __device__ constexpr ThreadGroupTensorSliceTransfer_v6r1r2(
+        const SrcDesc& src_desc,
+        const Index& src_block_slice_origin,
+        const DstDesc& dst_desc,
+        const Index& dst_block_slice_origin,
+        const ElementwiseOperation& element_op)
+        : threadwise_transfer_(src_desc,
+                               make_zero_multi_index<nDim>(),
+                               dst_desc,
+                               make_zero_multi_index<nDim>(),
+                               element_op)
+
+    {
+        static_assert(nDim == remove_cvref_t<SrcDesc>::GetNumOfDimension() &&
+                          nDim == remove_cvref_t<DstDesc>::GetNumOfDimension() &&
+                          nDim == ThreadClusterLengths::Size() &&
+                          nDim == ThreadClusterArrangeOrder::Size() &&
+                          nDim == DimAccessOrder::Size(),
+                      "wrong! nDim not consistent");
+
+        static_assert(
+            is_same<SliceLengths, decltype(thread_slice_lengths * ThreadClusterLengths{})>{},
+            "wrong! threads should be mapped to cover entire slicing window");
+
+        static_assert(ThreadGroup::GetNumOfThread() >= thread_cluster_desc_.GetElementSize(),
+                      "wrong! ThreadGroup::GetNumOfThread() too small");
+
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
+                make_multi_index(ThreadGroup::GetThreadId()));
+
+            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
+
+            threadwise_transfer_.SetSrcSliceOrigin(src_desc,
+                                                   src_block_slice_origin + thread_data_idx_begin);
+            threadwise_transfer_.SetDstSliceOrigin(dst_desc,
+                                                   dst_block_slice_origin + thread_data_idx_begin);
+        }
+    }
+
+    template <typename SrcBuffer, typename DstBuffer, InMemoryDataOperationEnum DstInMemOp>
+    __device__ void Run(const SrcDesc& src_desc,
+                        const SrcBuffer& src_buf,
+                        const DstDesc& dst_desc,
+                        DstBuffer& dst_buf)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.template Run<SrcBuffer, DstBuffer, DstInMemOp>(
+                src_desc, src_buf, dst_desc, dst_buf);
+        }
+    }
+
+    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc, const Index& step)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.MoveSrcSliceWindow(src_desc, step);
+        }
+    }
+
+    __device__ void MoveDstSliceWindow(const DstDesc& dst_desc, const Index& step)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.MoveDstSliceWindow(dst_desc, step);
+        }
+    }
+
+    __device__ void SetSrcSliceOrigin(const SrcDesc& src_desc, const Index& src_block_slice_origin)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
+                make_multi_index(ThreadGroup::GetThreadId()));
+
+            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
+
+            threadwise_transfer_.SetSrcSliceOrigin(src_desc,
+                                                   src_block_slice_origin + thread_data_idx_begin);
+        }
+    }
+
+    __device__ void SetDstSliceOrigin(const DstDesc& dst_desc, const Index& dst_block_slice_origin)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
+                make_multi_index(ThreadGroup::GetThreadId()));
+
+            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
+
+            threadwise_transfer_.SetDstSliceOrigin(dst_desc,
+                                                   dst_block_slice_origin + thread_data_idx_begin);
+        }
+    }
+
+    private:
+    static constexpr auto thread_cluster_desc_ =
+        make_cluster_descriptor(ThreadClusterLengths{}, ThreadClusterArrangeOrder{});
+
+    using ThreadwiseTransfer =
+        ThreadwiseTensorSliceTransfer_v6r1r2<SrcData,
+                                             DstData,
+                                             SrcDesc,
+                                             DstDesc,
+                                             ElementwiseOperation,
+                                             decltype(thread_slice_lengths),
+                                             DimAccessOrder,
+                                             VectorDim,
+                                             ScalarPerVector,
+                                             ThreadTransferSrcResetCoordinateAfterRun,
+                                             ThreadTransferDstResetCoordinateAfterRun>;
+
+    ThreadwiseTransfer threadwise_transfer_;
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_streamk.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_streamk.hpp
new file mode 100644
index 0000000000..ed081ad7fc
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_streamk.hpp
@@ -0,0 +1,64 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <vector>
+
+#include "device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
+struct DeviceGemmStreamK : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
+                                                              const void* p_b,
+                                                              void* p_c,
+                                                              ck::index_t M,
+                                                              ck::index_t N,
+                                                              ck::index_t K,
+                                                              ck::index_t StrideA,
+                                                              ck::index_t StrideB,
+                                                              ck::index_t StrideC,
+                                                              AElementwiseOperation a_element_op,
+                                                              BElementwiseOperation b_element_op,
+                                                              CElementwiseOperation c_element_op,
+                                                              ck::index_t NumSKBlocks = 0) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+template <typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
+using DeviceGemmStreamKPtr = std::unique_ptr<DeviceGemmStreamK<ALayout,
+                                                               BLayout,
+                                                               CLayout,
+                                                               ADataType,
+                                                               BDataType,
+                                                               CDataType,
+                                                               AElementwiseOperation,
+                                                               BElementwiseOperation,
+                                                               CElementwiseOperation>>;
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp
new file mode 100644
index 0000000000..6fe5784c28
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp
@@ -0,0 +1,357 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_streamk.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/hip_check_error.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerXDL,
+          ck::index_t NPerXDL,
+          ck::index_t MXdlPerWave,
+          ck::index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          ck::index_t ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          ck::index_t BBlockLdsAddExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CBlockTransferScalarPerVector_NWaveNPerXDL>
+struct DeviceGemmXdlStreamK : public DeviceGemmStreamK<ALayout,
+                                                       BLayout,
+                                                       CLayout,
+                                                       ADataType,
+                                                       BDataType,
+                                                       CDataType,
+                                                       AElementwiseOperation,
+                                                       BElementwiseOperation,
+                                                       CElementwiseOperation>
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+    using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_streamk<
+        BlockSize,
+        BlockToCTileMap_GemmStreamK<MPerBlock,
+                                    NPerBlock,
+                                    K0PerBlock * K1,
+                                    StreamKReductionStrategy::Atomic>,
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        CDataType,
+        ALayout,
+        BLayout,
+        CLayout,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerXDL,
+        NPerXDL,
+        K1,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CBlockTransferScalarPerVector_NWaveNPerXDL,
+        CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>;
+
+    using Argument = typename GridwiseGemm::Argument;
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        void Print(const Argument& karg) { karg.Print(); }
+
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                Print(karg);
+            }
+            if(!GridwiseGemm::CheckValidity(karg))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2 has invalid "
+                    "setting");
+            }
+
+            dim3 grid_dims = karg.block_mapping.get_grid_dims();
+
+            float ave_time = 0;
+
+            const auto kernel = kernel_gemm_xdlops_streamk<GridwiseGemm>;
+
+            // TODO: remove clear buffer for streamk kernels
+            if constexpr(GridwiseGemm::Block2CTileMap::ReductionStrategy ==
+                         StreamKReductionStrategy::Atomic)
+            {
+                hipGetErrorString(hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(CDataType)));
+                ave_time = launch_and_time_kernel(stream_config,
+                                                  kernel,
+                                                  grid_dims,
+                                                  dim3(BlockSize),
+                                                  0,
+                                                  karg.p_a_grid,
+                                                  karg.p_b_grid,
+                                                  karg.p_c_grid,
+                                                  karg.p_workspace_,
+                                                  karg.M,
+                                                  karg.N,
+                                                  karg.K,
+                                                  karg.StrideA,
+                                                  karg.StrideB,
+                                                  karg.StrideC,
+                                                  karg.block_mapping);
+            }
+            else if constexpr(GridwiseGemm::Block2CTileMap::ReductionStrategy ==
+                              StreamKReductionStrategy::Reduction)
+            {
+                char* workspace_semaphore = reinterpret_cast<char*>(karg.p_workspace_) +
+                                            karg.block_mapping.get_workspace_size_for_acc(
+                                                sizeof(typename GridwiseGemm::FloatAcc));
+                auto preprocess = [&]() {
+                    hipGetErrorString(
+                        hipMemsetAsync(workspace_semaphore,
+                                       0,
+                                       karg.block_mapping.get_workspace_size_for_semaphore(),
+                                       stream_config.stream_id_));
+                };
+
+                ave_time = launch_and_time_kernel_with_preprocess(stream_config,
+                                                                  preprocess,
+                                                                  kernel,
+                                                                  grid_dims,
+                                                                  dim3(BlockSize),
+                                                                  0,
+                                                                  karg.p_a_grid,
+                                                                  karg.p_b_grid,
+                                                                  karg.p_c_grid,
+                                                                  karg.p_workspace_,
+                                                                  karg.M,
+                                                                  karg.N,
+                                                                  karg.K,
+                                                                  karg.StrideA,
+                                                                  karg.StrideB,
+                                                                  karg.StrideC,
+                                                                  karg.block_mapping);
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    size_t GetWorkSpaceSize(const BaseArgument* pArg) const override
+    {
+        const Argument* p_arg = dynamic_cast<const Argument*>(pArg);
+        if constexpr(GridwiseGemm::Block2CTileMap::ReductionStrategy ==
+                     StreamKReductionStrategy::Reduction)
+        {
+            return p_arg->block_mapping.get_workspace_size(sizeof(typename GridwiseGemm::FloatAcc));
+        }
+        else
+        {
+            return 0;
+        }
+    }
+
+    void SetWorkSpacePointer(BaseArgument* pArg, void* p_workspace) const override
+    {
+        Argument* pArg_ = dynamic_cast<Argument*>(pArg);
+
+        pArg_->p_workspace_ = p_workspace;
+    }
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& karg)
+    {
+        if(!(ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx90a" ||
+             ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx941" ||
+             ck::get_device_name() == "gfx942"))
+        {
+            return false;
+        }
+        return GridwiseGemm::CheckValidity(karg);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const ADataType* p_a,
+                             const BDataType* p_b,
+                             CDataType* p_c,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             index_t StrideC,
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation,
+                             uint32_t NumSKBlocks = 0xffffffff)
+    {
+        const auto kernel = kernel_gemm_xdlops_streamk<GridwiseGemm>;
+        int occupancy, num_cu;
+        hipError_t rtn;
+        rtn = hipOccupancyMaxActiveBlocksPerMultiprocessor(
+            &occupancy, kernel, BlockSize, GridwiseGemm::GetSharedMemoryNumberOfByte());
+        hip_check_error(rtn);
+
+        hipDeviceProp_t dev_prop;
+        hipDevice_t dev;
+        rtn = hipGetDevice(&dev);
+        hip_check_error(rtn);
+        rtn = hipGetDeviceProperties(&dev_prop, dev);
+        hip_check_error(rtn);
+        num_cu = dev_prop.multiProcessorCount;
+
+        return Argument{p_a,
+                        p_b,
+                        p_c,
+                        M,
+                        N,
+                        K,
+                        StrideA,
+                        StrideB,
+                        StrideC,
+                        static_cast<uint32_t>(num_cu),
+                        static_cast<uint32_t>(occupancy),
+                        NumSKBlocks};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
+                                                      const void* p_b,
+                                                      void* p_c,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
+                                                      index_t StrideA,
+                                                      index_t StrideB,
+                                                      index_t StrideC,
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation,
+                                                      index_t NumSKBlocks = 0) override
+    {
+        const auto kernel = kernel_gemm_xdlops_streamk<GridwiseGemm>;
+        int occupancy, num_cu;
+        hipError_t rtn;
+        rtn = hipOccupancyMaxActiveBlocksPerMultiprocessor(
+            &occupancy, kernel, BlockSize, GridwiseGemm::GetSharedMemoryNumberOfByte());
+        hip_check_error(rtn);
+
+        hipDeviceProp_t dev_prop;
+        hipDevice_t dev;
+        rtn = hipGetDevice(&dev);
+        hip_check_error(rtn);
+        rtn = hipGetDeviceProperties(&dev_prop, dev);
+        hip_check_error(rtn);
+        num_cu = dev_prop.multiProcessorCount;
+
+        return std::make_unique<Argument>(reinterpret_cast<const ADataType*>(p_a),
+                                          reinterpret_cast<const BDataType*>(p_b),
+                                          reinterpret_cast<CDataType*>(p_c),
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideC,
+                                          static_cast<uint32_t>(num_cu),
+                                          static_cast<uint32_t>(occupancy),
+                                          static_cast<uint32_t>(NumSKBlocks));
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override { return GridwiseGemm::GetTypeString(); }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index 5c8b9f419a..c4f00e83b7 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -7,6 +7,8 @@
 #include "ck/utility/number.hpp"
 #include "ck/tensor_description/tensor_adaptor.hpp"
 #include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include <limits>
+#include <stdlib.h>
 
 namespace ck {
 
@@ -669,4 +671,406 @@ struct BlockToCTileMap_3DGrid_KSplit
     }
 };
 
+enum StreamKReductionStrategy
+{
+    Atomic = 0, // sk block use atomic to do reduction
+    Reduction,  // let some workgroup responsible for doing the reduction operation
+};
+
+template <uint32_t MPerBlock_,
+          uint32_t NPerBlock_,
+          uint32_t KPerBlock_,
+          StreamKReductionStrategy ReductionStrategy_ = StreamKReductionStrategy::Atomic,
+          uint32_t TileSwizzleSubM_                   = 8>
+struct BlockToCTileMap_GemmStreamK
+{
+    static constexpr uint32_t min_k_iters_per_sk_block          = 2;
+    static constexpr uint32_t MPerBlock                         = MPerBlock_;
+    static constexpr uint32_t NPerBlock                         = NPerBlock_;
+    static constexpr uint32_t KPerBlock                         = KPerBlock_;
+    static constexpr StreamKReductionStrategy ReductionStrategy = ReductionStrategy_;
+    static constexpr uint32_t tile_swizzle_sub_m                = TileSwizzleSubM_;
+
+    //--------------------------------------
+    // pass to device
+    uint32_t sk_num_blocks;
+    uint32_t sk_num_big_blocks;
+    uint32_t dp_start_block_idx;
+    uint32_t reduction_start_block_idx;
+    uint32_t k_iters_per_big_block;
+    MDiv2 n_tiles;
+    MDiv k_iters_per_tile;
+    MDiv eqav_tiles_big;    // for reduction
+    MDiv eqav_tiles_little; // for reduction
+
+    // MDiv tile_swizzle_sub_m_rem;
+    //--------------------------------------
+
+    // prefer construct on host
+    BlockToCTileMap_GemmStreamK(uint32_t m,
+                                uint32_t n,
+                                uint32_t k,
+                                uint32_t num_cu,
+                                uint32_t occupancy,
+                                uint32_t sk_blocks = 0xffffffff)
+    {
+        uint32_t num_tiles =
+            math::integer_divide_ceil(m, MPerBlock) * math::integer_divide_ceil(n, NPerBlock);
+        k_iters_per_tile = MDiv(math::integer_divide_ceil(k, KPerBlock));
+
+        // one cu can hold one wg at one time, from the whole chip's point of view
+        // if number of wg is same as num_cu, we call it 1 dispatch
+        // if number of wg is 2x num_cu, we call it 2 dispatches.
+        // one dispatch can deliver wg same as num_cu (full dispatch), or less than num_cu (partial
+        // dispatch)
+        //
+        uint32_t full_dispatches         = num_tiles / num_cu;
+        uint32_t full_dispatch_tiles     = full_dispatches * num_cu;
+        uint32_t partial_dispatche_tiles = num_tiles - full_dispatch_tiles;
+
+        uint32_t sk_occupancy = occupancy;
+        uint32_t dp_tiles     = full_dispatch_tiles;
+        uint32_t sk_tiles     = partial_dispatche_tiles;
+
+        if(full_dispatches < occupancy)
+        {
+            // in this case, we allocate all blocks as sk blocks
+            // sk_occupancy = occupancy - full_dispatches;
+            sk_occupancy = 1; // TODO: single occ seems better
+            dp_tiles     = full_dispatch_tiles;
+            sk_tiles     = partial_dispatche_tiles;
+        }
+        else if((occupancy > 1) && (full_dispatches % occupancy == occupancy - 1))
+        {
+            // e.g. occupancy = 2, full_dispatches = 3, 5, 7 ...
+            //      occupancy = 3, full_dispatches = 5, 8, 11 ...
+            //      occupancy = 4, full_dispatches = 7, 11 ...
+            sk_occupancy = 1; // left 1 slot for sk occupancy
+            dp_tiles     = full_dispatch_tiles;
+            sk_tiles     = partial_dispatche_tiles;
+        }
+        else
+        {
+            // others, we reduce 1 dispatch from dp, together with partial dispatch,
+            // to construct sk dispatch
+            sk_occupancy = occupancy - ((full_dispatches - 1) % occupancy);
+            dp_tiles     = full_dispatch_tiles - num_cu;
+            sk_tiles     = partial_dispatche_tiles + num_cu;
+        }
+
+        // uint32_t dp_iters_per_block = k_iters_per_tile.get();
+        uint32_t sk_total_iters = k_iters_per_tile.get() * sk_tiles;
+        uint32_t dp_num_blocks  = 0;
+
+        {
+            uint32_t min_sk_tiles = (sk_tiles >= num_cu) ? num_cu : (sk_tiles + 1);
+            uint32_t max_sk_tiles =
+                (sk_tiles >= num_cu) ? num_cu * sk_occupancy
+                                     : math::min(num_cu, sk_total_iters / min_k_iters_per_sk_block);
+
+            // if use dp for sk-block, how many iters do we need
+            uint32_t dp_for_sk_iters = k_iters_per_tile.get();
+
+            uint32_t best_sk_score =
+                std::numeric_limits<int>::max(); // we need to find the smallest sk iters
+            for(uint32_t tentative_sk_blocks = min_sk_tiles; tentative_sk_blocks < max_sk_tiles;
+                tentative_sk_blocks++)
+            {
+                uint32_t tentative_sk_iters_per_block =
+                    (sk_total_iters + tentative_sk_blocks - 1) / tentative_sk_blocks;
+                uint32_t tentative_sk_iters = tentative_sk_iters_per_block;
+                uint32_t sk_blocks_per_tile = (tentative_sk_blocks + sk_tiles - 1) / sk_tiles;
+
+                // TODO: carefully adjust this parameter
+                //       the more sk_blocks_per_tile, the worse the overhead
+                uint32_t cross_sk_blocks_overhead = sk_blocks_per_tile;
+                if(tentative_sk_blocks % sk_tiles != 0)
+                {
+                    // penalty for uneven divide
+                    cross_sk_blocks_overhead +=
+                        sk_blocks_per_tile * tentative_sk_iters_per_block / 50;
+                }
+
+                uint32_t tentative_sk_score = tentative_sk_iters + cross_sk_blocks_overhead;
+
+                if(tentative_sk_score < best_sk_score)
+                {
+                    best_sk_score = tentative_sk_score;
+                    sk_num_blocks = tentative_sk_blocks;
+                }
+            }
+
+            if(best_sk_score >= dp_for_sk_iters)
+            {
+                sk_num_blocks = 0;
+            }
+
+            // give a chance to control num of sk blocks
+            sk_num_blocks = sk_blocks != 0xffffffff ? sk_blocks : sk_num_blocks;
+
+            if(sk_num_blocks == 0)
+            {
+                sk_num_big_blocks     = 0;
+                k_iters_per_big_block = 0;
+
+                dp_num_blocks      = num_tiles; // all tile to be dp block
+                dp_start_block_idx = 0;
+                sk_total_iters     = 0; // clear this tiles
+            }
+            else
+            {
+                // k_iters_per_sk_block is the floor of avg each ck block loop over tiles.
+                // we need to decide how many iters for each sk block
+                // let m = k_iters_per_sk_block
+                // some of the sk block (little) will cover m iters, some (big) will cover m+1
+                // we have
+                // 1) l + b = sk_blocks
+                // 2) l * m + b * (m + 1) = sk_total_iters
+                //      => (l + b) * m + b = sk_total_iters
+                //      => sk_blocks * m + b = sk_total_iters
+                //      => b = sk_total_iters - m * sk_blocks
+                //      NOTE: big could be zero
+                uint32_t k_iters_per_sk_block = sk_total_iters / sk_num_blocks;
+                sk_num_big_blocks     = sk_total_iters - k_iters_per_sk_block * sk_num_blocks;
+                k_iters_per_big_block = k_iters_per_sk_block + 1;
+
+                dp_num_blocks      = dp_tiles;
+                dp_start_block_idx = (sk_num_blocks + num_cu - 1) / num_cu * num_cu;
+            }
+        }
+        n_tiles                   = MDiv2(math::integer_divide_ceil(n, NPerBlock));
+        reduction_start_block_idx = dp_start_block_idx + dp_num_blocks;
+
+        if constexpr(ReductionStrategy == StreamKReductionStrategy::Reduction)
+        {
+            uint32_t upper_big    = math::lcm(k_iters_per_big_block, k_iters_per_tile.get());
+            uint32_t upper_little = math::lcm(k_iters_per_big_block - 1, k_iters_per_tile.get());
+            eqav_tiles_big        = MDiv(upper_big / k_iters_per_tile.get());
+            eqav_tiles_little     = MDiv(upper_little / k_iters_per_tile.get());
+        }
+
+#if 0
+        printf("cu:%d, occupancy:%d, grids:%d, num_tiles:%d, dp_tiles:%d, sk_num_big_blocks:%d, "
+               "sk_num_blocks:%d, "
+               "sk_total_iters:%d, dp_start_block_idx:%d, dp_iters_per_block:%d, dp_num_blocks:%d, "
+               "k_iters_per_tile:%d, k_iters_per_big_block:%d, reduction_start_block_idx:%u, "
+               "sk_tiles:%u, workspace(acc float):%u\n",
+               num_cu,
+               occupancy,
+               get_grid_dims().x,
+               num_tiles,
+               dp_tiles,
+               sk_num_big_blocks,
+               sk_num_blocks,
+               sk_total_iters,
+               dp_start_block_idx,
+               dp_iters_per_block,
+               dp_num_blocks,
+               k_iters_per_tile.get(),
+               k_iters_per_big_block,
+               reduction_start_block_idx,
+               get_sk_tiles(),
+               get_workspace_size(sizeof(float)));
+#endif
+    }
+
+    __host__ __device__ uint32_t get_sk_total_iters() const
+    {
+        uint32_t sk_total_iters = sk_num_big_blocks * k_iters_per_big_block +
+                                  (sk_num_blocks - sk_num_big_blocks) * (k_iters_per_big_block - 1);
+        return sk_total_iters;
+    }
+
+    __host__ __device__ uint32_t get_sk_tiles() const
+    {
+        // tiles for sk
+        uint32_t sk_total_iters = get_sk_total_iters();
+        return k_iters_per_tile.div(sk_total_iters);
+    }
+
+    __host__ __device__ dim3 get_grid_dims() const
+    {
+        if constexpr(ReductionStrategy == StreamKReductionStrategy::Reduction)
+        {
+            return dim3(reduction_start_block_idx + get_sk_tiles(), 1, 1);
+        }
+        else
+            return dim3(reduction_start_block_idx, 1, 1);
+    }
+
+    __device__ uint32_t get_block_idx() const
+    {
+        // TODO: swizzle block index for better locality
+        return __builtin_amdgcn_readfirstlane(blockIdx.x);
+    }
+
+    __device__ void
+    get_block_itr(uint32_t block_idx, uint32_t& iter_start, uint32_t& iter_end) const
+    {
+        if(block_idx < sk_num_big_blocks)
+        {
+            iter_start = block_idx * k_iters_per_big_block;
+            iter_end   = iter_start + k_iters_per_big_block;
+        }
+        else if(block_idx < sk_num_blocks)
+        {
+            iter_start = (sk_num_big_blocks * k_iters_per_big_block) +
+                         (block_idx - sk_num_big_blocks) * (k_iters_per_big_block - 1);
+            iter_end = iter_start + (k_iters_per_big_block - 1);
+        }
+        else if(block_idx >= dp_start_block_idx)
+        {
+            uint32_t sk_total_iters     = get_sk_total_iters();
+            uint32_t dp_iters_per_block = k_iters_per_tile.get();
+            iter_start = sk_total_iters + (block_idx - dp_start_block_idx) * dp_iters_per_block;
+            iter_end   = iter_start + dp_iters_per_block;
+        }
+    }
+
+    __device__ uint32_t get_current_iter_length(uint32_t iter_start,
+                                                uint32_t iter_end,
+                                                uint32_t total_iter_length) const
+    {
+        uint32_t iter_length_mod, iter_length_quo /*unused*/;
+        k_iters_per_tile.divmod(iter_end, iter_length_quo, iter_length_mod);
+        uint32_t current_iter_length = math::min(
+            iter_length_mod == 0 ? (iter_end - iter_start) : iter_length_mod, total_iter_length);
+        return current_iter_length;
+    }
+
+    __device__ uint32_t get_tile_idx(uint32_t iter) const { return k_iters_per_tile.div(iter); }
+
+    __device__ void
+    get_tile_idx_with_offset(uint32_t iter, uint32_t& tile_idx, uint32_t& iter_offset) const
+    {
+        k_iters_per_tile.divmod(iter, tile_idx, iter_offset);
+    }
+
+    __device__ auto tile_to_spatial(uint32_t tile_idx, uint32_t m, uint32_t n) const
+    {
+        uint32_t m_tile_idx, n_tile_idx;
+        uint32_t n_tiles_value = math::integer_divide_ceil(n, NPerBlock);
+        n_tiles.divmod(tile_idx, n_tiles_value, m_tile_idx, n_tile_idx);
+
+        // swizzle tile
+        uint32_t m_tiles = math::integer_divide_ceil(m, MPerBlock);
+
+        uint32_t tile_swizzle_sub_m_rem = m_tiles % tile_swizzle_sub_m;
+
+        const auto sub_m_adapt = (m_tile_idx < (m_tiles - tile_swizzle_sub_m_rem))
+                                     ? tile_swizzle_sub_m
+                                     : tile_swizzle_sub_m_rem;
+
+        uint32_t m_tile_idx_sub0, m_tile_idx_sub1;
+        m_tile_idx_sub0 = m_tile_idx / tile_swizzle_sub_m;
+        m_tile_idx_sub1 = m_tile_idx % tile_swizzle_sub_m;
+
+        uint32_t tile_idx_local = n_tile_idx + m_tile_idx_sub1 * n_tiles_value;
+
+        uint32_t m_tile_idx_with_adapt, n_tile_idx_with_adapt;
+
+        n_tile_idx_with_adapt = tile_idx_local / sub_m_adapt;
+        m_tile_idx_with_adapt = tile_idx_local % sub_m_adapt;
+        return make_tuple(m_tile_idx_with_adapt + m_tile_idx_sub0 * tile_swizzle_sub_m,
+                          n_tile_idx_with_adapt);
+    }
+
+    __host__ __device__ uint32_t get_workspace_size_for_acc(uint32_t acc_element_bytes) const
+    {
+        static constexpr uint32_t alignment = 128;
+        uint32_t acc_buffer_bytes =
+            MPerBlock * NPerBlock * get_total_acc_buffers() * acc_element_bytes;
+        return (acc_buffer_bytes + alignment - 1) / alignment * alignment;
+    }
+
+    __host__ __device__ uint32_t get_workspace_size_for_semaphore() const
+    {
+        return get_sk_tiles() * sizeof(uint32_t);
+    }
+
+    __host__ __device__ uint32_t get_workspace_size(uint32_t acc_element_bytes) const
+    {
+        return get_workspace_size_for_acc(acc_element_bytes) + get_workspace_size_for_semaphore();
+    }
+
+    __host__ __device__ uint32_t get_tile_intersections(uint32_t tiles_,
+                                                        const MDiv& eqav_tiles_) const
+    {
+        uint32_t tile_idx_       = tiles_ == 0 ? 0 : (tiles_ - 1);
+        uint32_t max_eqav_tiles_ = eqav_tiles_.get() - 1;
+        uint32_t quo_, rem_;
+        eqav_tiles_.divmod(tile_idx_, quo_, rem_);
+        return quo_ * max_eqav_tiles_ + rem_;
+    }
+
+    __host__ __device__ uint32_t get_tiles_cover_sk_block(uint32_t num_sk_blocks_,
+                                                          uint32_t iters_per_sk_block_) const
+    {
+        return k_iters_per_tile.div(num_sk_blocks_ * iters_per_sk_block_ + k_iters_per_tile.get() -
+                                    1);
+    }
+
+    __host__ __device__ uint32_t get_total_acc_buffers() const
+    {
+        uint32_t tiles_cover_big_blocks =
+            get_tiles_cover_sk_block(sk_num_big_blocks, k_iters_per_big_block);
+        uint32_t tiles_cover_little_blocks =
+            get_tiles_cover_sk_block(sk_num_blocks - sk_num_big_blocks, k_iters_per_big_block - 1);
+
+        uint32_t total_intersec_big =
+            get_tile_intersections(tiles_cover_big_blocks, eqav_tiles_big);
+        uint32_t total_intersec_little =
+            get_tile_intersections(tiles_cover_little_blocks, eqav_tiles_little);
+
+        return sk_num_blocks + total_intersec_big + total_intersec_little;
+    }
+
+    __device__ uint32_t get_acc_buffer_offset_from_tile(uint32_t tile_idx_) const
+    {
+        // TODO: from big to little
+        uint32_t tiles_cover_big_blocks =
+            get_tiles_cover_sk_block(sk_num_big_blocks, k_iters_per_big_block);
+        if(tile_idx_ < tiles_cover_big_blocks)
+        {
+            uint32_t touched_sk_blocks =
+                (tile_idx_ * k_iters_per_tile.get() + k_iters_per_big_block - 1) /
+                k_iters_per_big_block;
+            uint32_t current_intersec = get_tile_intersections(tile_idx_, eqav_tiles_big);
+            return touched_sk_blocks + current_intersec;
+        }
+        else
+        {
+            uint32_t iters_per_little_sk_block = k_iters_per_big_block - 1;
+            uint32_t tile_idx_little_reverse   = get_sk_tiles() - tile_idx_;
+            uint32_t touched_sk_blocks =
+                (tile_idx_little_reverse * k_iters_per_tile.get() + iters_per_little_sk_block - 1) /
+                iters_per_little_sk_block;
+            uint32_t current_intersec =
+                get_tile_intersections(tile_idx_little_reverse, eqav_tiles_little);
+            return get_total_acc_buffers() - (touched_sk_blocks + current_intersec);
+        }
+    }
+
+    __device__ uint32_t get_acc_buffer_offset_from_block(uint32_t block_idx_) const
+    {
+        uint32_t iters_per_big_sk_block    = k_iters_per_big_block;
+        uint32_t iters_per_little_sk_block = k_iters_per_big_block - 1;
+        if(block_idx_ < sk_num_big_blocks)
+        {
+            uint32_t touched_tiles    = k_iters_per_tile.div(block_idx_ * iters_per_big_sk_block +
+                                                          k_iters_per_tile.get() - 1);
+            uint32_t current_intersec = get_tile_intersections(touched_tiles, eqav_tiles_big);
+            return block_idx_ + current_intersec;
+        }
+        else
+        {
+            uint32_t block_idx_little_reverse = sk_num_blocks - block_idx_;
+            uint32_t touched_tiles            = k_iters_per_tile.div(
+                block_idx_little_reverse * iters_per_little_sk_block + k_iters_per_tile.get() - 1);
+            uint32_t current_intersec = get_tile_intersections(touched_tiles, eqav_tiles_little);
+            return get_total_acc_buffers() - (block_idx_little_reverse + current_intersec);
+        }
+    }
+};
+
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v3.hpp
new file mode 100644
index 0000000000..ced62241cd
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v3.hpp
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+
+namespace ck {
+
+struct GridwiseGemmPipeline_v3
+{
+    __host__ __device__ static constexpr bool IsSupported(index_t)
+    {
+        // TODO: improve applicability
+        return true;
+    }
+
+    template <typename AGridDesc,
+              typename ABlockDesc,
+              typename ABlockTransfer,
+              typename AGridBuffer,
+              typename ABlockBuffer,
+              typename ABlockTransferStep,
+              typename BGridDesc,
+              typename BBlockDesc,
+              typename BBlockTransfer,
+              typename BGridBuffer,
+              typename BBlockBuffer,
+              typename BBlockTransferStep,
+              typename BlockwiseGemm,
+              typename CThreadBuffer>
+    __device__ static void Run(const AGridDesc& a_grid_desc,
+                               const ABlockDesc& a_block_desc,
+                               ABlockTransfer& a_blockwise_copy,
+                               const AGridBuffer& a_grid_buf,
+                               ABlockBuffer& a_block_buf,
+                               const ABlockTransferStep& a_block_copy_step,
+                               const BGridDesc& b_grid_desc,
+                               const BBlockDesc& b_block_desc,
+                               BBlockTransfer& b_blockwise_copy,
+                               const BGridBuffer& b_grid_buf,
+                               BBlockBuffer& b_block_buf,
+                               const BBlockTransferStep& b_block_copy_step,
+                               const BlockwiseGemm& blockwise_gemm,
+                               CThreadBuffer& c_thread_buf,
+                               index_t num_loop)
+    {
+        // global read 0
+        a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+        b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+        a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+        b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+
+        // Initialize C
+        c_thread_buf.Clear();
+
+        // LDS write 0
+        a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+        b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+        num_loop--;
+
+        while(num_loop > 0)
+        {
+            a_blockwise_copy.RunRead(a_grid_desc, a_grid_buf);
+            block_sync_lds();
+            b_blockwise_copy.RunRead(b_grid_desc, b_grid_buf);
+
+            blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
+
+            block_sync_lds();
+
+            a_blockwise_copy.MoveSrcSliceWindow(a_grid_desc, a_block_copy_step);
+            b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc, b_block_copy_step);
+            a_blockwise_copy.RunWrite(a_block_desc, a_block_buf);
+            b_blockwise_copy.RunWrite(b_block_desc, b_block_buf);
+
+            num_loop--;
+        }
+        // tail
+        {
+            block_sync_lds();
+            blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
new file mode 100644
index 0000000000..4408b34870
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
@@ -0,0 +1,1183 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v6r1r2.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v3.hpp"
+#include "ck/utility/workgroup_barrier.hpp"
+#include "ck/utility/reduction_functions_accumulate.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_xdlops_streamk(const typename GridwiseGemm::FloatAB* p_a_grid,
+                                   const typename GridwiseGemm::FloatAB* p_b_grid,
+                                   typename GridwiseGemm::FloatC* p_c_grid,
+                                   void* p_workspace,
+                                   index_t M,
+                                   index_t N,
+                                   index_t K,
+                                   index_t StrideA,
+                                   index_t StrideB,
+                                   index_t StrideC,
+                                   typename GridwiseGemm::Block2CTileMap block_mapping)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+    constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
+
+    __shared__ uint8_t p_shared[shared_size];
+
+    GridwiseGemm::Run(p_a_grid,
+                      p_b_grid,
+                      p_c_grid,
+                      p_workspace,
+                      M,
+                      N,
+                      K,
+                      StrideA,
+                      StrideB,
+                      StrideC,
+                      block_mapping,
+                      static_cast<void*>(p_shared));
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_c_grid;
+    ignore = p_workspace;
+    ignore = M;
+    ignore = N;
+    ignore = K;
+    ignore = StrideA;
+    ignore = StrideB;
+    ignore = StrideC;
+    ignore = block_mapping;
+#endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
+}
+
+template <index_t BlockSize,
+          typename Block2CTileMap_,
+          typename FloatAB_,
+          typename FloatAcc_,
+          typename FloatC_,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t K1Value,
+          index_t MRepeat,
+          index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_K1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_K1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
+          typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock>
+struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_streamk
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto K1        = Number<K1Value>{};
+    static constexpr auto M01       = 1;
+    static constexpr auto N01       = 1;
+    static constexpr auto KPerBlock = K0PerBlock * K1;
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+    using FloatAcc        = FloatAcc_;
+    using FloatCShuffle   = FloatAcc;
+
+    using Block2CTileMap = Block2CTileMap_;
+    using FloatAB        = FloatAB_;
+    using FloatC         = FloatC_;
+
+    struct Argument : public ck::tensor_operation::device::BaseArgument
+    {
+        const FloatAB* p_a_grid;
+        const FloatAB* p_b_grid;
+        FloatC* p_c_grid;
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        Block2CTileMap block_mapping;
+
+        Argument(const FloatAB* p_a_grid_,
+                 const FloatAB* p_b_grid_,
+                 FloatC* p_c_grid_,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_,
+                 uint32_t num_cu,
+                 uint32_t occupancy,
+                 uint32_t num_sk_blocks_)
+            : p_a_grid(p_a_grid_),
+              p_b_grid(p_b_grid_),
+              p_c_grid(p_c_grid_),
+              M(M_),
+              N(N_),
+              K(K_),
+              StrideA(StrideA_),
+              StrideB(StrideB_),
+              StrideC(StrideC_),
+              block_mapping(M, N, K, num_cu, occupancy, num_sk_blocks_)
+        {
+        }
+
+        void Print() const
+        {
+            std::cout << "arg {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << std::endl;
+        }
+    };
+
+    __host__ __device__ static auto CalculateGridSize(const Argument& karg)
+    {
+        return std::make_tuple(math::integer_divide_ceil(karg.N, NPerBlock),
+                               math::integer_divide_ceil(karg.M, MPerBlock),
+                               karg.k_batch);
+    }
+
+    __host__ __device__ static auto CalculateK0(index_t KPad) { return KPad / K1; }
+
+    __host__ __device__ static auto
+    MakeAGridDescriptor_K0_M_K1(index_t M, index_t MPad, index_t K, index_t KPad, index_t StrideA)
+    {
+        const index_t K0 = CalculateK0(KPad);
+
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return transform_tensor_descriptor(a_grid_desc_m_kpad,
+                                           make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
+                                                      make_right_pad_transform(M, MPad - M)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    __host__ __device__ static auto
+    MakeBGridDescriptor_K0_N_K1(index_t K, index_t KPad, index_t N, index_t NPad, index_t StrideB)
+    {
+        const index_t K0 = CalculateK0(KPad);
+
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
+            b_grid_desc_k_n,
+            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return transform_tensor_descriptor(b_grid_desc_kpad_n,
+                                           make_tuple(make_unmerge_transform(make_tuple(K0, K1)),
+                                                      make_right_pad_transform(N, NPad - N)),
+                                           make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    __host__ __device__ static auto
+    MakeCGridDescriptor_M_N(index_t M, index_t MPad, index_t N, index_t NPad, index_t StrideC)
+    {
+        const auto c_grid_desc_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        return transform_tensor_descriptor(c_grid_desc_m_n,
+                                           make_tuple(make_right_pad_transform(M, MPad - M),
+                                                      make_right_pad_transform(N, NPad - N)),
+                                           make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                           make_tuple(Sequence<0>{}, Sequence<1>{}));
+    }
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(Number<K0PerBlock>{}, Number<MPerBlock>{}, K1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * K1, K1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(Number<K0PerBlock>{}, Number<NPerBlock>{}, K1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * K1, K1, I1));
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        constexpr auto max_lds_align = K1;
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_k0_m_k1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_k0_n_k1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        constexpr auto a_block_space_size_aligned =
+            math::integer_least_multiple(a_block_desc_k0_m_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned =
+            math::integer_least_multiple(b_block_desc_k0_n_k1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto c_block_size =
+            GetCBlockDescriptor_MBlock_MPerShuffle_NBlock_NPerShuffle().GetElementSpaceSize();
+
+        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
+                             sizeof(FloatAB),
+                         c_block_size * sizeof(FloatCShuffle));
+    }
+
+    __host__ __device__ static constexpr bool CheckValidity(const Argument& karg)
+    {
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(karg.K % ABlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+        else
+        {
+            if(karg.M % ABlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(karg.N % BBlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+        else
+        {
+            if(karg.K % BBlockTransferSrcScalarPerVector != 0)
+                return false;
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+        {
+            if(karg.N % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+                return false;
+        }
+        else
+        {
+            if(karg.M % CBlockTransferScalarPerVector_NWaveNPerXDL != 0)
+                return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainK0BlockLoop(index_t K0)
+    {
+        const bool has_main_k0_block_loop = K0 > K0PerBlock;
+
+        return has_main_k0_block_loop;
+    }
+
+    template <typename CGridDesc>
+    __host__ __device__ static constexpr auto
+    MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(const CGridDesc& c_m_n_grid_desc)
+    {
+        const auto M = c_m_n_grid_desc.GetLength(I0);
+        const auto N = c_m_n_grid_desc.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        return transform_tensor_descriptor(
+            c_m_n_grid_desc,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+    }
+
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    template <typename CGridDesc>
+    __host__ __device__ static constexpr auto MakeCBlockClusterAdaptor(
+        const CGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
+    {
+        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc>(
+            c_m_n_grid_desc, 8, KBatch);
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCBlockDescriptor_MBlock_MPerShuffle_NBlock_NPerShuffle()
+    {
+        constexpr index_t MWave = MPerBlock / (MRepeat * MPerXDL);
+        constexpr index_t NWave = NPerBlock / (NRepeat * NPerXDL);
+
+        return make_naive_tensor_descriptor_packed(
+            make_tuple(I1,
+                       Number<CShuffleMRepeatPerShuffle * MWave * MPerXDL>{},
+                       I1,
+                       Number<CShuffleNRepeatPerShuffle * NWave * NPerXDL>{}));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCBlockDescriptor_MShuffleRepeat_MPerShuffle_NShuffleRepeat_NPerShuffle()
+    {
+        constexpr index_t MWave = MPerBlock / (MRepeat * MPerXDL);
+        constexpr index_t NWave = NPerBlock / (NRepeat * NPerXDL);
+
+        return make_naive_tensor_descriptor_packed(
+            make_tuple(Number<MRepeat / CShuffleMRepeatPerShuffle>{},
+                       Number<CShuffleMRepeatPerShuffle * MWave * MPerXDL>{},
+                       Number<NRepeat / CShuffleNRepeatPerShuffle>{},
+                       Number<CShuffleNRepeatPerShuffle * NWave * NPerXDL>{}));
+    }
+
+    __host__ __device__ static constexpr auto GetClusterLengthReduction()
+    {
+        // TODO: assume C is row major
+        // TODO: we always first loop over N, then M
+        constexpr auto NPerBlockPow2 = math::next_power_of_two<NPerBlock>();
+        constexpr auto NPerBlockReduction =
+            NPerBlockPow2 / CBlockTransferScalarPerVector_NWaveNPerXDL;
+        constexpr auto MPerBlockReduction =
+            (BlockSize + NPerBlockReduction - 1) / NPerBlockReduction;
+        return Sequence<MPerBlockReduction, NPerBlockReduction>{};
+    }
+
+    __host__ __device__ static constexpr auto GetPartialAccBlockDescriptor()
+    {
+        const auto c_partial_acc_block_m_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MPerBlock, NPerBlock),
+                                                    make_tuple(NPerBlock, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MPerBlock, NPerBlock),
+                                                    make_tuple(I1, MPerBlock));
+            }
+        }();
+        return c_partial_acc_block_m_n;
+    }
+
+    using CGridDesc_M_N = remove_cvref_t<decltype(MakeCGridDescriptor_M_N(1, 1, 1, 1, 1))>;
+
+    __device__ static void Run(const FloatAB* p_a_grid,
+                               const FloatAB* p_b_grid,
+                               FloatC* p_c_grid,
+                               void* p_workspace,
+                               index_t M,
+                               index_t N,
+                               index_t K,
+                               index_t StrideA,
+                               index_t StrideB,
+                               index_t StrideC,
+                               Block2CTileMap block_mapping,
+                               void* __restrict__ p_shared_block)
+    {
+        uint32_t m        = M;
+        uint32_t n        = N;
+        uint32_t k        = K;
+        uint32_t pad_m    = (m + MPerBlock - 1) / MPerBlock * MPerBlock;
+        uint32_t pad_n    = (n + NPerBlock - 1) / NPerBlock * NPerBlock;
+        uint32_t pad_k    = (k + KPerBlock - 1) / KPerBlock * KPerBlock;
+        uint32_t stride_a = StrideA;
+        uint32_t stride_b = StrideB;
+        uint32_t stride_c = StrideC;
+
+        const auto a_k0_m_k1_grid_desc = MakeAGridDescriptor_K0_M_K1(m, pad_m, k, pad_k, stride_a);
+        const auto b_k0_n_k1_grid_desc = MakeBGridDescriptor_K0_N_K1(k, pad_k, n, pad_n, stride_b);
+        const auto c_grid_desc_m_n     = MakeCGridDescriptor_M_N(m, pad_m, n, pad_n, stride_c);
+
+        const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
+        const AElementwiseOperation a_element_op = AElementwiseOperation{};
+        const BElementwiseOperation b_element_op = BElementwiseOperation{};
+        const CElementwiseOperation c_element_op = CElementwiseOperation{};
+
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_k0_m_k1_grid_desc.GetElementSpaceSize());
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_k0_n_k1_grid_desc.GetElementSpaceSize());
+        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+        // lds max alignment
+        constexpr auto max_lds_align = K1;
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_k0_m_k1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_k0_n_k1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        auto blockwise_gemm =
+            BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatAB,
+                                                                FloatAcc,
+                                                                decltype(a_block_desc_k0_m_k1),
+                                                                decltype(b_block_desc_k0_n_k1),
+                                                                MPerXDL,
+                                                                NPerXDL,
+                                                                MRepeat,
+                                                                NRepeat,
+                                                                K1>{};
+
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size =
+            math::integer_least_multiple(a_block_desc_k0_m_k1.GetElementSpaceSize(), max_lds_align);
+
+        FloatAB* p_a_block = static_cast<FloatAB*>(p_shared_block);
+        FloatAB* p_b_block = static_cast<FloatAB*>(p_shared_block) + a_block_space_size;
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(K0PerBlock, 0, 0);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            p_a_block, a_block_desc_k0_m_k1.GetElementSpaceSize());
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            p_b_block, b_block_desc_k0_n_k1.GetElementSpaceSize());
+
+        // gridwise GEMM pipeline
+        const auto gridwise_gemm_pipeline = GridwiseGemmPipeline_v3();
+
+        uint32_t block_idx = block_mapping.get_block_idx();
+        bool is_sk_block   = block_idx < block_mapping.sk_num_blocks;
+        bool is_dp_block   = block_idx >= block_mapping.dp_start_block_idx &&
+                           block_idx < block_mapping.reduction_start_block_idx;
+        bool is_reduction_block = block_idx >= block_mapping.reduction_start_block_idx;
+        bool is_padding_block   = block_idx >= block_mapping.sk_num_blocks &&
+                                block_idx < block_mapping.dp_start_block_idx;
+        uint32_t iter_start, iter_end;
+        block_mapping.get_block_itr(block_idx, iter_start, iter_end);
+        uint32_t total_iter_length = iter_end - iter_start;
+
+        if(is_padding_block)
+            return;
+
+        uint32_t* p_semaphore =
+            reinterpret_cast<uint32_t*>(reinterpret_cast<char*>(p_workspace) +
+                                        block_mapping.get_workspace_size_for_acc(sizeof(FloatAcc)));
+
+        if constexpr(Block2CTileMap::ReductionStrategy == StreamKReductionStrategy::Reduction)
+        {
+            if(is_reduction_block)
+            {
+                // descriptors
+                constexpr auto cluster_length_reduce = GetClusterLengthReduction();
+                constexpr auto reduce_desc = make_cluster_descriptor(cluster_length_reduce);
+                const auto reduce_thread_cluster_idx =
+                    reduce_desc.CalculateBottomIndex(make_multi_index(get_thread_local_1d_id()));
+                const auto thread_m_cluster_id = reduce_thread_cluster_idx[I0];
+                const auto thread_n_cluster_id = reduce_thread_cluster_idx[I1];
+
+                constexpr auto MReduceIters =
+                    math::integer_divide_ceil(Number<MPerBlock>{}, cluster_length_reduce.At(I0));
+                constexpr auto NReduceIters = math::integer_divide_ceil(
+                    Number<NPerBlock>{},
+                    cluster_length_reduce.At(I1) *
+                        Number<CBlockTransferScalarPerVector_NWaveNPerXDL>{});
+
+                constexpr auto acc_thread_buf_load_desc = make_naive_tensor_descriptor_packed(
+                    make_tuple(I1, Number<CBlockTransferScalarPerVector_NWaveNPerXDL>{}));
+                constexpr auto acc_thread_buf_store_desc = make_naive_tensor_descriptor_packed(
+                    make_tuple(I1, I1, I1, Number<CBlockTransferScalarPerVector_NWaveNPerXDL>{}));
+
+                constexpr auto c_partial_acc_block_m_n = GetPartialAccBlockDescriptor();
+
+                constexpr auto partial_acc_load_step_n = make_multi_index(
+                    0, cluster_length_reduce.At(I1) * CBlockTransferScalarPerVector_NWaveNPerXDL);
+                constexpr auto partial_acc_load_step_n_reverse =
+                    make_multi_index(0,
+                                     -1 * cluster_length_reduce.At(I1).value * (NReduceIters - 1) *
+                                         CBlockTransferScalarPerVector_NWaveNPerXDL);
+                constexpr auto partial_acc_load_step_m =
+                    make_multi_index(cluster_length_reduce.At(I0), 0);
+
+                constexpr auto partial_acc_store_step_n = make_multi_index(
+                    0,
+                    0,
+                    0,
+                    cluster_length_reduce.At(I1) * CBlockTransferScalarPerVector_NWaveNPerXDL);
+                constexpr auto partial_acc_store_step_n_reverse =
+                    make_multi_index(0,
+                                     0,
+                                     0,
+                                     -1 * cluster_length_reduce.At(I1).value * (NReduceIters - 1) *
+                                         CBlockTransferScalarPerVector_NWaveNPerXDL);
+                constexpr auto partial_acc_store_step_m =
+                    make_multi_index(0, cluster_length_reduce.At(I0), 0, 0);
+
+                StaticBuffer<AddressSpaceEnum::Vgpr,
+                             FloatAcc,
+                             CBlockTransferScalarPerVector_NWaveNPerXDL,
+                             true>
+                    parcial_acc_buf;
+                StaticBuffer<AddressSpaceEnum::Vgpr,
+                             FloatAcc,
+                             CBlockTransferScalarPerVector_NWaveNPerXDL,
+                             true>
+                    acc_buf;
+
+                // start to compute
+                auto reduction_idx = blockIdx.x - block_mapping.reduction_start_block_idx;
+                auto spatial_idx   = block_mapping.tile_to_spatial(reduction_idx, m, n);
+
+                workgroup_barrier wg_barrier(p_semaphore);
+
+                uint32_t tile_acc_offset_start =
+                    block_mapping.get_acc_buffer_offset_from_tile(reduction_idx);
+                uint32_t tile_acc_offset_end =
+                    block_mapping.get_acc_buffer_offset_from_tile(reduction_idx + 1);
+
+                auto acc_load = ThreadwiseTensorSliceTransfer_v2<
+                    FloatAcc,                                                // SrcData,
+                    FloatAcc,                                                // DstData,
+                    decltype(c_partial_acc_block_m_n),                       // SrcDesc,
+                    decltype(acc_thread_buf_load_desc),                      // DstDesc,
+                    Sequence<1, CBlockTransferScalarPerVector_NWaveNPerXDL>, // SliceLengths,
+                    Sequence<0, 1>,                                          // DimAccessOrder,
+                    1,                                                       // SrcVectorDim,
+                    CBlockTransferScalarPerVector_NWaveNPerXDL,              // SrcScalarPerVector,
+                    1,    // SrcScalarStrideInVector,
+                    false // SrcResetCoordinateAfterRun,
+                    >{c_partial_acc_block_m_n,
+                      make_multi_index(thread_m_cluster_id,
+                                       thread_n_cluster_id *
+                                           CBlockTransferScalarPerVector_NWaveNPerXDL)};
+
+                auto acc_store = ThreadwiseTensorSliceTransfer_v1r3<
+                    FloatAcc,                                                // SrcData,
+                    FloatC,                                                  // DstData,
+                    decltype(acc_thread_buf_store_desc),                     // SrcDesc,
+                    decltype(c_grid_desc_mblock_mperblock_nblock_nperblock), // DstDesc,
+                    CElementwiseOperation, // ElementwiseOperation,
+                    Sequence<1, 1, 1, CBlockTransferScalarPerVector_NWaveNPerXDL>, // SliceLengths,
+                    Sequence<0, 1, 2, 3>,                       // DimAccessOrder,
+                    3,                                          // DstVectorDim,
+                    CBlockTransferScalarPerVector_NWaveNPerXDL, // DstScalarPerVector,
+                    InMemoryDataOperationEnum::Set, // InMemoryDataOperationEnum DstInMemOp,
+                    1,                              // DstScalarStrideInVector,
+                    false                           // DstResetCoordinateAfterRun,
+                    >{c_grid_desc_mblock_mperblock_nblock_nperblock,
+                      make_multi_index(__builtin_amdgcn_readfirstlane(spatial_idx[I0]),
+                                       thread_m_cluster_id,
+                                       __builtin_amdgcn_readfirstlane(spatial_idx[I1]),
+                                       thread_n_cluster_id *
+                                           CBlockTransferScalarPerVector_NWaveNPerXDL),
+                      CElementwiseOperation{}};
+
+                // block synchronization
+                wg_barrier.wait_eq(reduction_idx, tile_acc_offset_end - tile_acc_offset_start);
+
+#if 0
+                if(threadIdx.x == 0) {
+                    printf("bid:%d, rid:%d, os:%d,%d, spatial:%d,%d\n", static_cast<int>(blockIdx.x),
+                        reduction_idx, __builtin_amdgcn_readfirstlane(tile_acc_offset_start), __builtin_amdgcn_readfirstlane(tile_acc_offset_end),
+                        __builtin_amdgcn_readfirstlane(spatial_idx[I0]),
+                        __builtin_amdgcn_readfirstlane(spatial_idx[I1]));
+                }
+#endif
+
+                using Accumulation = ck::detail::
+                    AccumulateWithNanCheck<false /*PropagateNan*/, reduce::Add, FloatAcc>;
+
+                for(int i_m = 0; i_m < MReduceIters; i_m++)
+                {
+                    static_for<0, NReduceIters, 1>{}([&](auto i_n_reduce) {
+                        acc_buf.Clear();
+                        for(auto i = tile_acc_offset_start; i < tile_acc_offset_end; i++)
+                        {
+                            auto c_partial_acc_buf =
+                                make_dynamic_buffer<AddressSpaceEnum::Global,
+                                                    AmdBufferCoherenceEnum::GLC>(
+                                    reinterpret_cast<FloatAcc*>(p_workspace) +
+                                        i * c_partial_acc_block_m_n.GetElementSpaceSize(),
+                                    c_partial_acc_block_m_n.GetElementSpaceSize());
+
+                            acc_load.Run(c_partial_acc_block_m_n,
+                                         c_partial_acc_buf,
+                                         acc_thread_buf_load_desc,
+                                         make_tuple(I0, I0),
+                                         parcial_acc_buf);
+
+                            static_for<0, CBlockTransferScalarPerVector_NWaveNPerXDL, 1>{}(
+                                [&](auto i_vec) {
+                                    constexpr auto offset =
+                                        acc_thread_buf_load_desc.CalculateOffset(
+                                            make_tuple(0, i_vec));
+                                    Accumulation::Calculate(acc_buf(Number<offset>{}),
+                                                            parcial_acc_buf[Number<offset>{}]);
+                                });
+                        }
+
+                        if(thread_n_cluster_id * CBlockTransferScalarPerVector_NWaveNPerXDL <
+                           NPerBlock)
+                        {
+                            acc_store.Run(acc_thread_buf_store_desc,
+                                          make_tuple(I0, I0, I0, I0),
+                                          acc_buf,
+                                          c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                          c_grid_buf);
+                        }
+                        if constexpr(NReduceIters != 1)
+                        {
+                            if constexpr(i_n_reduce != (NReduceIters - 1))
+                            {
+                                acc_load.MoveSrcSliceWindow(c_partial_acc_block_m_n,
+                                                            partial_acc_load_step_n);
+                                acc_store.MoveDstSliceWindow(
+                                    c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                    partial_acc_store_step_n);
+                            }
+                            else
+                            {
+                                acc_load.MoveSrcSliceWindow(c_partial_acc_block_m_n,
+                                                            partial_acc_load_step_n_reverse);
+                                acc_store.MoveDstSliceWindow(
+                                    c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                    partial_acc_store_step_n_reverse);
+                            }
+                        }
+                    });
+                    {
+                        acc_load.MoveSrcSliceWindow(c_partial_acc_block_m_n,
+                                                    partial_acc_load_step_m);
+                        acc_store.MoveDstSliceWindow(c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                     partial_acc_store_step_m);
+                    }
+                }
+                return;
+            }
+        }
+
+        // offset for last acc buffer of this block
+        uint32_t block_acc_offset =
+            (block_mapping.get_acc_buffer_offset_from_block(block_idx + 1) - 1) * MPerBlock *
+            NPerBlock;
+
+        while(true)
+        {
+            uint32_t current_iter_length = __builtin_amdgcn_readfirstlane(
+                block_mapping.get_current_iter_length(iter_start, iter_end, total_iter_length));
+            uint32_t tile_idx, iter_offset;
+            block_mapping.get_tile_idx_with_offset(iter_end - 1, tile_idx, iter_offset);
+            iter_offset = __builtin_amdgcn_readfirstlane(iter_offset - current_iter_length + 1);
+            auto spatial_idx = block_mapping.tile_to_spatial(tile_idx, m, n);
+
+            const index_t m_block_data_idx_on_grid =
+                __builtin_amdgcn_readfirstlane(spatial_idx[I0] * MPerBlock);
+
+            const index_t n_block_data_idx_on_grid =
+                __builtin_amdgcn_readfirstlane(spatial_idx[I1] * NPerBlock);
+
+            const index_t k0_block_data_idx_on_grid =
+                __builtin_amdgcn_readfirstlane(iter_offset * K0PerBlock);
+
+            // A matrix blockwise copy
+            auto a_blockwise_copy =
+                ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                    AElementwiseOperation,
+                                                    ck::tensor_operation::element_wise::PassThrough,
+                                                    InMemoryDataOperationEnum::Set,
+                                                    Sequence<K0PerBlock, MPerBlock, K1>,
+                                                    ABlockTransferThreadClusterLengths_K0_M_K1,
+                                                    ABlockTransferThreadClusterArrangeOrder,
+                                                    FloatAB,
+                                                    FloatAB,
+                                                    decltype(a_k0_m_k1_grid_desc),
+                                                    decltype(a_block_desc_k0_m_k1),
+                                                    ABlockTransferSrcAccessOrder,
+                                                    Sequence<1, 0, 2>,
+                                                    ABlockTransferSrcVectorDim,
+                                                    2,
+                                                    ABlockTransferSrcScalarPerVector,
+                                                    ABlockTransferDstScalarPerVector_K1,
+                                                    1,
+                                                    1,
+                                                    AThreadTransferSrcResetCoordinateAfterRun,
+                                                    true>(
+                    a_k0_m_k1_grid_desc,
+                    make_multi_index(k0_block_data_idx_on_grid, m_block_data_idx_on_grid, 0),
+                    a_element_op,
+                    a_block_desc_k0_m_k1,
+                    make_multi_index(0, 0, 0),
+                    ck::tensor_operation::element_wise::PassThrough{});
+
+            // B matrix blockwise copy
+            auto b_blockwise_copy =
+                ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                    BElementwiseOperation,
+                                                    ck::tensor_operation::element_wise::PassThrough,
+                                                    InMemoryDataOperationEnum::Set,
+                                                    Sequence<K0PerBlock, NPerBlock, K1>,
+                                                    BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                    BBlockTransferThreadClusterArrangeOrder,
+                                                    FloatAB,
+                                                    FloatAB,
+                                                    decltype(b_k0_n_k1_grid_desc),
+                                                    decltype(b_block_desc_k0_n_k1),
+                                                    BBlockTransferSrcAccessOrder,
+                                                    Sequence<1, 0, 2>,
+                                                    BBlockTransferSrcVectorDim,
+                                                    2,
+                                                    BBlockTransferSrcScalarPerVector,
+                                                    BBlockTransferDstScalarPerVector_K1,
+                                                    1,
+                                                    1,
+                                                    BThreadTransferSrcResetCoordinateAfterRun,
+                                                    true>(
+                    b_k0_n_k1_grid_desc,
+                    make_multi_index(k0_block_data_idx_on_grid, n_block_data_idx_on_grid, 0),
+                    b_element_op,
+                    b_block_desc_k0_n_k1,
+                    make_multi_index(0, 0, 0),
+                    ck::tensor_operation::element_wise::PassThrough{});
+
+            const index_t num_k_block_main_loop = current_iter_length;
+
+            gridwise_gemm_pipeline.Run(a_k0_m_k1_grid_desc,
+                                       a_block_desc_k0_m_k1,
+                                       a_blockwise_copy,
+                                       a_grid_buf,
+                                       a_block_buf,
+                                       a_block_slice_copy_step,
+                                       b_k0_n_k1_grid_desc,
+                                       b_block_desc_k0_n_k1,
+                                       b_blockwise_copy,
+                                       b_grid_buf,
+                                       b_block_buf,
+                                       b_block_slice_copy_step,
+                                       blockwise_gemm,
+                                       c_thread_buf,
+                                       num_k_block_main_loop);
+
+            // output: register to global memory
+            {
+                constexpr index_t MWave = MPerBlock / (MRepeat * MPerXDL);
+                constexpr index_t NWave = NPerBlock / (NRepeat * NPerXDL);
+
+                constexpr auto c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc =
+                    blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+                constexpr auto c_m0_n0_m1_n1_m2_m3_m4_n2_thread_desc =
+                    blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+                constexpr auto M0 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I0);
+                constexpr auto N0 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I1);
+                constexpr auto M1 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I2);
+                constexpr auto N1 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I3);
+                constexpr auto M2 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I4);
+                constexpr auto M3 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I5);
+                constexpr auto M4 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I6);
+                constexpr auto N2 = c_m0_n0_m1_n1_m2_m3_m4_n2_block_desc.GetLength(I7);
+
+                constexpr auto c_block_desc_mblock_mpershuffle_nblock_npershuffle =
+                    GetCBlockDescriptor_MBlock_MPerShuffle_NBlock_NPerShuffle();
+
+                constexpr auto c_block_desc_mshuffle_mpershuffle_nshuffle_npershuffle =
+                    GetCBlockDescriptor_MShuffleRepeat_MPerShuffle_NShuffleRepeat_NPerShuffle();
+
+                auto c_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                    reinterpret_cast<FloatCShuffle*>(p_shared_block),
+                    c_block_desc_mblock_mpershuffle_nblock_npershuffle.GetElementSpaceSize());
+
+                auto c_partial_acc_buf =
+                    make_dynamic_buffer<AddressSpaceEnum::Global, AmdBufferCoherenceEnum::GLC>(
+                        reinterpret_cast<FloatAcc*>(p_workspace) + block_acc_offset,
+                        c_block_desc_mshuffle_mpershuffle_nshuffle_npershuffle
+                            .GetElementSpaceSize());
+
+                constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2 = transform_tensor_descriptor(
+                    c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                    make_tuple(make_freeze_transform(I0), // freeze mblock
+                               make_unmerge_transform(
+                                   make_tuple(CShuffleMRepeatPerShuffle,
+                                              M1,
+                                              M2,
+                                              M3,
+                                              M4)),       // M1 = MWave, M2 * M3 * M4 = MPerXDL
+                               make_freeze_transform(I0), // freeze nblock
+                               make_unmerge_transform(
+                                   make_tuple(CShuffleNRepeatPerShuffle,
+                                              N1,
+                                              N2))), // M1 = MWave, M2 * M3 * M4 = MPerXDL
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<>{},
+                               Sequence<0, 2, 4, 5, 6>{},
+                               Sequence<>{},
+                               Sequence<1, 3, 7>{}));
+
+                // calculate origin of thread output tensor on global memory
+                //     blockwise GEMM c matrix starting index
+                const auto c_thread_mtx_on_block =
+                    blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+                const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+                const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+                const auto m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor =
+                    make_single_stage_tensor_adaptor(
+                        make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
+                        make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                        make_tuple(Sequence<0>{}));
+
+                const auto m_thread_data_on_block_idx =
+                    m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
+                        make_multi_index(m_thread_data_on_block));
+
+                const auto n_thread_data_on_block_to_n0_n1_n2_adaptor =
+                    make_single_stage_tensor_adaptor(
+                        make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                        make_tuple(Sequence<0, 1, 2>{}),
+                        make_tuple(Sequence<0>{}));
+
+                const auto n_thread_data_on_block_idx =
+                    n_thread_data_on_block_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                        make_multi_index(n_thread_data_on_block));
+
+                // VGPR to LDS
+                auto c_thread_copy_vgpr_to_lds = ThreadwiseTensorSliceTransfer_v1r3<
+                    FloatAcc,
+                    FloatCShuffle,
+                    decltype(c_m0_n0_m1_n1_m2_m3_m4_n2_thread_desc),
+                    decltype(c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                    ck::tensor_operation::element_wise::PassThrough,
+                    Sequence<CShuffleMRepeatPerShuffle,
+                             CShuffleNRepeatPerShuffle,
+                             I1,
+                             I1,
+                             M2,
+                             I1,
+                             M4,
+                             I1>,
+                    Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                    7,
+                    1,
+                    InMemoryDataOperationEnum::Set,
+                    1,
+                    true>{c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                          make_multi_index(0,
+                                           0,
+                                           m_thread_data_on_block_idx[I1],
+                                           n_thread_data_on_block_idx[I1],
+                                           m_thread_data_on_block_idx[I2],
+                                           m_thread_data_on_block_idx[I3],
+                                           m_thread_data_on_block_idx[I4],
+                                           n_thread_data_on_block_idx[I2]),
+                          ck::tensor_operation::element_wise::PassThrough{}};
+
+                // LDS to global
+                auto c_block_copy_lds_to_global = ThreadGroupTensorSliceTransfer_v6r1r2<
+                    ThisThreadBlock,       // index_t BlockSize,
+                    CElementwiseOperation, // ElementwiseOperation,
+                                           // InMemoryDataOperationEnum::Set, // DstInMemOp,
+                    Sequence<1,
+                             CShuffleMRepeatPerShuffle * MWave * MPerXDL,
+                             1,
+                             CShuffleNRepeatPerShuffle * NWave * NPerXDL>, // BlockSliceLengths,
+                    CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                    Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                    FloatCShuffle,        // typename SrcData,
+                    FloatC,               // typename DstData,
+                    decltype(c_block_desc_mblock_mpershuffle_nblock_npershuffle),
+                    decltype(c_grid_desc_mblock_mperblock_nblock_nperblock),
+                    Sequence<0, 1, 2, 3>,                       // typename DimAccessOrder,
+                    3,                                          // index_t VectorDim,
+                    CBlockTransferScalarPerVector_NWaveNPerXDL, // index_t ScalarPerVector,
+                    false, // bool ThreadTransferSrcResetCoordinateAfterRun,
+                    false> // bool ThreadTransferDstResetCoordinateAfterRun
+                    {c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                     make_multi_index(0, 0, 0, 0),
+                     c_grid_desc_mblock_mperblock_nblock_nperblock,
+                     make_multi_index(__builtin_amdgcn_readfirstlane(spatial_idx[I0]),
+                                      0,
+                                      __builtin_amdgcn_readfirstlane(spatial_idx[I1]),
+                                      0),
+                     c_element_op};
+
+                // LDS to global partial acc
+                auto c_block_copy_lds_to_partial_acc = ThreadGroupTensorSliceTransfer_v6r1r2<
+                    ThisThreadBlock,       // index_t BlockSize,
+                    CElementwiseOperation, // ElementwiseOperation,
+                                           // InMemoryDataOperationEnum::Set, // DstInMemOp,
+                    Sequence<1,
+                             CShuffleMRepeatPerShuffle * MWave * MPerXDL,
+                             1,
+                             CShuffleNRepeatPerShuffle * NWave * NPerXDL>, // BlockSliceLengths,
+                    CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                    Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                    FloatCShuffle,        // typename SrcData,
+                    FloatCShuffle,        // typename DstData,
+                    decltype(c_block_desc_mblock_mpershuffle_nblock_npershuffle),
+                    decltype(c_block_desc_mshuffle_mpershuffle_nshuffle_npershuffle),
+                    Sequence<0, 1, 2, 3>,                       // typename DimAccessOrder,
+                    3,                                          // index_t VectorDim,
+                    CBlockTransferScalarPerVector_NWaveNPerXDL, // index_t ScalarPerVector,
+                    false, // bool ThreadTransferSrcResetCoordinateAfterRun, => need to be false,
+                           // othre wise has scratch
+                    false> // bool ThreadTransferDstResetCoordinateAfterRun, => need to be false,
+                           // othre wise has scratch
+                    {c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                     make_multi_index(0, 0, 0, 0),
+                     c_block_desc_mshuffle_mpershuffle_nshuffle_npershuffle,
+                     make_multi_index(0, 0, 0, 0),
+                     c_element_op};
+
+                constexpr auto mxdlperwave_forward_step =
+                    make_multi_index(0, CShuffleMRepeatPerShuffle * MWave * MPerXDL, 0, 0);
+                constexpr auto nxdlperwave_forward_step =
+                    make_multi_index(0, 0, 0, CShuffleNRepeatPerShuffle * NWave * NPerXDL);
+                constexpr auto nxdlperwave_backward_step =
+                    make_multi_index(0, 0, 0, -CShuffleNRepeatPerShuffle * NWave * NPerXDL);
+
+                static_for<0, MRepeat, CShuffleMRepeatPerShuffle>{}([&](auto mxdlperwave_iter) {
+                    constexpr auto mxdlperwave = mxdlperwave_iter;
+
+                    static_for<0, NRepeat, CShuffleNRepeatPerShuffle>{}([&](auto nxdlperwave_iter) {
+                        constexpr bool nxdlperwave_forward_sweep =
+                            (mxdlperwave % (2 * CShuffleMRepeatPerShuffle) == 0);
+
+                        constexpr index_t nxdlperwave_value =
+                            nxdlperwave_forward_sweep
+                                ? nxdlperwave_iter
+                                : (NRepeat - nxdlperwave_iter - CShuffleNRepeatPerShuffle);
+
+                        constexpr auto nxdlperwave = Number<nxdlperwave_value>{};
+
+                        // make sure it's safe to do ds_write
+                        block_sync_lds();
+
+                        // VGPR to LDS
+                        c_thread_copy_vgpr_to_lds.Run(
+                            c_m0_n0_m1_n1_m2_m3_m4_n2_thread_desc,
+                            make_tuple(mxdlperwave, nxdlperwave, I0, I0, I0, I0, I0, I0),
+                            c_thread_buf,
+                            c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                            c_block_buf);
+
+                        // make sure it's safe to do ds_read
+                        block_sync_lds();
+
+                        c_block_copy_lds_to_global.SetSrcSliceOrigin(
+                            c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                            make_tuple(0, 0, 0, 0));
+
+                        // LDS to global
+                        if(is_dp_block)
+                            c_block_copy_lds_to_global.template Run<decltype(c_block_buf),
+                                                                    decltype(c_grid_buf),
+                                                                    InMemoryDataOperationEnum::Set>(
+                                c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                                c_block_buf,
+                                c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                c_grid_buf);
+                        else if(is_sk_block)
+                        {
+                            if constexpr(Block2CTileMap::ReductionStrategy ==
+                                         StreamKReductionStrategy::Reduction)
+                            {
+                                // constexpr offset
+                                c_block_copy_lds_to_partial_acc.SetSrcSliceOrigin(
+                                    c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                                    make_tuple(0, 0, 0, 0));
+
+                                c_block_copy_lds_to_partial_acc.SetDstSliceOrigin(
+                                    c_block_desc_mshuffle_mpershuffle_nshuffle_npershuffle,
+                                    make_tuple(mxdlperwave.value, 0, nxdlperwave.value, 0));
+
+                                c_block_copy_lds_to_partial_acc
+                                    .template Run<decltype(c_block_buf),
+                                                  decltype(c_partial_acc_buf),
+                                                  InMemoryDataOperationEnum::Set>(
+                                        c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                                        c_block_buf,
+                                        c_block_desc_mshuffle_mpershuffle_nshuffle_npershuffle,
+                                        c_partial_acc_buf);
+                            }
+                            else if constexpr(Block2CTileMap::ReductionStrategy ==
+                                              StreamKReductionStrategy::Atomic)
+                            {
+                                c_block_copy_lds_to_global
+                                    .template Run<decltype(c_block_buf),
+                                                  decltype(c_grid_buf),
+                                                  InMemoryDataOperationEnum::AtomicAdd>(
+                                        c_block_desc_mblock_mpershuffle_nblock_npershuffle,
+                                        c_block_buf,
+                                        c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                        c_grid_buf);
+                            }
+                        }
+
+                        // move on nxdlperwave dimension
+                        if constexpr(nxdlperwave_forward_sweep &&
+                                     (nxdlperwave < NRepeat - CShuffleNRepeatPerShuffle))
+                        {
+                            c_block_copy_lds_to_global.MoveDstSliceWindow(
+                                c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                nxdlperwave_forward_step);
+                        }
+                        else if constexpr((!nxdlperwave_forward_sweep) && (nxdlperwave > 0))
+                        {
+                            c_block_copy_lds_to_global.MoveDstSliceWindow(
+                                c_grid_desc_mblock_mperblock_nblock_nperblock,
+                                nxdlperwave_backward_step);
+                        }
+                    });
+
+                    // move on mxdlperwave dimension
+                    if constexpr(mxdlperwave < MRepeat - CShuffleMRepeatPerShuffle)
+                    {
+                        c_block_copy_lds_to_global.MoveDstSliceWindow(
+                            c_grid_desc_mblock_mperblock_nblock_nperblock,
+                            mxdlperwave_forward_step);
+                    }
+                });
+
+                if constexpr(Block2CTileMap::ReductionStrategy ==
+                             StreamKReductionStrategy::Reduction)
+                {
+                    if(is_sk_block)
+                    {
+                        // increase the counter for this tile
+                        workgroup_barrier wg_barrier(p_semaphore);
+                        wg_barrier.inc(tile_idx);
+                    }
+                }
+            }
+
+            // exit condition
+            iter_end -= current_iter_length;
+            if(iter_end <= iter_start)
+                break;
+
+            if constexpr(Block2CTileMap::ReductionStrategy == StreamKReductionStrategy::Reduction)
+            {
+                block_acc_offset -= MPerBlock * NPerBlock;
+            }
+            // make sure next loop LDS is ready for use
+            block_sync_lds();
+        }
+    }
+
+    template <typename Layout>
+    struct LStr
+    {
+        static std::string Get() { return ""; }
+    };
+
+    template <>
+    struct LStr<ck::tensor_layout::gemm::RowMajor>
+    {
+        static std::string Get() { return "R"; }
+    };
+
+    template <>
+    struct LStr<ck::tensor_layout::gemm::ColumnMajor>
+    {
+        static std::string Get() { return "C"; }
+    };
+
+    static std::string GetTypeString()
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "GemmXdlStreamK_"
+            << std::string(ALayout::name)[0]
+            << std::string(BLayout::name)[0]
+            << std::string(CLayout::name)[0]
+            << "_"
+            << "B" << BlockSize << "_"
+            << "Vec" << ABlockTransferSrcScalarPerVector << "x"
+            << BBlockTransferSrcScalarPerVector << "x"
+            << CBlockTransferScalarPerVector_NWaveNPerXDL << "_"
+            << MPerBlock << "x"
+            << NPerBlock << "x"
+            << K0PerBlock << "x"
+            << K1 ;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1r2.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1r2.hpp
new file mode 100644
index 0000000000..88ed217547
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1r2.hpp
@@ -0,0 +1,213 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_description/tensor_space_filling_curve.hpp"
+
+namespace ck {
+
+// Do following things to avoid "alloca" in LLVM-IR, which would cause scratch memory
+// and sometimes useless instructions:
+//   1. Don't save a reference to tensor descriptor in class, pass in tensor descriptor as argument
+//   instead
+//   2. Don't construct a new tensor coordinate everytime when using it, update and reuse the same
+//   tensor coordinate instead
+//   3. Don't use a pointer to VGPR buffer, use vector instead
+
+// Assume:
+//   1. src_desc and dst_desc are not known at compile-time
+//   2. SrcBuffer and DstBuffer are DynamicBuffer
+//   3. src_slice_origin and dst_slice_origin are not known at compile-time,
+template <typename SrcData,
+          typename DstData,
+          typename SrcDesc,
+          typename DstDesc,
+          typename ElementwiseOperation,
+          typename SliceLengths,
+          typename DimAccessOrder,
+          index_t VectorDim,
+          index_t ScalarPerVector,
+          bool SrcResetCoordinateAfterRun,
+          bool DstResetCoordinateAfterRun>
+struct ThreadwiseTensorSliceTransfer_v6r1r2
+{
+    static constexpr index_t nDim = SliceLengths::Size();
+
+    using Index = MultiIndex<nDim>;
+
+    using SrcCoord = decltype(make_tensor_coordinate(SrcDesc{}, Index{}));
+    using DstCoord = decltype(make_tensor_coordinate(DstDesc{}, Index{}));
+
+    static constexpr auto I0 = Number<0>{};
+
+    __device__ constexpr ThreadwiseTensorSliceTransfer_v6r1r2(
+        const SrcDesc& src_desc,
+        const Index& src_slice_origin,
+        const DstDesc& dst_desc,
+        const Index& dst_slice_origin,
+        const ElementwiseOperation& element_op)
+        : src_coord_(make_tensor_coordinate(src_desc, src_slice_origin)),
+          dst_coord_(make_tensor_coordinate(dst_desc, dst_slice_origin)),
+          element_op_(element_op)
+    {
+        static_assert(SliceLengths::At(Number<VectorDim>{}) % ScalarPerVector == 0,
+                      "wrong! cannot evenly divide");
+    }
+
+    __device__ void SetSrcSliceOrigin(const SrcDesc& src_desc, const Index& src_slice_origin_idx)
+    {
+        src_coord_ = make_tensor_coordinate(src_desc, src_slice_origin_idx);
+    }
+
+    __device__ void SetDstSliceOrigin(const DstDesc& dst_desc, const Index& dst_slice_origin_idx)
+    {
+        dst_coord_ = make_tensor_coordinate(dst_desc, dst_slice_origin_idx);
+    }
+
+    template <typename SrcBuffer, typename DstBuffer, InMemoryDataOperationEnum DstInMemOp>
+    __device__ void Run(const SrcDesc& src_desc,
+                        const SrcBuffer& src_buf,
+                        const DstDesc& dst_desc,
+                        DstBuffer& dst_buf)
+    {
+        // scalar per access on each dim
+        // TODO: don't use lambda_scalar_per_access
+        constexpr auto scalar_per_access = generate_sequence(
+            detail::lambda_scalar_per_access<VectorDim, ScalarPerVector>{}, Number<nDim>{});
+
+        using SpaceFillingCurve = SpaceFillingCurve<SliceLengths,
+                                                    DimAccessOrder,
+                                                    remove_cv_t<decltype(scalar_per_access)>>;
+
+        // loop over space-filling curve
+        constexpr auto num_access = SpaceFillingCurve::GetNumOfAccess();
+
+        static_for<0, num_access, 1>{}([&](auto idx_1d) {
+            using src_vector_type = vector_type_maker_t<SrcData, ScalarPerVector>;
+            using src_vector_t    = typename src_vector_type::type;
+
+            using dst_vector_type = vector_type_maker_t<DstData, ScalarPerVector>;
+            using dst_vector_t    = typename dst_vector_type::type;
+
+            const bool is_src_valid =
+                coordinate_has_valid_offset_assuming_visible_index_is_valid(src_desc, src_coord_);
+
+            // copy data from src_buf into src_vector_container
+            auto src_vector_container = src_vector_type{
+                src_buf.template Get<src_vector_t>(src_coord_.GetOffset(), is_src_valid)};
+
+            auto dst_vector_container = dst_vector_type{};
+
+            // apply pointwise operation
+            static_for<0, ScalarPerVector, 1>{}([&](auto i) {
+                SrcData v;
+
+                // apply element-wise operation
+                element_op_(v, src_vector_container.template AsType<SrcData>()[i]);
+
+                // apply type convert
+                dst_vector_container.template AsType<DstData>()(i) = type_convert<DstData>(v);
+            });
+
+            const bool is_dst_valid =
+                coordinate_has_valid_offset_assuming_visible_index_is_valid(dst_desc, dst_coord_);
+
+            // copy data from dst_vector into dst_buf
+            dst_buf.template Update<DstInMemOp, dst_vector_t>(
+                dst_coord_.GetOffset(),
+                is_dst_valid,
+                dst_vector_container.template AsType<dst_vector_t>()[I0]);
+
+            // move coordinate
+            if constexpr(idx_1d.value != num_access - 1)
+            {
+                constexpr auto forward_step = SpaceFillingCurve::GetForwardStep(idx_1d);
+                move_tensor_coordinate(
+                    src_desc, src_coord_, make_tensor_coordinate_step(src_desc, forward_step));
+                move_tensor_coordinate(
+                    dst_desc, dst_coord_, make_tensor_coordinate_step(dst_desc, forward_step));
+            }
+        });
+
+        // move coordinate back to slice origin (or not)
+        if constexpr(SrcResetCoordinateAfterRun)
+        {
+            const auto src_reset_step =
+                make_tensor_coordinate_step(src_desc, GetCoordinateResetStep());
+
+            move_tensor_coordinate(src_desc, src_coord_, src_reset_step);
+        }
+
+        if constexpr(DstResetCoordinateAfterRun)
+        {
+            const auto dst_reset_step =
+                make_tensor_coordinate_step(dst_desc, GetCoordinateResetStep());
+
+            move_tensor_coordinate(dst_desc, dst_coord_, dst_reset_step);
+        }
+    }
+
+    __device__ static constexpr auto GetCoordinateResetStep()
+    {
+        constexpr auto scalar_per_access = generate_sequence(
+            detail::lambda_scalar_per_access<VectorDim, ScalarPerVector>{}, Number<nDim>{});
+
+        using SpaceFillingCurve = SpaceFillingCurve<SliceLengths,
+                                                    DimAccessOrder,
+                                                    remove_cv_t<decltype(scalar_per_access)>>;
+
+        constexpr auto num_access = SpaceFillingCurve::GetNumOfAccess();
+        if constexpr(num_access == 0)
+        {
+            return typename SpaceFillingCurve::Index{};
+        }
+        else
+        {
+            constexpr auto reset_step =
+                SpaceFillingCurve::GetStepBetween(Number<num_access - 1>{}, Number<0>{});
+
+            return reset_step;
+        }
+    }
+
+    // src_slice_origin_step_idx need to be known at compile-time, for performance reason
+    __device__ void MoveSrcSliceWindow(const SrcDesc& src_desc,
+                                       const Index& src_slice_origin_step_idx)
+    {
+        // if src coord was not reset by RunRead(), then need to adjust the step here
+        const auto adjusted_step_idx = SrcResetCoordinateAfterRun
+                                           ? src_slice_origin_step_idx
+                                           : src_slice_origin_step_idx + GetCoordinateResetStep();
+
+        // is it OK to construct a new step every time?
+        const auto adjusted_step = make_tensor_coordinate_step(src_desc, adjusted_step_idx);
+
+        move_tensor_coordinate(src_desc, src_coord_, adjusted_step);
+    }
+
+    // dst_slice_origin_step_idx need to be known at compile-time, for performance reason
+    __device__ void MoveDstSliceWindow(const DstDesc& dst_desc,
+                                       const Index& dst_slice_origin_step_idx)
+    {
+        // if dst coord was not reset by Run(), then need to adjust the step here
+        const auto adjusted_step_idx = DstResetCoordinateAfterRun
+                                           ? dst_slice_origin_step_idx
+                                           : dst_slice_origin_step_idx + GetCoordinateResetStep();
+
+        // is it OK to construct a new step every time?
+        const auto adjusted_step = make_tensor_coordinate_step(dst_desc, adjusted_step_idx);
+
+        move_tensor_coordinate(dst_desc, dst_coord_, adjusted_step);
+    }
+
+    private:
+    SrcCoord src_coord_;
+    DstCoord dst_coord_;
+    const ElementwiseOperation element_op_;
+};
+
+} // namespace ck
diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index ea231154ea..897cb4f249 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -629,7 +629,7 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
 {
     static_assert(
         (is_same<T, double>::value && (N == 1 || N == 2)) ||
-            (is_same<T, float>::value && (N == 1 || N == 2 || N == 4)) ||
+            (is_same<T, float>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
             (is_same<T, half_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
             (is_same<T, bhalf_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
             (is_same<T, int32_t>::value && (N == 1 || N == 2 || N == 4)) ||
@@ -682,6 +682,20 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                                 dst_wave_addr_offset,
                                                 static_cast<index_t>(coherence));
         }
+        else if constexpr(N == 8)
+        {
+            vector_type<float, 8> tmp{src_thread_data};
+            llvm_amdgcn_raw_buffer_store_fp32x4(tmp.AsType<float4_t>()[Number<0>{}],
+                                                dst_wave_buffer_resource,
+                                                dst_thread_addr_offset,
+                                                dst_wave_addr_offset,
+                                                static_cast<index_t>(coherence));
+            llvm_amdgcn_raw_buffer_store_fp32x4(tmp.AsType<float4_t>()[Number<1>{}],
+                                                dst_wave_buffer_resource,
+                                                dst_thread_addr_offset,
+                                                dst_wave_addr_offset + 4 * sizeof(float),
+                                                static_cast<index_t>(coherence));
+        }
     }
     else if constexpr(is_same<T, half_t>::value)
     {
diff --git a/include/ck/utility/magic_division.hpp b/include/ck/utility/magic_division.hpp
index f19030d4e9..1d1f914c66 100644
--- a/include/ck/utility/magic_division.hpp
+++ b/include/ck/utility/magic_division.hpp
@@ -157,4 +157,76 @@ struct MagicDivision
     }
 };
 
+struct MDiv
+{
+    // 1 dword -> 3 dword storage
+    uint32_t divisor;
+    uint32_t multiplier;
+    uint32_t shift; // TODO: 8 bit is enough
+
+    // prefer construct on host
+    __host__ __device__ MDiv(uint32_t divisor_) : divisor(divisor_)
+    {
+        auto tmp = MagicDivision::CalculateMagicNumbers(divisor_);
+
+        multiplier = tmp[Number<0>{}];
+        shift      = tmp[Number<1>{}];
+    }
+
+    __host__ __device__ MDiv() : divisor(0), multiplier(0), shift(0) {}
+
+    __host__ __device__ void update(uint32_t divisor_)
+    {
+        divisor  = divisor_;
+        auto tmp = MagicDivision::CalculateMagicNumbers(divisor_);
+
+        multiplier = tmp[Number<0>{}];
+        shift      = tmp[Number<1>{}];
+    }
+
+    __host__ __device__ uint32_t div(uint32_t dividend_) const
+    {
+        return MagicDivision::DoMagicDivision(dividend_, multiplier, shift);
+    }
+
+    __host__ __device__ void
+    divmod(uint32_t dividend_, uint32_t& quotient_, uint32_t& remainder_) const
+    {
+        quotient_  = div(dividend_);
+        remainder_ = dividend_ - (quotient_ * divisor);
+    }
+
+    __host__ __device__ uint32_t get() const { return divisor; }
+};
+
+struct MDiv2
+{
+    // 1 dword -> 2 dword storage, divisor need compute from runtime
+    uint32_t multiplier;
+    uint32_t shift; // TODO: 8 bit is enough
+
+    // prefer construct on host
+    __host__ __device__ MDiv2(uint32_t divisor_)
+    {
+        auto tmp = MagicDivision::CalculateMagicNumbers(divisor_);
+
+        multiplier = tmp[Number<0>{}];
+        shift      = tmp[Number<1>{}];
+    }
+
+    __host__ __device__ MDiv2() : multiplier(0), shift(0) {}
+
+    __host__ __device__ uint32_t div(uint32_t dividend_) const
+    {
+        return MagicDivision::DoMagicDivision(dividend_, multiplier, shift);
+    }
+
+    __host__ __device__ void
+    divmod(uint32_t dividend_, uint32_t divisor_, uint32_t& quotient_, uint32_t& remainder_) const
+    {
+        quotient_  = div(dividend_);
+        remainder_ = dividend_ - (quotient_ * divisor_);
+    }
+};
+
 } // namespace ck
diff --git a/include/ck/utility/math.hpp b/include/ck/utility/math.hpp
index 326b0e61ef..c5e967c8f4 100644
--- a/include/ck/utility/math.hpp
+++ b/include/ck/utility/math.hpp
@@ -240,5 +240,21 @@ struct less
     __host__ __device__ constexpr bool operator()(T x, T y) const { return x < y; }
 };
 
+template <index_t X>
+__host__ __device__ constexpr auto next_power_of_two()
+{
+    // TODO: X need to be 2 ~ 0x7fffffff. 0, 1, or larger than 0x7fffffff will compile fail
+    constexpr index_t Y = 1 << (32 - __builtin_clz(X - 1));
+    return Y;
+}
+
+template <index_t X>
+__host__ __device__ constexpr auto next_power_of_two(Number<X> x)
+{
+    // TODO: X need to be 2 ~ 0x7fffffff. 0, 1, or larger than 0x7fffffff will compile fail
+    constexpr index_t Y = 1 << (32 - __builtin_clz(x.value - 1));
+    return Number<Y>{};
+}
+
 } // namespace math
 } // namespace ck
diff --git a/include/ck/utility/workgroup_barrier.hpp b/include/ck/utility/workgroup_barrier.hpp
new file mode 100644
index 0000000000..ec9151fd1b
--- /dev/null
+++ b/include/ck/utility/workgroup_barrier.hpp
@@ -0,0 +1,73 @@
+#pragma once
+#include <hip/hip_runtime.h>
+#include <stdint.h>
+
+namespace ck {
+struct workgroup_barrier
+{
+    __device__ workgroup_barrier(uint32_t* ptr) : base_ptr(ptr) {}
+
+    __device__ uint32_t ld(uint32_t offset)
+    {
+#if 0
+        float d = llvm_amdgcn_raw_buffer_load_fp32(
+                        amdgcn_make_buffer_resource(base_ptr),
+                        0,
+                        offset,
+                        AMDGCN_BUFFER_GLC);
+        union cvt {
+            float f32;
+            uint32_t u32;
+        };
+        cvt x;
+        x.f32 = d;
+        return x.u32;
+#endif
+        return __atomic_load_n(base_ptr + offset, __ATOMIC_RELAXED);
+    }
+
+    __device__ void wait_eq(uint32_t offset, uint32_t value)
+    {
+        if(threadIdx.x == 0)
+        {
+            while(ld(offset) != value) {}
+        }
+        __syncthreads();
+    }
+
+    __device__ void wait_lt(uint32_t offset, uint32_t value)
+    {
+        if(threadIdx.x == 0)
+        {
+            while(ld(offset) < value) {}
+        }
+        __syncthreads();
+    }
+
+    __device__ void wait_set(uint32_t offset, uint32_t compare, uint32_t value)
+    {
+        if(threadIdx.x == 0)
+        {
+            while(atomicCAS(base_ptr + offset, compare, value) != compare) {}
+        }
+        __syncthreads();
+    }
+
+    // enter critical zoon, assume buffer is zero when launch kernel
+    __device__ void aquire(uint32_t offset) { wait_set(offset, 0, 1); }
+
+    // exit critical zoon, assume buffer is zero when launch kernel
+    __device__ void release(uint32_t offset) { wait_set(offset, 1, 0); }
+
+    __device__ void inc(uint32_t offset)
+    {
+        __syncthreads();
+        if(threadIdx.x == 0)
+        {
+            atomicAdd(base_ptr + offset, 1);
+        }
+    }
+
+    uint32_t* base_ptr;
+};
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
new file mode 100644
index 0000000000..c40b607cc2
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
@@ -0,0 +1,121 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_streamk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmStreamK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+template <typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmStreamK<
+    ALayout,
+    BLayout,
+    CLayout,
+    ADataType,
+    BDataType,
+    CDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough>>
+{
+    using DeviceOp = DeviceGemmStreamK<ALayout,
+                                       BLayout,
+                                       CLayout,
+                                       ADataType,
+                                       BDataType,
+                                       CDataType,
+                                       ck::tensor_operation::element_wise::PassThrough,
+                                       ck::tensor_operation::element_wise::PassThrough,
+                                       ck::tensor_operation::element_wise::PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+#if 0
+        if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
+                     is_same_v<CDataType, float>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                          is_same_v<CDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+            }
+        }
+#endif
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<CDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+        }
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/utility/device_memory.hpp b/library/include/ck/library/utility/device_memory.hpp
index 1c16ff5916..b2099acfaa 100644
--- a/library/include/ck/library/utility/device_memory.hpp
+++ b/library/include/ck/library/utility/device_memory.hpp
@@ -20,8 +20,9 @@ __global__ void set_buffer_value(T* p, T x, uint64_t buffer_element_size)
  */
 struct DeviceMem
 {
-    DeviceMem() = delete;
+    DeviceMem() : mpDeviceBuf(nullptr), mMemSize(0) {}
     DeviceMem(std::size_t mem_size);
+    void Realloc(std::size_t mem_size);
     void* GetDeviceBuffer() const;
     std::size_t GetBufferSize() const;
     void ToDevice(const void* p) const;
diff --git a/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
new file mode 100644
index 0000000000..8dd0112a6b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
@@ -0,0 +1,10 @@
+add_instance_library(device_gemm_streamk_instance
+   # device_gemm_xdl_streamk_f32_f32_f32_mk_kn_mn_instance.cpp
+   # device_gemm_xdl_streamk_f32_f32_f32_mk_nk_mn_instance.cpp
+   # device_gemm_xdl_streamk_f32_f32_f32_km_kn_mn_instance.cpp
+   # device_gemm_xdl_streamk_f32_f32_f32_km_nk_mn_instance.cpp
+   device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
+   # device_gemm_xdl_streamk_f16_f16_f16_mk_nk_mn_instance.cpp
+   # device_gemm_xdl_streamk_f16_f16_f16_km_kn_mn_instance.cpp
+   # device_gemm_xdl_streamk_f16_f16_f16_km_nk_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..7e058a895d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+// static constexpr auto GemmMNPadding =
+// ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //##################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|
+        //##################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
+        //##################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##################|     |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   192,     4,  8,   32,   32,    1,    3,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 48, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   192,    64,     4,  8,   32,   32,    3,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4,  8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    64,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    64,   128,     4,  8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,                   S<1, 16, 1, 4>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 32, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,   192,     4,  8,   32,   32,    1,    3,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 24, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   192,    32,     4,  8,   32,   32,    3,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    64,    32,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,                   S<1, 32, 1, 4>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,   128,     4,  8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               8>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,    32,     4,  8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,                   S<1, 32, 1, 4>,               8>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmStreamK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/utility/device_memory.cpp b/library/src/utility/device_memory.cpp
index 11166783e8..e054216dc9 100644
--- a/library/src/utility/device_memory.cpp
+++ b/library/src/utility/device_memory.cpp
@@ -10,20 +10,57 @@ DeviceMem::DeviceMem(std::size_t mem_size) : mMemSize(mem_size)
     hip_check_error(hipMalloc(static_cast<void**>(&mpDeviceBuf), mMemSize));
 }
 
+void DeviceMem::Realloc(std::size_t mem_size)
+{
+    if(mpDeviceBuf)
+    {
+        hip_check_error(hipFree(mpDeviceBuf));
+    }
+    mMemSize = mem_size;
+    hip_check_error(hipMalloc(static_cast<void**>(&mpDeviceBuf), mMemSize));
+}
+
 void* DeviceMem::GetDeviceBuffer() const { return mpDeviceBuf; }
 
 std::size_t DeviceMem::GetBufferSize() const { return mMemSize; }
 
 void DeviceMem::ToDevice(const void* p) const
 {
-    hip_check_error(hipMemcpy(mpDeviceBuf, const_cast<void*>(p), mMemSize, hipMemcpyHostToDevice));
+    if(mpDeviceBuf)
+    {
+        hip_check_error(
+            hipMemcpy(mpDeviceBuf, const_cast<void*>(p), mMemSize, hipMemcpyHostToDevice));
+    }
+    else
+    {
+        throw std::runtime_error("ToDevice with an empty pointer");
+    }
 }
 
 void DeviceMem::FromDevice(void* p) const
 {
-    hip_check_error(hipMemcpy(p, mpDeviceBuf, mMemSize, hipMemcpyDeviceToHost));
+    if(mpDeviceBuf)
+    {
+        hip_check_error(hipMemcpy(p, mpDeviceBuf, mMemSize, hipMemcpyDeviceToHost));
+    }
+    else
+    {
+        throw std::runtime_error("FromDevice with an empty pointer");
+    }
 }
 
-void DeviceMem::SetZero() const { hip_check_error(hipMemset(mpDeviceBuf, 0, mMemSize)); }
+void DeviceMem::SetZero() const
+{
+    if(mpDeviceBuf)
+    {
+        hip_check_error(hipMemset(mpDeviceBuf, 0, mMemSize));
+    }
+}
 
-DeviceMem::~DeviceMem() { hip_check_error(hipFree(mpDeviceBuf)); }
+DeviceMem::~DeviceMem()
+{
+    if(mpDeviceBuf)
+    {
+        hip_check_error(hipFree(mpDeviceBuf));
+    }
+}
diff --git a/profiler/include/profiler/profile_gemm_streamk_impl.hpp b/profiler/include/profiler/profile_gemm_streamk_impl.hpp
new file mode 100644
index 0000000000..b5d24367d4
--- /dev/null
+++ b/profiler/include/profiler/profile_gemm_streamk_impl.hpp
@@ -0,0 +1,265 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+#include <iostream>
+#include <typeinfo>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_streamk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout>
+bool profile_gemm_streamk_impl(int do_verification,
+                               int init_method,
+                               bool do_log,
+                               bool time_kernel,
+                               int M,
+                               int N,
+                               int K,
+                               int StrideA,
+                               int StrideB,
+                               int StrideC,
+                               uint32_t NumSKBlocks = 0xffffffff)
+{
+    bool pass = true;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(is_same<decltype(layout), tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<CDataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+    Tensor<CDataType> c_m_n_device_result(f_host_tensor_descriptor(M, N, StrideC, CLayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "c_m_n: " << c_m_n_device_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-3, 3});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+    }
+
+    using AElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using BElementOp = ck::tensor_operation::element_wise::PassThrough;
+    using CElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+    const auto c_element_op = CElementOp{};
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem c_device_buf(sizeof(CDataType) * c_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    c_device_buf.ToDevice(c_m_n_device_result.mData.data());
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmStreamK<ALayout,
+                                                                     BLayout,
+                                                                     CLayout,
+                                                                     ADataType,
+                                                                     BDataType,
+                                                                     CDataType,
+                                                                     AElementOp,
+                                                                     BElementOp,
+                                                                     CElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances, "
+              << (do_verification ? "with verification" : "without verification") << std::endl;
+
+    // Run reference GEMM
+    if(do_verification)
+    {
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                CElementOp>;
+
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument = ref_gemm.MakeArgument(
+            a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op, c_element_op);
+
+        ref_invoker.Run(ref_argument);
+    }
+
+    std::string best_op_name;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device GEMM instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                        static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                        static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                        M,
+                                        N,
+                                        K,
+                                        StrideA,
+                                        StrideB,
+                                        StrideC,
+                                        a_element_op,
+                                        b_element_op,
+                                        c_element_op,
+                                        NumSKBlocks);
+        DeviceMem workspace;
+        std::size_t workspace_size = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+        if(workspace_size != 0)
+        {
+            workspace.Realloc(workspace_size);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+        }
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init C to zero before profiling next kernel
+            c_device_buf.SetZero();
+
+            std::string op_name = op_ptr->GetTypeString();
+
+            float ave_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+
+            std::size_t num_btype =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+                pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_device: ", c_m_n_device_result.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    }
+
+    if constexpr(is_same<CDataType, float>::value)
+    {
+        std::cout << "Best Perf for datatype = f32";
+    }
+    else if constexpr(is_same<CDataType, half_t>::value)
+    {
+        std::cout << "Best Perf for datatype = f16";
+    }
+    else if constexpr(is_same<CDataType, bhalf_t>::value)
+    {
+        std::cout << "Best Perf for datatype = bf16";
+    }
+    else if constexpr(is_same<CDataType, int8_t>::value)
+    {
+        std::cout << "Best Perf for datatype = int8";
+    }
+
+    if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " ALayout =  RowMajor";
+    }
+    else if constexpr(is_same<ALayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " ALayout =  ColumnMajor";
+    }
+
+    if constexpr(is_same<BLayout, tensor_layout::gemm::RowMajor>::value)
+    {
+        std::cout << " BLayout =  RowMajor";
+    }
+    else if constexpr(is_same<BLayout, tensor_layout::gemm::ColumnMajor>::value)
+    {
+        std::cout << " BLayout =  ColumnMajor";
+    }
+
+    std::cout << " M = " << M << " N = " << N << " K = " << K << " StrideA = " << StrideA
+              << " StrideB = " << StrideB << " StrideC = " << StrideC << " : " << best_ave_time
+              << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec << " GB/s, "
+              << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index d057dccb40..27b7c1c891 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -3,6 +3,7 @@ set(PROFILER_SOURCES
     profiler.cpp
     profile_gemm.cpp
     profile_gemm_splitk.cpp
+    profile_gemm_streamk.cpp
     profile_gemm_bilinear.cpp
     profile_gemm_bias_add_reduce.cpp
     profile_gemm_add_add_fastgelu.cpp
@@ -48,6 +49,7 @@ target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
diff --git a/profiler/src/profile_gemm_streamk.cpp b/profiler/src/profile_gemm_streamk.cpp
new file mode 100644
index 0000000000..a0a49eb36d
--- /dev/null
+++ b/profiler/src/profile_gemm_streamk.cpp
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_gemm_streamk_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+enum struct GemmMatrixLayout
+{
+    MK_KN_MN, // 0
+    MK_NK_MN, // 1
+    KM_KN_MN, // 2
+    KM_NK_MN, // 3
+};
+
+enum struct GemmDataType
+{
+    F32_F32_F32,    // 0
+    F16_F16_F16,    // 1
+    BF16_BF16_BF16, // 2
+    INT8_INT8_INT8, // 3
+};
+
+#define OP_NAME "gemm_streamk"
+#define OP_DESC "StreamK GEMM"
+
+int profile_gemm_streamk(int argc, char* argv[])
+{
+    if(argc < 14)
+    {
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
+        printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
+        printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
+        printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
+        printf("                     3: A[k, m] * B[n, k] = C[m, n])\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=no, 1=yes)\n");
+        printf("arg8 to 13: M, N, K, StrideA, StrideB, StrideC\n");
+        printf("arg14: num_sk_blocks (optional)\n");
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<GemmDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<GemmMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const int M = std::stoi(argv[8]);
+    const int N = std::stoi(argv[9]);
+    const int K = std::stoi(argv[10]);
+
+    const int StrideA = std::stoi(argv[11]);
+    const int StrideB = std::stoi(argv[12]);
+    const int StrideC = std::stoi(argv[13]);
+    const uint32_t NumSKBlocks =
+        argc >= 15 ? static_cast<uint32_t>(std::stoul(std::string(argv[14]))) : 0xffffffff;
+
+    using F32 = float;
+    using F16 = ck::half_t;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    auto profile = [&](auto a_type,
+                       auto b_type,
+                       auto acc_type,
+                       auto c_type,
+                       auto a_layout,
+                       auto b_layout,
+                       auto c_layout) {
+        using ADataType   = decltype(a_type);
+        using BDataType   = decltype(b_type);
+        using AccDataType = decltype(acc_type);
+        using CDataType   = decltype(c_type);
+
+        using ALayout = decltype(a_layout);
+        using BLayout = decltype(b_layout);
+        using CLayout = decltype(c_layout);
+
+        const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
+        const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
+        const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
+
+        bool pass = ck::profiler::profile_gemm_streamk_impl<ADataType,
+                                                            BDataType,
+                                                            AccDataType,
+                                                            CDataType,
+                                                            ALayout,
+                                                            BLayout,
+                                                            CLayout>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            M,
+            N,
+            K,
+            (StrideA <= 0) ? DefaultStrideA : StrideA,
+            (StrideB <= 0) ? DefaultStrideB : StrideB,
+            (StrideC <= 0) ? DefaultStrideC : StrideC,
+            NumSKBlocks);
+
+        return pass ? 0 : 1;
+    };
+
+    if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F32{}, F32{}, F32{}, F32{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F32{}, F32{}, F32{}, F32{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F32{}, F32{}, F32{}, F32{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F32{}, F32{}, F32{}, F32{}, Col{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_gemm_streamk);
diff --git a/test/block_swizzle_test/block_swizzle_test.cpp b/test/block_swizzle_test/block_swizzle_test.cpp
new file mode 100644
index 0000000000..29e118c2ad
--- /dev/null
+++ b/test/block_swizzle_test/block_swizzle_test.cpp
@@ -0,0 +1,406 @@
+#include <stdio.h>
+#include <string>
+#include <algorithm>
+#include <vector>
+#include <limits>
+#include "simple_args.h"
+
+simple_args_t create_arg(int argc, char** argv)
+{
+    simple_args_t args;
+    args.insert("m", "1024", "matrix m")
+        .insert("n", "1024", "matrix n")
+        .insert("k", "1024", "matrix k")
+        .insert("m_per_block", "128", "m_per_block")
+        .insert("n_per_block", "128", "n_per_block")
+        .insert("k_per_block", "32", "k_per_block")
+        .insert("num_cu", "104", "num cu")
+        .insert("occupancy", "2", "occupancy")
+        .parse(argc, argv);
+    return args;
+}
+
+namespace impl {
+template <typename T>
+T integer_divide_ceil(T n, T d)
+{
+    return (n + d - 1) / d;
+}
+
+template <typename T>
+T min(T a, T b)
+{
+    return a > b ? b : a;
+}
+
+template <typename T>
+T max(T a, T b)
+{
+    return a > b ? a : b;
+}
+
+} // namespace impl
+
+struct block_dispatcher_t
+{
+    public:
+    uint32_t m_per_block;
+    uint32_t n_per_block;
+    uint32_t k_per_block;
+    uint32_t num_cu;
+    uint32_t occupancy;
+    uint32_t m;
+    uint32_t n;
+    uint32_t k;
+
+    //--------------------------------------
+
+    uint32_t sk_num_blocks;
+    uint32_t sk_num_big_blocks;
+    uint32_t sk_total_iters;
+
+    // uint32_t sk_num_blocks_per_tile;    // how many
+
+    uint32_t dp_start_block_idx;
+    uint32_t dp_iters_per_block;
+    uint32_t dp_num_blocks;
+
+    uint32_t k_iters_per_tile;
+    uint32_t k_iters_per_big_block;
+    //--------------------------------------
+
+    static constexpr uint32_t min_k_iters_per_sk_block = 1;
+
+    void dump()
+    {
+        printf("%dx%dx%d(%dx%dx%d), cu:%d, occ:%d, grids:%d, sk_num_big_blocks:%d, "
+               "sk_num_blocks:%d, sk_total_iters:%d, dp_start_block_idx:%d, dp_iters_per_block:%d, "
+               "dp_num_blocks:%d, k_iters_per_tile:%d, k_iters_per_big_block:%d\n",
+               m,
+               n,
+               k,
+               m_per_block,
+               n_per_block,
+               k_per_block,
+               num_cu,
+               occupancy,
+               get_grid_dims_x(),
+               sk_num_big_blocks,
+               sk_num_blocks,
+               sk_total_iters,
+               dp_start_block_idx,
+               dp_iters_per_block,
+               dp_num_blocks,
+               k_iters_per_tile,
+               k_iters_per_big_block);
+    }
+
+    block_dispatcher_t(uint32_t m_per_block_,
+                       uint32_t n_per_block_,
+                       uint32_t k_per_block_,
+                       uint32_t num_cu_,
+                       uint32_t occupancy_,
+                       uint32_t m_,
+                       uint32_t n_,
+                       uint32_t k_)
+        : m_per_block(m_per_block_),
+          n_per_block(n_per_block_),
+          k_per_block(k_per_block_),
+          num_cu(num_cu_),
+          occupancy(occupancy_),
+          m(m_),
+          n(n_),
+          k(k_)
+    {
+        init();
+    }
+
+    uint32_t get_grid_dims_x() { return dp_start_block_idx + dp_num_blocks; }
+
+    uint32_t get_block_idx(uint32_t bid)
+    {
+        // block id is linearily allocated along sk blocks (dp blocks are fine)
+        // this function will compute blockIdx.x and the linear sk block mapping
+        // uint32_t block_idx = 0;
+        // if(bid < sk_num_big_blocks) {
+        //     uint32_t current_k_iter = bid * k_iters_per_big_block;
+        //     tile_idx = current_k_iter / k_iters_per_tile;
+        // }
+        return bid;
+    }
+
+    uint32_t get_current_itr(uint32_t block_idx)
+    {
+        uint32_t current_itr = 0;
+        if(block_idx < sk_num_big_blocks)
+        {
+            current_itr = block_idx * k_iters_per_big_block;
+        }
+        else if(block_idx < sk_num_blocks)
+        {
+            current_itr = (sk_num_big_blocks * k_iters_per_big_block) +
+                          (block_idx - sk_num_big_blocks) * (k_iters_per_big_block - 1);
+        }
+        else if(block_idx >= dp_start_block_idx)
+        {
+            current_itr = sk_total_iters + (block_idx - dp_start_block_idx) * dp_iters_per_block;
+        }
+        return current_itr;
+    }
+
+    void get_block_itr(uint32_t block_idx, uint32_t& iter_start, uint32_t& iter_end)
+    {
+        if(block_idx < sk_num_big_blocks)
+        {
+            iter_start = block_idx * k_iters_per_big_block;
+            iter_end   = iter_start + k_iters_per_big_block;
+        }
+        else if(block_idx < sk_num_blocks)
+        {
+            iter_start = (sk_num_big_blocks * k_iters_per_big_block) +
+                         (block_idx - sk_num_big_blocks) * (k_iters_per_big_block - 1);
+            iter_end = iter_start + (k_iters_per_big_block - 1);
+        }
+        else if(block_idx >= dp_start_block_idx)
+        {
+            iter_start = sk_total_iters + (block_idx - dp_start_block_idx) * dp_iters_per_block;
+            iter_end   = iter_start + dp_iters_per_block;
+        }
+    }
+
+    private:
+    void init()
+    {
+        uint32_t num_tiles =
+            impl::integer_divide_ceil(m, m_per_block) * impl::integer_divide_ceil(n, n_per_block);
+        k_iters_per_tile = impl::integer_divide_ceil(k, k_per_block);
+
+        // one cu can hold one wg at one time, from the whole chip's point of view
+        // if number of wg is same as num_cu, we call it 1 dispatch
+        // if number of wg is 2x num_cu, we call it 2 dispatches.
+        // one dispatch can deliever wg same as num_cu (full dispatch), or less than num_cu (partial
+        // dispatch)
+        //
+        uint32_t full_dispatches         = num_tiles / num_cu;
+        uint32_t full_dispatch_tiles     = full_dispatches * num_cu;
+        uint32_t partial_dispatche_tiles = num_tiles - full_dispatch_tiles;
+
+        uint32_t sk_occupancy = occupancy;
+        uint32_t dp_tiles     = full_dispatch_tiles;
+        uint32_t sk_tiles     = partial_dispatche_tiles;
+
+        if(full_dispatches < occupancy)
+        {
+            // in this case, we allocate all blocks as sk blocks
+            // sk_occupancy = occupancy - full_dispatches;
+            sk_occupancy = 1; // TODO: single occ seems better
+            dp_tiles     = full_dispatch_tiles;
+            sk_tiles     = partial_dispatche_tiles;
+        }
+        else if((occupancy > 1) && (full_dispatches % occupancy == occupancy - 1))
+        {
+            // e.g. occupancy = 2, full_dispatches = 3, 5, 7 ...
+            //      occupancy = 3, full_dispatches = 5, 8, 11 ...
+            //      occupancy = 4, full_dispatches = 7, 11 ...
+            sk_occupancy = 1; // left 1 slot for sk occupancy
+            dp_tiles     = full_dispatch_tiles;
+            sk_tiles     = partial_dispatche_tiles;
+        }
+        else
+        {
+            // others, we reduce 1 dispatch from dp, together with partial dispatch,
+            // to construct sk dispatch
+            sk_occupancy = occupancy - ((full_dispatches - 1) % occupancy);
+            dp_tiles     = full_dispatch_tiles - num_cu;
+            sk_tiles     = partial_dispatche_tiles + num_cu;
+        }
+
+        // dp_num_blocks = dp_tiles;
+        // dp_start_block_idx = num_cu * sk_occupancy;
+        dp_iters_per_block = k_iters_per_tile;
+
+        sk_total_iters = k_iters_per_tile * sk_tiles;
+
+        // printf("num_tiles:%d, full_dispatches:%d, full_dispatch_tiles:%d,
+        // partial_dispatche_tiles:%d\n",
+        //         num_tiles, full_dispatches, full_dispatch_tiles, partial_dispatche_tiles);
+
+        {
+            uint32_t min_sk_tiles = (sk_tiles >= num_cu) ? num_cu : (sk_tiles + 1);
+            uint32_t max_sk_tiles =
+                (sk_tiles >= num_cu) ? num_cu * sk_occupancy
+                                     : impl::min(num_cu, sk_total_iters / min_k_iters_per_sk_block);
+
+            // if use dp for sk-block, how many iters do we need
+            uint32_t dp_for_sk_iters = k_iters_per_tile;
+
+            uint32_t best_sk_score =
+                std::numeric_limits<int>::max(); // we need to find the smallest sk iters
+            for(uint32_t tentative_sk_blocks = min_sk_tiles; tentative_sk_blocks < max_sk_tiles;
+                tentative_sk_blocks++)
+            {
+                uint32_t tentative_sk_iters_per_block =
+                    (sk_total_iters + tentative_sk_blocks - 1) / tentative_sk_blocks;
+                uint32_t tentative_sk_iters = tentative_sk_iters_per_block;
+                uint32_t sk_blocks_per_tile = (tentative_sk_blocks + sk_tiles - 1) / sk_tiles;
+
+                // TODO: carefully adjust this parameter
+                //       the more sk_blocks_per_tile, the worse the overhead
+                uint32_t cross_sk_blocks_overhead = sk_blocks_per_tile;
+                if(tentative_sk_blocks % sk_tiles != 0)
+                {
+                    // penalty for uneven divide
+                    cross_sk_blocks_overhead +=
+                        sk_blocks_per_tile * tentative_sk_iters_per_block / 50;
+                }
+
+                uint32_t tentative_sk_score = tentative_sk_iters + cross_sk_blocks_overhead;
+
+                if(tentative_sk_score < best_sk_score)
+                {
+                    best_sk_score = tentative_sk_score;
+                    sk_num_blocks = tentative_sk_blocks;
+                }
+            }
+
+            if(best_sk_score >= dp_for_sk_iters)
+            {
+                sk_num_blocks = 0;
+            }
+
+            if(sk_num_blocks == 0)
+            {
+                sk_num_big_blocks     = 0;
+                k_iters_per_big_block = 0;
+
+                dp_num_blocks      = num_tiles; // all tile to be dp block
+                dp_start_block_idx = 0;
+                sk_total_iters     = 0; // clear this tiles
+            }
+            else
+            {
+                uint32_t k_iters_per_sk_block = sk_total_iters / sk_num_blocks;
+                sk_num_big_blocks     = sk_total_iters - k_iters_per_sk_block * sk_num_blocks;
+                k_iters_per_big_block = k_iters_per_sk_block + 1;
+
+                dp_num_blocks      = dp_tiles;
+                dp_start_block_idx = (sk_num_blocks + num_cu - 1) / num_cu * num_cu;
+            }
+        }
+    }
+};
+
+struct tile_work_t
+{
+    uint32_t tile_idx;
+    uint32_t iter_begin;
+    uint32_t k_begin;
+    uint32_t k_end;
+    uint32_t k_iters_remaining;
+};
+
+int main(int argc, char** argv)
+{
+    simple_args_t arg = create_arg(argc, argv);
+    block_dispatcher_t block_dispatcher{arg.get_uint32("m_per_block"),
+                                        arg.get_uint32("n_per_block"),
+                                        arg.get_uint32("k_per_block"),
+                                        arg.get_uint32("num_cu"),
+                                        arg.get_uint32("occupancy"),
+                                        arg.get_uint32("m"),
+                                        arg.get_uint32("n"),
+                                        arg.get_uint32("k")};
+    block_dispatcher.dump();
+    // simulate actual kernel launch
+    uint32_t dim_x = block_dispatcher.get_grid_dims_x();
+    uint32_t total_k_iters =
+        impl::integer_divide_ceil(arg.get_uint32("k"), arg.get_uint32("k_per_block"));
+    uint32_t num_tiles =
+        impl::integer_divide_ceil(arg.get_uint32("m"), arg.get_uint32("m_per_block")) *
+        impl::integer_divide_ceil(arg.get_uint32("n"), arg.get_uint32("n_per_block"));
+
+    std::vector<int> valid_tile_record(num_tiles * total_k_iters);
+
+    for(uint32_t bid = 0; bid < dim_x; bid++)
+    {
+        uint32_t block_idx = block_dispatcher.get_block_idx(bid);
+        bool is_sk_block   = block_idx < (block_dispatcher.sk_num_blocks);
+        bool is_dp_block   = block_idx >= block_dispatcher.dp_start_block_idx;
+        uint32_t iter_start, iter_end;
+        block_dispatcher.get_block_itr(block_idx, iter_start, iter_end);
+        uint32_t total_iter_length = iter_end - iter_start;
+
+        while(true)
+        {
+            uint32_t iter_length_mod = iter_end % block_dispatcher.k_iters_per_tile;
+            uint32_t current_iter_length =
+                impl::min(iter_length_mod == 0 ? (iter_end - iter_start) : iter_length_mod,
+                          total_iter_length);
+            uint32_t tile_idx = (iter_end - 1) / block_dispatcher.k_iters_per_tile;
+            uint32_t tile_iter_start =
+                ((iter_end - 1) % block_dispatcher.k_iters_per_tile) - current_iter_length + 1;
+
+            if(is_sk_block)
+            {
+                printf("[sk_block] bid:%3d, block_idx:%3d, tile_idx:%3d, iter_start:%d(%d | %d), "
+                       "iter_end:%d (len:%d)\n",
+                       bid,
+                       block_idx,
+                       tile_idx,
+                       iter_end - current_iter_length,
+                       tile_iter_start,
+                       iter_start,
+                       iter_end,
+                       current_iter_length);
+            }
+            else if(is_dp_block)
+            {
+                printf("[dp_block] bid:%3d, block_idx:%3d, tile_idx:%3d, iter_start:%d(%d | %d), "
+                       "iter_end:%d (len:%d)\n",
+                       bid,
+                       block_idx,
+                       tile_idx,
+                       iter_end - current_iter_length,
+                       tile_iter_start,
+                       iter_start,
+                       iter_end,
+                       current_iter_length);
+            }
+            else
+            {
+                printf("[other   ] bid:%3d, block_idx:%3d\n", bid, block_idx);
+            }
+
+            // some validation check
+            for(auto i = iter_end - current_iter_length; i < iter_end; i++)
+            {
+                if(i >= valid_tile_record.size())
+                {
+                    printf("unexpected, current iter:%d larger than max:%d\n",
+                           i,
+                           valid_tile_record.size());
+                    return -1;
+                }
+                valid_tile_record[i] = 1;
+            }
+
+            iter_end -= current_iter_length;
+            if(iter_end <= iter_start)
+                break;
+        }
+    }
+
+    int untouched = 0;
+    for(auto i = 0; i < valid_tile_record.size(); i++)
+    {
+        if(valid_tile_record[i] != 1)
+        {
+            printf("untouched at %d (%d)\n", i, valid_tile_record.size());
+            untouched++;
+        }
+    }
+    printf("untouched %d/%d, %s\n",
+           untouched,
+           valid_tile_record.size(),
+           untouched == 0 ? "valid" : "fail");
+}
diff --git a/test/block_swizzle_test/rebuild.sh b/test/block_swizzle_test/rebuild.sh
new file mode 100644
index 0000000000..b07eb55048
--- /dev/null
+++ b/test/block_swizzle_test/rebuild.sh
@@ -0,0 +1,3 @@
+CC=g++
+
+$CC -Wall -std=c++17 -Iinclude -O3 block_swizzle_test.cpp -o block_swizzle_test.exe
\ No newline at end of file
diff --git a/test/block_swizzle_test/simple_args.h b/test/block_swizzle_test/simple_args.h
new file mode 100644
index 0000000000..7b10456ce9
--- /dev/null
+++ b/test/block_swizzle_test/simple_args.h
@@ -0,0 +1,159 @@
+#pragma once
+
+#include <iomanip>
+#include <iostream>
+#include <stdlib.h>
+#include <string>
+#include <unordered_map>
+#include <vector>
+#include <assert.h>
+
+struct arg_content_t
+{
+    std::string name; // key
+    std::string value;
+    std::string help_text;
+};
+
+class simple_args_t
+{
+    public:
+    simple_args_t() {}
+    simple_args_t& insert(const std::string& name_,
+                          const std::string& default_value_,
+                          const std::string& help_text_)
+    {
+        arg_content_t arg{name_, default_value_, help_text_};
+
+        if(arg_map.count(arg.name) != 0)
+        {
+            std::cout << "arg:" << arg.name << "already exist" << std::endl;
+        }
+        else
+        {
+            arg_map[arg.name] = arg;
+        }
+        return *this;
+    }
+    void usage()
+    {
+        for(auto& content : arg_map)
+        {
+            std::vector<std::string> help_text_lines;
+            size_t pos = 0;
+            for(size_t next_pos = content.second.help_text.find('\n', pos);
+                next_pos != std::string::npos;)
+            {
+                help_text_lines.push_back(
+                    std::string(content.second.help_text.begin() + pos,
+                                content.second.help_text.begin() + next_pos++));
+                pos      = next_pos;
+                next_pos = content.second.help_text.find('\n', pos);
+            }
+            help_text_lines.push_back(std::string(content.second.help_text.begin() + pos,
+                                                  content.second.help_text.end()));
+
+            int arg_name_width = 16 - content.second.name.length();
+            arg_name_width     = arg_name_width > 0 ? arg_name_width : 2;
+            std::cout << std::setw(4) << "-" << content.second.name << std::setw(arg_name_width)
+                      << " " << help_text_lines[0] << std::endl;
+
+            for(auto help_next_line = std::next(help_text_lines.begin());
+                help_next_line != help_text_lines.end();
+                ++help_next_line)
+            {
+                std::cout << std::setw(28) << " " << *help_next_line << std::endl;
+            }
+        }
+    }
+    bool parse(int argc, char* argv[], int start_index = 1)
+    {
+        if(argc <= start_index)
+        {
+            // std::cout << "not enough args (" << argc << ") with starting index " << start_index
+            // << std::endl;
+            return true;
+        }
+        for(int i = start_index; i < argc; i++)
+        {
+            std::string cur_arg = std::string(argv[i]);
+            if(cur_arg[0] != '-')
+            {
+                std::cout << "illegal input" << std::endl;
+                usage();
+                return false;
+            }
+            else if(cur_arg[0] == '-' && cur_arg[1] == '?')
+            {
+                usage();
+                return false;
+            }
+            else
+            {
+                size_t found_equal = cur_arg.find('=');
+                if(found_equal == std::string::npos || found_equal == (cur_arg.length() - 1))
+                {
+                    std::cout << "failed while parsing \"" << cur_arg << "\", "
+                              << "arg must be in the form \"-name=value\"" << std::endl;
+                    return false;
+                }
+                std::string arg_name  = cur_arg.substr(1, found_equal - 1);
+                std::string arg_value = cur_arg.substr(found_equal + 1);
+                if(arg_map.count(arg_name) == 0)
+                {
+                    std::cout << "no such arg \"" << arg_name << "\" registered" << std::endl;
+                    return false;
+                }
+                arg_map[arg_name].value = arg_value;
+            }
+        }
+        return true;
+    }
+
+    std::string get(const std::string& name) const { return get_str(name); }
+
+    std::string get_str(const std::string& name) const
+    {
+        assert(arg_map.count(name) != 0);
+        std::string value = arg_map.at(name).value;
+        return value;
+    }
+
+    int get_int(const std::string& name) const
+    {
+        assert(arg_map.count(name) != 0);
+        int value = atoi(arg_map.at(name).value.c_str());
+        return value;
+    }
+
+    uint32_t get_uint32(const std::string& name) const
+    {
+        assert(arg_map.count(name) != 0);
+        uint32_t value = strtoul(arg_map.at(name).value.c_str(), nullptr, 10);
+        return value;
+    }
+
+    uint64_t get_uint64(const std::string& name) const
+    {
+        assert(arg_map.count(name) != 0);
+        uint64_t value = strtoull(arg_map.at(name).value.c_str(), nullptr, 10);
+        return value;
+    }
+
+    double get_double(const std::string& name) const
+    {
+        assert(arg_map.count(name) != 0);
+        double value = atof(arg_map.at(name).value.c_str());
+        return value;
+    }
+
+    float get_float(const std::string& name) const
+    {
+        assert(arg_map.count(name) != 0);
+        float value = atof(arg_map.at(name).value.c_str());
+        return value;
+    }
+
+    private:
+    std::unordered_map<std::string, arg_content_t> arg_map;
+};

From 7761e5232c1d1451d17df14f63dfb48285e60c0d Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Thu, 27 Jul 2023 20:29:44 +0200
Subject: [PATCH 073/194] Add s_nops after v_dot to avoid hazard (#808)

* Add s_nops after v_dot to avoid hazard

* Fix builtin for inner_produxt fp16

* Skip inline version to builtin

* Add comments regarding isa

* Fix comment regarding s_nop
---
 include/ck/ck.hpp                              |  8 ++++++--
 .../impl/device_batched_gemm_multiple_d_dl.hpp | 15 +++++++--------
 include/ck/utility/inner_product.hpp           | 18 +++++++++++++-----
 script/profile_batched_gemm.sh                 |  7 -------
 test/batched_gemm_multi_d/CMakeLists.txt       |  6 ++----
 5 files changed, 28 insertions(+), 26 deletions(-)

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 6a1ac09021..aa40955b39 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -118,8 +118,12 @@
 // inline asm
 #define CK_USE_AMD_INLINE_ASM 1
 
-// inner product (DLOP)
-#define CK_USE_AMD_INNER_PRODUCT_INLINE_ASM 1
+// inner product (V_MAC/V_FMAC)
+#define CK_USE_AMD_V_MAC_INLINE_ASM 1
+
+// V_DOT inline instructions, less efficient since they require adding
+// `s_nop`s to avoid hazard
+#define CK_USE_AMD_V_DOT_INLINE_ASM 0
 
 // block synchronization only s_wait lgkmcnt(0), not vmcnt(0)
 #define CK_EXPERIMENTAL_BLOCK_SYNC_LDS_WITHOUT_SYNC_VMEM 1
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
index 0df365ceb2..b51c600476 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
@@ -70,10 +70,9 @@ __global__ void
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch,
             const Block2CTileMap block_2_ctile_map)
 {
-// TODO: Enable for gfx90a after complier fix
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||              \
-    defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || defined(__gfx1101__) || \
-    defined(__gfx1102__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__))
 
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
@@ -650,10 +649,10 @@ struct DeviceBatchedGemmMultipleD_Dl : public DeviceBatchedGemmMultiD<ALayout,
     static bool IsSupportedArgument(const Argument& arg)
     {
         // TODO: Enable for gfx90a after complier fix
-        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx908" ||
-           ck::get_device_name() == "gfx1030" || ck::get_device_name() == "gfx940" ||
-           ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
-           ck::get_device_name() == "gfx1102")
+        if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx90a" ||
+           ck::get_device_name() == "gfx908" || ck::get_device_name() == "gfx1030" ||
+           ck::get_device_name() == "gfx940" || ck::get_device_name() == "gfx1100" ||
+           ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102")
         {
             bool pass = true;
             pass      = pass && arg.K_ % K1 == 0;
diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index b13bccb5ab..0087ae0f7a 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -13,13 +13,13 @@ __device__ void inner_product(const TA& a, const TB& b, TC& c);
 template <>
 __device__ void inner_product<float, float, float>(const float& a, const float& b, float& c)
 {
-#if CK_USE_AMD_INNER_PRODUCT_INLINE_ASM && defined(CK_USE_AMD_V_MAC_F32)
+#if CK_USE_AMD_V_MAC_INLINE_ASM && defined(CK_USE_AMD_V_MAC_F32)
     asm volatile("\n \
             v_mac_f32 %0, %1, %2 \n \
             "
                  : "=v"(c)
                  : "v"(a), "v"(b), "0"(c));
-#elif CK_USE_AMD_INNER_PRODUCT_INLINE_ASM && defined(CK_USE_AMD_V_FMAC_F32)
+#elif CK_USE_AMD_V_MAC_INLINE_ASM && defined(CK_USE_AMD_V_FMAC_F32)
     asm volatile("\n \
             v_fmac_f32 %0, %1, %2 \n \
             "
@@ -76,14 +76,18 @@ template <>
 __device__ void inner_product<half2_t, half2_t, float>(const half2_t& a, const half2_t& b, float& c)
 {
 #if defined(CK_USE_AMD_V_DOT2_F32_F16)
-#if CK_USE_AMD_INNER_PRODUCT_INLINE_ASM
+#if CK_USE_AMD_V_DOT_INLINE_ASM
+    // Use 3 x s_nop to avoid hazard (mi200 cdna2 isa page 47
+    // https://www.amd.com/system/files/TechDocs/instinct-mi200-cdna2-instruction-set-architecture.pdf
+    // ) s_nop with parameter 2 is equal to 3 x s_nop
     asm volatile("\n \
             v_dot2_f32_f16 %0, %1, %2, %0\n \
+            s_nop 2 \n \
             "
                  : "=v"(c)
                  : "v"(a), "v"(b), "0"(c));
 #else
-    c = __builtin_amdgcn_sdot2(a, b, c, false);
+    c = __builtin_amdgcn_fdot2(a, b, c, false);
 #endif
 #else
     const vector_type<half_t, 2> a_vector{a};
@@ -163,9 +167,13 @@ __device__ void
 inner_product<int8x4_t, int8x4_t, int32_t>(const int8x4_t& a, const int8x4_t& b, int32_t& c)
 {
 #if defined(CK_USE_AMD_V_DOT4_I32_I8)
-#if CK_USE_AMD_INNER_PRODUCT_INLINE_ASM
+#if CK_USE_AMD_V_DOT_INLINE_ASM
+    // Use 3 x s_nop to avoid hazard (mi200 cdna2 isa page 47
+    // https://www.amd.com/system/files/TechDocs/instinct-mi200-cdna2-instruction-set-architecture.pdf
+    // ) s_nop with parameter 2 is equal to 3 x s_nop
     asm volatile("\n \
             v_dot4_i32_i8 %0, %1, %2, %0\n \
+            s_nop 2 \n \
             "
                  : "=v"(c)
                  : "v"(bit_cast<int32_t>(a)), "v"(bit_cast<int32_t>(b)), "0"(c));
diff --git a/script/profile_batched_gemm.sh b/script/profile_batched_gemm.sh
index d19ddd0c65..f90baaed68 100755
--- a/script/profile_batched_gemm.sh
+++ b/script/profile_batched_gemm.sh
@@ -3,13 +3,6 @@
 ## GPU visibility
 export HIP_VISIBLE_DEVICES=0
 DRIVER="../build/bin/ckProfiler"
-OP=$1
-DATATYPE=$2
-LAYOUT=$3
-VERIFY=$4
-INIT=$5
-LOG=$6
-TIME=$7
  
 OP=$1
 DATATYPE=$2
diff --git a/test/batched_gemm_multi_d/CMakeLists.txt b/test/batched_gemm_multi_d/CMakeLists.txt
index f8ceb545c2..102a5b8f27 100644
--- a/test/batched_gemm_multi_d/CMakeLists.txt
+++ b/test/batched_gemm_multi_d/CMakeLists.txt
@@ -1,7 +1,5 @@
 # TODO: Enable for gfx90a after complier fix
 if(DL_KERNELS)
-  if(NOT GPU_TARGETS MATCHES "gfx90a")
-    add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
-    target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
-  endif()
+  add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
+  target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
 endif()

From f7cc8c3b03c53000eeebf3aa5c4d3a51f84e9cb7 Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Wed, 2 Aug 2023 23:32:22 +0800
Subject: [PATCH 074/194] Update tuning parameter & compilation options of
 DeviceGemmXdl<> instance (layout=TT) (#819)

* Enable pipeline v2 opt for layout=TT instance

* Use better thread mapping for reading A tile

* Conditionally enable pipeline v2 opt

* Allow enabling only fp16 gemm instances in profiler

* Fix formatting error

* Fix compilation error if we enable fp32 in profiler
---
 .../gpu/gemm/CMakeLists.txt                   | 65 ++++++++++---------
 ...kn_mn_default_pipeline_v2_opt_instance.cpp |  2 +-
 profiler/src/profile_gemm.cpp                 |  8 ++-
 3 files changed, 42 insertions(+), 33 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index b6f63f39d7..266a5d4b50 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -95,36 +95,39 @@ endif()
 
 add_instance_library(device_gemm_instance ${GEMM_INSTANCES})
 
-set(ENABLE_PIPELINE_V2_OPT OFF)
 
-if (ENABLE_PIPELINE_V2_OPT)
-set(MAX_ILP_OPTS
-   -mllvm
-   -amdgpu-enable-max-ilp-scheduling-strategy
-)
-set(WAVES_PER_EU_DEFS
-   CK_USE_WAVES_PER_EU=1
-   CK_MIN_WAVES_PER_EU=1
-   CK_MAX_WAVES_PER_EU=1
-)
-set(IGLP_OPT_DEFS
-   CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT=1
-)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  set(ENABLE_PIPELINE_V2_OPT OFF)
 
-# layout=NT
-set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-   COMPILE_OPTIONS ";;"
-   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
-# layout=NN
-set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
-   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
-# layout=TT
-set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
-   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS}")
-# layout=TN
-set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-   COMPILE_OPTIONS "${MAX_ILP_OPTS}"
-   COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
-endif(ENABLE_PIPELINE_V2_OPT)
+  if (ENABLE_PIPELINE_V2_OPT)
+    set(MAX_ILP_OPTS
+       -mllvm
+       -amdgpu-enable-max-ilp-scheduling-strategy
+    )
+    set(WAVES_PER_EU_DEFS
+       CK_USE_WAVES_PER_EU=1
+       CK_MIN_WAVES_PER_EU=1
+       CK_MAX_WAVES_PER_EU=1
+    )
+    set(IGLP_OPT_DEFS
+       CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT=1
+    )
+
+    # layout=NT
+    set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+       COMPILE_OPTIONS ";;"
+       COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+    # layout=NN
+    set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+       COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+       COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+    # layout=TT
+    set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+       COMPILE_OPTIONS ";;"
+       COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+    # layout=TN
+    set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
+       COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+       COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
+  endif(ENABLE_PIPELINE_V2_OPT)
+endif(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
index 5314c7fc20..b931b8fdfd 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
@@ -18,7 +18,7 @@ using Instances =
         //##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|            |                       |                      |
         //##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|            |                       |                      |
         //##########|      |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |            |                       |                      |
-        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
+        DeviceGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     8,  8,   32,   32,    1,    2,     S<8, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,           1,  LoopScheduler::Default,  PipelineVersion::v2>
 #endif
         // clang-format on
         >;
diff --git a/profiler/src/profile_gemm.cpp b/profiler/src/profile_gemm.cpp
index fbef123860..d1f5debc7d 100644
--- a/profiler/src/profile_gemm.cpp
+++ b/profiler/src/profile_gemm.cpp
@@ -121,7 +121,10 @@ int profile_gemm(int argc, char* argv[])
         return pass ? 0 : 1;
     };
 
-    if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
+    if(false)
+        ;
+#ifdef __fp32__
+    else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, F32{}, F32{}, F32{}, F32{});
     }
@@ -137,6 +140,8 @@ int profile_gemm(int argc, char* argv[])
     {
         return profile(Col{}, Col{}, Row{}, F32{}, F32{}, F32{}, F32{});
     }
+#endif
+#ifdef __fp16__
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, F16{}, F16{}, F32{}, F16{});
@@ -153,6 +158,7 @@ int profile_gemm(int argc, char* argv[])
     {
         return profile(Col{}, Col{}, Row{}, F16{}, F16{}, F32{}, F16{});
     }
+#endif
 #ifdef __bf16__
     else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN)
     {

From 8c13df07bf1fce9780fe1292579b4ee4a78a8e07 Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Thu, 3 Aug 2023 15:50:38 +0200
Subject: [PATCH 075/194] Improve formatting of docs; Add a note about the
 DL_KERNELS flag (#825)

* Improve formatting of docs; Add a note about the DL_KERNELS flag

* Change the recommended version of ROCm to 5.6
---
 docs/API_Reference_Guide.rst        |  8 +--
 docs/Supported_Primitives_Guide.rst | 13 ++---
 docs/dockerhub.rst                  | 31 ++++++-----
 docs/index.rst                      |  7 ++-
 docs/tutorial_hello_world.rst       | 80 +++++++++++++++++++++--------
 5 files changed, 91 insertions(+), 48 deletions(-)

diff --git a/docs/API_Reference_Guide.rst b/docs/API_Reference_Guide.rst
index b59c6e3026..f21d43c593 100644
--- a/docs/API_Reference_Guide.rst
+++ b/docs/API_Reference_Guide.rst
@@ -7,8 +7,8 @@ API Reference Guide
 Introduction
 =================
 
-This document contains details of the APIs for the Composable Kernel (CK) library and introduces some of the key design
-principles that are used to write new classes that extend CK functionality.
+This document contains details of the APIs for the Composable Kernel (CK) library and introduces
+some of the key design principles that are used to write new classes that extend CK functionality.
 
 =================
 Using CK API
@@ -30,8 +30,8 @@ DeviceMem
 Kernels For Flashattention
 ---------------------------
 
-The Flashattention algorithm is defined in :cite:t:`dao2022flashattention`.  This sections lists the classes that are
-used in the CK GPU implementation of Flashattention.
+The Flashattention algorithm is defined in :cite:t:`dao2022flashattention`. This sections lists
+the classes that are used in the CK GPU implementation of Flashattention.
 
 **Gridwise classes**
 
diff --git a/docs/Supported_Primitives_Guide.rst b/docs/Supported_Primitives_Guide.rst
index 4c3adf67d7..3462283d90 100644
--- a/docs/Supported_Primitives_Guide.rst
+++ b/docs/Supported_Primitives_Guide.rst
@@ -2,15 +2,16 @@
 Supported Primitives Guide
 ==========================
 
-This document contains details of supported primitives in Composable Kernel (CK). In contrast to the API Reference
-Guide, the Supported Primitives Guide is an introduction to the math which underpins the algorithms implemented in CK.
+This document contains details of supported primitives in Composable Kernel (CK). In contrast to the
+API Reference Guide, the Supported Primitives Guide is an introduction to the math which underpins
+the algorithms implemented in CK.
 
 ------------
 Softmax
 ------------
 
-For vectors :math:`x^{(1)}, x^{(2)}, \ldots, x^{(T)}` of size :math:`B` we can decompose the softmax of concatenated
-:math:`x = [ x^{(1)}\ | \ \ldots \ | \ x^{(T)} ]` as,
+For vectors :math:`x^{(1)}, x^{(2)}, \ldots, x^{(T)}` of size :math:`B` we can decompose the
+softmax of concatenated :math:`x = [ x^{(1)}\ | \ \ldots \ | \ x^{(T)} ]` as,
 
 .. math::
    :nowrap:
@@ -25,8 +26,8 @@ For vectors :math:`x^{(1)}, x^{(2)}, \ldots, x^{(T)}` of size :math:`B` we can d
 where :math:`f(x^{(j)}) = \exp( x^{(j)} - m(x^{(j)}) )` is of size :math:`B` and
 :math:`z(x^{(j)}) = f(x_1^{(j)})+ \ldots+ f(x_B^{(j)})` is a scalar.
 
-For a matrix :math:`X` composed of :math:`T_r \times T_c` tiles, :math:`X_{ij}`, of size :math:`B_r \times B_c` we can
-compute the row-wise softmax as follows.
+For a matrix :math:`X` composed of :math:`T_r \times T_c` tiles, :math:`X_{ij}`, of size
+:math:`B_r \times B_c` we can compute the row-wise softmax as follows.
 
 For :math:`j` from :math:`1` to :math:`T_c`, and :math:`i` from :math:`1` to :math:`T_r` calculate,
 
diff --git a/docs/dockerhub.rst b/docs/dockerhub.rst
index b51226cfeb..66ec91096e 100644
--- a/docs/dockerhub.rst
+++ b/docs/dockerhub.rst
@@ -1,27 +1,27 @@
 ===================
-CK docker hub
+CK Docker Hub
 ===================
 
-`Docker hub <https://hub.docker.com/r/rocm/composable_kernel>`_
-
 -------------------------------------
 Why do I need this?
 -------------------------------------
 
-To make our lives easier and bring Composable Kernel dependencies together, we recommend using docker images.
+To make our lives easier and bring Composable Kernel dependencies together, we recommend using
+docker images that can be found on `Docker Hub <https://hub.docker.com/r/rocm/composable_kernel>`_.
 
 -------------------------------------
 So what is Composable Kernel?
 -------------------------------------
 
-Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
+Composable Kernel (CK) library aims to provide a programming model for writing performance critical
+kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc,
+through general purpose kernel languages, like HIP C++.
 
 To get the CK library::
 
     git clone https://github.com/ROCmSoftwarePlatform/composable_kernel.git
 
 
-
 run a docker container::
 
     docker run                                                            \
@@ -30,7 +30,7 @@ run a docker container::
     --group-add sudo                                                      \
     -w /root/workspace                                                    \
     -v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace                         \
-    rocm/composable_kernel:ck_ub20.04_rocm5.3_release                     \
+    rocm/composable_kernel:ck_ub20.04_rocm5.6                             \
     /bin/bash
 
 and build the CK::
@@ -58,7 +58,9 @@ We can also run specific examples or tests like::
     ./bin/example_gemm_xdl_fp16
     ./bin/test_gemm_fp16
 
-For more details visit `CK github repo <https://github.com/ROCmSoftwarePlatform/composable_kernel>`_, `CK examples <https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/example)>`_, `even more CK examples <https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/client_example>`_.
+For more details visit `CK github repository <https://github.com/ROCmSoftwarePlatform/composable_kernel>`_,
+`CK examples <https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/example)>`_,
+`even more CK examples <https://github.com/ROCmSoftwarePlatform/composable_kernel/tree/develop/client_example>`_.
 
 -------------------------------------
 And what is inside?
@@ -74,12 +76,11 @@ The docker images have everything you need for running CK including:
 Which image is right for me?
 -------------------------------------
 
-Let's take a look at the image naming, for example "ck_ub20.04_rocm5.4_release". The image specs are:
+Let's take a look at the image naming, for example ``ck_ub20.04_rocm5.6``. The image specs are:
 
-* "ck" - made for running Composable Kernel
-* "ub20.04" - based on Ubuntu 20.04
-* "rocm5.4" - ROCm platform version 5.4
-* "release" - compiler version is release
+* ``ck`` - made for running Composable Kernel;
+* ``ub20.04`` - based on Ubuntu 20.04;
+* ``rocm5.6`` - ROCm platform version 5.6.
 
 So just pick the right image for your project dependencies and you're all set.
 
@@ -87,7 +88,9 @@ So just pick the right image for your project dependencies and you're all set.
 DIY starts here
 -------------------------------------
 
-If you need to customize a docker image or just can't stop tinkering, feel free to adjust the `Dockerfile <https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/Dockerfile>`_ for your needs.
+If you need to customize a docker image or just can't stop tinkering, feel free to adjust the
+`Dockerfile <https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/Dockerfile>`_
+for your needs.
 
 -------------------------------------
 License
diff --git a/docs/index.rst b/docs/index.rst
index f4e66c1b51..51c0c862ae 100644
--- a/docs/index.rst
+++ b/docs/index.rst
@@ -12,12 +12,15 @@ This document contains instructions for installing, using, and contributing to C
 Methodology
 -----------
 
-Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
+Composable Kernel (CK) library aims to provide a programming model for writing performance critical
+kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc,
+through general purpose kernel languages, like HIP C++.
 
 CK utilizes two concepts to achieve performance portability and code maintainability:
 
 * A tile-based programming model
-* Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
+* Algorithm complexity reduction for complex ML operators, using innovative technique we call
+  "Tensor Coordinate Transformation".
 
 .. image:: data/ck_component.png
    :alt: CK Components
diff --git a/docs/tutorial_hello_world.rst b/docs/tutorial_hello_world.rst
index b8fd094654..bfb197e085 100644
--- a/docs/tutorial_hello_world.rst
+++ b/docs/tutorial_hello_world.rst
@@ -6,15 +6,26 @@ CK Hello world
 Motivation
 -------------------------------------
 
-This tutorial is aimed at engineers dealing with artificial intelligence and machine learning who would like to optimize their pipelines and squeeze every performance drop by adding Composable Kernel (CK) library to their projects. We would like to make the CK library approachable so the tutorial is not based on the latest release and doesn't have all the bleeding edge features, but it will be reproducible now and forever.
+This tutorial is aimed at engineers dealing with artificial intelligence and machine learning who
+would like to optimize their pipelines and squeeze every performance drop by adding Composable
+Kernel (CK) library to their projects. We would like to make the CK library approachable so
+the tutorial is not based on the latest release and doesn't have all the bleeding edge features,
+but it will be reproducible now and forever.
 
-During this tutorial we will have an introduction to the CK library, we will build it and run some examples and tests, so to say we will run a "Hello world" example. In future tutorials we will go in depth and breadth and get familiar with other tools and ways to integrate CK into your project.
+During this tutorial we will have an introduction to the CK library, we will build it and run some
+examples and tests, so to say we will run a "Hello world" example. In future tutorials we will go
+in depth and breadth and get familiar with other tools and ways to integrate CK into your project.
 
 -------------------------------------
 Description
 -------------------------------------
 
-Modern AI technology solves more and more problems in all imaginable fields, but crafting fast and efficient workflows is still challenging. CK is one of the tools to make AI heavy lifting as fast and efficient as possible. CK is a collection of optimized AI operator kernels and tools to create new ones. The library has components required for majority of modern neural networks architectures including matrix multiplication, convolution, contraction, reduction, attention modules, variety of activation functions, fused operators and many more.
+Modern AI technology solves more and more problems in all imaginable fields, but crafting fast and
+efficient workflows is still challenging. CK is one of the tools to make AI heavy lifting as fast
+and efficient as possible. CK is a collection of optimized AI operator kernels and tools to create
+new ones. The library has components required for majority of modern neural networks architectures
+including matrix multiplication, convolution, contraction, reduction, attention modules, variety of
+activation functions, fused operators and many more.
 
 So how do we (almost) reach the speed of light? CK acceleration abilities are based on:
 
@@ -24,15 +35,18 @@ So how do we (almost) reach the speed of light? CK acceleration abilities are ba
 * Hardware acceleration use.
 * Support of low precision data types including fp16, bf16, int8 and int4.
 
-If you are excited and need more technical details and benchmarking results - read this awesome `blog post <https://community.amd.com/t5/instinct-accelerators/amd-composable-kernel-library-efficient-fused-kernels-for-ai/ba-p/553224>`_.
+If you are excited and need more technical details and benchmarking results - read this awesome
+`blog post <https://community.amd.com/t5/instinct-accelerators/amd-composable-kernel-library-efficient-fused-kernels-for-ai/ba-p/553224>`_.
 
-For more details visit our `github repo <https://github.com/ROCmSoftwarePlatform/composable_kernel>`_.
+For more details visit our `github repository <https://github.com/ROCmSoftwarePlatform/composable_kernel>`_.
 
 -------------------------------------
 Hardware targets
 -------------------------------------
 
-CK library fully supports "gfx908" and "gfx90a" GPU architectures and only some operators are supported for "gfx1030". Let's check the hardware you have at hand and decide on the target GPU architecture
+CK library fully supports `gfx908` and `gfx90a` GPU architectures and only some operators are
+supported for `gfx1030`. Let's check the hardware you have at hand and decide on the target
+GPU architecture.
 
 ==========     =========
 GPU Target     AMD GPU
@@ -42,7 +56,8 @@ gfx90a 	       Radeon Instinct MI210, MI250, MI250X
 gfx1030        Radeon PRO V620, W6800, W6800X, W6800X Duo, W6900X, RX 6800, RX 6800 XT, RX 6900 XT, RX 6900 XTX, RX 6950 XT
 ==========     =========
 
-There are also `cloud options <https://aws.amazon.com/ec2/instance-types/g4/>`_ you can find if you don't have an AMD GPU at hand.
+There are also `cloud options <https://aws.amazon.com/ec2/instance-types/g4/>`_ you can find if
+you don't have an AMD GPU at hand.
 
 -------------------------------------
 Build the library
@@ -54,9 +69,13 @@ First let's clone the library and rebase to the tested version::
     cd composable_kernel/
     git checkout tutorial_hello_world
 
-To make our lives easier we prepared `docker images <https://hub.docker.com/r/rocm/composable_kernel>`_ with all the necessary dependencies. Pick the right image and create a container. In this tutorial we use "rocm/composable_kernel:ck_ub20.04_rocm5.3_release" image, it is based on Ubuntu 20.04, ROCm v5.3, compiler release version.
+To make our lives easier we prepared
+`docker images <https://hub.docker.com/r/rocm/composable_kernel>`_ with all the necessary
+dependencies. Pick the right image and create a container. In this tutorial we use
+``rocm/composable_kernel:ck_ub20.04_rocm5.6`` image, it is based on Ubuntu 20.04 and
+ROCm v5.6.
 
-If your current folder is ${HOME}, start the docker container with::
+If your current folder is ``${HOME}``, start the docker container with::
 
     docker run  \
     -it  \
@@ -64,20 +83,23 @@ If your current folder is ${HOME}, start the docker container with::
     --group-add sudo  \
     -w /root/workspace  \
     -v ${HOME}:/root/workspace  \
-    rocm/composable_kernel:ck_ub20.04_rocm5.3_release  \
+    rocm/composable_kernel:ck_ub20.04_rocm5.6  \
     /bin/bash
 
-If your current folder is different from ${HOME}, adjust the line `-v ${HOME}:/root/workspace` to fit your folder structure.
+If your current folder is different from ``${HOME}``, adjust the line ``-v ${HOME}:/root/workspace``
+to fit your folder structure.
 
-Inside the docker container current folder is "~/workspace", library path is "~/workspace/composable_kernel", navigate to the library::
+Inside the docker container current folder is ``~/workspace``, library path is
+``~/workspace/composable_kernel``, navigate to the library::
 
     cd composable_kernel/
 
-Create and go to the "build" directory::
+Create and go to the ``build`` directory::
 
     mkdir build && cd build
 
-In the previous section we talked about target GPU architecture. Once you decide which one is right for you, run cmake using the right GPU_TARGETS flag::
+In the previous section we talked about target GPU architecture. Once you decide which one is right
+for you, run CMake using the right ``GPU_TARGETS`` flag::
 
     cmake  \
     -D CMAKE_PREFIX_PATH=/opt/rocm  \
@@ -87,7 +109,7 @@ In the previous section we talked about target GPU architecture. Once you decide
     -D BUILD_DEV=OFF  \
     -D GPU_TARGETS="gfx908;gfx90a;gfx1030" ..
 
-If everything went well the cmake run will end up with::
+If everything went well the CMake run will end up with::
 
     -- Configuring done
     -- Generating done
@@ -118,9 +140,12 @@ We can also run them separately, here is a separate example execution::
 
     ./bin/example_gemm_xdl_fp16 1 1 1
 
-The arguments "1 1 1" mean that we want to run this example in the mode: verify results with CPU, initialize matrices with integers and benchmark the kernel execution. You can play around with these parameters and see how output and execution results change.
+The arguments ``1 1 1`` mean that we want to run this example in the mode: verify results with CPU,
+initialize matrices with integers and benchmark the kernel execution. You can play around with
+these parameters and see how output and execution results change.
 
-If everything goes well and you have a device based on gfx908 or gfx90a architecture you should see something like::
+If everything goes well and you have a device based on `gfx908` or `gfx90a` architecture you should see
+something like::
 
     a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
     b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
@@ -130,14 +155,15 @@ If everything goes well and you have a device based on gfx908 or gfx90a architec
     Start running 10 times...
     Perf: 1.10017 ms, 117.117 TFlops, 87.6854 GB/s, DeviceGemmXdl<256, 256, 128, 4, 8, 32, 32, 4, 2> NumPrefetch: 1, LoopScheduler: Default, PipelineVersion: v1
 
-Meanwhile, running it on a gfx1030 device should result in::
+Meanwhile, running it on a `gfx1030` device should result in::
 
     a_m_k: dim 2, lengths {3840, 4096}, strides {4096, 1}
     b_k_n: dim 2, lengths {4096, 4096}, strides {1, 4096}
     c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
     DeviceGemmXdl<256, 256, 128, 4, 8, 32, 32, 4, 2> NumPrefetch: 1, LoopScheduler: Default, PipelineVersion: v1 does not support this problem
 
-But don't panic, some of the operators are supported on gfx1030 architecture, so you can run a separate example like::
+But don't panic, some of the operators are supported on `gfx1030` architecture, so you can run a
+separate example like::
 
     ./bin/example_gemm_dl_fp16 1 1 1
 
@@ -154,7 +180,14 @@ and it should result in something nice similar to::
     Start running 10 times...
     Perf: 3.65695 ms, 35.234 TFlops, 26.3797 GB/s, DeviceGemmDl<256, 128, 128, 16, 2, 4, 4, 1>
 
-Or we can run a separate test::
+.. note::
+
+    There was a new CMake flag ``DL_KERNELS`` added in the latest versions of CK. If you use one of
+    the newest versions of the library and do not see the above results when running
+    ``example_gemm_dl_fp16``, it might be necessary to add ``-D DL_KERNELS=ON`` to your CMake command
+    in order to build the operators supported on the `gfx1030` architecture.
+
+We can also run a separate test::
 
     ctest -R test_gemm_fp16
 
@@ -169,6 +202,9 @@ If everything goes well you should see something like::
 Summary
 -----------
 
-In this tutorial we took the first look at the Composable Kernel library, built it on your system and ran some examples and tests. Stay tuned, in the next tutorial we will run kernels with different configs to find out the best one for your hardware and task.
+In this tutorial we took the first look at the Composable Kernel library, built it on your system
+and ran some examples and tests. Stay tuned, in the next tutorial we will run kernels with different
+configs to find out the best one for your hardware and task.
 
-P.S.: Don't forget to switch out the cloud instance if you have launched one, you can find better ways to spend your money for sure!
+P.S.: Don't forget to switch off the cloud instance if you have launched one, you can find better
+ways to spend your money for sure!

From e6a826d35af7012504b6f333b547295eeabdfa92 Mon Sep 17 00:00:00 2001
From: Bartlomiej Kocot <barkocot@amd.com>
Date: Tue, 1 Aug 2023 10:43:27 +0000
Subject: [PATCH 076/194] Rename the workaround to a proper issue name

---
 include/ck/ck.hpp                             |  2 +-
 ...ice_grouped_conv_bwd_data_xdl_instance.hpp | 84 +++++++++----------
 2 files changed, 43 insertions(+), 43 deletions(-)

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index aa40955b39..cd43001d6c 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -198,7 +198,7 @@
 #define CK_WORKAROUND_SWDEV_388832 1
 
 // workaround: Grouped Conv2d_bwd_data fails for already implemented instance
-#define CK_WORKAROUND_SWDEV_3318619 0
+#define CK_WORKAROUND_GITHUB_CK_824 1
 
 // flag to enable (1) or disable (0) the debugging output in some kernels
 #define DEBUG_LOG 0
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
index c6c1697f0b..acccfdf2e2 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -49,20 +49,20 @@ using device_grouped_conv_bwd_data_xdl_f16_instances =
         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
         
-#ifdef CK_WORKAROUND_SWDEV_3318619
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#ifndef CK_WORKAROUND_GITHUB_CK_824
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
 #endif
         // clang-format on
         >;
@@ -83,20 +83,20 @@ using device_grouped_conv_bwd_data_xdl_bf16_instances =
         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
         
-#ifdef CK_WORKAROUND_SWDEV_3318619
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        //     DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
+#ifndef CK_WORKAROUND_GITHUB_CK_824
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
 #endif
         // clang-format on
         >;
@@ -117,20 +117,20 @@ using device_grouped_conv_bwd_data_xdl_f32_instances =
          // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
          DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
 
-#ifdef CK_WORKAROUND_SWDEV_3318619
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
-        // DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+#ifndef CK_WORKAROUND_GITHUB_CK_824
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
 #endif
         // clang-format on
         >;

From aac65a031ebc76c7e1ab3aa84aa9eb273eae2122 Mon Sep 17 00:00:00 2001
From: Bartlomiej Kocot <barkocot@amd.com>
Date: Tue, 1 Aug 2023 22:05:28 +0000
Subject: [PATCH 077/194] Change to github_issue prefix

---
 include/ck/ck.hpp                                           | 2 +-
 .../device_grouped_conv_bwd_data_xdl_instance.hpp           | 6 +++---
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index cd43001d6c..950121feb3 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -198,7 +198,7 @@
 #define CK_WORKAROUND_SWDEV_388832 1
 
 // workaround: Grouped Conv2d_bwd_data fails for already implemented instance
-#define CK_WORKAROUND_GITHUB_CK_824 1
+#define CK_WORKAROUND_GITHUB_ISSUE_824 1
 
 // flag to enable (1) or disable (0) the debugging output in some kernels
 #define DEBUG_LOG 0
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
index acccfdf2e2..e18a58b5e1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -49,7 +49,7 @@ using device_grouped_conv_bwd_data_xdl_f16_instances =
         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
         
-#ifndef CK_WORKAROUND_GITHUB_CK_824
+#ifndef CK_WORKAROUND_GITHUB_ISSUE_824
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
@@ -83,7 +83,7 @@ using device_grouped_conv_bwd_data_xdl_bf16_instances =
         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
         
-#ifndef CK_WORKAROUND_GITHUB_CK_824
+#ifndef CK_WORKAROUND_GITHUB_ISSUE_824
             DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
             DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
             DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
@@ -117,7 +117,7 @@ using device_grouped_conv_bwd_data_xdl_f32_instances =
          // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
          DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
 
-#ifndef CK_WORKAROUND_GITHUB_CK_824
+#ifndef CK_WORKAROUND_GITHUB_ISSUE_824
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,

From 2474dddbee9c49e2ef9c42f5a6ce631f46245d8d Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 3 Aug 2023 10:10:27 -0700
Subject: [PATCH 078/194] add an option to build ckProfiler package for
 specific architectures (#828)

---
 CMakeLists.txt | 43 +++++++++++++++++++++++++++++++++++++------
 1 file changed, 37 insertions(+), 6 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 514b98fde6..7fab38cd27 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -60,11 +60,34 @@ message("checking which targets are supported")
 #This is the list of targets to be used in case GPU_TARGETS is not set on command line
 #These targets will be filtered and only supported ones will be used
 #Setting GPU_TARGETS on command line will override this list
-rocm_check_target_ids(DEFAULT_GPU_TARGETS
-    TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102"
-)
+if(NOT PROFILER_ONLY)
+    rocm_check_target_ids(DEFAULT_GPU_TARGETS
+        TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
+else()
+    add_definitions(-DPROFILER_ONLY)
+    if(GPU_TARGETS)
+        message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx9, gfx10, or gfx11")
+    endif()
+    if(GPU_ARCH MATCHES "gfx9")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942")
+    elseif(GPU_ARCH MATCHES "gfx10")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1030")
+    elseif(GPU_ARCH MATCHES "gfx11")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1100;gfx1101;gfx1102")
+    else()
+        message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx9, gfx10, or gfx11")
+    endif()
+endif()
+
 message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}")
+
 set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " ")
+
+if(GPU_TARGETS)
+    message("Building CK for the following targets: ${GPU_TARGETS}")
+else()
+    message("Building CK for the following targets: ${AMDGPU_TARGETS}")
+endif()
 find_package(hip)
 
 option(USE_BITINT_EXTENSION_INT4, "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
@@ -347,6 +370,7 @@ add_custom_target(instances DEPENDS utility;${CK_DEVICE_INSTANCES}  SOURCES ${IN
 add_subdirectory(library)
 
 if(NOT DEFINED INSTANCES_ONLY)
+ if(NOT DEFINED PROFILER_ONLY)
    rocm_package_setup_component(tests
         LIBRARY_NAME composablekernel
         PACKAGE_NAME tests # Prevent -static suffix on package name
@@ -356,15 +380,22 @@ if(NOT DEFINED INSTANCES_ONLY)
         LIBRARY_NAME composablekernel
         PACKAGE_NAME examples
    )
+   add_subdirectory(example)
+   add_subdirectory(test)
 
    rocm_package_setup_component(profiler
         LIBRARY_NAME composablekernel
         PACKAGE_NAME ckProfiler
    )
-
-   add_subdirectory(example)
-   add_subdirectory(test)
    add_subdirectory(profiler)
+  else()
+    #When building PROFILER_ONLY, label the package with GPU_ARCH
+    rocm_package_setup_component(profiler
+       LIBRARY_NAME composablekernel
+       PACKAGE_NAME ckProfiler_${GPU_ARCH}
+    )
+    add_subdirectory(profiler)
+  endif()
 endif()
 
 #Create an interface target for the include only files and call it "composablekernels"

From 22443f7aaec112080e9a884bb047bc01db1c9ffd Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Mon, 7 Aug 2023 17:23:45 +0200
Subject: [PATCH 079/194] Add wei_strides to grouped conv3d wei to keep
 consistency (#817)

* Add wei_strides to grouped conv3d wei to keep consistency

* Fix strides in client examples

* Unify backward weight api with forward

* Fix for example

* Fixes for examples

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 .../11_grouped_conv_bwd_weight/common.hpp     | 103 +++---
 .../grouped_conv1d_bwd_weight_fp16.cpp        |  22 +-
 .../grouped_conv2d_bwd_weight_fp16.cpp        |  23 +-
 .../grouped_conv3d_bwd_weight_fp16.cpp        |  23 +-
 .../grouped_conv3d_bwd_weight_fp32.cpp        |  36 +-
 .../run_grouped_conv_bwd_weight_example.inc   |  25 +-
 .../device/device_grouped_conv_bwd_weight.hpp |  15 +-
 ...uped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp | 159 ++++-----
 ...e_grouped_conv_bwd_weight_xdl_cshuffle.hpp | 334 ++++++++----------
 .../profile_grouped_conv_bwd_weight_impl.hpp  |  25 +-
 ...st_grouped_convnd_bwd_weight_interface.cpp |  25 +-
 11 files changed, 357 insertions(+), 433 deletions(-)

diff --git a/client_example/11_grouped_conv_bwd_weight/common.hpp b/client_example/11_grouped_conv_bwd_weight/common.hpp
index f63e5f2157..4292cded20 100644
--- a/client_example/11_grouped_conv_bwd_weight/common.hpp
+++ b/client_example/11_grouped_conv_bwd_weight/common.hpp
@@ -32,63 +32,49 @@ struct SimpleDeviceMem
 };
 
 template <ck::index_t NumDimSpatial>
-std::size_t GetFlops(ck::index_t G,
-                     ck::index_t N,
-                     ck::index_t K,
-                     ck::index_t C,
-                     const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths,
-                     const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths)
+std::size_t GetFlops(const std::array<ck::index_t, NumDimSpatial>& output_lengths,
+                     const std::array<ck::index_t, NumDimSpatial>& filter_lengths)
 {
+    constexpr ck::index_t spatial_offset = 3;
+    const auto C                         = filter_lengths[2];
     // 2 * G * N * K * C * <output spatial lengths product> * <filter spatial lengths product>
-    return static_cast<std::size_t>(2) * G * N * K * C *
-           std::accumulate(std::begin(output_spatial_lengths),
-                           std::end(output_spatial_lengths),
+    return static_cast<std::size_t>(2) * C *
+           std::accumulate(std::begin(output_lengths),
+                           std::end(output_lengths),
                            static_cast<std::size_t>(1),
                            std::multiplies<>()) *
-           std::accumulate(std::begin(filter_spatial_lengths),
-                           std::end(filter_spatial_lengths),
+           std::accumulate(std::begin(filter_lengths) + spatial_offset,
+                           std::end(filter_lengths),
                            static_cast<std::size_t>(1),
                            std::multiplies<>());
 }
 
 template <typename InDataType, ck::index_t NumDimSpatial>
-std::size_t GetInputByte(ck::index_t G,
-                         ck::index_t N,
-                         ck::index_t C,
-                         const std::array<ck::index_t, NumDimSpatial>& input_spatial_lengths)
+std::size_t GetInputByte(const std::array<ck::index_t, NumDimSpatial>& input_lengths)
 {
     // sizeof(InDataType) * (G * N * C * <input spatial lengths product>) +
-    return sizeof(InDataType) * (G * N * C *
-                                 std::accumulate(std::begin(input_spatial_lengths),
-                                                 std::end(input_spatial_lengths),
+    return sizeof(InDataType) * (std::accumulate(std::begin(input_lengths),
+                                                 std::end(input_lengths),
                                                  static_cast<std::size_t>(1),
                                                  std::multiplies<>()));
 }
 
 template <typename WeiDataType, ck::index_t NumDimSpatial>
-std::size_t GetWeightByte(ck::index_t G,
-                          ck::index_t K,
-                          ck::index_t C,
-                          const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths)
+std::size_t GetWeightByte(const std::array<ck::index_t, NumDimSpatial>& filter_lengths)
 {
     // sizeof(WeiDataType) * (G * K * C * <filter spatial lengths product>) +
-    return sizeof(WeiDataType) * (G * K * C *
-                                  std::accumulate(std::begin(filter_spatial_lengths),
-                                                  std::end(filter_spatial_lengths),
+    return sizeof(WeiDataType) * (std::accumulate(std::begin(filter_lengths),
+                                                  std::end(filter_lengths),
                                                   static_cast<std::size_t>(1),
                                                   std::multiplies<>()));
 }
 
 template <typename OutDataType, ck::index_t NumDimSpatial>
-std::size_t GetOutputByte(ck::index_t G,
-                          ck::index_t N,
-                          ck::index_t K,
-                          const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths)
+std::size_t GetOutputByte(const std::array<ck::index_t, NumDimSpatial>& output_lengths)
 {
     // sizeof(OutDataType) * (G * N * K * <output spatial lengths product>);
-    return sizeof(OutDataType) * (G * N * K *
-                                  std::accumulate(std::begin(output_spatial_lengths),
-                                                  std::end(output_spatial_lengths),
+    return sizeof(OutDataType) * (std::accumulate(std::begin(output_lengths),
+                                                  std::end(output_lengths),
                                                   static_cast<std::size_t>(1),
                                                   std::multiplies<std::size_t>()));
 }
@@ -101,14 +87,11 @@ template <ck::index_t NumDimSpatial,
           typename WeiLayout,
           typename OutLayout>
 bool run_grouped_conv_bwd_weight(
-    const ck::index_t G,
-    const ck::index_t N,
-    const ck::index_t K,
-    const ck::index_t C,
-    const std::array<ck::index_t, NumDimSpatial>& input_spatial_lengths,
-    const std::array<ck::index_t, NumDimSpatial>& filter_spatial_lengths,
-    const std::array<ck::index_t, NumDimSpatial>& output_spatial_lengths,
+    const std::array<ck::index_t, NumDimSpatial + 3>& input_lengths,
     const std::array<ck::index_t, NumDimSpatial + 3>& input_strides,
+    const std::array<ck::index_t, NumDimSpatial + 3>& filter_lengths,
+    const std::array<ck::index_t, NumDimSpatial + 3>& weights_strides,
+    const std::array<ck::index_t, NumDimSpatial + 3>& output_lengths,
     const std::array<ck::index_t, NumDimSpatial + 3>& output_strides,
     const std::array<ck::index_t, NumDimSpatial>& conv_filter_strides,
     const std::array<ck::index_t, NumDimSpatial>& conv_filter_dilations,
@@ -117,9 +100,9 @@ bool run_grouped_conv_bwd_weight(
 {
 
     ck::index_t split_k = 2;
-    SimpleDeviceMem in(GetInputByte<InDataType, NumDimSpatial>(G, N, C, input_spatial_lengths));
-    SimpleDeviceMem wei(GetWeightByte<WeiDataType, NumDimSpatial>(G, K, C, filter_spatial_lengths));
-    SimpleDeviceMem out(GetOutputByte<OutDataType, NumDimSpatial>(G, N, K, output_spatial_lengths));
+    SimpleDeviceMem in(GetInputByte<InDataType, NumDimSpatial + 3>(input_lengths));
+    SimpleDeviceMem wei(GetWeightByte<WeiDataType, NumDimSpatial + 3>(filter_lengths));
+    SimpleDeviceMem out(GetOutputByte<OutDataType, NumDimSpatial + 3>(output_lengths));
 
     using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdWeight<NumDimSpatial,
                                                                               InLayout,
@@ -143,6 +126,10 @@ bool run_grouped_conv_bwd_weight(
     float best_gb_per_sec = 0;
     float best_tflops     = 0;
 
+    std::array<ck::index_t, NumDimSpatial + 3> a_g_n_c_wis_lengths{};
+    std::array<ck::index_t, NumDimSpatial + 3> a_g_n_c_wis_strides{};
+    std::array<ck::index_t, NumDimSpatial + 3> b_g_k_c_xs_lengths{};
+
     // profile device operation instances
     std::cout << "Run all instances and do timing" << std::endl;
 
@@ -152,14 +139,11 @@ bool run_grouped_conv_bwd_weight(
         auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         wei.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
-                                                        G,
-                                                        N,
-                                                        K,
-                                                        C,
-                                                        input_spatial_lengths,
-                                                        filter_spatial_lengths,
-                                                        output_spatial_lengths,
+                                                        input_lengths,
                                                         input_strides,
+                                                        filter_lengths,
+                                                        weights_strides,
+                                                        output_lengths,
                                                         output_strides,
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
@@ -176,12 +160,10 @@ bool run_grouped_conv_bwd_weight(
         {
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
-            std::size_t flop =
-                GetFlops<NumDimSpatial>(G, N, K, C, output_spatial_lengths, filter_spatial_lengths);
-            std::size_t num_bytes =
-                GetInputByte<InDataType, NumDimSpatial>(G, N, C, input_spatial_lengths) +
-                GetWeightByte<WeiDataType, NumDimSpatial>(G, K, C, filter_spatial_lengths) +
-                GetOutputByte<OutDataType, NumDimSpatial>(G, N, K, output_spatial_lengths);
+            std::size_t flop      = GetFlops<NumDimSpatial + 3>(output_lengths, filter_lengths);
+            std::size_t num_bytes = GetInputByte<InDataType, NumDimSpatial + 3>(input_lengths) +
+                                    GetWeightByte<WeiDataType, NumDimSpatial + 3>(filter_lengths) +
+                                    GetOutputByte<OutDataType, NumDimSpatial + 3>(output_lengths);
 
             float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
@@ -221,14 +203,11 @@ bool run_grouped_conv_bwd_weight(
         auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         wei.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
-                                                        G,
-                                                        N,
-                                                        K,
-                                                        C,
-                                                        input_spatial_lengths,
-                                                        filter_spatial_lengths,
-                                                        output_spatial_lengths,
+                                                        input_lengths,
                                                         input_strides,
+                                                        filter_lengths,
+                                                        weights_strides,
+                                                        output_lengths,
                                                         output_strides,
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
index 1c6f485da2..e6d427faf4 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv1d_bwd_weight_fp16.cpp
@@ -22,11 +22,12 @@ static constexpr ck::index_t C             = 192;
 static constexpr ck::index_t X             = 3;
 static constexpr ck::index_t Wi            = 28;
 static constexpr ck::index_t Wo            = 28;
-static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Wi};
-static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{X};
-static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Wo};
-static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{N * Wi * C, Wi* C, C, 1};
-static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{N * Wo * K, Wo* K, K, 1};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_lengths{G, N, C, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> filter_lengths{G, K, C, X};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_lengths{G, N, K, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{N * Wi * C, Wi* C, 1, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> weights_strides{K * X * C, X* C, 1, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{N * Wo * K, Wo* K, 1, K};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1};
 static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1};
@@ -40,14 +41,11 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G,
-                                                  N,
-                                                  K,
-                                                  C,
-                                                  input_spatial_lengths,
-                                                  filter_spatial_lengths,
-                                                  output_spatial_lengths,
+                                       OutLayout>(input_lengths,
                                                   input_strides,
+                                                  filter_lengths,
+                                                  weights_strides,
+                                                  output_lengths,
                                                   output_strides,
                                                   conv_filter_strides,
                                                   conv_filter_dilations,
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
index 25e82f3896..4201ea61b4 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv2d_bwd_weight_fp16.cpp
@@ -25,13 +25,15 @@ static constexpr ck::index_t Hi            = 28;
 static constexpr ck::index_t Wi            = 28;
 static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 28;
-static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Hi, Wi};
-static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Y, X};
-static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_lengths{G, N, C, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> filter_lengths{G, K, C, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_lengths{G, N, K, Ho, Wo};
 static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
-    N * Hi * Wi * C, Hi* Wi* C, Wi* C, C, 1};
+    N * Hi * Wi * C, Hi* Wi* C, 1, Wi* C, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> weights_strides{
+    K * Y * X * C, Y* X* C, 1, X* C, C};
 static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
-    N * Ho * Wo * K, Ho* Wo* K, Wo* K, K, 1};
+    N * Ho * Wo * K, Ho* Wo* K, 1, Wo* K, K};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1};
 static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
@@ -45,14 +47,11 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G,
-                                                  N,
-                                                  K,
-                                                  C,
-                                                  input_spatial_lengths,
-                                                  filter_spatial_lengths,
-                                                  output_spatial_lengths,
+                                       OutLayout>(input_lengths,
                                                   input_strides,
+                                                  filter_lengths,
+                                                  weights_strides,
+                                                  output_lengths,
                                                   output_strides,
                                                   conv_filter_strides,
                                                   conv_filter_dilations,
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
index a5f5e628ff..3ae46bcd55 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp16.cpp
@@ -28,13 +28,15 @@ static constexpr ck::index_t Wi            = 3;
 static constexpr ck::index_t Do            = 28;
 static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 3;
-static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Di, Hi, Wi};
-static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Z, Y, X};
-static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Do, Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_lengths{G, N, C, Di, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> filter_lengths{G, K, C, Z, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_lengths{G, N, K, Do, Ho, Wo};
 static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
-    N * Di * Hi * Wi * C, Di* Hi* Wi* C, Hi* Wi* C, Wi* C, C, 1};
+    N * Di * Hi * Wi * C, Di* Hi* Wi* C, 1, Hi* Wi* C, Wi* C, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> weights_strides{
+    K * Z * Y * X * C, Z* Y* X* C, 1, Y* X* C, X* C, C};
 static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
-    N * Do * Ho * Wo * K, Do* Ho* Wo* K, Ho* Wo* K, Wo* K, K, 1};
+    N * Do * Ho * Wo * K, Do* Ho* Wo* K, 1, Ho* Wo* K, Wo* K, K};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
 static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
@@ -48,14 +50,11 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(G,
-                                                  N,
-                                                  K,
-                                                  C,
-                                                  input_spatial_lengths,
-                                                  filter_spatial_lengths,
-                                                  output_spatial_lengths,
+                                       OutLayout>(input_lengths,
                                                   input_strides,
+                                                  filter_lengths,
+                                                  weights_strides,
+                                                  output_lengths,
                                                   output_strides,
                                                   conv_filter_strides,
                                                   conv_filter_dilations,
diff --git a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
index d95e8a205e..2eb869f392 100644
--- a/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
+++ b/client_example/11_grouped_conv_bwd_weight/grouped_conv3d_bwd_weight_fp32.cpp
@@ -28,13 +28,15 @@ static constexpr ck::index_t Wi            = 3;
 static constexpr ck::index_t Do            = 28;
 static constexpr ck::index_t Ho            = 28;
 static constexpr ck::index_t Wo            = 3;
-static constexpr std::array<ck::index_t, NumDimSpatial> input_spatial_lengths{Di, Hi, Wi};
-static constexpr std::array<ck::index_t, NumDimSpatial> filter_spatial_lengths{Z, Y, X};
-static constexpr std::array<ck::index_t, NumDimSpatial> output_spatial_lengths{Do, Ho, Wo};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_lengths{G, N, C, Di, Hi, Wi};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> filter_lengths{G, K, C, Z, Y, X};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_lengths{G, N, K, Do, Ho, Wo};
 static constexpr std::array<ck::index_t, NumDimSpatial + 3> input_strides{
-    N * Di * Hi * Wi * C, Di* Hi* Wi* C, Hi* Wi* C, Wi* C, C, 1};
+    N * Di * Hi * Wi * C, Di* Hi* Wi* C, 1, Hi* Wi* C, Wi* C, C};
+static constexpr std::array<ck::index_t, NumDimSpatial + 3> weights_strides{
+    K * Z * Y * X * C, Z* Y* X* C, 1, Y* X* C, X* C, C};
 static constexpr std::array<ck::index_t, NumDimSpatial + 3> output_strides{
-    N * Do * Ho * Wo * K, Do* Ho* Wo* K, Ho* Wo* K, Wo* K, K, 1};
+    N * Do * Ho * Wo * K, Do* Ho* Wo* K, 1, Ho* Wo* K, Wo* K, K};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_strides{1, 1, 1};
 static constexpr std::array<ck::index_t, NumDimSpatial> conv_filter_dilations{1, 1, 1};
 static constexpr std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
@@ -48,20 +50,16 @@ int main()
                                        OutDataType,
                                        InLayout,
                                        WeiLayout,
-                                       OutLayout>(
-               G,
-               N,
-               K,
-               C,
-               {Di, Hi, Wi},
-               {Z, Y, X},
-               {Do, Ho, Wo},
-               {N * Di * Hi * Wi * C, Di * Hi * Wi * C, Hi * Wi * C, Wi * C, C, 1},
-               {N * Do * Ho * Wo * K, Do * Ho * Wo * K, Ho * Wo * K, Wo * K, K, 1},
-               {1, 1, 1},
-               {1, 1, 1},
-               {1, 1, 1},
-               {1, 1, 1})
+                                       OutLayout>(input_lengths,
+                                                  input_strides,
+                                                  filter_lengths,
+                                                  weights_strides,
+                                                  output_lengths,
+                                                  output_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads)
                ? EXIT_SUCCESS
                : EXIT_FAILURE;
 }
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 49bd9fc7f0..29ce0324ab 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -72,10 +72,11 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     // init to 0
     wei_device_buf.SetZero();
 
-    std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
-    std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
-    std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_lengths{};
     std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> filter_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> weights_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> output_lengths{};
     std::array<ck::index_t, NDimSpatial + 3> output_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
@@ -84,10 +85,11 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
 
     auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
 
-    range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
-    range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
-    range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+    range_copy(in_g_n_c_wis_desc.GetLengths(), begin(input_lengths));
     range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+    range_copy(wei_g_k_c_xs_desc.GetLengths(), begin(filter_lengths));
+    range_copy(wei_g_k_c_xs_desc.GetStrides(), begin(weights_strides));
+    range_copy(out_g_n_k_wos_desc.GetLengths(), begin(output_lengths));
     range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
     range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
     range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
@@ -100,14 +102,11 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
     auto argument = conv.MakeArgument(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
                                       static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
                                       static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-                                      conv_param.G_,
-                                      conv_param.N_,
-                                      conv_param.K_,
-                                      conv_param.C_,
-                                      input_spatial_lengths,
-                                      filter_spatial_lengths,
-                                      output_spatial_lengths,
+                                      input_lengths,
                                       input_strides,
+                                      filter_lengths,
+                                      weights_strides,
+                                      output_lengths,
                                       output_strides,
                                       conv_filter_strides,
                                       conv_filter_dilations,
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
index a103bdff35..ab9e6adb41 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
@@ -27,15 +27,12 @@ struct DeviceGroupedConvBwdWeight : public BaseOperator
     MakeArgumentPointer(const void* p_in,
                         void* p_wei,
                         const void* p_out,
-                        const ck::index_t G,
-                        const ck::index_t N,
-                        const ck::index_t K,
-                        const ck::index_t C,
-                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
-                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
                         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
index 003a508f07..198751cdf3 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
@@ -784,15 +784,12 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         Argument(const InDataType* p_in_grid,
                  WeiDataType* p_wei_grid,
                  const OutDataType* p_out_grid,
-                 const ck::index_t G,
-                 const ck::index_t N,
-                 const ck::index_t K,
-                 const ck::index_t C,
-                 const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                 const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                 const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                 const std::array<ck::index_t, NDimSpatial + 3>& /*input_strides*/,
-                 const std::array<ck::index_t, NDimSpatial + 3>& /*output_strides*/,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                 const std::array<index_t, NDimSpatial + 3>& /*a_g_n_c_wis_strides*/,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                 const std::array<index_t, NDimSpatial + 3>& /*b_g_k_c_xs_strides*/,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                 const std::array<index_t, NDimSpatial + 3>& /*e_g_n_k_wos_strides*/,
                  const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                  const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                  const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -812,27 +809,38 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
               a_element_op_{out_element_op},
               b_element_op_{wei_element_op},
               c_element_op_{in_element_op},
-              Conv_G_{G},
-              Conv_N_{N},
-              Conv_K_{K},
-              Conv_C_{C},
-              input_spatial_lengths_{input_spatial_lengths},
-              filter_spatial_lengths_{filter_spatial_lengths},
-              output_spatial_lengths_{output_spatial_lengths},
+              Conv_G_{a_g_n_c_wis_lengths[0]},
+              Conv_N_{a_g_n_c_wis_lengths[1]},
+              Conv_K_{b_g_k_c_xs_lengths[1]},
+              Conv_C_{a_g_n_c_wis_lengths[2]},
+              input_spatial_lengths_{},
+              filter_spatial_lengths_{},
+              output_spatial_lengths_{},
               conv_filter_strides_{conv_filter_strides},
               conv_filter_dilations_{conv_filter_dilations},
               input_left_pads_{input_left_pads},
               input_right_pads_{input_right_pads},
               k_batch_{split_k}
         {
+            constexpr index_t spatial_offset = 3;
+            std::copy(begin(a_g_n_c_wis_lengths) + spatial_offset,
+                      end(a_g_n_c_wis_lengths),
+                      begin(input_spatial_lengths_));
+            std::copy(begin(b_g_k_c_xs_lengths) + spatial_offset,
+                      end(b_g_k_c_xs_lengths),
+                      begin(filter_spatial_lengths_));
+            std::copy(begin(e_g_n_k_wos_lengths) + spatial_offset,
+                      end(e_g_n_k_wos_lengths),
+                      begin(output_spatial_lengths_));
+
             const auto descs =
                 DeviceOp::MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<NDimSpatial>(
-                    N,
-                    K,
-                    C,
-                    input_spatial_lengths,
-                    filter_spatial_lengths,
-                    output_spatial_lengths,
+                    Conv_N_,
+                    Conv_K_,
+                    Conv_C_,
+                    input_spatial_lengths_,
+                    filter_spatial_lengths_,
+                    output_spatial_lengths_,
                     conv_filter_strides,
                     conv_filter_dilations,
                     input_left_pads,
@@ -856,21 +864,21 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
 
             // A/B/C Batch Stride
             compute_ptr_offset_of_batch_.BatchStrideA_ =
-                N * K *
-                std::accumulate(begin(output_spatial_lengths),
-                                end(output_spatial_lengths),
+                Conv_N_ * Conv_K_ *
+                std::accumulate(begin(output_spatial_lengths_),
+                                end(output_spatial_lengths_),
                                 index_t{1},
                                 std::multiplies<>{});
             compute_ptr_offset_of_batch_.BatchStrideB_ =
-                N * C *
-                std::accumulate(begin(input_spatial_lengths),
-                                end(input_spatial_lengths),
+                Conv_N_ * Conv_C_ *
+                std::accumulate(begin(input_spatial_lengths_),
+                                end(input_spatial_lengths_),
                                 index_t{1},
                                 std::multiplies<>{});
             compute_ptr_offset_of_batch_.BatchStrideC_ =
-                K * C *
-                std::accumulate(begin(filter_spatial_lengths),
-                                end(filter_spatial_lengths),
+                Conv_K_ * Conv_C_ *
+                std::accumulate(begin(filter_spatial_lengths_),
+                                end(filter_spatial_lengths_),
                                 index_t{1},
                                 std::multiplies<>{});
         }
@@ -904,9 +912,9 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         const index_t Conv_K_;
         const index_t Conv_C_;
 
-        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths_;
-        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths_;
-        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial> input_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial> output_spatial_lengths_;
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations_;
         const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
@@ -1110,39 +1118,34 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
     }
 
-    static auto MakeArgument(const InDataType* p_in_grid,
-                             WeiDataType* p_wei_grid,
-                             const OutDataType* p_out_grid,
-                             const ck::index_t G,
-                             const ck::index_t N,
-                             const ck::index_t K,
-                             const ck::index_t C,
-                             const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                             const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                             const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                             const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
-                             const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
-                             const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
-                             const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
-                             const std::array<ck::index_t, NDimSpatial>& input_left_pads,
-                             const std::array<ck::index_t, NDimSpatial>& input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op,
-                             ck::index_t split_k)
+    static auto
+    MakeArgument(const InDataType* p_in_grid,
+                 WeiDataType* p_wei_grid,
+                 const OutDataType* p_out_grid,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                 const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+                 InElementwiseOperation in_element_op,
+                 WeiElementwiseOperation wei_element_op,
+                 OutElementwiseOperation out_element_op,
+                 ck::index_t split_k)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
                         p_out_grid,
-                        G,
-                        N,
-                        K,
-                        C,
-                        input_spatial_lengths,
-                        filter_spatial_lengths,
-                        output_spatial_lengths,
-                        input_strides,
-                        output_strides,
+                        a_g_n_c_wis_lengths, // input
+                        a_g_n_c_wis_strides,
+                        b_g_k_c_xs_lengths, // weight
+                        b_g_k_c_xs_strides,
+                        e_g_n_k_wos_lengths, // output
+                        e_g_n_k_wos_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -1159,15 +1162,12 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     MakeArgumentPointer(const void* p_in_grid,
                         void* p_wei_grid,
                         const void* p_out_grid,
-                        const ck::index_t G,
-                        const ck::index_t N,
-                        const ck::index_t K,
-                        const ck::index_t C,
-                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
-                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
                         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -1180,15 +1180,12 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
                                           static_cast<WeiDataType*>(p_wei_grid),
                                           static_cast<const OutDataType*>(p_out_grid),
-                                          G,
-                                          N,
-                                          K,
-                                          C,
-                                          input_spatial_lengths,
-                                          filter_spatial_lengths,
-                                          output_spatial_lengths,
-                                          input_strides,
-                                          output_strides,
+                                          a_g_n_c_wis_lengths, // input
+                                          a_g_n_c_wis_strides,
+                                          b_g_k_c_xs_lengths, // weight
+                                          b_g_k_c_xs_strides,
+                                          e_g_n_k_wos_lengths, // output
+                                          e_g_n_k_wos_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
index a55da9add2..5a8bb6188b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -245,21 +245,10 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                        const ck::index_t K,
                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides)
     {
-        if constexpr(is_GNHWK_GKYXC_GNHWC)
-        {
-            return make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-        }
-        else if constexpr(is_NHWGK_GKYXC_NHWGC)
-        {
-            const index_t WoStride = output_strides[4];
-            const auto KStride     = Number<1>{};
-            return make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, K),
-                                                make_tuple(WoStride, KStride));
-        }
-        else
-        {
-            throw std::runtime_error("wrong! unsupported layout: " + OutLayout::name());
-        }
+        const index_t WoStride = output_strides[4];
+        const auto KStride     = Number<1>{};
+        return make_naive_tensor_descriptor(make_tuple(N * Ho * Wo, K),
+                                            make_tuple(WoStride, KStride));
     }
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
@@ -270,42 +259,36 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                       const ck::index_t C,
                       const std::array<ck::index_t, NDimSpatial + 3>& input_strides)
     {
-        if constexpr(is_GNHWK_GKYXC_GNHWC)
+        const index_t NStride  = input_strides[1];
+        const index_t HiStride = input_strides[3];
+        const index_t WiStride = input_strides[4];
+        const auto CStride     = input_strides[2];
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
-            if constexpr(ConvBackwardWeightSpecialization ==
-                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
-            {
-                return make_naive_tensor_descriptor_packed(make_tuple(N * Hi * Wi, C));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
-            }
-        }
-        else if constexpr(is_NHWGK_GKYXC_NHWGC)
-        {
-            const index_t NStride  = input_strides[1];
-            const index_t HiStride = input_strides[3];
-            const index_t WiStride = input_strides[4];
-            const auto CStride     = input_strides[2];
-            if constexpr(ConvBackwardWeightSpecialization ==
-                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
-            {
-                return make_naive_tensor_descriptor(make_tuple(N * Hi * Wi, C),
-                                                    make_tuple(WiStride, CStride));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor(
-                    make_tuple(N, Hi, Wi, C), make_tuple(NStride, HiStride, WiStride, CStride));
-            }
+            return make_naive_tensor_descriptor(make_tuple(N * Hi * Wi, C),
+                                                make_tuple(WiStride, CStride));
         }
         else
         {
-            throw std::runtime_error("wrong! unsupported layout: " + InLayout::name());
+            return make_naive_tensor_descriptor(make_tuple(N, Hi, Wi, C),
+                                                make_tuple(NStride, HiStride, WiStride, CStride));
         }
     }
 
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
+    constexpr static auto
+    make_wei_grid_desc(const ck::index_t K,
+                       const ck::index_t Y,
+                       const ck::index_t X,
+                       const ck::index_t C,
+                       const std::array<ck::index_t, NDimSpatial + 3>& weights_strides)
+    {
+        const auto CStride = Number<1>{};
+        const auto KStride = weights_strides[1];
+        return make_naive_tensor_descriptor(make_tuple(K, Y * X * C), make_tuple(KStride, CStride));
+    }
+
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     constexpr static auto
     make_out_grid_desc(const ck::index_t N,
@@ -315,21 +298,10 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                        const ck::index_t K,
                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides)
     {
-        if constexpr(is_GNDHWK_GKZYXC_GNDHWC)
-        {
-            return make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
-        }
-        else if constexpr(is_NDHWGK_GKZYXC_NDHWGC)
-        {
-            const index_t WoStride = output_strides[5];
-            const auto KStride     = Number<1>{};
-            return make_naive_tensor_descriptor(make_tuple(N * Do * Ho * Wo, K),
-                                                make_tuple(WoStride, KStride));
-        }
-        else
-        {
-            throw std::runtime_error("wrong! unsupported layout: " + OutLayout::name());
-        }
+        const index_t WoStride = output_strides[5];
+        const auto KStride     = Number<1>{};
+        return make_naive_tensor_descriptor(make_tuple(N * Do * Ho * Wo, K),
+                                            make_tuple(WoStride, KStride));
     }
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
@@ -341,44 +313,40 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                       const ck::index_t C,
                       const std::array<ck::index_t, NDimSpatial + 3>& input_strides)
     {
-        if constexpr(is_GNDHWK_GKZYXC_GNDHWC)
+        const index_t NStride  = input_strides[1];
+        const index_t DiStride = input_strides[3];
+        const index_t HiStride = input_strides[4];
+        const index_t WiStride = input_strides[5];
+        const auto CStride     = input_strides[2];
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
-            if constexpr(ConvBackwardWeightSpecialization ==
-                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
-            {
-                return make_naive_tensor_descriptor_packed(make_tuple(N * Di * Hi * Wi, C));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
-            }
-        }
-        else if constexpr(is_NDHWGK_GKZYXC_NDHWGC)
-        {
-            const index_t NStride  = input_strides[1];
-            const index_t DiStride = input_strides[3];
-            const index_t HiStride = input_strides[4];
-            const index_t WiStride = input_strides[5];
-            const auto CStride     = input_strides[2];
-            if constexpr(ConvBackwardWeightSpecialization ==
-                         ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
-            {
-                return make_naive_tensor_descriptor(make_tuple(N * Di * Hi * Wi, C),
-                                                    make_tuple(WiStride, CStride));
-            }
-            else
-            {
-                return make_naive_tensor_descriptor(
-                    make_tuple(N, Di, Hi, Wi, C),
-                    make_tuple(NStride, DiStride, HiStride, WiStride, CStride));
-            }
+            return make_naive_tensor_descriptor(make_tuple(N * Di * Hi * Wi, C),
+                                                make_tuple(WiStride, CStride));
         }
         else
         {
-            throw std::runtime_error("wrong! unsupported layout: " + InLayout::name());
+            return make_naive_tensor_descriptor(
+                make_tuple(N, Di, Hi, Wi, C),
+                make_tuple(NStride, DiStride, HiStride, WiStride, CStride));
         }
     }
 
+    template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    constexpr static auto
+    make_wei_grid_desc(const ck::index_t K,
+                       const ck::index_t Z,
+                       const ck::index_t Y,
+                       const ck::index_t X,
+                       const ck::index_t C,
+                       const std::array<ck::index_t, NDimSpatial + 3>& weights_strides)
+    {
+        const auto CStride = Number<1>{};
+        const auto KStride = weights_strides[1];
+        return make_naive_tensor_descriptor(make_tuple(K, Z * Y * X * C),
+                                            make_tuple(KStride, CStride));
+    }
+
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
         const ck::index_t N,
@@ -388,6 +356,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial + 3>& /* input_strides */,
+        const std::array<ck::index_t, NDimSpatial + 3>& /* weights_strides */,
         const std::array<ck::index_t, NDimSpatial + 3>& /* output_strides */,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
@@ -542,6 +511,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+        const std::array<ck::index_t, NDimSpatial + 3>& weights_strides,
         const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
@@ -584,6 +554,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
         const auto out_grid_desc = make_out_grid_desc<NDim>(N, Ho, Wo, K, output_strides);
         const auto in_grid_desc  = make_in_grid_desc<NDim>(N, Hi, Wi, C, input_strides);
+        const auto wei_grid_desc = make_wei_grid_desc<NDim>(K, Y, X, C, weights_strides);
 
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
@@ -618,13 +589,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
-
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_grid_desc);
         }
         else
         {
@@ -684,13 +651,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
-
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_grid_desc);
         }
     }
 
@@ -703,6 +666,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
         const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
+        const std::array<ck::index_t, NDimSpatial + 3>& weights_strides,
         const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
@@ -752,6 +716,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
         const auto out_grid_desc = make_out_grid_desc<NDim>(N, Do, Ho, Wo, K, output_strides);
         const auto in_grid_desc  = make_in_grid_desc<NDim>(N, Di, Hi, Wi, C, input_strides);
+        const auto wei_grid_desc = make_wei_grid_desc<NDim>(K, Z, Y, X, C, weights_strides);
 
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
@@ -786,13 +751,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Z * Y * X * C));
-
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_grid_desc);
         }
         else
         {
@@ -861,13 +822,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Z * Y * X * C));
-
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_grid_desc);
         }
     } // function end
 
@@ -887,6 +844,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                                                                   lengths,
                                                                   strides,
                                                                   strides,
+                                                                  strides,
                                                                   params,
                                                                   params,
                                                                   params,
@@ -910,6 +868,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                                                                   lengths,
                                                                   strides,
                                                                   strides,
+                                                                  strides,
                                                                   params,
                                                                   params,
                                                                   params,
@@ -933,6 +892,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                                                                   lengths,
                                                                   strides,
                                                                   strides,
+                                                                  strides,
                                                                   params,
                                                                   params,
                                                                   params,
@@ -1051,15 +1011,12 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         Argument(const InDataType* p_in_grid,
                  WeiDataType* p_wei_grid,
                  const OutDataType* p_out_grid,
-                 const ck::index_t G,
-                 const ck::index_t N,
-                 const ck::index_t K,
-                 const ck::index_t C,
-                 const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                 const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                 const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                 const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
-                 const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
                  const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                  const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                  const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -1084,27 +1041,40 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
               a_element_op_{out_element_op},
               b_element_op_{in_element_op},
               c_element_op_{wei_element_op},
-              Conv_G_{G},
-              Conv_N_{N},
-              Conv_K_{K},
-              Conv_C_{C},
-              output_spatial_lengths_{output_spatial_lengths},
-              filter_spatial_lengths_{filter_spatial_lengths},
+              Conv_G_{a_g_n_c_wis_lengths[0]},
+              Conv_N_{a_g_n_c_wis_lengths[1]},
+              Conv_K_{b_g_k_c_xs_lengths[1]},
+              Conv_C_{a_g_n_c_wis_lengths[2]},
+              input_spatial_lengths_{},
+              filter_spatial_lengths_{},
+              output_spatial_lengths_{},
               conv_filter_strides_{conv_filter_strides},
               input_left_pads_{input_left_pads},
               input_right_pads_{input_right_pads},
               k_batch_{split_k}
         {
+            constexpr index_t spatial_offset = 3;
+            std::copy(begin(a_g_n_c_wis_lengths) + spatial_offset,
+                      end(a_g_n_c_wis_lengths),
+                      begin(input_spatial_lengths_));
+            std::copy(begin(b_g_k_c_xs_lengths) + spatial_offset,
+                      end(b_g_k_c_xs_lengths),
+                      begin(filter_spatial_lengths_));
+            std::copy(begin(e_g_n_k_wos_lengths) + spatial_offset,
+                      end(e_g_n_k_wos_lengths),
+                      begin(output_spatial_lengths_));
+
             const auto descs =
                 DeviceOp::MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<NDimSpatial>(
-                    N,
-                    K,
-                    C,
-                    input_spatial_lengths,
-                    filter_spatial_lengths,
-                    output_spatial_lengths,
-                    input_strides,
-                    output_strides,
+                    Conv_N_,
+                    Conv_K_,
+                    Conv_C_,
+                    input_spatial_lengths_,
+                    filter_spatial_lengths_,
+                    output_spatial_lengths_,
+                    a_g_n_c_wis_strides,
+                    b_g_k_c_xs_strides,
+                    e_g_n_k_wos_strides,
                     conv_filter_strides,
                     conv_filter_dilations,
                     input_left_pads,
@@ -1119,12 +1089,12 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
 
             // A/B/C Batch Stride
-            compute_ptr_offset_of_batch_.BatchStrideA_ = output_strides[0];
-            compute_ptr_offset_of_batch_.BatchStrideB_ = input_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideA_ = e_g_n_k_wos_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = a_g_n_c_wis_strides[0];
             compute_ptr_offset_of_batch_.BatchStrideC_ =
-                K * C *
-                std::accumulate(begin(filter_spatial_lengths),
-                                end(filter_spatial_lengths),
+                Conv_K_ * Conv_C_ *
+                std::accumulate(begin(filter_spatial_lengths_),
+                                end(filter_spatial_lengths_),
                                 index_t{1},
                                 std::multiplies<>{});
 
@@ -1163,8 +1133,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const index_t Conv_N_;
         const index_t Conv_K_;
         const index_t Conv_C_;
-        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths_;
-        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial> input_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial> output_spatial_lengths_;
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
         const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
         const std::array<ck::index_t, NDimSpatial>& input_right_pads_;
@@ -1339,39 +1310,34 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
     }
 
-    static auto MakeArgument(const InDataType* p_in_grid,
-                             WeiDataType* p_wei_grid,
-                             const OutDataType* p_out_grid,
-                             const ck::index_t G,
-                             const ck::index_t N,
-                             const ck::index_t K,
-                             const ck::index_t C,
-                             const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                             const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                             const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                             const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
-                             const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
-                             const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
-                             const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
-                             const std::array<ck::index_t, NDimSpatial>& input_left_pads,
-                             const std::array<ck::index_t, NDimSpatial>& input_right_pads,
-                             InElementwiseOperation in_element_op,
-                             WeiElementwiseOperation wei_element_op,
-                             OutElementwiseOperation out_element_op,
-                             const ck::index_t split_k)
+    static auto
+    MakeArgument(const InDataType* p_in_grid,
+                 WeiDataType* p_wei_grid,
+                 const OutDataType* p_out_grid,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<ck::index_t, NDimSpatial>& input_left_pads,
+                 const std::array<ck::index_t, NDimSpatial>& input_right_pads,
+                 InElementwiseOperation in_element_op,
+                 WeiElementwiseOperation wei_element_op,
+                 OutElementwiseOperation out_element_op,
+                 const ck::index_t split_k)
     {
         return Argument{p_in_grid,
                         p_wei_grid,
                         p_out_grid,
-                        G,
-                        N,
-                        K,
-                        C,
-                        input_spatial_lengths,
-                        filter_spatial_lengths,
-                        output_spatial_lengths,
-                        input_strides,
-                        output_strides,
+                        a_g_n_c_wis_lengths, // input
+                        a_g_n_c_wis_strides,
+                        b_g_k_c_xs_lengths, // weight
+                        b_g_k_c_xs_strides,
+                        e_g_n_k_wos_lengths, // output
+                        e_g_n_k_wos_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -1390,15 +1356,12 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
     MakeArgumentPointer(const void* p_in_grid,
                         void* p_wei_grid,
                         const void* p_out_grid,
-                        const ck::index_t G,
-                        const ck::index_t N,
-                        const ck::index_t K,
-                        const ck::index_t C,
-                        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
-                        const std::array<ck::index_t, NDimSpatial + 3>& input_strides,
-                        const std::array<ck::index_t, NDimSpatial + 3>& output_strides,
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
                         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -1411,15 +1374,12 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
                                           static_cast<WeiDataType*>(p_wei_grid),
                                           static_cast<const OutDataType*>(p_out_grid),
-                                          G,
-                                          N,
-                                          K,
-                                          C,
-                                          input_spatial_lengths,
-                                          filter_spatial_lengths,
-                                          output_spatial_lengths,
-                                          input_strides,
-                                          output_strides,
+                                          a_g_n_c_wis_lengths, // input
+                                          a_g_n_c_wis_strides,
+                                          b_g_k_c_xs_lengths, // weight
+                                          b_g_k_c_xs_strides,
+                                          e_g_n_k_wos_lengths, // output
+                                          e_g_n_k_wos_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
index 3a4295eeb3..48bf639a70 100644
--- a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
@@ -136,10 +136,11 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
     // profile device Conv instances
     bool all_pass = true;
 
-    std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
-    std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
-    std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> filter_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> output_lengths{};
     std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> weights_strides{};
     std::array<ck::index_t, NDimSpatial + 3> output_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
@@ -148,10 +149,11 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
 
     auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
 
-    range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
-    range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
-    range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+    range_copy(in_g_n_c_wis_desc.GetLengths(), begin(input_lengths));
     range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+    range_copy(wei_g_k_c_xs_desc.GetLengths(), begin(filter_lengths));
+    range_copy(wei_g_k_c_xs_desc.GetStrides(), begin(weights_strides));
+    range_copy(out_g_n_k_wos_desc.GetLengths(), begin(output_lengths));
     range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
     range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
     range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
@@ -164,14 +166,11 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
             op_ptr->MakeArgumentPointer(static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
                                         static_cast<WeiDataType*>(wei_device_buf.GetDeviceBuffer()),
                                         static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-                                        conv_param.G_,
-                                        conv_param.N_,
-                                        conv_param.K_,
-                                        conv_param.C_,
-                                        input_spatial_lengths,
-                                        filter_spatial_lengths,
-                                        output_spatial_lengths,
+                                        input_lengths,
                                         input_strides,
+                                        filter_lengths,
+                                        weights_strides,
+                                        output_lengths,
                                         output_strides,
                                         conv_filter_strides,
                                         conv_filter_dilations,
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
index 57d4e4186c..cfbf13f00e 100644
--- a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
@@ -70,10 +70,11 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
             ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
                 conv_param);
 
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> input_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> filter_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> output_lengths{};
         std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> weights_strides{};
         std::array<ck::index_t, NDimSpatial + 3> output_strides{};
         std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
         std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
@@ -82,10 +83,11 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
 
         auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
 
-        range_copy(conv_param.input_spatial_lengths_, begin(input_spatial_lengths));
-        range_copy(conv_param.filter_spatial_lengths_, begin(filter_spatial_lengths));
-        range_copy(conv_param.output_spatial_lengths_, begin(output_spatial_lengths));
+        range_copy(in_g_n_c_wis_desc.GetLengths(), begin(input_lengths));
         range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+        range_copy(wei_g_k_c_xs_desc.GetLengths(), begin(filter_lengths));
+        range_copy(wei_g_k_c_xs_desc.GetStrides(), begin(weights_strides));
+        range_copy(out_g_n_k_wos_desc.GetLengths(), begin(output_lengths));
         range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
         range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
         range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
@@ -97,14 +99,11 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
         auto argument = conv.MakeArgument(nullptr,
                                           nullptr,
                                           nullptr,
-                                          conv_param.G_,
-                                          conv_param.N_,
-                                          conv_param.K_,
-                                          conv_param.C_,
-                                          input_spatial_lengths,
-                                          filter_spatial_lengths,
-                                          output_spatial_lengths,
+                                          input_lengths,
                                           input_strides,
+                                          filter_lengths,
+                                          weights_strides,
+                                          output_lengths,
                                           output_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,

From 08eb17692978f2af709deb59c98aae6db4c82b6b Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 7 Aug 2023 14:56:10 -0700
Subject: [PATCH 080/194] Allow building CK for specific data types and split
 off last remaining DL instances. (#830)

* properly split conv_nd_bwd_data instances

* split conv2d_fwd instance data types

* split the gemm, conv2d_fwd and batched_gemm_softamx_gemm

* split the tests by data types where possible

* filter examples by DTYPES

* split few remaining examples by DTYPES

* filter most instances by DTYPES

* add new lines at end of headers, fix grouped_gemm profiler

* fix syntax

* split the ckprofiler instances by DTYPES

* split the conv2d and quantization DL and XDL instances

* fix the splitting of conv2d DL instances

* split softmax and pool_fwd tests for fp16 and fp32 types

* fix syntax

* fix the dl_int8 quantization instances isolation
---
 example/01_gemm/CMakeLists.txt                | 78 ++++++++--------
 example/02_gemm_bilinear/CMakeLists.txt       |  2 +
 example/03_gemm_bias_relu/CMakeLists.txt      |  2 +
 .../04_gemm_add_add_fastgelu/CMakeLists.txt   | 30 ++++---
 example/09_convnd_fwd/CMakeLists.txt          | 34 +++++--
 .../CMakeLists.txt                            | 24 +++--
 example/13_pool2d_fwd/CMakeLists.txt          |  9 +-
 example/14_gemm_quantization/CMakeLists.txt   |  4 +-
 example/15_grouped_gemm/CMakeLists.txt        | 38 ++++----
 .../CMakeLists.txt                            | 54 +++++------
 example/17_convnd_bwd_data/CMakeLists.txt     |  8 +-
 example/18_batched_gemm_reduce/CMakeLists.txt |  2 +
 .../20_grouped_conv_bwd_weight/CMakeLists.txt | 26 +++---
 example/21_gemm_layernorm/CMakeLists.txt      |  2 +
 example/22_cgemm/CMakeLists.txt               | 25 +++---
 example/24_batched_gemm/CMakeLists.txt        | 28 +++---
 example/25_gemm_bias_e_permute/CMakeLists.txt |  6 +-
 example/26_contraction/CMakeLists.txt         | 13 +--
 example/27_layernorm/CMakeLists.txt           |  6 +-
 .../CMakeLists.txt                            |  4 +-
 .../CMakeLists.txt                            |  8 +-
 .../CMakeLists.txt                            | 40 +++++----
 example/31_batched_gemm_gemm/CMakeLists.txt   | 17 ++--
 .../CMakeLists.txt                            | 36 +++++---
 example/35_splitK_gemm/CMakeLists.txt         | 27 +++---
 .../CMakeLists.txt                            |  4 +-
 .../CMakeLists.txt                            |  4 +-
 example/39_permute/CMakeLists.txt             | 16 ++--
 .../40_conv2d_fwd_quantization/CMakeLists.txt | 30 +++----
 .../41_grouped_conv_conv_fwd/CMakeLists.txt   | 16 +++-
 example/42_groupnorm/CMakeLists.txt           |  8 +-
 .../CMakeLists.txt                            |  8 +-
 example/44_elementwise_permute/CMakeLists.txt |  6 +-
 example/46_gemm_add_multiply/CMakeLists.txt   |  8 +-
 example/48_pool3d_fwd/CMakeLists.txt          |  5 +-
 example/49_maxpool2d_bwd/CMakeLists.txt       | 12 ++-
 example/50_put_element/CMakeLists.txt         |  4 +-
 .../gpu/batched_gemm.hpp                      | 35 +++++---
 .../gpu/batched_gemm_add_relu_gemm_add.hpp    |  3 +-
 .../gpu/batched_gemm_bias_permute.hpp         |  3 +-
 ...batched_gemm_bias_softmax_gemm_permute.hpp | 12 ++-
 .../gpu/batched_gemm_gemm.hpp                 |  3 +-
 .../gpu/batched_gemm_multi_d.hpp              | 13 +--
 .../gpu/batched_gemm_softmax_gemm.hpp         |  3 +-
 .../gpu/batched_gemm_softmax_gemm_permute.hpp | 11 ++-
 .../gpu/contraction_bilinear.hpp              | 14 +--
 .../gpu/contraction_scale.hpp                 | 14 +--
 .../gpu/convolution_backward_data.hpp         | 89 +++++++++++++------
 .../gpu/convolution_forward.hpp               | 26 ++++--
 .../gpu/elementwise_normalization.hpp         |  3 +-
 .../tensor_operation_instance/gpu/gemm.hpp    |  4 +
 .../gpu/gemm_add_relu_add_layernorm.hpp       |  3 +-
 .../gpu/gemm_bilinear.hpp                     |  3 +-
 .../gpu/gemm_fastgelu.hpp                     |  3 +-
 .../gpu/gemm_streamk.hpp                      |  3 +-
 .../gpu/grouped_gemm.hpp                      |  3 +-
 .../gpu/normalization.hpp                     | 17 ++--
 .../gpu/pool2d_fwd.hpp                        | 17 ++--
 .../gpu/pool3d_fwd.hpp                        | 17 ++--
 .../gpu/quantization/gemm_quantization.hpp    | 15 +++-
 ...n_bias_forward_perchannel_quantization.hpp | 13 ++-
 ...ion_bias_forward_perlayer_quantization.hpp | 13 ++-
 ...lution_forward_perchannel_quantization.hpp | 11 ++-
 ...volution_forward_perlayer_quantization.hpp | 11 ++-
 .../tensor_operation_instance/gpu/softmax.hpp | 10 ++-
 .../gpu/batched_gemm/CMakeLists.txt           | 44 +++++----
 .../CMakeLists.txt                            |  2 +
 .../batched_gemm_bias_permute/CMakeLists.txt  |  3 +-
 .../gpu/batched_gemm_gemm/CMakeLists.txt      |  2 +
 .../gpu/batched_gemm_reduce/CMakeLists.txt    |  3 +-
 .../batched_gemm_softmax_gemm/CMakeLists.txt  |  3 +-
 .../CMakeLists.txt                            | 16 ++--
 .../gpu/contraction_bilinear/CMakeLists.txt   | 24 ++---
 .../gpu/contraction_scale/CMakeLists.txt      | 24 ++---
 .../gpu/conv2d_bwd_data/CMakeLists.txt        | 12 ++-
 ...wd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp |  3 +-
 ...wd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp |  3 +-
 ...d_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp |  3 +-
 ..._data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp |  3 +-
 .../gpu/conv2d_fwd/CMakeLists.txt             | 23 +++--
 ...d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp |  3 +-
 ...2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp |  3 +-
 ...2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp |  3 +-
 ...d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp |  3 +-
 .../elementwise_normalization/CMakeLists.txt  |  2 +
 .../gpu/gemm_add_add_fastgelu/CMakeLists.txt  |  2 +
 .../gpu/gemm_add_fastgelu/CMakeLists.txt      |  2 +
 .../CMakeLists.txt                            |  2 +
 .../gpu/gemm_bilinear/CMakeLists.txt          |  2 +
 .../gpu/gemm_fastgelu/CMakeLists.txt          |  2 +
 .../gpu/gemm_streamk/CMakeLists.txt           |  2 +
 .../gpu/grouped_gemm/CMakeLists.txt           |  2 +
 .../gpu/grouped_gemm_fastgelu/CMakeLists.txt  |  2 +
 .../gpu/normalization/CMakeLists.txt          | 20 +++--
 .../gpu/pool_fwd/CMakeLists.txt               | 24 ++---
 .../gpu/quantization/CMakeLists.txt           | 38 ++++----
 .../gpu/softmax/CMakeLists.txt                | 16 ++--
 profiler/src/CMakeLists.txt                   | 48 +++++-----
 profiler/src/profile_grouped_gemm.cpp         |  4 +-
 test/CMakeLists.txt                           |  2 +-
 test/batched_gemm/CMakeLists.txt              | 35 ++++----
 test/batched_gemm_gemm/CMakeLists.txt         | 12 +--
 test/batched_gemm_multi_d/CMakeLists.txt      |  5 +-
 test/batched_gemm_reduce/CMakeLists.txt       | 10 ++-
 test/batched_gemm_softmax_gemm/CMakeLists.txt | 12 +--
 .../CMakeLists.txt                            | 34 +++----
 test/elementwise_normalization/CMakeLists.txt | 13 ++-
 test/gemm_layernorm/CMakeLists.txt            |  2 +
 test/gemm_reduce/CMakeLists.txt               |  8 +-
 test/grouped_gemm/CMakeLists.txt              |  2 +
 test/normalization/CMakeLists.txt             | 35 ++++----
 test/pool_fwd/test_avg_pool2d_fwd.cpp         |  4 +
 test/pool_fwd/test_avg_pool3d_fwd.cpp         |  6 +-
 test/pool_fwd/test_max_pool2d_fwd.cpp         |  6 +-
 test/pool_fwd/test_max_pool3d_fwd.cpp         |  6 +-
 test/softmax/test_softmax_rank3.cpp           |  5 +-
 test/softmax/test_softmax_rank4.cpp           |  5 +-
 117 files changed, 976 insertions(+), 590 deletions(-)

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 42a3499385..7ce866931c 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -1,31 +1,44 @@
-add_custom_target(example_gemm_dl)
+if(DL_KERNELS)
+  add_custom_target(example_gemm_dl)
 
-add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
-add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
+  add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
+  add_dependencies(example_gemm_dl example_gemm_dl_fp32)
+  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
+    add_dependencies(example_gemm_dl example_gemm_dl_fp16)
+  endif()
+  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+    add_dependencies(example_gemm_dl example_gemm_dl_int8)
+  endif()
 
-add_dependencies(example_gemm_dl example_gemm_dl_fp32)
-add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
-   add_dependencies(example_gemm_dl example_gemm_dl_int8)
+  if(USE_BITINT_EXTENSION_INT4)
+    add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
+    add_dependencies(example_gemm_dl example_gemm_dl_int4)
+  endif(USE_BITINT_EXTENSION_INT4)
 endif()
 
-if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
-  add_dependencies(example_gemm_dl example_gemm_dl_int4)
-endif(USE_BITINT_EXTENSION_INT4)
-
-
 add_custom_target(example_gemm_xdl)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
+  add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
+  add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
 
-add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
-add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
-add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
+  if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+    add_custom_target(example_gemm_wmma)
+    add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
+    add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
+  endif()
 
-add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
-add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
-add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
+endif()
+
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
+endif()
 
 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
   add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
@@ -37,22 +50,17 @@ if(USE_BITINT_EXTENSION_INT4)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
 endif(USE_BITINT_EXTENSION_INT4)
 
-add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
-# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
-
-add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
-add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
-
-if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
-  add_custom_target(example_gemm_wmma)
-  add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
-  add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
+if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+  # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
+  add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
 endif()
 
 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
 
-if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
+if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
+  if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+    add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
+  endif()
 endif()
diff --git a/example/02_gemm_bilinear/CMakeLists.txt b/example/02_gemm_bilinear/CMakeLists.txt
index 7ea4564d1b..9989e45e0c 100644
--- a/example/02_gemm_bilinear/CMakeLists.txt
+++ b/example/02_gemm_bilinear/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
 list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
@@ -15,3 +16,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
+endif()
diff --git a/example/03_gemm_bias_relu/CMakeLists.txt b/example/03_gemm_bias_relu/CMakeLists.txt
index 2f5cba924d..a247a052cb 100644
--- a/example/03_gemm_bias_relu/CMakeLists.txt
+++ b/example/03_gemm_bias_relu/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -6,3 +7,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
+endif()
diff --git a/example/04_gemm_add_add_fastgelu/CMakeLists.txt b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
index 447fa9871b..15ec62c89f 100644
--- a/example/04_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
@@ -3,22 +3,26 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_custom_target(example_gemm_add_add_fastgelu_xdl)
-
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
+    endif()
+    if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
+    endif()
+    if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
+    endif()
     if(USE_BITINT_EXTENSION_INT4)
        add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
+       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
     endif(USE_BITINT_EXTENSION_INT4)
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
-
-    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
-    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
-    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
-    if(USE_BITINT_EXTENSION_INT4)
-        add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
-    endif(USE_BITINT_EXTENSION_INT4)
-    add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+    if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+    endif()
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/example/09_convnd_fwd/CMakeLists.txt b/example/09_convnd_fwd/CMakeLists.txt
index 90104c1630..1af1e6c858 100644
--- a/example/09_convnd_fwd/CMakeLists.txt
+++ b/example/09_convnd_fwd/CMakeLists.txt
@@ -2,16 +2,34 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-    add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
-    add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
-    add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
-    add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
+    if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
+    endif()
+    if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
+    endif()
+    if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
+    endif()
+    if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
+    endif()
     # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-    add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
+    if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+      add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
+    endif()
    set(target 1)
  endif()
 endforeach()
 
-add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
-add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
-add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
+if(DL_KERNELS)
+  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
+  endif()
+  if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
+  endif()
+  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
+  endif()
+endif()
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
index 9577b4569a..e7d941ae6b 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
@@ -3,14 +3,22 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_convnd_fwd_reduce_xdl)
-   add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
-   add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
-   add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
-   add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
-   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
-   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
-   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
-   add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
+   endif()
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
+   endif()
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
+   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
       add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
diff --git a/example/13_pool2d_fwd/CMakeLists.txt b/example/13_pool2d_fwd/CMakeLists.txt
index db09c03321..d0f356757b 100644
--- a/example/13_pool2d_fwd/CMakeLists.txt
+++ b/example/13_pool2d_fwd/CMakeLists.txt
@@ -1,3 +1,6 @@
-add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
-add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
-
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
+endif()
diff --git a/example/14_gemm_quantization/CMakeLists.txt b/example/14_gemm_quantization/CMakeLists.txt
index ca06265e31..3b3ad80dd8 100644
--- a/example/14_gemm_quantization/CMakeLists.txt
+++ b/example/14_gemm_quantization/CMakeLists.txt
@@ -1,6 +1,8 @@
 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 # dlops
-add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
+if(DL_KERNELS)
+  add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
+endif()
 
 # xdlops
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index 9df256c38d..dca60b0e73 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -1,21 +1,25 @@
 add_custom_target(example_grouped_gemm_xdl)
-
-add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
-add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
-add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
-add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
-add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
-add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
-
-
-add_dependencies(example_grouped_gemm_xdl
-                 example_grouped_gemm_xdl_fp32
-                 example_grouped_gemm_xdl_fp16
-                 example_grouped_gemm_xdl_bfp16
-                 example_grouped_gemm_xdl_int8
-                 example_grouped_gemm_multiple_d_dl_fp16
-                 example_grouped_gemm_xdl_splitk_fp16)
-
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+  add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
+  add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
+  add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
+  add_dependencies(example_grouped_gemm_xdl 
+                   example_grouped_gemm_xdl_fp16
+                   example_grouped_gemm_multiple_d_dl_fp16
+                   example_grouped_gemm_xdl_splitk_fp16)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bfp16)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+  add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
+endif()
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
diff --git a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
index a42b427c61..00786d34a3 100644
--- a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+++ b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
@@ -6,33 +6,33 @@ foreach(gpu IN LISTS GPU_TARGETS)
    add_custom_target(example_gemm_reduce_xdl_max)
    add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
    add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
-
-   add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
-   add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
-   add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
-   add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
-
-   add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
-
-   add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
-   add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
-   add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
-   add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
-
-   add_dependencies(example_gemm_reduce_xdl_max
-                 example_gemm_max_xdl_bf16
-                 example_gemm_max_xdl_fp16
-                 example_gemm_max_xdl_fp32
-                 example_gemm_max_xdl_int8)
-
-   add_dependencies(example_gemm_reduce_xdl_mean_meansquare
-                 example_gemm_mean_meansquare_xdl_fp16
-                 example_gemm_mean_meansquare_xdl_fp32
-                 example_gemm_mean_meansquare_xdl_bf16
-                 example_gemm_add_addsquare_xdl_int8)
-
-   add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
-
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
+    add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
+    add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
+    add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
+    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
+    add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
+    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
+    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
+   endif()
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
+    add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
+    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
+    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
+    add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
+    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
+    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
+   endif()
+   
    add_dependencies(example_gemm_reduce_xdl
                  example_gemm_reduce_xdl_mean_meansquare
                  example_gemm_reduce_xdl_max
diff --git a/example/17_convnd_bwd_data/CMakeLists.txt b/example/17_convnd_bwd_data/CMakeLists.txt
index 8ab9f37bb3..e187bd4337 100644
--- a/example/17_convnd_bwd_data/CMakeLists.txt
+++ b/example/17_convnd_bwd_data/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -7,5 +8,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
-target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
+  if(DL_KERNELS)
+    add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
+    target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
+  endif()
+endif()
diff --git a/example/18_batched_gemm_reduce/CMakeLists.txt b/example/18_batched_gemm_reduce/CMakeLists.txt
index 94ed129dc0..a1bb398af0 100644
--- a/example/18_batched_gemm_reduce/CMakeLists.txt
+++ b/example/18_batched_gemm_reduce/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -6,3 +7,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
+endif()
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index db170decce..d649567ed2 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -3,18 +3,22 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_weight)
-
-   add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
-   add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
-
-   add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16
-                                                 example_grouped_conv_bwd_weight_xdl_bf16)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
+    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
+    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
+   endif()
    set(target 1)
  endif()
 endforeach()
 
-add_custom_target(example_grouped_conv_bwd_weight_dl)
-
-add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
-
-add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  if(DL_KERNELS)
+    add_custom_target(example_grouped_conv_bwd_weight_dl)
+    add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
+    add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
+  endif()
+endif()
\ No newline at end of file
diff --git a/example/21_gemm_layernorm/CMakeLists.txt b/example/21_gemm_layernorm/CMakeLists.txt
index ff870e7c6b..6a6735efd6 100644
--- a/example/21_gemm_layernorm/CMakeLists.txt
+++ b/example/21_gemm_layernorm/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -9,3 +10,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
+endif()
diff --git a/example/22_cgemm/CMakeLists.txt b/example/22_cgemm/CMakeLists.txt
index 1564561156..854f07fda6 100644
--- a/example/22_cgemm/CMakeLists.txt
+++ b/example/22_cgemm/CMakeLists.txt
@@ -1,16 +1,21 @@
 add_custom_target(example_cgemm_xdl)
 
-add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
-add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
+  add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
+  add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
 add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp)
-add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
-
-add_dependencies(example_cgemm_xdl 
-                 example_cgemm_xdl_bf16
-                 example_cgemm_xdl_fp16
-                 example_cgemm_xdl_fp32
-                 example_cgemm_xdl_int8)
-
+add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+  add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
+  add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
+endif()
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp)
   add_dependencies(example_cgemm_xdl example_cgemm_xdl_int4)
diff --git a/example/24_batched_gemm/CMakeLists.txt b/example/24_batched_gemm/CMakeLists.txt
index 7962576e87..48a3b58ff5 100644
--- a/example/24_batched_gemm/CMakeLists.txt
+++ b/example/24_batched_gemm/CMakeLists.txt
@@ -1,16 +1,20 @@
 add_custom_target(example_batched_gemm_xdl)
-
-add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
-add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
-add_example_executable(example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp)
-add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
-
-add_dependencies(example_batched_gemm_xdl
-                 example_batched_gemm_xdl_fp32
-                 example_batched_gemm_xdl_fp16
-                 example_batched_gemm_xdl_bfp16
-                 example_batched_gemm_xdl_int8)
-
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+  add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
+  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
+  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+  add_example_executable(example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp)
+  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bfp16)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+  add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
+  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8)
+endif()
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp)
   add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
diff --git a/example/25_gemm_bias_e_permute/CMakeLists.txt b/example/25_gemm_bias_e_permute/CMakeLists.txt
index cbc3c007bc..eb274b2338 100644
--- a/example/25_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/25_gemm_bias_e_permute/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp)
-add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp)
+    add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp)
+endif()
diff --git a/example/26_contraction/CMakeLists.txt b/example/26_contraction/CMakeLists.txt
index c58751f0dd..6cab88b13e 100644
--- a/example/26_contraction/CMakeLists.txt
+++ b/example/26_contraction/CMakeLists.txt
@@ -1,5 +1,8 @@
-add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
-add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
-
-add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
-add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
+    add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
+endif()
+if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+    add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
+    add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
+endif()
diff --git a/example/27_layernorm/CMakeLists.txt b/example/27_layernorm/CMakeLists.txt
index 94c23ce774..9cb2cd0766 100644
--- a/example/27_layernorm/CMakeLists.txt
+++ b/example/27_layernorm/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
-add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
+    add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
+endif()
diff --git a/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt b/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
index 44ab16894c..2fda1f62a9 100644
--- a/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
@@ -1 +1,3 @@
-add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
+endif()
diff --git a/example/29_batched_gemm_bias_e_permute/CMakeLists.txt b/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
index 32a87dd200..09c3e6c608 100644
--- a/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
@@ -1,5 +1,7 @@
-add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
 
-if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
-    add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
+    if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+        add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
+    endif()
 endif()
diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index 1c07538b09..e37413c098 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -5,23 +5,29 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
    add_custom_target(example_grouped_conv_fwd_multiple_d)
-
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
-
-   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
-   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
-   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
-   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
-
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
+      add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
+   endif()
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
+      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
+   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
       add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
    endif() # USE_BITINT_EXTENSION_INT4
-   add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
-   add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
+
    set(target 1)
  endif()
 endforeach()
@@ -29,8 +35,12 @@ endforeach()
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
+   endif()
    set(target 1)
  endif()
 endforeach()
diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index 83989bcd97..2074520f8c 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -3,10 +3,15 @@ list(APPEND gpu_list2 gfx908 gfx90a)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
-   add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
-   add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
-   add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
-
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
+   endif()
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
+   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
@@ -15,5 +20,7 @@ foreach(gpu IN LISTS GPU_TARGETS)
 endforeach()
 
 if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
-   add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
+   endif()
 endif()
diff --git a/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt b/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
index 8d9aaec85a..0463bf6bd3 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
+++ b/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
@@ -1,16 +1,24 @@
-add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
-add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
-add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
-add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
-add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
-add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
-add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
+    add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
+    add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
+    add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+    add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
+    add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
+endif()
 
 add_custom_target(example_gemm_scale_softmax_gemm)
-add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
-add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
-add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
-add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
-add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
-add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
-add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
+    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
+    add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
+    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
+    add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
+    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
+endif()
diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index 57ac33fc9a..fa66a5a520 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -3,17 +3,22 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_splitK_gemm_xdl)
-   add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
-   add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
-   add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
-   add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
-
-   add_dependencies(example_splitK_gemm_xdl
-                 example_splitK_gemm_xdl_fp32
-                 example_splitK_gemm_xdl_fp16
-                 example_splitK_gemm_xdl_bfp16
-                 example_splitK_gemm_xdl_int8)
-
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
+    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
+    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
+    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bfp16)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
+    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
+   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
       add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
diff --git a/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt b/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
index a9be3a7108..36bb5720d5 100644
--- a/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
+++ b/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
@@ -1 +1,3 @@
-add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
+endif()
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index de48093ac2..3821f8aaca 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -10,4 +11,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
    add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
    set(target 1)
  endif()
-endforeach()
\ No newline at end of file
+endforeach()
+endif()
diff --git a/example/39_permute/CMakeLists.txt b/example/39_permute/CMakeLists.txt
index 573ad7239e..5b43de9725 100644
--- a/example/39_permute/CMakeLists.txt
+++ b/example/39_permute/CMakeLists.txt
@@ -1,9 +1,11 @@
-add_custom_target(example_permute)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_custom_target(example_permute)
 
-add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
-add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
-add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
+    add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
+    add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
+    add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
 
-add_dependencies(example_permute example_permute_1xHxW_fp16)
-add_dependencies(example_permute example_permute_NxHxW_fp16)
-add_dependencies(example_permute example_permute_HxWx4_fp16)
+    add_dependencies(example_permute example_permute_1xHxW_fp16)
+    add_dependencies(example_permute example_permute_NxHxW_fp16)
+    add_dependencies(example_permute example_permute_HxWx4_fp16)
+endif()
diff --git a/example/40_conv2d_fwd_quantization/CMakeLists.txt b/example/40_conv2d_fwd_quantization/CMakeLists.txt
index 1962b6013b..55464957ac 100644
--- a/example/40_conv2d_fwd_quantization/CMakeLists.txt
+++ b/example/40_conv2d_fwd_quantization/CMakeLists.txt
@@ -10,21 +10,19 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-# Conv perlayer quantization
-add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
 
-# Conv perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
-
-# Conv + bias + relu perlayer quantization
-add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
-
-# Conv + bias + relu perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
-
-# Conv + bias + tanh perlayer quantization
-add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
-
-# Conv + bias + tanh perchannel quantization
-add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
+  if(DL_KERNELS)
+  # Conv perlayer quantization
+  add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
+  # Conv perchannel quantization
+  add_example_executable(example_conv2d_fwd_dl_perchannel_quantization_int8 conv2d_fwd_dl_perchannel_quantization_int8.cpp)
+  # Conv + bias + relu perlayer quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_relu_perlayer_quantization_int8 conv2d_fwd_dl_bias_relu_perlayer_quantization_int8.cpp)
+  # Conv + bias + relu perchannel quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_relu_perchannel_quantization_int8 conv2d_fwd_dl_bias_relu_perchannel_quantization_int8.cpp)
+  # Conv + bias + tanh perlayer quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
+  # Conv + bias + tanh perchannel quantization
+  add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
+  endif()
 endif()
\ No newline at end of file
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index ae251e88d2..085d90a9e9 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -3,9 +3,15 @@ list(APPEND gpu_list2 gfx908 gfx90a)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
+   endif()
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
+   endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
@@ -14,5 +20,7 @@ foreach(gpu IN LISTS GPU_TARGETS)
 endforeach()
 
 if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
-   add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
+   endif()
 endif()
diff --git a/example/42_groupnorm/CMakeLists.txt b/example/42_groupnorm/CMakeLists.txt
index e8c306ac58..bc2246a2bf 100644
--- a/example/42_groupnorm/CMakeLists.txt
+++ b/example/42_groupnorm/CMakeLists.txt
@@ -1,3 +1,5 @@
-add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
-add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
-add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
+    add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
+    add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
+endif()
diff --git a/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt b/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt
index c29f18f162..7e070f5357 100644
--- a/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt
@@ -1,2 +1,6 @@
-add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
-add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
+endif()
diff --git a/example/44_elementwise_permute/CMakeLists.txt b/example/44_elementwise_permute/CMakeLists.txt
index 0e0091a986..877a820315 100644
--- a/example/44_elementwise_permute/CMakeLists.txt
+++ b/example/44_elementwise_permute/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
-add_example_executable(example_elementwise_permute_4D_fp16_2d elementwise_permute_4D_fp16_2d.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
+    add_example_executable(example_elementwise_permute_4D_fp16_2d elementwise_permute_4D_fp16_2d.cpp)
+endif()
diff --git a/example/46_gemm_add_multiply/CMakeLists.txt b/example/46_gemm_add_multiply/CMakeLists.txt
index bfe057e8da..cf7d81f895 100644
--- a/example/46_gemm_add_multiply/CMakeLists.txt
+++ b/example/46_gemm_add_multiply/CMakeLists.txt
@@ -1,2 +1,6 @@
-add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
-add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    if(DL_KERNELS)
+        add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
+    endif()
+    add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
+endif()
diff --git a/example/48_pool3d_fwd/CMakeLists.txt b/example/48_pool3d_fwd/CMakeLists.txt
index 5d58d6a0b1..f821f2c97b 100644
--- a/example/48_pool3d_fwd/CMakeLists.txt
+++ b/example/48_pool3d_fwd/CMakeLists.txt
@@ -1,2 +1,3 @@
-add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
-
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
+endif()
diff --git a/example/49_maxpool2d_bwd/CMakeLists.txt b/example/49_maxpool2d_bwd/CMakeLists.txt
index b29cf9ccbc..fe98b7c997 100644
--- a/example/49_maxpool2d_bwd/CMakeLists.txt
+++ b/example/49_maxpool2d_bwd/CMakeLists.txt
@@ -1,3 +1,9 @@
-add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
-add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
-add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
+endif()
diff --git a/example/50_put_element/CMakeLists.txt b/example/50_put_element/CMakeLists.txt
index 1b0020ebcf..eca410008f 100644
--- a/example/50_put_element/CMakeLists.txt
+++ b/example/50_put_element/CMakeLists.txt
@@ -1 +1,3 @@
-add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
+endif()
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
index c3c8c0e5a7..a89358538e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __bf16__
 void add_device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
@@ -36,7 +36,8 @@ void add_device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __fp16__
 void add_device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -56,7 +57,8 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __fp32__
 void add_device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -76,7 +78,8 @@ void add_device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __int8__
 void add_device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemm<Col,
                                                   Row,
@@ -120,7 +123,7 @@ void add_device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
 template <typename ALayout,
           typename BLayout,
           typename CLayout,
@@ -151,7 +154,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp32__
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<CDataType, float>)
         {
@@ -176,8 +179,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
                 add_device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
-                          is_same_v<CDataType, half_t>)
+#endif
+#ifdef __fp16__
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<CDataType, half_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -200,8 +205,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
                 add_device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, bhalf_t> && is_same_v<BDataType, bhalf_t> &&
-                          is_same_v<CDataType, bhalf_t>)
+#endif
+#ifdef __bf16__
+        if constexpr(is_same_v<ADataType, bhalf_t> && is_same_v<BDataType, bhalf_t> &&
+                     is_same_v<CDataType, bhalf_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -224,8 +231,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
                 add_device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
-                          is_same_v<CDataType, int8_t>)
+#endif
+#ifdef __int8__
+        if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
+                     is_same_v<CDataType, int8_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -248,7 +257,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
                 add_device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
index 73f6004257..7f326879f3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
@@ -14,7 +14,7 @@
 
 using CDE0ElementOp = ck::tensor_operation::element_wise::AddRelu;
 using CDE1ElementOp = ck::tensor_operation::element_wise::Add;
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -137,3 +137,4 @@ struct DeviceOperationInstanceFactory<
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
index 70bff27899..0b2a0c729b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
@@ -13,7 +13,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -91,3 +91,4 @@ struct DeviceOperationInstanceFactory<
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
index 33653a3080..bfe221af7b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp16__
 void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -58,7 +58,8 @@ void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_
                                                             PassThrough,
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
-
+#endif
+#ifdef __bf16__
 void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -100,7 +101,7 @@ void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf
                                                             PassThrough,
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
-
+#endif
 template <typename ADataType,
           typename B0DataType,
           typename B1DataType,
@@ -147,7 +148,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<B0DataType, half_t> &&
                      is_same_v<B1DataType, half_t> && is_same_v<CDataType, half_t> &&
                      Acc0BiasDataType::Size() == 1 &&
@@ -164,6 +165,8 @@ struct DeviceOperationInstanceFactory<
                     op_ptrs);
             }
         }
+#endif
+#ifdef __bf16__
         else if constexpr(is_same_v<ADataType, BF16> && is_same_v<B0DataType, BF16> &&
                           is_same_v<B1DataType, BF16> && is_same_v<CDataType, BF16> &&
                           Acc0BiasDataType::Size() == 1 &&
@@ -180,6 +183,7 @@ struct DeviceOperationInstanceFactory<
                     op_ptrs);
             }
         }
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
index 28ccf61a3a..e57a4206d5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp16__
 void add_device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance(
     std::vector<std::unique_ptr<DeviceBatchedGemmGemm<Row,
                                                       Col,
@@ -111,3 +111,4 @@ struct DeviceOperationInstanceFactory<
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
index 2d7aaf8729..4cb81e8770 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
@@ -19,7 +19,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp16__
 void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
                                                         Row,
@@ -123,7 +123,8 @@ void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instan
                                                         PassThrough,
                                                         PassThrough,
                                                         PassThrough>>>& instances);
-
+#endif
+#ifdef __int8__
 void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
                                                         Row,
@@ -227,7 +228,7 @@ void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instances
                                                         PassThrough,
                                                         PassThrough,
                                                         PassThrough>>>& instances);
-
+#endif
 template <typename ALayout,
           typename BLayout,
           typename ELayout,
@@ -262,7 +263,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<EDataType, half_t>)
         {
@@ -295,6 +296,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
                     op_ptrs);
             }
         }
+#endif
+#ifdef __int8__
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<EDataType, int8_t>)
         {
@@ -327,7 +330,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
                     op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
index 0a30c210dd..c6df0d2441 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -119,3 +119,4 @@ struct DeviceOperationInstanceFactory<
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
index 2ff64675c9..c1d7bf9002 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp16__
 void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -58,7 +58,8 @@ void add_device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_g
                                                             PassThrough,
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
-
+#endif
+#ifdef __bf16__
 void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -100,6 +101,7 @@ void add_device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf
                                                             PassThrough,
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
+#endif
 
 template <typename ADataType,
           typename B0DataType,
@@ -146,7 +148,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<B0DataType, half_t> &&
                      is_same_v<B1DataType, half_t> && is_same_v<CDataType, half_t>)
         {
@@ -161,6 +163,8 @@ struct DeviceOperationInstanceFactory<
                     op_ptrs);
             }
         }
+#endif
+#ifdef __bf16__
         else if constexpr(is_same_v<ADataType, BF16> && is_same_v<B0DataType, BF16> &&
                           is_same_v<B1DataType, BF16> && is_same_v<CDataType, BF16>)
         {
@@ -175,6 +179,7 @@ struct DeviceOperationInstanceFactory<
                     op_ptrs);
             }
         }
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index 2ed8255f6d..da391feef3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp32__
 // float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -65,7 +65,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            PassThrough,
                                                            PassThrough,
                                                            Bilinear>>>& instances);
-
+#endif
+#ifdef __fp64__
 // double
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -114,7 +115,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn
                                                            PassThrough,
                                                            PassThrough,
                                                            Bilinear>>>& instances);
-
+#endif
 // Contraction + Bilinear
 template <index_t NumDimM,
           index_t NumDimN,
@@ -149,7 +150,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp32__
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<DDataType, float> && is_same_v<EDataType, float>)
         {
@@ -165,7 +166,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                     op_ptrs);
             }
         }
-
+#endif
+#ifdef __fp64__
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<DDataType, double> && is_same_v<EDataType, double>)
         {
@@ -181,7 +183,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                     op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index 5d9567731d..6b11523392 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp32__
 // float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -65,7 +65,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            PassThrough,
                                                            PassThrough,
                                                            Scale>>>& instances);
-
+#endif
+#ifdef __fp64__
 // double
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -114,7 +115,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instanc
                                                            PassThrough,
                                                            PassThrough,
                                                            Scale>>>& instances);
-
+#endif
 // Contraction + Scale
 template <index_t NumDimM,
           index_t NumDimN,
@@ -148,7 +149,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp32__
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<EDataType, float>)
         {
@@ -164,7 +165,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                     op_ptrs);
             }
         }
-
+#endif
+#ifdef __fp64__
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<EDataType, double>)
         {
@@ -180,7 +182,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                     op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index 628fb5e479..8ec9832896 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __bf16__
 // conv1d backward data
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
@@ -29,16 +29,19 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef __fp16__
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __fp32__
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
+#endif
 #ifdef __int8__
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
@@ -52,6 +55,7 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
+#ifdef __bf16__
 // conv2d backward data
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -64,7 +68,8 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef __fp16__
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -76,7 +81,8 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef __fp32__
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -88,6 +94,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
+#endif
 #ifdef __int8__
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -101,6 +108,8 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
+#ifdef DL_KERNELS
+#ifdef __fp16__
 // conv2d dl
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -113,7 +122,8 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef __fp32__
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -125,6 +135,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
+#endif
 #ifdef __int8__
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -138,6 +149,8 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
+#endif
+#ifdef __bf16__
 // conv3d backward data
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
@@ -150,7 +163,8 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef __fp16__
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -162,7 +176,8 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef __fp32__
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -174,6 +189,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough,
                                                   PassThrough>>>& instances);
+#endif
 #ifdef __int8__
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
@@ -229,20 +245,23 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#ifdef __fp16__
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef __bf16__
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(op_ptrs);
             }
+#endif
 #ifdef __int8__
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(op_ptrs);
             }
@@ -255,26 +274,35 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
                          is_same_v<OutDataType, float>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#ifdef __fp16__
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef __bf16__
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
+#endif
 #ifdef __int8__
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(op_ptrs);
+#endif
             }
 #endif
         }
@@ -286,20 +314,23 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#ifdef __fp16__
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef __bf16__
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(op_ptrs);
             }
+#endif
 #ifdef __int8__
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(op_ptrs);
             }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
index dd8c1987cf..af4c2aee66 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
@@ -18,11 +18,17 @@ namespace device {
 namespace instance {
 
 // conv2d forward
+#ifdef __fp16__
 void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(
+    std::vector<std::unique_ptr<
+        DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+#endif
+#ifdef __bf16__
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvFwd<2,
                                               NHWC,
@@ -34,17 +40,14 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
                                               PassThrough,
                                               PassThrough,
                                               PassThrough>>>& instances);
-
-void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(
-    std::vector<std::unique_ptr<
-        DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances);
-
+#endif
+#ifdef __fp32__
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef __int8__
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvFwd<2,
                                               NHWC,
@@ -56,6 +59,7 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
                                               PassThrough,
                                               PassThrough,
                                               PassThrough>>>& instances);
+#endif
 
 template <ck::index_t NumDimSpatial,
           typename InLayout,
@@ -99,23 +103,29 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
             }
+#ifdef __fp16__
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
                 add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
             }
+#endif
+#ifdef __bf16__
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&
                               is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
+#endif
+#ifdef __int8__
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(op_ptrs);
             }
+#endif
         }
 
         return op_ptrs;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
index a1c006cf62..b863243c1b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -77,3 +77,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 56a74f7770..538ca758f7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -343,6 +343,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(op_ptrs);
             }
         }
+#ifdef __fp16__
         else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -388,6 +389,8 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+#endif
+#ifdef __bf16__
         else if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
                           is_same_v<CDataType, ck::bhalf_t>)
         {
@@ -412,6 +415,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instances(op_ptrs);
             }
         }
+#endif
 #ifdef __int8__
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<CDataType, int8_t>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
index de21f325a6..4b11a9b262 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
@@ -9,7 +9,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -170,3 +170,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index f80efd5511..09dc40e3ec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -144,3 +144,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
index 09b1c2190f..c30ba4a1f5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
@@ -10,7 +10,7 @@
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -136,3 +136,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
index c40b607cc2..963a18cb44 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp16__
 void add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmStreamK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -119,3 +119,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmSt
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
index b482e97ee0..0cd2a262ec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
@@ -10,7 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -192,3 +192,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
index 778f625d84..21d3b88799 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef __fp16__
 // FP16
 void add_device_normalization_rank_2_1_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 2, 1>>>&);
@@ -26,7 +26,8 @@ void add_device_normalization_rank_4_3_f16_instances(
 
 void add_device_normalization_rank_5_3_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 5, 3>>>&);
-
+#endif
+#ifdef __fp32__
 // FP32
 void add_device_normalization_rank_2_1_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, PassThrough, 2, 1>>>&);
@@ -36,7 +37,7 @@ void add_device_normalization_rank_4_3_f32_instances(
 
 void add_device_normalization_rank_5_3_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, PassThrough, 5, 3>>>&);
-
+#endif
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
@@ -65,7 +66,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
                      is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
         {
@@ -82,8 +83,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
                 add_device_normalization_rank_5_3_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
-                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
+#endif
+#ifdef __fp32__
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
+                     is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
         {
             if constexpr(Rank == 2 && NumReduceDim == 1)
             {
@@ -98,7 +101,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
                 add_device_normalization_rank_5_3_f32_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
index ccb5cb5a9e..ebdc5f9308 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
@@ -22,7 +22,7 @@ static constexpr auto WindowRank = 2;
 
 static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
 static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-
+#ifdef __fp16__
 // FP16
 void add_device_pool2d_fwd_nhwc_f16_instances(
     std::vector<
@@ -36,7 +36,8 @@ void add_device_pool2d_fwd_nhwc_f16_instances(
 void add_device_pool2d_fwd_nhwc_index_f16_instances(
     std::vector<
         std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
-
+#endif
+#ifdef __fp32__
 // FP32
 void add_device_pool2d_fwd_nhwc_f32_instances(
     std::vector<
@@ -50,7 +51,7 @@ void add_device_pool2d_fwd_nhwc_f32_instances(
 void add_device_pool2d_fwd_nhwc_index_f32_instances(
     std::vector<
         std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
-
+#endif
 template <typename InDataType,
           typename OutDataType,
           typename IndexDataType,
@@ -75,7 +76,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
                      is_same_v<IndexDataType, I32>)
         {
@@ -88,8 +89,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 add_device_pool2d_fwd_nhwc_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                          is_same_v<IndexDataType, I32>)
+#endif
+#ifdef __fp32__
+        if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                     is_same_v<IndexDataType, I32>)
         {
             if constexpr(OutputIndex && ReduceOpId == MaxOp)
             {
@@ -100,7 +103,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 add_device_pool2d_fwd_nhwc_f32_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
index 3a006b00a5..b4b4c140ec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -22,7 +22,7 @@ static constexpr auto WindowRank = 3;
 
 static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
 static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-
+#ifdef __fp16__
 // FP16
 void add_device_pool3d_fwd_ndhwc_f16_instances(
     std::vector<
@@ -36,7 +36,8 @@ void add_device_pool3d_fwd_ndhwc_f16_instances(
 void add_device_pool3d_fwd_ndhwc_index_f16_instances(
     std::vector<
         std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
-
+#endif
+#ifdef __fp32__
 // FP32
 void add_device_pool3d_fwd_ndhwc_f32_instances(
     std::vector<
@@ -50,7 +51,7 @@ void add_device_pool3d_fwd_ndhwc_f32_instances(
 void add_device_pool3d_fwd_ndhwc_index_f32_instances(
     std::vector<
         std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
-
+#endif
 template <typename InDataType,
           typename OutDataType,
           typename IndexDataType,
@@ -75,7 +76,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
                      is_same_v<IndexDataType, I32>)
         {
@@ -88,8 +89,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                          is_same_v<IndexDataType, I32>)
+#endif
+#ifdef __fp32__
+        if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                     is_same_v<IndexDataType, I32>)
         {
             if constexpr(OutputIndex && ReduceOpId == MaxOp)
             {
@@ -100,7 +103,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
index 2ed4b0d5f8..04d184f981 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // Layout(A, B, C) = [Col, Row, Row]
 void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
@@ -76,7 +76,7 @@ void add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
                                                     PassThrough,
                                                     Activation_Mul_Clamp<PassThrough>>>>&
         instances);
-
+#endif
 // Layout(A, B, C) = [Col, Row, Row]
 void add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
@@ -181,7 +181,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instances(op_ptrs);
                 }
             }
@@ -190,7 +192,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(op_ptrs);
                 }
             }
@@ -199,7 +203,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(op_ptrs);
                 }
             }
@@ -208,7 +214,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
+#endif
                     add_device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instances(op_ptrs);
                 }
             }
@@ -222,3 +230,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
\ No newline at end of file
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
index 8a96f6707e..733ac516fb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(
     std::vector<
@@ -64,7 +64,7 @@ void add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(
                                                       PassThrough,
                                                       Add_Mul2_Activation_Mul_Clamp<TanH>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -163,12 +163,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_perchannel_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_relu_perchannel_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -229,7 +233,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, TanH>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_tanh_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_tanh_perchannel_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -243,3 +249,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
index e17bea5fdf..e487c5fab9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(
     std::vector<
@@ -63,7 +63,7 @@ void add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(
                                                               PassThrough,
                                                               Add_Mul_Activation_Mul_Clamp<TanH>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(
     std::vector<
         std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -161,12 +161,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_perlayer_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_relu_perlayer_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -227,7 +231,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, TanH>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_bias_tanh_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_bias_tanh_perlayer_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -241,3 +247,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
index 9236b5c79e..a883b2b935 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -47,7 +47,7 @@ void add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(
                                                               PassThrough,
                                                               Activation_Mul2_Clamp<Relu>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_perchannel_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -128,12 +128,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_perchannel_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_relu_perchannel_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_relu_perchannel_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -147,3 +151,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
index 7c1eb4e429..0a71571ba7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef DL_KERNELS
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_conv2d_dl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -47,7 +47,7 @@ void add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(
                                                               PassThrough,
                                                               Activation_Mul_Clamp<Relu>>>>&
         instances);
-
+#endif
 void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -125,12 +125,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 if constexpr(is_same_v<Activation, PassThrough>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_perlayer_quantization_int8_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<Activation, Relu>)
                 {
+#ifdef DL_KERNELS
                     add_device_conv2d_dl_relu_perlayer_quantization_int8_instances(op_ptrs);
+#endif
                     add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(op_ptrs);
                 }
             }
@@ -144,3 +148,4 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
index 26815f1447..f03eb36d31 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
@@ -40,7 +40,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftma
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef __fp16__
         if constexpr(std::is_same_v<InDataType, F16> && std::is_same_v<AccDataType, F32> &&
                      std::is_same_v<OutDataType, F16>)
         {
@@ -65,8 +65,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftma
                     add_device_softmax_f16_f16_rank4_reduce4_instances(op_ptrs);
             }
         }
-        else if constexpr(std::is_same_v<InDataType, F32> && std::is_same_v<AccDataType, F32> &&
-                          std::is_same_v<OutDataType, F32>)
+#endif
+#ifdef __fp32__
+        if constexpr(std::is_same_v<InDataType, F32> && std::is_same_v<AccDataType, F32> &&
+                     std::is_same_v<OutDataType, F32>)
         {
             if constexpr(Rank == 3)
             {
@@ -89,7 +91,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftma
                     add_device_softmax_f32_f32_rank4_reduce4_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt
index 0f2a739199..5b342595d8 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt
@@ -1,18 +1,26 @@
-add_instance_library(device_batched_gemm_instance
-   device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
-   device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
-   device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp
-)
+set(BATCHED_GEMM_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp)
+endif()
+add_instance_library(device_batched_gemm_instance ${BATCHED_GEMM_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt
index d0e9b265af..6710035ec6 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt
@@ -1,4 +1,6 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_add_relu_gemm_add_instance
     device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
     device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
 )
+endif()
\ No newline at end of file
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
index cd9c95c066..b0f37e68fd 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
@@ -1,4 +1,5 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_bias_permute_instance
     device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
 )
-
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
index 865a31e79a..cdb1a53901 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
@@ -1,4 +1,6 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_gemm_instance
     device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
     device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
index db3719cff8..728e35fc3b 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
@@ -1,7 +1,8 @@
+if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_reduce_instance
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
 )
-
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
index 29fce56610..5ac55655d4 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -1,4 +1,5 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_softmax_gemm_instance
     device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
 )
-
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index eba248e599..d5110e4505 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -1,7 +1,11 @@
-add_instance_library(device_batched_gemm_softmax_gemm_permute_instance
-    device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
-    device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
-    device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
-    device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
-)
+set(DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp)
+    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp)
+    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp)
+endif()
+add_instance_library(device_batched_gemm_softmax_gemm_permute_instance ${DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES})
 
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
index d2a0a3d0fb..1db6985f61 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
@@ -1,13 +1,17 @@
-add_instance_library(device_contraction_bilinear_instance
+set(DEVICE_CONTRACTION_BILINEAR_INSTANCES)
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
     #float
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
+    list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
     #double
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
-    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
-)
+    list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
+                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
+endif()
+add_instance_library(device_contraction_bilinear_instance ${DEVICE_CONTRACTION_BILINEAR_INSTANCES})
 
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
index 31f6a0fcdc..aed8bef2a8 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
@@ -1,13 +1,17 @@
-add_instance_library(device_contraction_scale_instance
+set(DEVICE_CONTRACTION_SCALE_INSTANCES)
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
     #float
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
+    list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
+                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
+                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
+                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
     #double
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
-    device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
-)
+    list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
+                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
+                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
+                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp)
+endif()
+add_instance_library(device_contraction_scale_instance ${DEVICE_CONTRACTION_SCALE_INSTANCES})
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
index 582eab3064..54ef9cc7a6 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
@@ -1,17 +1,23 @@
 set(CONV2D_BWD_DATA_INSTANCES)
 if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
    list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp)
+   if(DL_KERNELS)
+      list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp)
+   endif()
 endif()
 if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
    list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp)
 endif()
 if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp)
    list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp)
+   if(DL_KERNELS)
+      list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp)
+   endif()
 endif()
 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
    list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp)
+   if(DL_KERNELS)
+      list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp)
+   endif()
 endif()
 add_instance_library(device_conv2d_bwd_data_instance ${CONV2D_BWD_DATA_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
index fcb858728e..6ca909c35e 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -9,7 +9,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef DL_KERNELS
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -81,3 +81,4 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
index 2baa6ae061..d263e98851 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -9,7 +9,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef DL_KERNELS
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -81,3 +81,4 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
index 28867fe19d..bc949e757c 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -9,7 +9,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef DL_KERNELS
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -81,3 +81,4 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index 40656e382b..ca8ceb2fa1 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __bf16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -155,3 +155,4 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
index 5b646852fc..96ecc95658 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
@@ -1,7 +1,16 @@
-add_instance_library(device_conv2d_fwd_instance
-   device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
-   device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
-   device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
-   device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
-   device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
-)
+set(DEVICE_CONV2D_FWD_INSTANCES)
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp)
+   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp)
+endif()
+if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp)
+endif()
+   
+add_instance_library(device_conv2d_fwd_instance ${DEVICE_CONV2D_FWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index 78e9c893c9..8eb423f6f1 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __bf16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -126,3 +126,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
index 4663a9ac32..93d3cfe57b 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __fp16__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -118,3 +118,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
index 0b1df52f8c..204612d877 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __fp32__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -117,3 +117,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 9969a8bbc9..25bec9f8a6 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
+#ifdef __int8__
 namespace ck {
 namespace tensor_operation {
 namespace device {
@@ -123,3 +123,4 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
index 0c7cc2cd31..ea53b82be5 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
@@ -1,3 +1,5 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_elementwise_normalization_instance
     device_elementwise_normalization_f16_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
index bbf81a5fa2..9028829fe9 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
@@ -1,6 +1,8 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_add_fastgelu_instance
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
index 0beb10e379..1085966807 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
@@ -1,6 +1,8 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_fastgelu_instance
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
index 97693a2566..6079f90130 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
@@ -1,6 +1,8 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_relu_add_layernorm_instance
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
index cb1b3a486f..b9f1210b2e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
@@ -1,6 +1,8 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_bilinear_instance
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
index 17d27ab150..772373dcb0 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
@@ -1,6 +1,8 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_fastgelu_instance
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
index 8dd0112a6b..d9baf3f062 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_streamk_instance
    # device_gemm_xdl_streamk_f32_f32_f32_mk_kn_mn_instance.cpp
    # device_gemm_xdl_streamk_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -8,3 +9,4 @@ add_instance_library(device_gemm_streamk_instance
    # device_gemm_xdl_streamk_f16_f16_f16_km_kn_mn_instance.cpp
    # device_gemm_xdl_streamk_f16_f16_f16_km_nk_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
index b973b70aac..f1553e3d5b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -8,3 +9,4 @@ add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
    device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
index 648f2146cb..a45bf36399 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
@@ -1,6 +1,8 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_grouped_gemm_fastgelu_instance
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
 )
+endif()
diff --git a/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
index 176fb2fbee..4892a9c183 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
@@ -1,11 +1,15 @@
-add_instance_library(device_normalization_instance
-    device_layernorm2d_f16_instance.cpp
-    device_layernorm2d_f32_instance.cpp
+set(DEVICE_NORMALIZATION_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f16_instance.cpp
     device_layernorm4d_f16_instance.cpp
-    device_layernorm4d_f32_instance.cpp
     device_groupnorm_f16_instance.cpp
-    device_groupnorm_f32_instance.cpp
     device_groupnorm_swish_f16_instance.cpp
-    device_groupnorm_swish_f32_instance.cpp
-    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
-)
+    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f32_instance.cpp
+    device_layernorm4d_f32_instance.cpp
+    device_groupnorm_f32_instance.cpp
+    device_groupnorm_swish_f32_instance.cpp)
+endif()
+add_instance_library(device_normalization_instance ${DEVICE_NORMALIZATION_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
index 0d0f896c8d..ba2a8991bd 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
@@ -1,10 +1,14 @@
-add_instance_library(device_pool_fwd_instance
-    device_avg_pool2d_fwd_nhwc_f16_instance.cpp
-    device_avg_pool2d_fwd_nhwc_f32_instance.cpp
-    device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
-    device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
-    device_max_pool2d_fwd_nhwc_f16_instance.cpp
-    device_max_pool2d_fwd_nhwc_f32_instance.cpp
-    device_max_pool3d_fwd_ndhwc_f16_instance.cpp
-    device_max_pool3d_fwd_ndhwc_f32_instance.cpp
-)
+set(DEVICE_POOL_FWD_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+                                          device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+                                          device_max_pool2d_fwd_nhwc_f16_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+                                          device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+                                          device_max_pool2d_fwd_nhwc_f32_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_f32_instance.cpp)
+endif()
+add_instance_library(device_pool_fwd_instance ${DEVICE_POOL_FWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
index 3b4c8935a4..00eb6ff1c1 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
@@ -1,34 +1,26 @@
 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-set(CONV2D_PERLAYER_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
-)
-
-set(CONV2D_PERCHANNEL_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp
-)
-
-set(CONV2D_BIAS_PERLAYER_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp
-)
-
-set(CONV2D_BIAS_PERCHANNEL_QUANT_SRC
-    conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp
-    conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp
-)
 
+set(CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp)
+set(CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp)
+set(CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp)
+set(CONV2D_BIAS_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_xdl_bias_perchannel_quantization_int8_instance.cpp)
 set(GEMM_QUANT_SRC
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
-    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
 )
+if(DL_KERNELS)
+    list(APPEND CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp)
+    list(APPEND CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp)
+    list(APPEND CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp)
+    list(APPEND CONV2D_BIAS_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp)
+    list(APPEND GEMM_QUANT_SRC
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
+endif()
 
 add_instance_library(device_quantization_instance
     ${CONV2D_PERLAYER_QUANT_SRC}
diff --git a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
index 202ad12b97..ba0197477f 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
@@ -1,16 +1,20 @@
-add_instance_library(device_softmax_instance
-    device_softmax_f16_f16_instance_rank3_reduce1.cpp
+set(DEVICE_SOFTMAX_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_SOFTMAX_INSTANCES device_softmax_f16_f16_instance_rank3_reduce1.cpp
     device_softmax_f16_f16_instance_rank3_reduce2.cpp
     device_softmax_f16_f16_instance_rank3_reduce3.cpp
     device_softmax_f16_f16_instance_rank4_reduce1.cpp
     device_softmax_f16_f16_instance_rank4_reduce2.cpp
     device_softmax_f16_f16_instance_rank4_reduce3.cpp
-    device_softmax_f16_f16_instance_rank4_reduce4.cpp
-    device_softmax_f32_f32_instance_rank3_reduce1.cpp
+    device_softmax_f16_f16_instance_rank4_reduce4.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_SOFTMAX_INSTANCES device_softmax_f32_f32_instance_rank3_reduce1.cpp
     device_softmax_f32_f32_instance_rank3_reduce2.cpp
     device_softmax_f32_f32_instance_rank3_reduce3.cpp
     device_softmax_f32_f32_instance_rank4_reduce1.cpp
     device_softmax_f32_f32_instance_rank4_reduce2.cpp
     device_softmax_f32_f32_instance_rank4_reduce3.cpp
-    device_softmax_f32_f32_instance_rank4_reduce4.cpp
-)
+    device_softmax_f32_f32_instance_rank4_reduce4.cpp)
+endif()
+add_instance_library(device_softmax_instance ${DEVICE_SOFTMAX_INSTANCES})
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 27b7c1c891..a4ea5c399c 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -3,20 +3,11 @@ set(PROFILER_SOURCES
     profiler.cpp
     profile_gemm.cpp
     profile_gemm_splitk.cpp
-    profile_gemm_streamk.cpp
-    profile_gemm_bilinear.cpp
     profile_gemm_bias_add_reduce.cpp
-    profile_gemm_add_add_fastgelu.cpp
     profile_gemm_add_multiply.cpp
-    profile_gemm_add_fastgelu.cpp
-    profile_gemm_add_relu_add_layernorm.cpp
-    profile_gemm_fastgelu.cpp
     profile_gemm_reduce.cpp
     profile_batched_gemm.cpp
-    profile_batched_gemm_gemm.cpp
-    profile_batched_gemm_add_relu_gemm_add.cpp
     profile_batched_gemm_reduce.cpp
-    profile_grouped_gemm.cpp
     profile_conv_fwd.cpp
     profile_conv_fwd_bias_relu.cpp
     profile_conv_fwd_bias_relu_add.cpp
@@ -32,7 +23,6 @@ set(PROFILER_SOURCES
     profile_batchnorm_fwd.cpp
     profile_batchnorm_bwd.cpp
     profile_batchnorm_infer.cpp
-    profile_grouped_gemm_fastgelu.cpp
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
     profile_grouped_conv_bwd_data.cpp
@@ -40,6 +30,18 @@ set(PROFILER_SOURCES
 if(DL_KERNELS)
   list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
 endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  list(APPEND PROFILER_SOURCES profile_batched_gemm_gemm.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_fastgelu.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_streamk.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_bilinear.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_add_fastgelu.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_add_add_fastgelu.cpp)
+  list(APPEND PROFILER_SOURCES profile_gemm_add_relu_add_layernorm.cpp)
+  list(APPEND PROFILER_SOURCES profile_batched_gemm_add_relu_gemm_add.cpp)
+  list(APPEND PROFILER_SOURCES profile_grouped_gemm.cpp)
+  list(APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp)
+endif()
 
 set(PROFILER_EXECUTABLE ckProfiler)
 
@@ -49,20 +51,11 @@ target_compile_options(${PROFILER_EXECUTABLE} PRIVATE -Wno-global-constructors)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_reduce_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_conv2d_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv1d_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_fwd_instance)
@@ -79,13 +72,24 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instan
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 if(DL_KERNELS)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 endif()
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bilinear_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_add_fastgelu_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_streamk_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_fastgelu_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_gemm_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_add_relu_gemm_add_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_gemm_fastgelu_instance)
+endif()
 rocm_install(TARGETS ${PROFILER_EXECUTABLE} COMPONENT profiler)
diff --git a/profiler/src/profile_grouped_gemm.cpp b/profiler/src/profile_grouped_gemm.cpp
index d023db54de..1d4707a6c8 100644
--- a/profiler/src/profile_grouped_gemm.cpp
+++ b/profiler/src/profile_grouped_gemm.cpp
@@ -88,7 +88,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     const auto StrideBs = argToIntArray(argv[12]);
     const auto StrideCs = argToIntArray(argv[13]);
     const int kbatch    = argc == 15 ? std::stoi(argv[14]) : 1;
-
+#ifdef __fp16__
     if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         ck::profiler::profile_grouped_gemm_impl<ck::half_t,
@@ -173,7 +173,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     {
         throw std::runtime_error("wrong! this GEMM data_type & layout is not implemented");
     }
-
+#endif
     return 0;
 }
 
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index ad08e94702..cd1f305313 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -60,6 +60,6 @@ add_subdirectory(contraction)
 add_subdirectory(pool_fwd)
 add_subdirectory(batched_gemm_multi_d)
 add_subdirectory(grouped_convnd_bwd_data)
-if(GPU_TARGETS MATCHES "gfx1100")
+if(GPU_TARGETS MATCHES "gfx11")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/batched_gemm/CMakeLists.txt b/test/batched_gemm/CMakeLists.txt
index 12142c33e7..8a3269e90f 100644
--- a/test/batched_gemm/CMakeLists.txt
+++ b/test/batched_gemm/CMakeLists.txt
@@ -2,21 +2,26 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
-   target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
-   target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
-
-   add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
-   target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
-   target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
-
-   add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
-   target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
-   target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
-
-   add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
-   target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
-   target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
+      target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
+      target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
+   endif()
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+      add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
+      target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
+      target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+      add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
+      target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
+      target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
+   endif()
+   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+      add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
+      target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
+      target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+   endif()
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_gemm/CMakeLists.txt b/test/batched_gemm_gemm/CMakeLists.txt
index 858efd8379..404e74f3bc 100644
--- a/test/batched_gemm_gemm/CMakeLists.txt
+++ b/test/batched_gemm_gemm/CMakeLists.txt
@@ -2,10 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(test_batched_gemm_gemm)
-   add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
-   target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
-   add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
-   set(target 1)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+      add_custom_target(test_batched_gemm_gemm)
+      add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
+      target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
+      add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
+      set(target 1)
+   endif()
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_multi_d/CMakeLists.txt b/test/batched_gemm_multi_d/CMakeLists.txt
index 102a5b8f27..825f0dd235 100644
--- a/test/batched_gemm_multi_d/CMakeLists.txt
+++ b/test/batched_gemm_multi_d/CMakeLists.txt
@@ -1,5 +1,4 @@
-# TODO: Enable for gfx90a after complier fix
 if(DL_KERNELS)
-  add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
-  target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
+    add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
+    target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
 endif()
diff --git a/test/batched_gemm_reduce/CMakeLists.txt b/test/batched_gemm_reduce/CMakeLists.txt
index 0710f4647a..af95a50eab 100644
--- a/test/batched_gemm_reduce/CMakeLists.txt
+++ b/test/batched_gemm_reduce/CMakeLists.txt
@@ -2,9 +2,11 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
-   target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
-   target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
-   set(target 1)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+     add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
+     target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
+     target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
+     set(target 1)
+   endif()
  endif()
 endforeach()
diff --git a/test/batched_gemm_softmax_gemm/CMakeLists.txt b/test/batched_gemm_softmax_gemm/CMakeLists.txt
index 984cc3c160..c49175a2eb 100644
--- a/test/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -2,10 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(test_batched_gemm_softmax_gemm)
-   add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
-   target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
-   add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
-   set(target 1)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+     add_custom_target(test_batched_gemm_softmax_gemm)
+     add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
+     target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
+     add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
+     set(target 1)
+   endif()
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index 4d04be3faa..ae28daa80e 100644
--- a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -2,21 +2,25 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(test_batched_gemm_softmax_gemm_permute)
-
-   add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
-   add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
-   target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-   target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
-   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
-
-   add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
-   add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
-   target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-   target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
-   add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
+   if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+     add_custom_target(test_batched_gemm_softmax_gemm_permute)
+   endif()
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+     add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
+     add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
+     target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+     target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
+     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
+   endif()
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+     add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
+     add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
+     target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+     target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
+     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
+   endif()
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/elementwise_normalization/CMakeLists.txt b/test/elementwise_normalization/CMakeLists.txt
index a20eb26325..74a3e4999e 100644
--- a/test/elementwise_normalization/CMakeLists.txt
+++ b/test/elementwise_normalization/CMakeLists.txt
@@ -1,7 +1,6 @@
-add_custom_target(test_elementwise_normalization)
-
-add_gtest_executable(test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp)
-
-target_link_libraries(test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance)
-
-add_dependencies(test_elementwise_normalization test_elementwise_layernorm_fp16)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_custom_target(test_elementwise_normalization)
+  add_gtest_executable(test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp)
+  target_link_libraries(test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance)
+  add_dependencies(test_elementwise_normalization test_elementwise_layernorm_fp16)
+endif()
\ No newline at end of file
diff --git a/test/gemm_layernorm/CMakeLists.txt b/test/gemm_layernorm/CMakeLists.txt
index 56b8d7737a..ba0a99b67f 100644
--- a/test/gemm_layernorm/CMakeLists.txt
+++ b/test/gemm_layernorm/CMakeLists.txt
@@ -2,10 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
+  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
    add_custom_target(test_gemm_layernorm)
    add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
    target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
    add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
    set(target 1)
+  endif()
  endif()
 endforeach()
diff --git a/test/gemm_reduce/CMakeLists.txt b/test/gemm_reduce/CMakeLists.txt
index 349f892c19..43c8d60745 100644
--- a/test/gemm_reduce/CMakeLists.txt
+++ b/test/gemm_reduce/CMakeLists.txt
@@ -1,3 +1,5 @@
-add_test_executable(test_gemm_reduce_fp16 gemm_reduce_fp16.cpp)
-target_link_libraries(test_gemm_reduce_fp16 PRIVATE utility)
-target_link_libraries(test_gemm_reduce_fp16 PRIVATE device_gemm_reduce_instance)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_test_executable(test_gemm_reduce_fp16 gemm_reduce_fp16.cpp)
+  target_link_libraries(test_gemm_reduce_fp16 PRIVATE utility)
+  target_link_libraries(test_gemm_reduce_fp16 PRIVATE device_gemm_reduce_instance)
+endif()
\ No newline at end of file
diff --git a/test/grouped_gemm/CMakeLists.txt b/test/grouped_gemm/CMakeLists.txt
index 8c57b667e2..476d953ed8 100644
--- a/test/grouped_gemm/CMakeLists.txt
+++ b/test/grouped_gemm/CMakeLists.txt
@@ -1,3 +1,4 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -12,3 +13,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
+endif()
diff --git a/test/normalization/CMakeLists.txt b/test/normalization/CMakeLists.txt
index a5d7fb2982..2beda4dd74 100644
--- a/test/normalization/CMakeLists.txt
+++ b/test/normalization/CMakeLists.txt
@@ -1,16 +1,19 @@
-add_custom_target(test_normalization)
-
-add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
-add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
-add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
-add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
-
-target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
-target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
-target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
-target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
-
-add_dependencies(test_normalization test_layernorm2d_fp32)
-add_dependencies(test_normalization test_layernorm2d_fp16)
-add_dependencies(test_normalization test_groupnorm_fp16)
-add_dependencies(test_normalization test_groupnorm_fp32)
+if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+  add_custom_target(test_normalization)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+  add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
+  add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
+  target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
+  target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
+  add_dependencies(test_normalization test_layernorm2d_fp32)
+  add_dependencies(test_normalization test_groupnorm_fp32)
+endif()
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
+  add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
+  target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
+  target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
+  add_dependencies(test_normalization test_layernorm2d_fp16)
+  add_dependencies(test_normalization test_groupnorm_fp16)
+endif()
diff --git a/test/pool_fwd/test_avg_pool2d_fwd.cpp b/test/pool_fwd/test_avg_pool2d_fwd.cpp
index 72749fd6e1..31f4a166f2 100644
--- a/test/pool_fwd/test_avg_pool2d_fwd.cpp
+++ b/test/pool_fwd/test_avg_pool2d_fwd.cpp
@@ -41,8 +41,12 @@ class TestAvgPool2dFwd : public ::testing::Test
     }
 };
 
+#ifdef __fp16__
 using KernelTypes =
     ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
+#else
+using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
+#endif
 
 TYPED_TEST_SUITE(TestAvgPool2dFwd, KernelTypes);
 TYPED_TEST(TestAvgPool2dFwd, Test_Pool)
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool_fwd/test_avg_pool3d_fwd.cpp
index 00cc3740fa..a99a3c698e 100644
--- a/test/pool_fwd/test_avg_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_avg_pool3d_fwd.cpp
@@ -40,10 +40,12 @@ class TestAvgPool3dFwd : public ::testing::Test
         }
     }
 };
-
+#ifdef __fp16__
 using KernelTypes =
     ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
-
+#else
+using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
+#endif
 TYPED_TEST_SUITE(TestAvgPool3dFwd, KernelTypes);
 TYPED_TEST(TestAvgPool3dFwd, Test_Pool)
 {
diff --git a/test/pool_fwd/test_max_pool2d_fwd.cpp b/test/pool_fwd/test_max_pool2d_fwd.cpp
index 1cf1314f46..c9050848d2 100644
--- a/test/pool_fwd/test_max_pool2d_fwd.cpp
+++ b/test/pool_fwd/test_max_pool2d_fwd.cpp
@@ -59,10 +59,12 @@ class TestMaxPool2dFwd : public ::testing::Test
         }
     }
 };
-
+#ifdef __fp16__
 using KernelTypes =
     ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
-
+#else
+using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
+#endif
 TYPED_TEST_SUITE(TestMaxPool2dFwd, KernelTypes);
 TYPED_TEST(TestMaxPool2dFwd, Test_Pool)
 {
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool_fwd/test_max_pool3d_fwd.cpp
index 0b0de4d90e..7c0aada473 100644
--- a/test/pool_fwd/test_max_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_max_pool3d_fwd.cpp
@@ -60,8 +60,12 @@ class TestMaxPool3dFwd : public ::testing::Test
     }
 };
 
+#ifdef __fp16__
 using KernelTypes =
-    ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
+    ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
+#else
+using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
+#endif
 
 TYPED_TEST_SUITE(TestMaxPool3dFwd, KernelTypes);
 TYPED_TEST(TestMaxPool3dFwd, Test_Pool)
diff --git a/test/softmax/test_softmax_rank3.cpp b/test/softmax/test_softmax_rank3.cpp
index 43ae11bf1f..f17672286f 100644
--- a/test/softmax/test_softmax_rank3.cpp
+++ b/test/softmax/test_softmax_rank3.cpp
@@ -10,8 +10,9 @@
 
 template <ck::index_t N>
 using I = ck::Number<N>;
-
+#ifdef __fp16__
 using F16 = ck::half_t;
+#endif
 using F32 = float;
 
 template <typename Tuple>
@@ -22,7 +23,9 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 // clang-format off
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
+#ifdef __fp16__
     std::tuple<       F16,         F32,         F16,    I<3>>,
+#endif
     std::tuple<       F32,         F32,         F32,    I<3>>
     >;
 // clang-format on
diff --git a/test/softmax/test_softmax_rank4.cpp b/test/softmax/test_softmax_rank4.cpp
index 5cf96bbaa8..161aebf617 100644
--- a/test/softmax/test_softmax_rank4.cpp
+++ b/test/softmax/test_softmax_rank4.cpp
@@ -10,8 +10,9 @@
 
 template <ck::index_t N>
 using I = ck::Number<N>;
-
+#ifdef __fp16__
 using F16 = ck::half_t;
+#endif
 using F32 = float;
 
 template <typename Tuple>
@@ -22,7 +23,9 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 // clang-format off
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
+#ifdef __fp16__
     std::tuple<       F16,         F32,         F16,    I<4>>,
+#endif
     std::tuple<       F32,         F32,         F32,    I<4>>
     >;
 // clang-format on

From 6802611334a25d35173b87e5f8386d2ad14714f6 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 8 Aug 2023 17:58:31 -0700
Subject: [PATCH 081/194] add no-offload-uniform-block flag for rocm5.7 and up
 (#838)

* add -fno-offload-uniform-block flag for rocm5.7 and up

* add a comment and compiler ticket number
---
 CMakeLists.txt | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7fab38cd27..27602cde86 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -89,6 +89,14 @@ else()
     message("Building CK for the following targets: ${AMDGPU_TARGETS}")
 endif()
 find_package(hip)
+# No assumption that HIP kernels are launched with uniform block size for backward compatibility
+# SWDEV-413293 and https://reviews.llvm.org/D155213
+math(EXPR hip_VERSION_FLAT "(${hip_VERSION_MAJOR} * 1000 + ${hip_VERSION_MINOR}) * 100000 + ${hip_VERSION_PATCH}")
+message("hip_version_flat=${hip_VERSION_FLAT}")
+if(${hip_VERSION_FLAT} GREATER 500700000)
+   message("Adding the fno-offload-uniform-block compiler flag")
+   add_compile_options(-fno-offload-uniform-block)
+endif()
 
 option(USE_BITINT_EXTENSION_INT4, "Whether to enable clang's BitInt extension to provide int4 data type." OFF)
 option(USE_OPT_NAVI3X, "Whether to enable LDS cumode and Wavefront32 mode for NAVI3X silicons." OFF)

From 9c54eaab04e6db605dc86f1d1ab16bd04f51fc89 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Wed, 9 Aug 2023 08:44:23 -0500
Subject: [PATCH 082/194] Enable f16/f8 mixed precision mode (#820)

* Enable f16/f8 mixed precision

* Add an argument to enable mixed precision

* Update for compatibility

* Add mixed precision example

* Introduce ComputeType argument
---
 example/01_gemm/CMakeLists.txt                |  3 ++
 example/01_gemm/gemm_xdl_fp16_f8.cpp          | 41 ++++++++++++++++
 .../impl/device_cgemm_4gemm_xdl_cshuffle.hpp  | 17 +++++--
 .../device/impl/device_gemm_xdl_cshuffle.hpp  |  9 ++--
 .../grid/gridwise_gemm_xdl_cshuffle_v1.hpp    | 48 +++++++++++--------
 5 files changed, 89 insertions(+), 29 deletions(-)
 create mode 100644 example/01_gemm/gemm_xdl_fp16_f8.cpp

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 7ce866931c..f8b886ebbf 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -64,3 +64,6 @@ if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
     add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
   endif()
 endif()
+
+add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
+add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
diff --git a/example/01_gemm/gemm_xdl_fp16_f8.cpp b/example/01_gemm/gemm_xdl_fp16_f8.cpp
new file mode 100644
index 0000000000..d3cf3d397a
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_fp16_f8.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::half_t;
+using CDataType        = ck::half_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto LoopSched   = ck::make_default_loop_scheduler();
+static constexpr auto PipelineVer = ck::PipelineVersion::v1;
+using ComputeType                 = ck::half_t;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|      Loop|    Pipeline|  ComputeType|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector| Scheduler|     Version|             |
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|          |            |             |
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |          |            |             |
+        < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,  LoopSched, PipelineVer, ComputeType>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
index 106746716d..a095521161 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_cgemm_4gemm_xdl_cshuffle.hpp
@@ -123,7 +123,8 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
         ALayout,
         BLayout,
         CLayout,
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
         GemmAccDataType,
         CShuffleDataType,
         CDataType,
@@ -284,8 +285,11 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
 
             if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
             {
-                const auto kernel =
-                    kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, ADataType, CDataType, true>;
+                const auto kernel = kernel_gemm_xdl_cshuffle_v1<GridwiseGemm,
+                                                                ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                true>;
 
                 ave_time += launch_and_time_kernel(stream_config,
                                                    kernel,
@@ -357,8 +361,11 @@ struct DeviceCGemm_4Gemm_Xdl_CShuffle
             }
             else
             {
-                const auto kernel =
-                    kernel_gemm_xdl_cshuffle_v1<GridwiseGemm, ADataType, CDataType, false>;
+                const auto kernel = kernel_gemm_xdl_cshuffle_v1<GridwiseGemm,
+                                                                ADataType,
+                                                                BDataType,
+                                                                CDataType,
+                                                                false>;
 
                 ave_time += launch_and_time_kernel(stream_config,
                                                    kernel,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index 4c9ddbab80..86bc7b5bbb 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -65,7 +65,8 @@ template <typename ALayout,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched     = make_default_loop_scheduler(),
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          PipelineVersion PipelineVer = PipelineVersion::v1,
+          typename ComputeType        = CDataType>
 struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
                                                    BLayout,
                                                    CLayout,
@@ -87,7 +88,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         ALayout,
         BLayout,
         CLayout,
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
         GemmAccDataType,
         CShuffleDataType,
         CDataType,
@@ -128,7 +130,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched,
-        PipelineVer>;
+        PipelineVer,
+        ComputeType>;
 
     using Argument = typename GridwiseGemm::Argument;
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index 7f1fb21d33..29cbf14a57 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -35,13 +35,17 @@ __global__ void
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
-template <typename GridwiseGemm, typename FloatAB, typename FloatC, bool HasMainKBlockLoop>
+template <typename GridwiseGemm,
+          typename FloatA,
+          typename FloatB,
+          typename FloatC,
+          bool HasMainKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdl_cshuffle_v1(const FloatAB* __restrict__ p_a_grid,
-                                    const FloatAB* __restrict__ p_b_grid,
+        kernel_gemm_xdl_cshuffle_v1(const FloatA* __restrict__ p_a_grid,
+                                    const FloatB* __restrict__ p_b_grid,
                                     FloatC* __restrict__ p_c_grid,
                                     typename GridwiseGemm::Problem problem)
 {
@@ -61,7 +65,8 @@ __global__ void
 template <typename ALayout,
           typename BLayout,
           typename CLayout,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatGemmAcc,
           typename FloatCShuffle,
           typename FloatC,
@@ -102,7 +107,8 @@ template <typename ALayout,
           typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched,
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          PipelineVersion PipelineVer = PipelineVersion::v1,
+          typename ComputeType        = FloatC>
 struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
 {
     static constexpr auto I0 = Number<0>{};
@@ -463,8 +469,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     // Argument
     struct Argument : public tensor_operation::device::BaseArgument, public Problem
     {
-        __host__ Argument(const FloatAB* p_a_grid_,
-                          const FloatAB* p_b_grid_,
+        __host__ Argument(const FloatA* p_a_grid_,
+                          const FloatB* p_b_grid_,
                           FloatC* p_c_grid_,
                           index_t M_,
                           index_t N_,
@@ -479,8 +485,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         {
         }
 
-        const FloatAB* p_a_grid;
-        const FloatAB* p_b_grid;
+        const FloatA* p_a_grid;
+        const FloatB* p_b_grid;
         FloatC* p_c_grid;
     };
 
@@ -541,8 +547,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         constexpr auto c_block_size =
             c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
 
-        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
-                             sizeof(FloatAB),
+        return math::max((a_block_space_size_aligned * sizeof(ComputeType) +
+                          b_block_space_size_aligned * sizeof(ComputeType)),
                          c_block_size * sizeof(FloatCShuffle));
     }
 
@@ -676,8 +682,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     using Block2CTileMap = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>;
 
     template <bool HasMainKBlockLoop>
-    __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
-                               const FloatAB* __restrict__ p_b_grid,
+    __device__ static void Run(const FloatA* __restrict__ p_a_grid,
+                               const FloatB* __restrict__ p_b_grid,
                                FloatC* __restrict__ p_c_grid,
                                void* __restrict__ p_shared,
                                const Problem& problem)
@@ -743,8 +749,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 Sequence<AK0Number, MPerBlock, AK1Number>,
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
-                                                FloatAB,
-                                                FloatAB,
+                                                FloatA,
+                                                ComputeType,
                                                 decltype(a_grid_desc_ak0_m_ak1),
                                                 decltype(a_block_desc_ak0_m_ak1),
                                                 ABlockTransferSrcAccessOrder,
@@ -774,8 +780,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 Sequence<BK0Number, NPerBlock, BK1Number>,
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
-                                                FloatAB,
-                                                FloatAB,
+                                                FloatB,
+                                                ComputeType,
                                                 decltype(b_grid_desc_bk0_n_bk1),
                                                 decltype(b_block_desc_bk0_n_bk1),
                                                 BBlockTransferSrcAccessOrder,
@@ -805,11 +811,11 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         // sanity check
         constexpr index_t KPack =
             math::max(math::lcm(AK1Number, BK1Number),
-                      MfmaSelector<FloatAB, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+                      MfmaSelector<ComputeType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            FloatAB,
+            ComputeType,
             FloatGemmAcc,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -827,10 +833,10 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<FloatAB*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+            static_cast<ComputeType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<FloatAB*>(p_shared) + a_block_space_size_aligned,
+            static_cast<ComputeType*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1Number, 0, 0);

From 472fa029ba274a325ffe4c9dd3abedba78f5cabd Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 9 Aug 2023 17:40:55 +0200
Subject: [PATCH 083/194] Enable grouped conv with small K or C (#822)

* Enable grouped conv with small K or C

* Add missing instances

* Refactor grouped conv fwd instances

* Fix fp16 instances since it supports src_per_vec %2 = 0

* Add generic instances
---
 ...e_grouped_conv_bwd_weight_xdl_cshuffle.hpp | 111 ++++++++++-
 ...ice_grouped_conv_bwd_data_xdl_instance.hpp |   3 +
 ...e_grouped_conv_bwd_weight_xdl_instance.hpp |  44 +++--
 .../device_grouped_conv_fwd_dl_instance.hpp}  |  33 +++-
 .../device_grouped_conv_fwd_xdl_instance.hpp  | 180 ++++++++++++++++++
 ...d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp | 119 ++----------
 ...1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp | 119 ++----------
 ...1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp | 118 ++----------
 ...d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp | 115 ++---------
 .../device_grouped_conv2d_fwd_common.hpp      |  53 ------
 ..._fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   2 +-
 ..._fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   2 +-
 ...wd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |  54 +++---
 ...fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  54 +++---
 ...fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |  54 +++---
 ...device_grouped_conv2d_fwd_xdl_instance.hpp | 105 ----------
 ...wd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |  54 +++---
 ...fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  54 +++---
 ...fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |  54 +++---
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp | 121 +++---------
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp | 121 +++---------
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp | 120 +++---------
 ...xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp | 117 +++---------
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp | 121 +++---------
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp | 121 +++---------
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp | 120 +++---------
 ...xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp | 118 +++---------
 .../test_grouped_convnd_bwd_weight.cpp        |  12 ++
 .../grouped_convnd_fwd/grouped_convnd_fwd.cpp |  16 +-
 29 files changed, 788 insertions(+), 1527 deletions(-)
 rename library/{src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp => include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp} (86%)
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
index 5a8bb6188b..71c528e4c9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -378,6 +378,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const index_t GemmM      = K;
         const index_t GemmN      = C * X;
 
+        const auto PadGemmM = (MPerBlock - GemmM % MPerBlock) % MPerBlock;
+        const auto PadGemmN = (NPerBlock - GemmN % NPerBlock) % NPerBlock;
+
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
@@ -496,9 +499,37 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
             const auto wei_gemmm_gemmn_grid_desc =
                 make_naive_tensor_descriptor_packed(make_tuple(K, X * C));
 
-            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+            // Padd
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmKBatch),
+                               make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmM, PadGemmM),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmKBatch),
+                               make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmN, PadGemmN),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto wei_gemmm_gemmn_pad_grid_desc =
+                transform_tensor_descriptor(wei_gemmm_gemmn_grid_desc,
+                                            make_tuple(make_right_pad_transform(GemmM, PadGemmM),
+                                                       make_right_pad_transform(GemmN, PadGemmN)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc,
+                              wei_gemmm_gemmn_pad_grid_desc);
         }
     }
 
@@ -546,6 +577,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const index_t GemmM      = K;
         const index_t GemmN      = C * X * Y;
 
+        const auto PadGemmM = (MPerBlock - GemmM % MPerBlock) % MPerBlock;
+        const auto PadGemmN = (NPerBlock - GemmN % NPerBlock) % NPerBlock;
+
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
@@ -651,9 +685,37 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_grid_desc);
+            // Padd
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmKBatch),
+                               make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmM, PadGemmM),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmKBatch),
+                               make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmN, PadGemmN),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto wei_gemmm_gemmn_pad_grid_desc =
+                transform_tensor_descriptor(wei_grid_desc,
+                                            make_tuple(make_right_pad_transform(GemmM, PadGemmM),
+                                                       make_right_pad_transform(GemmN, PadGemmN)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc,
+                              wei_gemmm_gemmn_pad_grid_desc);
         }
     }
 
@@ -708,6 +770,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         const index_t GemmM      = K;
         const index_t GemmN      = C * Z * X * Y;
 
+        const auto PadGemmM = (MPerBlock - GemmM % MPerBlock) % MPerBlock;
+        const auto PadGemmN = (NPerBlock - GemmN % NPerBlock) % NPerBlock;
+
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
@@ -822,9 +887,37 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
-            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
-                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_grid_desc);
+            // Padd
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmKBatch),
+                               make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmM, PadGemmM),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmKBatch),
+                               make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmN, PadGemmN),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
+
+            const auto wei_gemmm_gemmn_pad_grid_desc =
+                transform_tensor_descriptor(wei_grid_desc,
+                                            make_tuple(make_right_pad_transform(GemmM, PadGemmM),
+                                                       make_right_pad_transform(GemmN, PadGemmN)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc,
+                              wei_gemmm_gemmn_pad_grid_desc);
         }
     } // function end
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
index e18a58b5e1..d05e8e3a87 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -63,6 +63,7 @@ using device_grouped_conv_bwd_data_xdl_f16_instances =
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // TODO: After enable, add instance for small conv.K and conv.C
 #endif
         // clang-format on
         >;
@@ -97,6 +98,7 @@ using device_grouped_conv_bwd_data_xdl_bf16_instances =
             DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
             DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
             DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
+            // TODO: After enable, add instance for small conv.K and conv.C
 #endif
         // clang-format on
         >;
@@ -131,6 +133,7 @@ using device_grouped_conv_bwd_data_xdl_f32_instances =
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        // TODO: After enable, add instance for small conv.K and conv.C
 #endif
         // clang-format on
         >;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
index 0f44952810..6fc91565b9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
@@ -39,10 +39,16 @@ template <ck::index_t NDimSpatial,
           ConvolutionBackwardWeightSpecialization ConvSpec>
 using device_grouped_conv_bwd_weight_xdl_c_shuffle_f32_instances = std::tuple<
     // clang-format off
-        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  4,   32,   32,    2,    2,  S<1, 4, 16, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              1,              4,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              4,      true,           1,           1,   S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  4,   32,   32,    2,    1,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              1,      true,           1,           1,   S<1, 32, 1, 4>,               1>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  4,   32,   32,    1,    2,  S<1, 4, 8,  2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              1,              2,      true,  S<1, 4, 16, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 32, 2>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              2,      true,  S<1, 4, 64, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,   S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              4,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,
@@ -66,10 +72,18 @@ template <ck::index_t NDimSpatial,
           ConvolutionBackwardWeightSpecialization ConvSpec>
 using device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_instances = std::tuple<
     // clang-format off
-        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              2,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              2,              4,      true,           1,           1,   S<1, 16, 1, 4>,               2>,   
+        // instance for small conv.K
+        // for fp16 conv.K and conv.C must be divisible by 2
+        // since half_t atomic_add require scalar_per_x_vector % 2 == 0
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              2,              1,      true,           1,           1,   S<1, 32, 1, 4>,               2>,  
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              2,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8>,
+
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8>,   
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8>,   
@@ -93,10 +107,16 @@ template <ck::index_t NDimSpatial,
           ConvolutionBackwardWeightSpecialization ConvSpec>
 using device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances = std::tuple<
     // clang-format off
-        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|                      ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
-        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                            Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
-        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|                    Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
-        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                                  |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|               ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                     Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|             Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                           |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              1,              4,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              4,      true,           1,           1,   S<1, 16, 1, 4>,               1>,   
+        // instance for small conv.K
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 4,  8>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              1,              1,      true,           1,           1,   S<1, 32, 1, 4>,               1>, 
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              1,              2,      true,   S<1, 4, 8,  2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               4>,
+
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 4>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               4>,
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              2,      true,   S<1, 4, 32, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               4>,   
         DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                   ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>, S<0, 3, 1, 2>,   S<0, 2, 1, 3>,              2,              8,              4,      true,   S<1, 4, 16, 2>,   S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               4>,   
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
similarity index 86%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
rename to library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
index 631c2ce093..ae3bec621c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
@@ -1,14 +1,45 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp"
-#include "device_grouped_conv2d_fwd_common.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 template <typename InLayout,
           typename WeiLayout,
           typename DsLayout,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
new file mode 100644
index 0000000000..23edf35e98
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
@@ -0,0 +1,180 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using Empty_Tuple = ck::Tuple<>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 = ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 = ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = GemmSpecialization::MNKPadding;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_bf16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,  BF16,  BF16,     F32,     BF16,    DsLayout,  BF16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_f32_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F32,   F32,     F32,      F32,    DsLayout,   F32, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
+    // clang-format on
+    >;
+
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_int8_instances = std::tuple<
+    // clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|           |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|           |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout, int8_t, int8_t, int32_t,   int8_t,    DsLayout, int8_t, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index 5aa50adb31..032ebc1eff 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,94 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = ck::bhalf_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// Compilation parameters for in[g, n, wi, c] * wei[g, k, x, c] = out[g, n, wo, k]
-using device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -120,7 +24,26 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<1,
+                                                                              GNWC,
+                                                                              GKXC,
+                                                                              Empty_Tuple,
+                                                                              GNWK,
+                                                                              ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<1,
+                                                                              GNWC,
+                                                                              GKXC,
+                                                                              Empty_Tuple,
+                                                                              GNWK,
+                                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<1,
+                                                                              GNWC,
+                                                                              GKXC,
+                                                                              Empty_Tuple,
+                                                                              GNWK,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index 333b40c71c..3a0ddb736f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,94 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// Compilation parameters for in[g, n, wi, c] * wei[g, k, x, c] = out[g, n, wo, k]
-using device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -120,7 +24,26 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<1,
+                                                                             GNWC,
+                                                                             GKXC,
+                                                                             Empty_Tuple,
+                                                                             GNWK,
+                                                                             ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<1,
+                                                                             GNWC,
+                                                                             GKXC,
+                                                                             Empty_Tuple,
+                                                                             GNWK,
+                                                                             ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<1,
+                                                                             GNWC,
+                                                                             GKXC,
+                                                                             Empty_Tuple,
+                                                                             GNWK,
+                                                                             ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index 506a93ae9b..1db1226b7f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,93 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// Compilation parameters for in[g, n, wi, c] * wei[g, k, x, c] = out[g, n, wo, k]
-using device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -119,7 +24,26 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<1,
+                                                                             GNWC,
+                                                                             GKXC,
+                                                                             Empty_Tuple,
+                                                                             GNWK,
+                                                                             ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<1,
+                                                                             GNWC,
+                                                                             GKXC,
+                                                                             Empty_Tuple,
+                                                                             GNWK,
+                                                                             ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<1,
+                                                                             GNWC,
+                                                                             GKXC,
+                                                                             Empty_Tuple,
+                                                                             GNWK,
+                                                                             ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
index 30084f16d1..6fbf631766 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,90 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNWC = ck::tensor_layout::convolution::GNWC;
-using GKXC = ck::tensor_layout::convolution::GKXC;
-using GNWK = ck::tensor_layout::convolution::GNWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// Compilation parameters for in[g, n, wi, c] * wei[g, k, x, c] = out[g, n, wo, k]
-using device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances = std::tuple<
-    // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-                                                                                                                                                                                                                
-        // Filter1x1Stride1Pad0                                                                                                                                                                                 
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       1,   GNWC,   GKXC, Empty_Tuple,   GNWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -116,7 +24,26 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances{});
+                                   device_grouped_conv_fwd_xdl_int8_instances<1,
+                                                                              GNWC,
+                                                                              GKXC,
+                                                                              Empty_Tuple,
+                                                                              GNWK,
+                                                                              ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<1,
+                                                                              GNWC,
+                                                                              GKXC,
+                                                                              Empty_Tuple,
+                                                                              GNWK,
+                                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<1,
+                                                                              GNWC,
+                                                                              GKXC,
+                                                                              Empty_Tuple,
+                                                                              GNWK,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
deleted file mode 100644
index 85a7a5be25..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_common.hpp
+++ /dev/null
@@ -1,53 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using BF16 = ck::bhalf_t;
-using F16  = ck::half_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NHWGC = ck::tensor_layout::convolution::NHWGC;
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-
-using NHWGK = ck::tensor_layout::convolution::NHWGK;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 5eb881549c..1925989838 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_dl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 4157853c41..95ef4bbe37 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_dl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index bc7d577b69..db14ce4306 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,40 +24,36 @@ void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwdDefault>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              ConvFwdDefault>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwd1x1P0>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              ConvFwd1x1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwd1x1S1P0>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              ConvFwd1x1S1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwdOddC>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index 55bf6da9b9..debdb05b4d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,40 +24,36 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdDefault>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwdDefault>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1P0>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwd1x1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1S1P0>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwd1x1S1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdOddC>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 202cdd6b44..20326f5be1 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,40 +24,36 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdDefault>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwdDefault>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1P0>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwd1x1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1S1P0>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwd1x1S1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdOddC>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
deleted file mode 100644
index 07bea1c03c..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_instance.hpp
+++ /dev/null
@@ -1,105 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "device_grouped_conv2d_fwd_common.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          typename DsDatatype,
-          typename CDEElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-using device_grouped_conv2d_fwd_xdl_f16_instances =
-    std::tuple<
-        // clang-format off
-        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F16,   F16,     F32,      F16, DsDatatype,   F16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          typename DsDatatype,
-          typename CDEElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-using device_grouped_conv2d_fwd_xdl_bf16_instances =
-    std::tuple<
-        // clang-format off
-        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, DsDatatype,  BF16, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          typename DsDatatype,
-          typename CDEElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-using device_grouped_conv2d_fwd_xdl_f32_instances =
-    std::tuple<
-        // clang-format off
-        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData| AccData| CShuffle|         Ds| EData|           A|           B|          CDE|    ConvForward|            GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|    Type| DataType|   DataType|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |        |        |         |        |      |      |        |         |           |      |   Operation|   Operation|    Operation|               |                |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |        |        |         |        |      |      |        |         |           |      |            |            |             |               |                |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,   F32,   F32,     F32,      F32, DsDatatype,   F32, PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
-        // clang-format on
-        >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
index 82edf896db..af68c3b07d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,40 +24,36 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                NHWGK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwdDefault>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              ConvFwdDefault>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                NHWGK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwd1x1P0>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              ConvFwd1x1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                NHWGK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwd1x1S1P0>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              ConvFwd1x1S1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_bf16_instances<NHWGC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                NHWGK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwdOddC>{});
+                                   device_grouped_conv_fwd_xdl_bf16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
index 4bd3236caf..8b1506e0fb 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,40 +24,36 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdDefault>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwdDefault>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1P0>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwd1x1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1S1P0>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwd1x1S1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f16_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdOddC>{});
+                                   device_grouped_conv_fwd_xdl_f16_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
index 4f5bdb2023..c8bdfb8c7c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -2,7 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "device_grouped_conv2d_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -24,40 +24,36 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                               PassThrough>>>& instances)
 {
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdDefault>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwdDefault>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1P0>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwd1x1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1S1P0>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwd1x1S1P0>{});
 
     add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_xdl_f32_instances<NHWGC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               NHWGK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdOddC>{});
+                                   device_grouped_conv_fwd_xdl_f32_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             ConvFwdOddC>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index c3b100ea69..bd8443f5ec 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,94 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = ck::bhalf_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -119,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index ca488e9dc4..fdd5c3169d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,94 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -119,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index 6087b1b187..b486dd80b4 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,93 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -118,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
index fd8c47deb3..de3d866477 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,90 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using GNDHWC = ck::tensor_layout::convolution::GNDHWC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using GNDHWK = ck::tensor_layout::convolution::GNDHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances = std::tuple<
-    // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-                                                                                                                                                                                                                
-        // Filter1x1Stride1Pad0                                                                                                                                                                                 
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, GNDHWC, GKZYXC, Empty_Tuple, GNDHWK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -115,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
index d2cacf4539..b3c4c1d9c9 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,94 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using BF16 = ck::bhalf_t;
-using F32  = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -119,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_bf16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
index 6354f01112..d252c7513f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,94 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F16 = ck::half_t;
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -119,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f16_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
index 381ac5ef08..0d79c1c08a 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -1,15 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -17,93 +9,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32 = float;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances =
-    std::tuple<
-        // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-
-        // Filter1x1Stride1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,              4>
-        // clang-format on
-        >;
-
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -118,8 +23,27 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_f32_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
index 6bd53d8693..881b6df5c5 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
@@ -1,106 +1,13 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NDHWGC = ck::tensor_layout::convolution::NDHWGC;
-using GKZYXC = ck::tensor_layout::convolution::GKZYXC;
-using NDHWGK = ck::tensor_layout::convolution::NDHWGK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho, wo, k]
-using device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances = std::tuple<
-    // clang-format off
-        // Default
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwdDefault, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-
-        // Filter1x1Pad0
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough,   ConvFwd1x1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-                                                                                                                                                                                                                
-        // Filter1x1Stride1Pad0                                                                                                                                                                                 
-        //########################################|  NumDim|      A|      B|          Ds|      E|  AData|  BData| AccData| CShuffle|          Ds|  EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial| Layout| Layout|      Layout| Layout|   Type|   Type|    Type| DataType|    DataType|   Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |       |       |            |       |       |       |        |         |            |       |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<       3, NDHWGC, GKZYXC, Empty_Tuple, NDHWGK, int8_t, int8_t, int32_t,   int8_t, Empty_Tuple, int8_t, PassThrough, PassThrough, PassThrough, ConvFwd1x1S1P0, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
-    // clang-format on
-    >;
-
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -115,8 +22,27 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances)
 {
-    add_device_operation_instances(
-        instances, device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              ConvFwdDefault>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              ConvFwd1x1P0>{});
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_xdl_int8_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              ConvFwd1x1S1P0>{});
 }
 
 } // namespace instance
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
index 86d99c4a53..ae15f6c0c9 100644
--- a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -100,6 +100,9 @@ TYPED_TEST(TestGroupedConvndBwdWeight1d, Test1D)
     this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
     this->conv_params.push_back({1, 2, 32, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
     this->conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 1, 1, 1, 32, {3}, {32}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 1, 64, 3, {3}, {32}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 1, 1, 1, {3}, {32}, {1}, {1}, {1}, {1}});
     this->Run();
 }
 
@@ -112,6 +115,9 @@ TYPED_TEST(TestGroupedConvndBwdWeight2d, Test2D)
         {2, 2, 4, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     this->conv_params.push_back(
         {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     this->Run();
 }
 
@@ -124,5 +130,11 @@ TYPED_TEST(TestGroupedConvndBwdWeight3d, Test3D)
         {3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
     this->conv_params.push_back(
         {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
     this->Run();
 }
diff --git a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
index 4a804ef7f1..c856255ea3 100644
--- a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
+++ b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
@@ -22,6 +22,8 @@ TEST_F(TestGroupedConvNdFwd, GroupedConv1dFwdGNWC)
     conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
     conv_params.push_back({1, 2, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
     conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
+    conv_params.push_back({1, 1, 1, 1, 32, {3}, {32}, {1}, {1}, {1}, {1}});
+    conv_params.push_back({1, 1, 1, 64, 3, {3}, {32}, {1}, {1}, {1}, {1}});
 
     for(auto& param : conv_params)
     {
@@ -96,6 +98,9 @@ TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdGNHWC)
     conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
     conv_params.push_back({2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
 
     for(auto& param : conv_params)
     {
@@ -173,6 +178,12 @@ TEST_F(TestGroupedConvNdFwd, GroupedConv3dFwdGNDHWC)
         {3, 2, 128, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
     conv_params.push_back(
         {3, 2, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    conv_params.push_back(
+        {3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    conv_params.push_back(
+        {3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
 
     for(auto& param : conv_params)
     {
@@ -247,6 +258,9 @@ TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdNHWGC)
     conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
     conv_params.push_back({2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
 
     for(auto& param : conv_params)
     {
@@ -255,7 +269,7 @@ TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdNHWGC)
         // fp16
         pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
                                                            ck::tensor_layout::convolution::NHWGC,
-                                                           ck::tensor_layout::convolution::KYXGC,
+                                                           ck::tensor_layout::convolution::GKYXC,
                                                            ck::tensor_layout::convolution::NHWGK,
                                                            ck::half_t,
                                                            ck::half_t,

From 9af519ee86aacf3929d032cb5a50599569d4d1fa Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 9 Aug 2023 10:30:40 -0700
Subject: [PATCH 084/194] add gfx941 to the ckProfiler package (#840)

---
 Jenkinsfile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 6df9ea3889..1817207093 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -710,8 +710,8 @@ pipeline {
                     }
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')

From 1b7da171c981d640e6204ddc335251bf00971bc8 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 9 Aug 2023 13:44:13 -0700
Subject: [PATCH 085/194] Update the list of contributors. (#836)

* add linting and update contributors list

* skip the linting and doc changes

* add Astha

* add YanXing
---
 CONTRIBUTORS.md | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/CONTRIBUTORS.md b/CONTRIBUTORS.md
index 07d8368817..cdce5a4630 100644
--- a/CONTRIBUTORS.md
+++ b/CONTRIBUTORS.md
@@ -6,9 +6,11 @@ This is the list of developers and contributors to Composable Kernel library
 ## Developers
 [Chao Liu](https://github.com/asroy), [Jing Zhang](https://github.com/zjing14), 2018-2023
 
-[Letao Qin](https://github.com/ltqin), [Qianfeng Zhang](https://github.com/qianfengz), [Liang Huang](https://github.com/carlushuang), [Shaojie Wang](https://github.com/shaojiewang), 2019-2022
+[Letao Qin](https://github.com/ltqin), [Qianfeng Zhang](https://github.com/qianfengz), [Liang Huang](https://github.com/carlushuang), [Shaojie Wang](https://github.com/shaojiewang), 2019-2023
 
-[Anthony Chang](https://github.com/rosenrodt), [Chunyu Lai](https://github.com/rocking5566), [Illia Silin](https://github.com/illsilin), [Adam Osewski](https://github.com/aosewski), [Poyen Chen](https://github.com/poyenc), [Rosty Geyyer](https://github.com/geyyer), 2022
+[Anthony Chang](https://github.com/rosenrodt), [Chunyu Lai](https://github.com/rocking5566), [Illia Silin](https://github.com/illsilin), [Adam Osewski](https://github.com/aosewski), [Poyen Chen](https://github.com/poyenc), [Rosty Geyyer](https://github.com/geyyer), [Astha Rai](https://github.com/arai713), [Shi YanXing](https://github.com/Yanxing-Shi), 2022-2023
+
+[Hari Sadasivan](https://github.com/hsadasiv), [Bartlomiej Kocot](https://github.com/bartekxk), [Bartlomiej Wroblewski](https://github.com/bwroblew), 2023
 
 Hanwen Chang, 2019-2021,
 

From cbbd172fd67d39a48c7684f823354ccec6e1b964 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 9 Aug 2023 13:50:04 -0700
Subject: [PATCH 086/194] Update the rocm version threshold to apply the
 -fno-offload-uniform-block flag. (#839)

* add fno-offload-uniform-block flag for rocm5.7 and up

* add a comment and compiler ticket number

* update the threshold rocm version
---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 27602cde86..6eba503942 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -93,7 +93,7 @@ find_package(hip)
 # SWDEV-413293 and https://reviews.llvm.org/D155213
 math(EXPR hip_VERSION_FLAT "(${hip_VERSION_MAJOR} * 1000 + ${hip_VERSION_MINOR}) * 100000 + ${hip_VERSION_PATCH}")
 message("hip_version_flat=${hip_VERSION_FLAT}")
-if(${hip_VERSION_FLAT} GREATER 500700000)
+if(${hip_VERSION_FLAT} GREATER 500723302)
    message("Adding the fno-offload-uniform-block compiler flag")
    add_compile_options(-fno-offload-uniform-block)
 endif()

From 578142db3a9e2bc273c2a178ca3e550ee79e6050 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Thu, 10 Aug 2023 12:04:35 +0800
Subject: [PATCH 087/194] Average pool backward deviceOP and example (#797)

* Add avgpool bwd reference code

* Refine naming

* Fix invalid in_element op in ref_conv

* Add example (only reference now)

* Add the full example of avgpool bwd

* Fix copyright

* Imitate MakeDescriptor from  transform_conv_bwd_data_to_gemm_v1.hpp

* rename channel to c from k

* Arrange the code

* Imitate the argument from conv bwd

* Implement invoker

* Fix order of parameter in example

* Refactor reference code for different dimension

* Support different stride

* Check if argument is valid

* Fix kernel parameter for NDHWC, fastest dimension C is not reduced

* Add more data type in example

* Fix bug in example

* calculate Do Ho Wo according to the dilation

* Remove useless header

* Add comment in reference code

* Add layout parameter

* Remove layout in derived class

* Refine reference comment
---
 example/51_avgpool3d_bwd/CMakeLists.txt       |   3 +
 .../51_avgpool3d_bwd/avgpool3d_bwd_bf16.cpp   |  62 ++
 .../51_avgpool3d_bwd/avgpool3d_bwd_common.hpp | 147 +++++
 .../51_avgpool3d_bwd/avgpool3d_bwd_fp16.cpp   |  62 ++
 .../51_avgpool3d_bwd/avgpool3d_bwd_fp32.cpp   |  62 ++
 .../gpu/device/device_avgpool_bwd.hpp         |  39 ++
 .../gpu/device/device_put_element.hpp         |   2 +-
 .../impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp | 575 ++++++++++++++++++
 .../gpu/grid/gridwise_put_element_1d.hpp      |   2 +-
 .../cpu/reference_avgpool_bwd.hpp             | 354 +++++++++++
 .../cpu/reference_conv_bwd_data.hpp           |   6 +-
 11 files changed, 1309 insertions(+), 5 deletions(-)
 create mode 100644 example/51_avgpool3d_bwd/CMakeLists.txt
 create mode 100644 example/51_avgpool3d_bwd/avgpool3d_bwd_bf16.cpp
 create mode 100644 example/51_avgpool3d_bwd/avgpool3d_bwd_common.hpp
 create mode 100644 example/51_avgpool3d_bwd/avgpool3d_bwd_fp16.cpp
 create mode 100644 example/51_avgpool3d_bwd/avgpool3d_bwd_fp32.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_avgpool_bwd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp
 create mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp

diff --git a/example/51_avgpool3d_bwd/CMakeLists.txt b/example/51_avgpool3d_bwd/CMakeLists.txt
new file mode 100644
index 0000000000..fef0c66835
--- /dev/null
+++ b/example/51_avgpool3d_bwd/CMakeLists.txt
@@ -0,0 +1,3 @@
+add_example_executable(example_avgpool3d_bwd_bf16 avgpool3d_bwd_bf16.cpp)
+add_example_executable(example_avgpool3d_bwd_fp16 avgpool3d_bwd_fp16.cpp)
+add_example_executable(example_avgpool3d_bwd_fp32 avgpool3d_bwd_fp32.cpp)
diff --git a/example/51_avgpool3d_bwd/avgpool3d_bwd_bf16.cpp b/example/51_avgpool3d_bwd/avgpool3d_bwd_bf16.cpp
new file mode 100644
index 0000000000..a5ab6ef24a
--- /dev/null
+++ b/example/51_avgpool3d_bwd/avgpool3d_bwd_bf16.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp"
+
+#include "avgpool3d_bwd_common.hpp"
+
+using DOutDataType    = ck::bhalf_t;
+using DInDataType     = ck::bhalf_t;
+using ComputeDataType = float;
+
+#if 1
+using DOutLayout = ck::tensor_layout::convolution::NDHWC;
+using DInLayout  = ck::tensor_layout::convolution::NDHWC;
+#else
+using DOutLayout = ck::tensor_layout::convolution::NCDHW;
+using DInLayout  = ck::tensor_layout::convolution::NCDHW;
+#endif
+
+using DevicePoolBwdInstance =
+    ck::tensor_operation::device::DeviceAvgPool3dBwd_NDHWC_NDHWC<DOutDataType,
+                                                                 DInDataType,
+                                                                 ComputeDataType,
+                                                                 64, // BlockSize
+                                                                 64, // ReduceMThreadClusterSize
+                                                                 1,  // ReduceKThreadClusterSize
+                                                                 1,  // ReduceMThreadSliceSize
+                                                                 1,  // ReduceKThreadSliceSize
+                                                                 1>; // InSrcOutDstVectorSize
+
+int main()
+{
+    std::vector<ck::index_t> window_lengths    = {5, 5, 5};
+    std::vector<ck::index_t> window_strides    = {2, 2, 2};
+    std::vector<ck::index_t> window_dilations  = {2, 2, 2};
+    std::vector<ck::index_t> dinput_left_pads  = {0, 0, 0};
+    std::vector<ck::index_t> dinput_right_pads = {0, 0, 0};
+
+    ck::index_t N  = 1;
+    ck::index_t C  = 16;
+    ck::index_t Di = 40;
+    ck::index_t Hi = 40;
+    ck::index_t Wi = 40;
+
+    pool3d_bwd_test<DevicePoolBwdInstance, DOutDataType, DInDataType, DOutLayout, DInLayout>(
+        true,
+        false,
+        N,
+        C,
+        Di,
+        Hi,
+        Wi,
+        window_lengths,
+        window_strides,
+        window_dilations,
+        dinput_left_pads,
+        dinput_right_pads);
+}
diff --git a/example/51_avgpool3d_bwd/avgpool3d_bwd_common.hpp b/example/51_avgpool3d_bwd/avgpool3d_bwd_common.hpp
new file mode 100644
index 0000000000..394f046b1e
--- /dev/null
+++ b/example/51_avgpool3d_bwd/avgpool3d_bwd_common.hpp
@@ -0,0 +1,147 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp"
+
+template <typename TensorLayout>
+std::vector<ck::index_t> f_tensor_strides_ncdhw(ck::index_t N_,
+                                                ck::index_t C_,
+                                                ck::index_t D,
+                                                ck::index_t H,
+                                                ck::index_t W,
+                                                TensorLayout layout)
+{
+    using namespace ck::literals;
+    (void)N_;
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+        return {C_ * D * H * W, D * H * W, H * W, W, 1_uz};
+    else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+        return {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_};
+};
+
+template <typename TensorLayout>
+HostTensorDescriptor f_host_tensor_descriptor(std::size_t N_,
+                                              std::size_t C_,
+                                              std::size_t D,
+                                              std::size_t H,
+                                              std::size_t W,
+                                              TensorLayout layout)
+{
+    using namespace ck::literals;
+
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+    {
+        return HostTensorDescriptor({N_, C_, D, H, W}, {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
+    }
+    else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+    {
+        return HostTensorDescriptor({N_, C_, D, H, W},
+                                    {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+    }
+};
+
+template <typename DevicePoolBwdInstance,
+          typename DOutDataType,
+          typename DInDataType,
+          typename DOutLayout,
+          typename DInLayout>
+bool pool3d_bwd_test(bool do_verification,
+                     bool time_kernel,
+                     ck::index_t N,
+                     ck::index_t C,
+                     ck::index_t Di,
+                     ck::index_t Hi,
+                     ck::index_t Wi,
+                     std::vector<ck::index_t> window_lengths,
+                     std::vector<ck::index_t> window_strides,
+                     std::vector<ck::index_t> window_dilations,
+                     std::vector<ck::index_t> dinput_left_pads,
+                     std::vector<ck::index_t> dinput_right_pads)
+{
+    auto OutSpatialLength = [&](auto InSpatialLength, int index) {
+        ck::index_t left_pad   = dinput_left_pads[index];
+        ck::index_t right_pad  = dinput_right_pads[index];
+        ck::index_t window_len = window_lengths[index];
+        ck::index_t stride     = window_strides[index];
+        ck::index_t dilation   = window_dilations[index];
+        ck::index_t eff        = (window_len - 1) * dilation + 1;
+        return (InSpatialLength + left_pad + right_pad - eff) / stride + 1;
+    };
+
+    ck::index_t Do = OutSpatialLength(Di, 0);
+    ck::index_t Ho = OutSpatialLength(Hi, 1);
+    ck::index_t Wo = OutSpatialLength(Wi, 2);
+
+    Tensor<DOutDataType> dout(f_host_tensor_descriptor(N, C, Do, Ho, Wo, DOutLayout{}));
+    Tensor<DInDataType> din_dev(f_host_tensor_descriptor(N, C, Di, Hi, Wi, DInLayout{}));
+    Tensor<DInDataType> din_host(f_host_tensor_descriptor(N, C, Di, Hi, Wi, DInLayout{}));
+
+    std::cout << "dout: " << dout.mDesc << std::endl;
+    std::cout << "din_host: " << din_host.mDesc << std::endl;
+
+    dout.GenerateTensorValue(GeneratorTensor_3<DOutDataType>{0.0, 1.0});
+
+    DeviceMem dout_device_buf(sizeof(DOutDataType) * dout.mDesc.GetElementSpaceSize());
+    DeviceMem din_device_buf(sizeof(DInDataType) * din_dev.mDesc.GetElementSpaceSize());
+
+    dout_device_buf.ToDevice(dout.mData.data());
+    din_device_buf.SetZero();
+
+    auto pool        = DevicePoolBwdInstance{};
+    auto invoker_ptr = pool.MakeInvokerPointer();
+    auto argument_ptr =
+        pool.MakeArgumentPointer(static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+                                 static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+                                 {N, C, Do, Ho, Wo},
+                                 {N, C, Di, Hi, Wi},
+                                 f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, DOutLayout{}),
+                                 f_tensor_strides_ncdhw(N, C, Di, Hi, Wi, DInLayout{}),
+                                 window_lengths,
+                                 window_strides,
+                                 window_dilations,
+                                 dinput_left_pads,
+                                 dinput_right_pads);
+
+    if(!pool.IsSupportedArgument(argument_ptr.get()))
+    {
+        throw std::runtime_error("wrong! device_op with the specified compilation parameters does "
+                                 "not support this problem");
+    }
+
+    float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+    std::cout << "Perf: " << ave_time << std::endl;
+
+    bool pass = true;
+
+    if(do_verification)
+    {
+        auto ref_pool =
+            ck::tensor_operation::host::ReferenceAvgPoolBwd<3, DInDataType, DOutDataType>();
+
+        auto ref_invoker = ref_pool.MakeInvoker();
+
+        auto ref_argument = ref_pool.MakeArgument(din_host,
+                                                  dout,
+                                                  window_lengths,
+                                                  window_strides,
+                                                  window_dilations,
+                                                  dinput_left_pads,
+                                                  dinput_right_pads);
+
+        ref_invoker.Run(ref_argument);
+
+        din_device_buf.FromDevice(din_dev.mData.data());
+        pass = ck::utils::check_err(din_dev, din_host);
+    }
+
+    return pass;
+}
diff --git a/example/51_avgpool3d_bwd/avgpool3d_bwd_fp16.cpp b/example/51_avgpool3d_bwd/avgpool3d_bwd_fp16.cpp
new file mode 100644
index 0000000000..578f563d5c
--- /dev/null
+++ b/example/51_avgpool3d_bwd/avgpool3d_bwd_fp16.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp"
+
+#include "avgpool3d_bwd_common.hpp"
+
+using DOutDataType    = ck::half_t;
+using DInDataType     = ck::half_t;
+using ComputeDataType = float;
+
+#if 1
+using DOutLayout = ck::tensor_layout::convolution::NDHWC;
+using DInLayout  = ck::tensor_layout::convolution::NDHWC;
+#else
+using DOutLayout = ck::tensor_layout::convolution::NCDHW;
+using DInLayout  = ck::tensor_layout::convolution::NCDHW;
+#endif
+
+using DevicePoolBwdInstance =
+    ck::tensor_operation::device::DeviceAvgPool3dBwd_NDHWC_NDHWC<DOutDataType,
+                                                                 DInDataType,
+                                                                 ComputeDataType,
+                                                                 64, // BlockSize
+                                                                 64, // ReduceMThreadClusterSize
+                                                                 1,  // ReduceKThreadClusterSize
+                                                                 1,  // ReduceMThreadSliceSize
+                                                                 1,  // ReduceKThreadSliceSize
+                                                                 1>; // InSrcOutDstVectorSize
+
+int main()
+{
+    std::vector<ck::index_t> window_lengths    = {5, 5, 5};
+    std::vector<ck::index_t> window_strides    = {2, 2, 2};
+    std::vector<ck::index_t> window_dilations  = {2, 2, 2};
+    std::vector<ck::index_t> dinput_left_pads  = {0, 0, 0};
+    std::vector<ck::index_t> dinput_right_pads = {0, 0, 0};
+
+    ck::index_t N  = 1;
+    ck::index_t C  = 16;
+    ck::index_t Di = 40;
+    ck::index_t Hi = 40;
+    ck::index_t Wi = 40;
+
+    pool3d_bwd_test<DevicePoolBwdInstance, DOutDataType, DInDataType, DOutLayout, DInLayout>(
+        true,
+        false,
+        N,
+        C,
+        Di,
+        Hi,
+        Wi,
+        window_lengths,
+        window_strides,
+        window_dilations,
+        dinput_left_pads,
+        dinput_right_pads);
+}
diff --git a/example/51_avgpool3d_bwd/avgpool3d_bwd_fp32.cpp b/example/51_avgpool3d_bwd/avgpool3d_bwd_fp32.cpp
new file mode 100644
index 0000000000..c2910c5567
--- /dev/null
+++ b/example/51_avgpool3d_bwd/avgpool3d_bwd_fp32.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp"
+
+#include "avgpool3d_bwd_common.hpp"
+
+using DOutDataType    = float;
+using DInDataType     = float;
+using ComputeDataType = float;
+
+#if 1
+using DOutLayout = ck::tensor_layout::convolution::NDHWC;
+using DInLayout  = ck::tensor_layout::convolution::NDHWC;
+#else
+using DOutLayout = ck::tensor_layout::convolution::NCDHW;
+using DInLayout  = ck::tensor_layout::convolution::NCDHW;
+#endif
+
+using DevicePoolBwdInstance =
+    ck::tensor_operation::device::DeviceAvgPool3dBwd_NDHWC_NDHWC<DOutDataType,
+                                                                 DInDataType,
+                                                                 ComputeDataType,
+                                                                 64, // BlockSize
+                                                                 64, // ReduceMThreadClusterSize
+                                                                 1,  // ReduceKThreadClusterSize
+                                                                 1,  // ReduceMThreadSliceSize
+                                                                 1,  // ReduceKThreadSliceSize
+                                                                 1>; // InSrcOutDstVectorSize
+
+int main()
+{
+    std::vector<ck::index_t> window_lengths    = {5, 5, 5};
+    std::vector<ck::index_t> window_strides    = {2, 2, 2};
+    std::vector<ck::index_t> window_dilations  = {2, 2, 2};
+    std::vector<ck::index_t> dinput_left_pads  = {0, 0, 0};
+    std::vector<ck::index_t> dinput_right_pads = {0, 0, 0};
+
+    ck::index_t N  = 1;
+    ck::index_t C  = 16;
+    ck::index_t Di = 40;
+    ck::index_t Hi = 40;
+    ck::index_t Wi = 40;
+
+    pool3d_bwd_test<DevicePoolBwdInstance, DOutDataType, DInDataType, DOutLayout, DInLayout>(
+        true,
+        false,
+        N,
+        C,
+        Di,
+        Hi,
+        Wi,
+        window_lengths,
+        window_strides,
+        window_dilations,
+        dinput_left_pads,
+        dinput_right_pads);
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_avgpool_bwd.hpp b/include/ck/tensor_operation/gpu/device/device_avgpool_bwd.hpp
new file mode 100644
index 0000000000..5a2b082c31
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_avgpool_bwd.hpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <index_t NDimSpatial,
+          typename DOutDataType,
+          typename DInDataType,
+          typename DOutLayout,
+          typename DInLayout>
+struct DeviceAvgPoolBwd : public BaseOperator
+{
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_dout,
+                        void* p_din,
+                        std::vector<ck::index_t> dout_n_k_wos_lengths,
+                        std::vector<ck::index_t> dout_n_k_wos_strides,
+                        std::vector<ck::index_t> din_n_k_wos_length,
+                        std::vector<ck::index_t> din_n_k_wos_strides,
+                        std::vector<ck::index_t> window_k_c_xs_lengths,
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> window_dilations,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_put_element.hpp b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
index 9186827491..17df2de37b 100644
--- a/include/ck/tensor_operation/gpu/device/device_put_element.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_put_element.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp
new file mode 100644
index 0000000000..3a2280a75c
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp
@@ -0,0 +1,575 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
+#include "ck/tensor_operation/gpu/device/device_avgpool_bwd.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// In and Din = [N, C, Di, Hi, Wi]
+// Out and Dout = [N, C, Do, Ho, Wo]
+// Out = AvgPoolFwd(In)
+// Din = AvgPoolBwd(Dout)
+// Pooling dimension = D, H, W
+template <typename DOutDataType,
+          typename DInDataType,
+          typename ComputeDataType,
+          ck::index_t BlockSize,
+          ck::index_t MThreadClusterSize,
+          ck::index_t KThreadClusterSize,
+          ck::index_t MThreadSliceSize,
+          ck::index_t KThreadSliceSize,
+          ck::index_t InSrcOutDstVectorSize>
+struct DeviceAvgPool3dBwd_NDHWC_NDHWC : public DeviceAvgPoolBwd<3,
+                                                                DOutDataType,
+                                                                DInDataType,
+                                                                tensor_layout::convolution::NDHWC,
+                                                                tensor_layout::convolution::NDHWC>
+{
+    static constexpr ck::index_t NDimSpatial = 3;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr ck::index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
+    static constexpr ck::index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
+
+    static auto
+    Make3DGridDescriptor_Out_M_K_In_M(const std::vector<ck::index_t>& dout_n_c_wos_lengths,
+                                      const std::vector<ck::index_t>& din_n_c_wos_length,
+                                      const std::vector<ck::index_t>& dout_n_c_wos_strides,
+                                      const std::vector<ck::index_t>& din_n_c_wos_strides,
+                                      const std::vector<ck::index_t>& window_lengths,
+                                      const std::vector<ck::index_t>& window_strides,
+                                      const std::vector<ck::index_t>& window_dilations,
+                                      const std::vector<ck::index_t>& input_left_pads,
+                                      const std::vector<ck::index_t>& input_right_pads,
+                                      const std::vector<ck::index_t>& tildes)
+    {
+        index_t i_ztilde = tildes[0];
+        index_t i_ytilde = tildes[1];
+        index_t i_xtilde = tildes[2];
+
+        const index_t N = dout_n_c_wos_lengths[0];
+        const index_t C = dout_n_c_wos_lengths[1];
+
+        const index_t Di = din_n_c_wos_length[2];
+        const index_t Hi = din_n_c_wos_length[3];
+        const index_t Wi = din_n_c_wos_length[4];
+
+        const index_t Do = dout_n_c_wos_lengths[2];
+        const index_t Ho = dout_n_c_wos_lengths[3];
+        const index_t Wo = dout_n_c_wos_lengths[4];
+
+        const index_t Z = window_lengths[0];
+        const index_t Y = window_lengths[1];
+        const index_t X = window_lengths[2];
+
+        const index_t InLeftPadD = input_left_pads[0];
+        const index_t InLeftPadH = input_left_pads[1];
+        const index_t InLeftPadW = input_left_pads[2];
+
+        const index_t InRightPadD = input_right_pads[0];
+        const index_t InRightPadH = input_right_pads[1];
+        const index_t InRightPadW = input_right_pads[2];
+
+        const index_t ConvStrideD = window_strides[0];
+        const index_t ConvStrideH = window_strides[1];
+        const index_t ConvStrideW = window_strides[2];
+
+        const index_t ConvDilationD = window_dilations[0];
+        const index_t ConvDilationH = window_dilations[1];
+        const index_t ConvDilationW = window_dilations[2];
+
+        const auto out_n_do_ho_wo_c_grid_desc =
+            make_naive_tensor_descriptor(make_tuple(N, Do, Ho, Wo, C),
+                                         make_tuple(dout_n_c_wos_strides[0],
+                                                    dout_n_c_wos_strides[2],
+                                                    dout_n_c_wos_strides[3],
+                                                    dout_n_c_wos_strides[4],
+                                                    dout_n_c_wos_strides[1]));
+
+        const auto GcdStrideDilationD = math::gcd(ConvStrideD, ConvDilationD);
+        const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
+        const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
+
+        const auto ZTilde = ConvStrideD / GcdStrideDilationD;
+        const auto YTilde = ConvStrideH / GcdStrideDilationH;
+        const auto XTilde = ConvStrideW / GcdStrideDilationW;
+
+        const auto ZDot = math::integer_divide_ceil(Z, ZTilde);
+        const auto YDot = math::integer_divide_ceil(Y, YTilde);
+        const auto XDot = math::integer_divide_ceil(X, XTilde);
+
+        const auto DTilde = Do + math::integer_divide_ceil(ConvDilationD * (Z - I1), ConvStrideD);
+        const auto HTilde = Ho + math::integer_divide_ceil(ConvDilationH * (Y - I1), ConvStrideH);
+        const auto WTilde = Wo + math::integer_divide_ceil(ConvDilationW * (X - I1), ConvStrideW);
+
+        // only work on Tildes that contribute to non-padding area of input tensor
+        const auto IDTildeSliceBegin = math::integer_divide_floor(
+            math::max(I0, InLeftPadD - ConvDilationD * (ZTilde - I1)), ConvStrideD);
+        const auto IHTildeSliceBegin = math::integer_divide_floor(
+            math::max(I0, InLeftPadH - ConvDilationH * (YTilde - I1)), ConvStrideH);
+        const auto IWTildeSliceBegin = math::integer_divide_floor(
+            math::max(I0, InLeftPadW - ConvDilationW * (XTilde - I1)), ConvStrideW);
+
+        const auto IDTildeSliceEnd =
+            math::min(DTilde, math::integer_divide_ceil(InLeftPadD + Di - I1, ConvStrideD) + I1);
+        const auto IHTildeSliceEnd =
+            math::min(HTilde, math::integer_divide_ceil(InLeftPadH + Hi - I1, ConvStrideH) + I1);
+        const auto IWTildeSliceEnd =
+            math::min(WTilde, math::integer_divide_ceil(InLeftPadW + Wi - I1, ConvStrideW) + I1);
+
+        const auto DTildeSlice = IDTildeSliceEnd - IDTildeSliceBegin;
+        const auto HTildeSlice = IHTildeSliceEnd - IHTildeSliceBegin;
+        const auto WTildeSlice = IWTildeSliceEnd - IWTildeSliceBegin;
+
+        // ReduceK is different for each Reduce
+        const auto ZDotSlice = math::integer_divide_ceil(Z - i_ztilde, ZTilde);
+        const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
+        const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
+
+        // Problem size of reduction kernel
+        const index_t MRaw = N * DTildeSlice * HTildeSlice * WTildeSlice * C;
+        const index_t MPad = math::integer_least_multiple(MRaw, M_BlockTileSize) - MRaw;
+
+        const index_t KRaw = ZDotSlice * YDotSlice * XDotSlice;
+        const index_t KPad = math::integer_least_multiple(KRaw, K_BlockTileSize) - KRaw;
+
+        // Out[ReduceM, ReduceK]
+        const auto out_n_dop_hop_wop_c_grid_desc = transform_tensor_descriptor(
+            out_n_do_ho_wo_c_grid_desc,
+            make_tuple(make_pass_through_transform(N),
+                       make_pad_transform(Do, I0, I0),
+                       make_pad_transform(Ho, I0, I0),
+                       make_pad_transform(Wo, I0, I0),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+        const auto out_n_zdot_dtilde_ydot_htilde_xdot_wtilde_c_grid_desc =
+            transform_tensor_descriptor(
+                out_n_dop_hop_wop_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(ZDot, DTilde),
+                                         make_tuple(-ConvDilationD / GcdStrideDilationD, I1)),
+                    make_embed_transform(make_tuple(YDot, HTilde),
+                                         make_tuple(-ConvDilationH / GcdStrideDilationH, I1)),
+                    make_embed_transform(make_tuple(XDot, WTilde),
+                                         make_tuple(-ConvDilationW / GcdStrideDilationW, I1)),
+                    make_pass_through_transform(C)),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                make_tuple(Sequence<0>{},
+                           Sequence<1, 2>{},
+                           Sequence<3, 4>{},
+                           Sequence<5, 6>{},
+                           Sequence<7>{}));
+
+        const auto
+            out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_c_grid_desc =
+                transform_tensor_descriptor(
+                    out_n_zdot_dtilde_ydot_htilde_xdot_wtilde_c_grid_desc,
+                    make_tuple(make_pass_through_transform(N),
+                               make_slice_transform(ZDot, I0, ZDotSlice),
+                               make_slice_transform(DTilde, IDTildeSliceBegin, DTildeSlice),
+                               make_slice_transform(YDot, I0, YDotSlice),
+                               make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
+                               make_slice_transform(XDot, I0, XDotSlice),
+                               make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<0>{},
+                               Sequence<1>{},
+                               Sequence<2>{},
+                               Sequence<3>{},
+                               Sequence<4>{},
+                               Sequence<5>{},
+                               Sequence<6>{},
+                               Sequence<7>{}),
+                    make_tuple(Sequence<0>{},
+                               Sequence<1>{},
+                               Sequence<2>{},
+                               Sequence<3>{},
+                               Sequence<4>{},
+                               Sequence<5>{},
+                               Sequence<6>{},
+                               Sequence<7>{}));
+
+        const auto out_grid_desc_reducemraw_reducekraw = transform_tensor_descriptor(
+            out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_c_grid_desc,
+            make_tuple(
+                make_merge_transform(make_tuple(N, DTildeSlice, HTildeSlice, WTildeSlice, C)),
+                make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice))),
+            make_tuple(Sequence<0, 2, 4, 6, 7>{}, Sequence<1, 3, 5>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto out_grid_desc_reducem_reducek = transform_tensor_descriptor(
+            out_grid_desc_reducemraw_reducekraw,
+            make_tuple(make_right_pad_transform(MRaw, MPad), make_right_pad_transform(KRaw, KPad)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        // In[ReduceM]
+        const auto in_n_di_hi_wi_c_grid_desc =
+            make_naive_tensor_descriptor(make_tuple(N, Di, Hi, Wi, C),
+                                         make_tuple(din_n_c_wos_strides[0],
+                                                    din_n_c_wos_strides[2],
+                                                    din_n_c_wos_strides[3],
+                                                    din_n_c_wos_strides[4],
+                                                    din_n_c_wos_strides[1]));
+
+        const auto in_n_dip_hip_wip_c_grid_desc = transform_tensor_descriptor(
+            in_n_di_hi_wi_c_grid_desc,
+            make_tuple(make_pass_through_transform(N),
+                       make_pad_transform(Di, InLeftPadD, InRightPadD),
+                       make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                       make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                       make_pass_through_transform(C)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+        const auto in_n_ztilde_dtilde_ytilde_htilde_xtilde_wtilde_c_grid_desc =
+            transform_tensor_descriptor(
+                in_n_dip_hip_wip_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_embed_transform(make_tuple(XTilde, DTilde),
+                                                make_tuple(ConvDilationD, ConvStrideD)),
+                           make_embed_transform(make_tuple(YTilde, HTilde),
+                                                make_tuple(ConvDilationH, ConvStrideH)),
+                           make_embed_transform(make_tuple(XTilde, WTilde),
+                                                make_tuple(ConvDilationW, ConvStrideW)),
+                           make_pass_through_transform(C)),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                make_tuple(Sequence<0>{},
+                           Sequence<1, 2>{},
+                           Sequence<3, 4>{},
+                           Sequence<5, 6>{},
+                           Sequence<7>{}));
+
+        const auto in_n_dtildeslice_htildeslice_wtildeslice_c_grid_desc =
+            transform_tensor_descriptor(
+                in_n_ztilde_dtilde_ytilde_htilde_xtilde_wtilde_c_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_freeze_transform(i_ztilde),
+                           make_slice_transform(DTilde, IDTildeSliceBegin, DTildeSlice),
+                           make_freeze_transform(i_ytilde),
+                           make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
+                           make_freeze_transform(i_xtilde),
+                           make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
+                           make_pass_through_transform(C)),
+                make_tuple(Sequence<0>{},
+                           Sequence<1>{},
+                           Sequence<2>{},
+                           Sequence<3>{},
+                           Sequence<4>{},
+                           Sequence<5>{},
+                           Sequence<6>{},
+                           Sequence<7>{}),
+                make_tuple(Sequence<0>{},
+                           Sequence<>{},
+                           Sequence<1>{},
+                           Sequence<>{},
+                           Sequence<2>{},
+                           Sequence<>{},
+                           Sequence<3>{},
+                           Sequence<4>{}));
+
+        const auto in_grid_desc_reducemraw = transform_tensor_descriptor(
+            in_n_dtildeslice_htildeslice_wtildeslice_c_grid_desc,
+            make_tuple(
+                make_merge_transform(make_tuple(N, DTildeSlice, HTildeSlice, WTildeSlice, C))),
+            make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+            make_tuple(Sequence<0>{}));
+
+        const auto in_grid_desc_reducem =
+            transform_tensor_descriptor(in_grid_desc_reducemraw,
+                                        make_tuple(make_right_pad_transform(MRaw, MPad)),
+                                        make_tuple(Sequence<0>{}),
+                                        make_tuple(Sequence<0>{}));
+
+        return make_tuple(out_grid_desc_reducem_reducek, in_grid_desc_reducem);
+    }
+
+    using DoutDinGridDesc = decltype(Make3DGridDescriptor_Out_M_K_In_M({0, 0, 0, 0, 0},
+                                                                       {0, 0, 0, 0, 0},
+                                                                       {0, 0, 0, 0, 0},
+                                                                       {0, 0, 0, 0, 0},
+                                                                       {0, 0, 0},
+                                                                       {0, 0, 0},
+                                                                       {0, 0, 0},
+                                                                       {0, 0, 0},
+                                                                       {0, 0, 0},
+                                                                       {0, 0, 0}));
+
+    using DoutGridDesc_M_K = remove_cvref_t<tuple_element_t<0, DoutDinGridDesc>>;
+    using DinGridDesc_M    = remove_cvref_t<tuple_element_t<1, DoutDinGridDesc>>;
+
+    // FIXME
+    // for NDHWC, the dim C is the fastest dimension, and is not reduced.
+    // Hence, it is in M dimension for reduction kernel.
+    static constexpr index_t OutSrcInDstVectorDim = 0; // 0: M, 1: K
+
+    using PassThrough = tensor_operation::element_wise::PassThrough;
+    using Div         = tensor_operation::element_wise::UnaryDivide;
+
+    using gridwise_reduce = GridwiseReduction_mk_to_m_threadwise<DOutDataType,
+                                                                 DInDataType,
+                                                                 ComputeDataType,
+                                                                 int,
+                                                                 DoutGridDesc_M_K,
+                                                                 DinGridDesc_M,
+                                                                 reduce::Add,
+                                                                 PassThrough,
+                                                                 Div,
+                                                                 InMemoryDataOperationEnum::Set,
+                                                                 false, // propagate_nan
+                                                                 BlockSize,
+                                                                 MThreadSliceSize,
+                                                                 KThreadSliceSize,
+                                                                 OutSrcInDstVectorDim,
+                                                                 InSrcOutDstVectorSize,
+                                                                 InSrcOutDstVectorSize>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const DOutDataType* p_dout,
+                 DInDataType* p_din,
+                 std::vector<ck::index_t> dout_n_c_wos_lengths,
+                 std::vector<ck::index_t> din_n_c_wos_length,
+                 std::vector<ck::index_t> dout_n_c_wos_strides,
+                 std::vector<ck::index_t> din_n_c_wos_strides,
+                 std::vector<ck::index_t> window_lengths,
+                 std::vector<ck::index_t> window_strides,
+                 std::vector<ck::index_t> window_dilations,
+                 std::vector<ck::index_t> input_left_pads,
+                 std::vector<ck::index_t> input_right_pads)
+            : p_dout_grid_{p_dout},
+              p_din_grid_{p_din},
+              dout_n_c_wos_lengths_{dout_n_c_wos_lengths},
+              din_n_c_wos_length_{din_n_c_wos_length},
+              dout_n_c_wos_strides_{dout_n_c_wos_strides},
+              din_n_c_wos_strides_{din_n_c_wos_strides},
+              num_reduce_{1},
+              div_element_op_{window_lengths[0] * window_lengths[1] * window_lengths[2]}
+        {
+            std::vector<ck::index_t> Tildes(NDimSpatial);
+            for(int i = 0; i < NDimSpatial; ++i)
+            {
+                int GcdStrideDilation = math::gcd(window_strides[i], window_dilations[i]);
+                Tildes[i]             = window_strides[i] / GcdStrideDilation;
+                num_reduce_ *= Tildes[i];
+            }
+
+            for(index_t i_ztilde = 0; i_ztilde < Tildes[0]; ++i_ztilde)
+            {
+                for(index_t i_ytilde = 0; i_ytilde < Tildes[1]; ++i_ytilde)
+                {
+                    for(index_t i_xtilde = 0; i_xtilde < Tildes[2]; ++i_xtilde)
+                    {
+                        // check slice is valid
+                        const auto ZDotSlice =
+                            math::integer_divide_ceil(window_lengths[0] - i_ztilde, Tildes[0]);
+                        const auto YDotSlice =
+                            math::integer_divide_ceil(window_lengths[1] - i_ytilde, Tildes[1]);
+                        const auto XDotSlice =
+                            math::integer_divide_ceil(window_lengths[2] - i_xtilde, Tildes[2]);
+
+                        if(ZDotSlice * YDotSlice * XDotSlice <= 0)
+                        {
+                            continue;
+                        }
+
+                        const auto dout_din_grid_desc =
+                            Make3DGridDescriptor_Out_M_K_In_M(dout_n_c_wos_lengths,
+                                                              din_n_c_wos_length,
+                                                              dout_n_c_wos_strides,
+                                                              din_n_c_wos_strides,
+                                                              window_lengths,
+                                                              window_strides,
+                                                              window_dilations,
+                                                              input_left_pads,
+                                                              input_right_pads,
+                                                              {i_ztilde, i_ytilde, i_xtilde});
+
+                        dout_grid_desc_m_k_container_.push_back(dout_din_grid_desc[I0]);
+                        din_grid_desc_m_container_.push_back(dout_din_grid_desc[I1]);
+                    }
+                }
+            }
+        }
+
+        const DOutDataType* p_dout_grid_;
+        DInDataType* p_din_grid_;
+        std::vector<ck::index_t> dout_n_c_wos_lengths_;
+        std::vector<ck::index_t> din_n_c_wos_length_;
+        std::vector<ck::index_t> dout_n_c_wos_strides_;
+        std::vector<ck::index_t> din_n_c_wos_strides_;
+
+        int num_reduce_;
+        std::vector<DoutGridDesc_M_K> dout_grid_desc_m_k_container_;
+        std::vector<DinGridDesc_M> din_grid_desc_m_container_;
+
+        Div div_element_op_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            float ave_time = 0;
+
+            for(index_t i = 0; i < arg.num_reduce_; i++)
+            {
+                const auto kernel = kernel_reduce_threadwise<gridwise_reduce,
+                                                             false,
+                                                             false,
+                                                             false, // don't have index input
+                                                             DOutDataType,
+                                                             DInDataType,
+                                                             ComputeDataType,
+                                                             int,
+                                                             DoutGridDesc_M_K,
+                                                             DinGridDesc_M,
+                                                             PassThrough,
+                                                             Div>;
+
+                ck::index_t M           = arg.dout_grid_desc_m_k_container_[i].GetLength(I0);
+                const index_t grid_size = (M / M_BlockTileSize);
+
+                ave_time += launch_and_time_kernel(stream_config,
+                                                   kernel,
+                                                   dim3(grid_size),
+                                                   dim3(BlockSize),
+                                                   0,
+                                                   arg.dout_grid_desc_m_k_container_[i],
+                                                   arg.din_grid_desc_m_container_[i],
+                                                   PassThrough{},
+                                                   arg.div_element_op_,
+                                                   float(1),
+                                                   arg.p_dout_grid_,
+                                                   nullptr,
+                                                   float(0),
+                                                   arg.p_din_grid_,
+                                                   nullptr);
+            }
+
+            return ave_time;
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        constexpr index_t Rank = NDimSpatial + 2;
+        int doutFastestDim     = -1;
+        int dinFastestDim      = -1;
+
+        for(int i = 0; i < Rank; ++i)
+        {
+            if(arg.dout_n_c_wos_strides_[i] == 1)
+                doutFastestDim = i;
+            if(arg.din_n_c_wos_strides_[i] == 1)
+                dinFastestDim = i;
+        }
+
+        if(doutFastestDim == -1 || dinFastestDim == -1)
+        {
+            if constexpr(InSrcOutDstVectorSize != 1)
+                return false;
+        }
+        else
+        {
+            if(arg.dout_n_c_wos_lengths_[doutFastestDim] % InSrcOutDstVectorSize != 0)
+                return false;
+            if(arg.din_n_c_wos_length_[dinFastestDim] % InSrcOutDstVectorSize != 0)
+                return false;
+        }
+
+        return true;
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_dout,
+                        void* p_din,
+                        std::vector<ck::index_t> dout_n_c_wos_lengths,
+                        std::vector<ck::index_t> din_n_c_wos_length,
+                        std::vector<ck::index_t> dout_n_c_wos_strides,
+                        std::vector<ck::index_t> din_n_c_wos_strides,
+                        std::vector<ck::index_t> window_lengths,
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> window_dilations,
+                        std::vector<ck::index_t> input_left_pads,
+                        std::vector<ck::index_t> input_right_pads) override
+    {
+        constexpr index_t Rank = NDimSpatial + 2;
+
+        if(dout_n_c_wos_strides.size() != Rank || din_n_c_wos_strides.size() != Rank ||
+           dout_n_c_wos_lengths.size() != Rank || din_n_c_wos_length.size() != Rank)
+            throw std::runtime_error("dimension is incorrect");
+
+        if(window_lengths.size() != NDimSpatial || window_strides.size() != NDimSpatial ||
+           window_dilations.size() != NDimSpatial || input_left_pads.size() != NDimSpatial ||
+           input_right_pads.size() != NDimSpatial)
+            throw std::runtime_error("dimension is incorrect");
+
+        return std::make_unique<Argument>(static_cast<const DOutDataType*>(p_dout),
+                                          static_cast<DInDataType*>(p_din),
+                                          dout_n_c_wos_lengths,
+                                          din_n_c_wos_length,
+                                          dout_n_c_wos_strides,
+                                          din_n_c_wos_strides,
+                                          window_lengths,
+                                          window_strides,
+                                          window_dilations,
+                                          input_left_pads,
+                                          input_right_pads);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceAvgPool3dBwd<" << BlockSize << ",";
+        str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
+        str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
+        str <<"InSrcOutDstVectorSize_" << InSrcOutDstVectorSize << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
index 8f72f88b06..c41eef8c45 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp
new file mode 100644
index 0000000000..fa06e77560
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp
@@ -0,0 +1,354 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+#include "ck/library/utility/host_tensor.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+// dinput descriptor in [N, C, Do, Ho, Wo] order
+// doutput descriptor in [N, C, Di, Hi, Wi] order
+// phyiscal layout is irrelavent
+template <ck::index_t NDimSpatial,
+          typename DInDataType,
+          typename DOutDataType,
+          typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
+struct ReferenceAvgPoolBwd : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        Argument(Tensor<DInDataType>& dinput,
+                 const Tensor<DOutDataType>& doutput,
+                 std::vector<ck::index_t> window_spatial_lengths,
+                 std::vector<ck::index_t> window_strides,
+                 std::vector<ck::index_t> window_dilations,
+                 std::vector<ck::index_t> dinput_left_pads,
+                 std::vector<ck::index_t> dinput_right_pads)
+            : dinput_{dinput},
+              doutput_{doutput},
+              window_spatial_lengths_{window_spatial_lengths},
+              window_strides_{window_strides},
+              window_dilations_{window_dilations},
+              in_left_pads_{dinput_left_pads},
+              in_right_pads_{dinput_right_pads}
+        {
+        }
+
+        Tensor<DInDataType>& dinput_;
+        const Tensor<DOutDataType>& doutput_;
+
+        std::vector<ck::index_t> window_spatial_lengths_;
+        std::vector<index_t> window_strides_;
+        std::vector<index_t> window_dilations_;
+        std::vector<index_t> in_left_pads_;
+        std::vector<index_t> in_right_pads_;
+    };
+
+    // Invoker
+    struct Invoker : public device::BaseInvoker
+    {
+        using Argument = ReferenceAvgPoolBwd::Argument;
+
+        template <ck::index_t NDimSpatial_,
+                  typename std::enable_if<NDimSpatial_ == 1, bool>::type = false>
+        float RunAvgPoolBwd(const Argument& arg)
+        {
+            // Let input = x, outpu = y
+            // shape of x = [10], y = [6]
+            // window_size = 5, pad = 0, stride = 1, dilation = 1
+            // Forward:
+            // y0 = 1/5 * (x0 + x1 + x2 + x3 + x4)
+            // y1 = 1/5 * (x1 + x2 + x3 + x4 + x5)
+            // ...
+            // y5 = 1/5 * (x5 + x6 + x7 + x8 + x9)
+            // y6 = 1/5 * (x6 + x7 + x8 + x9)
+            // ...
+            // y9 = 1/5 * (x9)
+
+            // Backward:
+            // shape of dy = [6], dx = [10]
+            // dx0 = 1/5 * dy0
+            // dx1 = 1/5 * (dy0 + dy1)
+            // dx2 = 1/5 * (dy0 + dy1 + dy2)
+            // ...
+            // dx4 = 1/5 * (dy0 + dy1 + dy2 + dy3 + dy4)
+            // dx5 = 1/5 * (dy1 + dy2 + dy3 + dy4 + dy5)
+            // ...
+            // dx9 = 1/5 * (dy5 + dy6 + dy7 + dy8 + dy9)
+
+            auto f_ncw = [&](auto n, auto c, auto wi) {
+                std::size_t X  = arg.window_spatial_lengths_[0];
+                std::size_t Wo = arg.doutput_.GetLengths()[2];
+
+                float v_acc = 0;
+
+                for(std::size_t x = 0; x < X; ++x)
+                {
+                    // Out_Position = (In_Position + pad - x * dilation) / stride
+                    auto w_tmp = static_cast<ck::long_index_t>(wi) +
+                                 static_cast<ck::long_index_t>(arg.in_left_pads_[0]) -
+                                 static_cast<ck::long_index_t>(x * arg.window_dilations_[0]);
+
+                    // Check the input pixel validity (in perspective of being affected by some
+                    // doutput pixel)
+                    if(w_tmp % arg.window_strides_[0] == 0)
+                    {
+                        auto wo = static_cast<ck::long_index_t>(w_tmp) /
+                                  static_cast<ck::long_index_t>(arg.window_strides_[0]);
+
+                        // Get the doutput pixel in valid range to accumulate the gradients for this
+                        // input pixel
+                        if(wo >= 0 && ck::type_convert<std::size_t>(wo) < Wo)
+                        {
+                            v_acc += ck::type_convert<float>(arg.doutput_(n, c, wo));
+                        }
+                    }
+                }
+
+                v_acc /= ck::type_convert<float>(X);
+                arg.dinput_(n, c, wi) = ck::type_convert<DInDataType>(v_acc);
+            };
+
+            make_ParallelTensorFunctor(f_ncw,
+                                       arg.dinput_.GetLengths()[0],
+                                       arg.dinput_.GetLengths()[1],
+                                       arg.dinput_.GetLengths()[2])(
+                std::thread::hardware_concurrency());
+
+            return 0;
+        }
+
+        template <ck::index_t NDimSpatial_,
+                  typename std::enable_if<NDimSpatial_ == 2, bool>::type = false>
+        float RunAvgPoolBwd(const Argument& arg)
+        {
+            auto f_nchw = [&](auto n, auto c, auto hi, auto wi) {
+                std::size_t Y = arg.window_spatial_lengths_[0];
+                std::size_t X = arg.window_spatial_lengths_[1];
+
+                std::size_t Ho = arg.doutput_.GetLengths()[2];
+                std::size_t Wo = arg.doutput_.GetLengths()[3];
+
+                float v_acc = 0;
+
+                for(std::size_t y = 0; y < Y; ++y)
+                {
+                    // Out_Position = (In_Position + pad - x * dilation) / stride
+                    auto h_tmp = static_cast<ck::long_index_t>(hi) +
+                                 static_cast<ck::long_index_t>(arg.in_left_pads_[0]) -
+                                 static_cast<ck::long_index_t>(y * arg.window_dilations_[0]);
+
+                    // Check the input pixel validity (in perspective of being affected by some
+                    // doutput pixel)
+                    if(h_tmp % arg.window_strides_[0] == 0)
+                    {
+                        auto ho = static_cast<ck::long_index_t>(h_tmp) /
+                                  static_cast<ck::long_index_t>(arg.window_strides_[0]);
+
+                        // Get the doutput pixel in valid range to accumulate the gradients for this
+                        // input pixel
+                        if(ho >= 0 && ck::type_convert<std::size_t>(ho) < Ho)
+                        {
+                            for(std::size_t x = 0; x < X; ++x)
+                            {
+                                auto w_tmp =
+                                    static_cast<ck::long_index_t>(wi) +
+                                    static_cast<ck::long_index_t>(arg.in_left_pads_[1]) -
+                                    static_cast<ck::long_index_t>(x * arg.window_dilations_[1]);
+                                if(w_tmp % arg.window_strides_[1] == 0)
+                                {
+                                    auto wo = static_cast<ck::long_index_t>(w_tmp) /
+                                              static_cast<ck::long_index_t>(arg.window_strides_[1]);
+                                    if(wo >= 0 && ck::type_convert<std::size_t>(wo) < Wo)
+                                    {
+                                        v_acc +=
+                                            ck::type_convert<float>(arg.doutput_(n, c, ho, wo));
+                                    }
+                                }
+                            }
+                        }
+                    }
+                }
+
+                v_acc /= ck::type_convert<float>(Y * X);
+                arg.dinput_(n, c, hi, wi) = ck::type_convert<DInDataType>(v_acc);
+            };
+
+            make_ParallelTensorFunctor(f_nchw,
+                                       arg.dinput_.GetLengths()[0],
+                                       arg.dinput_.GetLengths()[1],
+                                       arg.dinput_.GetLengths()[2],
+                                       arg.dinput_.GetLengths()[3])(
+                std::thread::hardware_concurrency());
+
+            return 0;
+        }
+
+        template <ck::index_t NDimSpatial_,
+                  typename std::enable_if<NDimSpatial_ == 3, bool>::type = false>
+        float RunAvgPoolBwd(const Argument& arg)
+        {
+            auto f_ncdhw = [&](auto n, auto c, auto di, auto hi, auto wi) {
+                std::size_t Z = arg.window_spatial_lengths_[0];
+                std::size_t Y = arg.window_spatial_lengths_[1];
+                std::size_t X = arg.window_spatial_lengths_[2];
+
+                std::size_t Do = arg.doutput_.GetLengths()[2];
+                std::size_t Ho = arg.doutput_.GetLengths()[3];
+                std::size_t Wo = arg.doutput_.GetLengths()[4];
+
+                float v_acc = 0;
+
+                for(std::size_t z = 0; z < Z; ++z)
+                {
+                    // Out_Position = (In_Position + pad - x * dilation) / stride
+                    auto d_tmp = static_cast<ck::long_index_t>(di) +
+                                 static_cast<ck::long_index_t>(arg.in_left_pads_[0]) -
+                                 static_cast<ck::long_index_t>(z * arg.window_dilations_[0]);
+
+                    // Check the input pixel validity (in perspective of being affected by some
+                    // doutput pixel)
+                    if(d_tmp % arg.window_strides_[0] == 0)
+                    {
+                        auto do_ = static_cast<ck::long_index_t>(d_tmp) /
+                                   static_cast<ck::long_index_t>(arg.window_strides_[0]);
+
+                        // Get the doutput pixel in valid range to accumulate the gradients for this
+                        // input pixel
+                        if(do_ >= 0 && ck::type_convert<std::size_t>(do_) < Do)
+                        {
+                            for(std::size_t y = 0; y < Y; ++y)
+                            {
+                                auto h_tmp =
+                                    static_cast<ck::long_index_t>(hi) +
+                                    static_cast<ck::long_index_t>(arg.in_left_pads_[1]) -
+                                    static_cast<ck::long_index_t>(y * arg.window_dilations_[1]);
+                                if(h_tmp % arg.window_strides_[1] == 0)
+                                {
+                                    auto ho = static_cast<ck::long_index_t>(h_tmp) /
+                                              static_cast<ck::long_index_t>(arg.window_strides_[1]);
+                                    if(ho >= 0 && ck::type_convert<std::size_t>(ho) < Ho)
+                                    {
+                                        for(std::size_t x = 0; x < X; ++x)
+                                        {
+                                            auto w_tmp = static_cast<ck::long_index_t>(wi) +
+                                                         static_cast<ck::long_index_t>(
+                                                             arg.in_left_pads_[2]) -
+                                                         static_cast<ck::long_index_t>(
+                                                             x * arg.window_dilations_[2]);
+
+                                            if(w_tmp % arg.window_strides_[2] == 0)
+                                            {
+                                                auto wo = static_cast<ck::long_index_t>(w_tmp) /
+                                                          static_cast<ck::long_index_t>(
+                                                              arg.window_strides_[2]);
+                                                if(wo >= 0 &&
+                                                   ck::type_convert<std::size_t>(wo) < Wo)
+                                                {
+                                                    v_acc += ck::type_convert<float>(
+                                                        arg.doutput_(n, c, do_, ho, wo));
+                                                }
+                                            }
+                                        }
+                                    }
+                                }
+                            }
+                        }
+                    }
+                }
+
+                v_acc /= ck::type_convert<float>(Z * Y * X);
+                arg.dinput_(n, c, di, hi, wi) = ck::type_convert<DInDataType>(v_acc);
+            };
+
+            make_ParallelTensorFunctor(f_ncdhw,
+                                       arg.dinput_.GetLengths()[0],
+                                       arg.dinput_.GetLengths()[1],
+                                       arg.dinput_.GetLengths()[2],
+                                       arg.dinput_.GetLengths()[3],
+                                       arg.dinput_.GetLengths()[4])(
+                std::thread::hardware_concurrency());
+
+            return 0;
+        }
+
+        float Run(const Argument& arg)
+        {
+            if(!(arg.dinput_.GetNumOfDimension() == NDimSpatial + 2 &&
+                 arg.doutput_.GetNumOfDimension() == NDimSpatial + 2))
+            {
+                throw std::runtime_error("wrong! inconsistent dimension");
+            }
+
+            return RunAvgPoolBwd<NDimSpatial>(arg);
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /* stream_config */ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    bool IsSupportedArgument(const device::BaseArgument*) override { return true; }
+
+    static auto MakeArgument(Tensor<DInDataType>& dinput,
+                             const Tensor<DOutDataType>& doutput,
+                             std::vector<ck::index_t> window_spatial_lengths,
+                             std::vector<ck::index_t> window_strides,
+                             std::vector<ck::index_t> window_dilations,
+                             std::vector<ck::index_t> dinput_left_pads,
+                             std::vector<ck::index_t> dinput_right_pads)
+    {
+        if(window_spatial_lengths.size() != NDimSpatial || window_strides.size() != NDimSpatial ||
+           window_dilations.size() != NDimSpatial || dinput_left_pads.size() != NDimSpatial ||
+           dinput_right_pads.size() != NDimSpatial)
+            throw std::runtime_error("dimension is incorrect");
+
+        return Argument{dinput,
+                        doutput,
+                        window_spatial_lengths,
+                        window_strides,
+                        window_dilations,
+                        dinput_left_pads,
+                        dinput_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceAvgPoolBwd"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
index 449734f434..50040a2441 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_data.hpp
@@ -125,7 +125,7 @@ struct ReferenceConvBwdData : public device::BaseOperator
 
                     arg.in_element_op_(v_in, v_acc);
 
-                    arg.input_(g, n, c, wi) = ck::type_convert<InDataType>(v_acc);
+                    arg.input_(g, n, c, wi) = ck::type_convert<InDataType>(v_in);
                 };
 
                 make_ParallelTensorFunctor(f_ncw,
@@ -201,7 +201,7 @@ struct ReferenceConvBwdData : public device::BaseOperator
 
                     arg.in_element_op_(v_in, v_acc);
 
-                    arg.input_(g, n, c, hi, wi) = ck::type_convert<InDataType>(v_acc);
+                    arg.input_(g, n, c, hi, wi) = ck::type_convert<InDataType>(v_in);
                 };
 
                 make_ParallelTensorFunctor(f_nchw,
@@ -299,7 +299,7 @@ struct ReferenceConvBwdData : public device::BaseOperator
 
                     arg.in_element_op_(v_in, v_acc);
 
-                    arg.input_(g, n, c, di, hi, wi) = ck::type_convert<InDataType>(v_acc);
+                    arg.input_(g, n, c, di, hi, wi) = ck::type_convert<InDataType>(v_in);
                 };
 
                 make_ParallelTensorFunctor(f_ncdhw,

From 6237bd12476fa2f1c39ea47c9113e05faf6bf307 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 10 Aug 2023 09:25:21 -0700
Subject: [PATCH 088/194] Add the rocm5.7 RC1 compiler and use it for QA
 builds. (#842)

* add docker for rocm5.7 RC1

* fix rocm5.7 rc1 build

* build QA with rocm5.7 rc1 compiler
---
 Dockerfile  | 11 ++++++++---
 Jenkinsfile |  2 +-
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index 8d11a29853..b9c815f527 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -26,9 +26,14 @@ RUN if [ "$ROCMVERSION" != "5.7" ]; then \
         sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
         sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
     elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
-         sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
-         apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
-         amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
+        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
+        apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
+        amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
+    elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "rc1" ]; then \
+        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
+        apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
+        sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.7 rel-19 > /etc/apt/sources.list.d/rocm-build.list' && \
+        amdgpu-repo --amdgpu-build=1637781; \
     fi
 
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
diff --git a/Jenkinsfile b/Jenkinsfile
index 1817207093..3fce4dabec 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -612,7 +612,7 @@ def process_results(Map conf=[:]){
 }
 
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
-CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true
+CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=rc1
                                               0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
                                               0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 

From a5343db00ddf65c2d839c1058c8b17fd26047cbe Mon Sep 17 00:00:00 2001
From: "dependabot[bot]" <49699333+dependabot[bot]@users.noreply.github.com>
Date: Fri, 11 Aug 2023 11:27:56 -0600
Subject: [PATCH 089/194] Bump rocm-docs-core from 0.10.3 to 0.20.0 in
 /docs/sphinx (#844)

* Bump rocm-docs-core from 0.10.3 to 0.20.0 in /docs/sphinx

Bumps [rocm-docs-core](https://github.com/RadeonOpenCompute/rocm-docs-core) from 0.10.3 to 0.20.0.
- [Release notes](https://github.com/RadeonOpenCompute/rocm-docs-core/releases)
- [Changelog](https://github.com/RadeonOpenCompute/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/RadeonOpenCompute/rocm-docs-core/compare/v0.10.3...v0.20.0)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>

* set min version of rocm-docs-core

---------

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Sam Wu <sam.wu2@amd.com>
---
 docs/sphinx/requirements.in  |  2 +-
 docs/sphinx/requirements.txt | 23 +++++------------------
 2 files changed, 6 insertions(+), 19 deletions(-)

diff --git a/docs/sphinx/requirements.in b/docs/sphinx/requirements.in
index 4bdf41b959..a12a00a2b2 100644
--- a/docs/sphinx/requirements.in
+++ b/docs/sphinx/requirements.in
@@ -1,2 +1,2 @@
-rocm-docs-core==0.10.3
+rocm-docs-core>=0.20.0
 sphinxcontrib-bibtex==2.5.0
diff --git a/docs/sphinx/requirements.txt b/docs/sphinx/requirements.txt
index 097acba225..d5f67eeb58 100644
--- a/docs/sphinx/requirements.txt
+++ b/docs/sphinx/requirements.txt
@@ -38,6 +38,8 @@ docutils==0.16
     #   pydata-sphinx-theme
     #   sphinx
     #   sphinxcontrib-bibtex
+fastjsonschema==2.18.0
+    # via rocm-docs-core
 gitdb==4.0.10
     # via gitpython
 gitpython==3.1.31
@@ -46,20 +48,12 @@ idna==3.4
     # via requests
 imagesize==1.4.1
     # via sphinx
-importlib-metadata==6.0.0
-    # via
-    #   sphinx
-    #   sphinxcontrib-bibtex
-importlib-resources==5.12.0
-    # via rocm-docs-core
 jinja2==3.1.2
     # via
     #   myst-parser
     #   sphinx
 latexcodec==2.0.1
     # via pybtex
-linkify-it-py==1.0.3
-    # via myst-parser
 markdown-it-py==2.2.0
     # via
     #   mdit-py-plugins
@@ -70,7 +64,7 @@ mdit-py-plugins==0.3.5
     # via myst-parser
 mdurl==0.1.2
     # via markdown-it-py
-myst-parser[linkify]==1.0.0
+myst-parser==1.0.0
     # via rocm-docs-core
 packaging==23.0
     # via
@@ -99,18 +93,17 @@ pyjwt[crypto]==2.6.0
     # via pygithub
 pynacl==1.5.0
     # via pygithub
-pytz==2023.3
-    # via babel
 pyyaml==6.0
     # via
     #   myst-parser
     #   pybtex
+    #   rocm-docs-core
     #   sphinx-external-toc
 requests==2.28.2
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core==0.10.3
+rocm-docs-core>=0.20.0
     # via -r requirements.in
 six==1.16.0
     # via
@@ -160,13 +153,7 @@ sphinxcontrib-serializinghtml==1.1.5
     # via sphinx
 typing-extensions==4.5.0
     # via pydata-sphinx-theme
-uc-micro-py==1.0.1
-    # via linkify-it-py
 urllib3==1.26.15
     # via requests
 wrapt==1.15.0
     # via deprecated
-zipp==3.15.0
-    # via
-    #   importlib-metadata
-    #   importlib-resources

From 03b8119e2eb7c2312eda9d32772c67f598304d86 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Sat, 12 Aug 2023 01:31:31 +0800
Subject: [PATCH 090/194] Add Normalization splitk instances (#829)

* Add normalization splitK to layernorm and groupnorm instances

* Fix bug of GetKPerThread()

* Refine naming

* clang format
---
 .../gridwise_normalization_splitk_1st.hpp     | 26 +++---
 .../device_groupnorm_f16_instance.cpp         |  2 +
 .../device_groupnorm_f32_instance.cpp         |  2 +
 ...oupnorm_swish_f16_f32_f32_f16_instance.cpp |  2 +
 .../device_groupnorm_swish_f16_instance.cpp   |  2 +
 .../device_groupnorm_swish_f32_instance.cpp   |  2 +
 .../device_layernorm2d_f16_instance.cpp       |  2 +
 .../device_layernorm2d_f32_instance.cpp       |  2 +
 .../device_layernorm4d_f16_instance.cpp       |  2 +
 .../device_layernorm4d_f32_instance.cpp       |  2 +
 .../normalization_instance_common.hpp         | 79 +++++++++++++++++++
 .../profiler/profile_groupnorm_impl.hpp       |  4 +
 .../profiler/profile_layernorm_impl.hpp       |  4 +
 13 files changed, 120 insertions(+), 11 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
index fc42e97629..0fb961eb7a 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
@@ -78,17 +78,18 @@ struct GridwiseNormalizationSplitK1st
     static constexpr auto ThreadBufferNumber = Number<KThreadSliceSize / XSrcVectorSize>{};
 
     __device__ static int
-    GetKPerThread(int kRaw, int kGridSize, int block_k_cluster_id, int thread_k_cluster_id)
+    GetKPerThread(int k, int kRaw, int kGridSize, int block_k_cluster_id, int thread_k_cluster_id)
     {
         bool is_rightmost_block = block_k_cluster_id == kGridSize - 1;
 
         if(is_rightmost_block)
         {
-            int left_kPerBlock = math::integer_divide_ceil(kRaw, kGridSize);
-            int kPerBlock      = kRaw % kGridSize == 0 ? left_kPerBlock : kRaw % left_kPerBlock;
-            int kPerThread =
-                kPerBlock < K_BlockTileSize ? 0 : KThreadSliceSize * (kPerBlock / K_BlockTileSize);
-            int kPerBlockTail = kPerBlock - kPerThread * KThreadClusterSize;
+            int left_kPerBlock  = math::integer_divide_ceil(k, kGridSize);
+            int kRightmostBlock = kRaw - left_kPerBlock * (kGridSize - 1);
+            int kPerThread      = kRightmostBlock < K_BlockTileSize
+                                 ? 0
+                                 : KThreadSliceSize * (kRightmostBlock / K_BlockTileSize);
+            int kPerBlockTail = kRightmostBlock - kPerThread * KThreadClusterSize;
 
             if(kPerBlockTail > 0)
             {
@@ -105,7 +106,7 @@ struct GridwiseNormalizationSplitK1st
         }
         else
         {
-            int kPerBlock = math::integer_divide_ceil(kRaw, kGridSize);
+            int kPerBlock = math::integer_divide_ceil(k, kGridSize);
             return KThreadSliceSize * (kPerBlock / K_BlockTileSize);
         }
     }
@@ -193,10 +194,13 @@ struct GridwiseNormalizationSplitK1st
         auto var_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_variance_global, mean_var_grid_desc_m_kblock.GetElementSpaceSize());
 
-        auto threadwise_welford = ThreadwiseWelford();
-        int kRaw                = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
-        threadwise_welford.max_count_ =
-            GetKPerThread(kRaw, k_grid_size, block_k_cluster_id, thread_k_cluster_id);
+        auto threadwise_welford       = ThreadwiseWelford();
+        int kRaw                      = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
+        threadwise_welford.max_count_ = GetKPerThread(x_grid_desc_m_k.GetLength(I1),
+                                                      kRaw,
+                                                      k_grid_size,
+                                                      block_k_cluster_id,
+                                                      thread_k_cluster_id);
 
         static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
             mean_thread_buf(I) = type_convert<ComputeDataType>(0.0f);
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
index e3820462cf..762da1c6ae 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_5_3_f16_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f16_generic_instance<Pass, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 5, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f16_instances<Pass, 5, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
index d85817aad3..44b553bd16 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f32_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_5_3_f32_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f32_generic_instance<Pass, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 5, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f32_instances<Pass, 5, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
index a81f776c0f..aa662b7dfe 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
@@ -18,6 +18,8 @@ void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
         instances, device_normalization_f16_f32_f32_f16_generic_instance<Swish, 5, 3>{});
     add_device_operation_instances(instances,
                                    device_normalization_f16_f32_f32_f16_instances<Swish, 5, 3>{});
+    add_device_operation_instances(
+        instances, device_normalization_splitk_f16_f32_f32_f16_instances<Swish, 5, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
index f4bb8bda81..bc5cd801ae 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_5_3_swish_f16_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f16_generic_instance<Swish, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Swish, 5, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f16_instances<Swish, 5, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
index bbb9bd0fe8..4b2ab33570 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f32_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_5_3_swish_f32_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f32_generic_instance<Swish, 5, 3>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Swish, 5, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f32_instances<Swish, 5, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
index 3f7e4aff1a..0d235f1fa7 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_2_1_f16_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f16_generic_instance<Pass, 2, 1>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 2, 1>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f16_instances<Pass, 2, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
index 1f0db3a036..00039531e1 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f32_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_2_1_f32_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f32_generic_instance<Pass, 2, 1>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 2, 1>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f32_instances<Pass, 2, 1>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
index cb9d72e614..6bc3950062 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_4_3_f16_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f16_generic_instance<Pass, 4, 3>{});
     add_device_operation_instances(instances, device_normalization_f16_instances<Pass, 4, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f16_instances<Pass, 4, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
index ed555b840d..b387dc2f3f 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f32_instance.cpp
@@ -17,6 +17,8 @@ void add_device_normalization_rank_4_3_f32_instances(
     add_device_operation_instances(instances,
                                    device_normalization_f32_generic_instance<Pass, 4, 3>{});
     add_device_operation_instances(instances, device_normalization_f32_instances<Pass, 4, 3>{});
+    add_device_operation_instances(instances,
+                                   device_normalization_splitk_f32_instances<Pass, 4, 3>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
index b0684962f9..7aa3da8eed 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
@@ -5,6 +5,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp"
 #include "ck/utility/data_type.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
@@ -43,6 +44,32 @@ using device_normalization_f16_instances =
         // clang-format on
         >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_splitk_f16_instances =
+    // clang-format off
+    std::tuple <
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8>
+        // clang-format on
+        >;
+
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_generic_instance = std::tuple<
     // clang-format off
@@ -76,6 +103,32 @@ using device_normalization_f32_instances = std::tuple<
     // clang-format on
     >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_splitk_f32_instances = std::tuple<
+    // clang-format off
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+    // clang-format on
+    >;
+
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f32_generic_instance = std::tuple<
     // clang-format off
@@ -109,6 +162,32 @@ using device_normalization_f16_f32_f32_f16_instances = std::tuple<
     // clang-format on
     >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_splitk_f16_f32_f32_f16_instances = std::tuple<
+    // clang-format off
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+    // clang-format on
+    >;
+
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_f32_f32_f16_generic_instance = std::tuple<
     // clang-format off
diff --git a/profiler/include/profiler/profile_groupnorm_impl.hpp b/profiler/include/profiler/profile_groupnorm_impl.hpp
index ebefe3dad4..f88ba8453c 100644
--- a/profiler/include/profiler/profile_groupnorm_impl.hpp
+++ b/profiler/include/profiler/profile_groupnorm_impl.hpp
@@ -139,6 +139,10 @@ bool profile_groupnorm_impl(int do_verification,
             continue;
         }
 
+        size_t workspace_sz = inst_ptr->GetWorkSpaceSize(argument_ptr.get());
+        DeviceMem workspace_dev(workspace_sz);
+        inst_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
         auto invoker_ptr = inst_ptr->MakeInvokerPointer();
 
         float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
diff --git a/profiler/include/profiler/profile_layernorm_impl.hpp b/profiler/include/profiler/profile_layernorm_impl.hpp
index 2d87c8c8fe..f969646c2f 100644
--- a/profiler/include/profiler/profile_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_layernorm_impl.hpp
@@ -155,6 +155,10 @@ bool profile_layernorm_impl(int do_verification,
             continue;
         }
 
+        size_t workspace_sz = inst_ptr->GetWorkSpaceSize(argument_ptr.get());
+        DeviceMem workspace_dev(workspace_sz);
+        inst_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_dev.GetDeviceBuffer());
+
         auto invoker_ptr = inst_ptr->MakeInvokerPointer();
 
         float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});

From f60f0a5e03e64c365369b4db93596b9e927075c4 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Tue, 15 Aug 2023 02:25:28 +0800
Subject: [PATCH 091/194] Refactor pool fwd (#815)

* Do not hardcode stride

* devicePool2DFwd Inherit devicePool3DFwd

* Move instance declaration out of common

* Add dilation

* use the pool3d rank, because pool2d inherit pooo3d

* calculate Do Ho Wo for the dilation

* Fix header name

* Modify ckProfiler

* Remove pool2d instance

* Remove pool2d in profiler

* Remove pool2d and add dilation

* In to client example, this commit revise following:
1. Add dilation.
2. Use pool3d to implement pool2d

* Refine naming and IsSupportedArgument()

* Add dilation to maxpool bwd example

* clang format

* 1. Remove useless header
2. Fix copyright
3. Refine naming

* Add layout parameter to pool fwd

* clang format

* Fix merge error

* Fix compile error

* Remove layout parameter in derived class

* Refine changlog

* Fix compile error

* Fix compiler error

* Add layout to external api and profiler
---
 CHANGELOG.md                                  |   4 +-
 client_example/19_pool_fwd/avg_pool3d_fwd.cpp |  57 +--
 client_example/19_pool_fwd/max_pool2d_fwd.cpp |  93 ++++-
 example/13_pool2d_fwd/pool2d_fwd_common.hpp   |  40 +-
 example/13_pool2d_fwd/pool2d_fwd_fp16.cpp     |  58 +--
 example/13_pool2d_fwd/pool2d_fwd_fp32.cpp     |  58 +--
 example/48_pool3d_fwd/pool3d_fwd_common.hpp   |  99 ++---
 example/48_pool3d_fwd/pool3d_fwd_fp16.cpp     |  57 ++-
 .../49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp   |  28 +-
 .../49_maxpool2d_bwd/maxpool2d_bwd_common.hpp |  36 +-
 .../49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp   |  28 +-
 .../49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp   |  28 +-
 .../gpu/device/device_pool_fwd.hpp            |  17 +-
 .../impl/device_pool2d_fwd_nhwc_nhwc.hpp      | 347 +++---------------
 .../impl/device_pool3d_fwd_ndhwc_ndhwc.hpp    | 252 ++++++++-----
 .../cpu/reference_pool_fwd.hpp                |  33 +-
 .../gpu/pool2d_fwd.hpp                        | 114 ------
 .../gpu/pool3d_fwd.hpp                        |  74 ++--
 .../gpu/pool3d_fwd/CMakeLists.txt             |  10 +
 ...vice_avg_pool3d_fwd_ndhwc_f16_instance.cpp |   4 +-
 ...vice_avg_pool3d_fwd_ndhwc_f32_instance.cpp |   4 +-
 ...vice_max_pool3d_fwd_ndhwc_f16_instance.cpp |   8 +-
 ...vice_max_pool3d_fwd_ndhwc_f32_instance.cpp |   8 +-
 .../pool3d_fwd/pool_fwd_instance_common.hpp   |  41 +++
 .../gpu/pool_fwd/CMakeLists.txt               |  14 -
 ...evice_avg_pool2d_fwd_nhwc_f16_instance.cpp |  23 --
 ...evice_avg_pool2d_fwd_nhwc_f32_instance.cpp |  23 --
 ...evice_max_pool2d_fwd_nhwc_f16_instance.cpp |  30 --
 ...evice_max_pool2d_fwd_nhwc_f32_instance.cpp |  30 --
 .../gpu/pool_fwd/pool_fwd_instance_common.hpp |  55 ---
 .../profiler/profile_pool2d_fwd_impl.hpp      | 264 -------------
 .../profiler/profile_pool3d_fwd_impl.hpp      |  23 +-
 profiler/src/CMakeLists.txt                   |   3 +-
 profiler/src/profile_avg_pool2d_fwd.cpp       | 141 -------
 profiler/src/profile_max_pool3d_fwd.cpp       | 117 +++---
 test/pool_fwd/CMakeLists.txt                  |  10 +-
 test/pool_fwd/test_avg_pool2d_fwd.cpp         |  60 ---
 test/pool_fwd/test_avg_pool3d_fwd.cpp         |  12 +-
 test/pool_fwd/test_max_pool2d_fwd.cpp         |  77 ----
 test/pool_fwd/test_max_pool3d_fwd.cpp         |  15 +-
 test/pool_fwd/test_pool_fwd_common.hpp        |   3 +
 41 files changed, 824 insertions(+), 1574 deletions(-)
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
 rename library/src/tensor_operation_instance/gpu/{pool_fwd => pool3d_fwd}/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp (81%)
 rename library/src/tensor_operation_instance/gpu/{pool_fwd => pool3d_fwd}/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp (81%)
 rename library/src/tensor_operation_instance/gpu/{pool_fwd => pool3d_fwd}/device_max_pool3d_fwd_ndhwc_f16_instance.cpp (74%)
 rename library/src/tensor_operation_instance/gpu/{pool_fwd => pool3d_fwd}/device_max_pool3d_fwd_ndhwc_f32_instance.cpp (74%)
 create mode 100644 library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
 delete mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
 delete mode 100644 profiler/include/profiler/profile_pool2d_fwd_impl.hpp
 delete mode 100644 profiler/src/profile_avg_pool2d_fwd.cpp
 delete mode 100644 test/pool_fwd/test_avg_pool2d_fwd.cpp
 delete mode 100644 test/pool_fwd/test_max_pool2d_fwd.cpp

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9001c2f4e7..31f129b581 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -25,8 +25,8 @@ Full documentation for Composable Kernel is not yet available.
 - Added multi-embeddings support (#542).
 - Added Navi3x blockwise GEMM and real GEMM support (#541).
 - Added Navi grouped ConvBwdWeight support (#505).
-- Added pool3d forward (#697).
-- Added maxpool backward (#750).
+- Added MaxPool, AvgPool forward (#815).
+- Added MaxPool backward (#750).
 
 ### Changed
 - Changed ...
diff --git a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
index 2edaf474b5..db8e0569d7 100644
--- a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
+++ b/client_example/19_pool_fwd/avg_pool3d_fwd.cpp
@@ -16,6 +16,9 @@ using InDataType    = ck::half_t;
 using OutDataType   = ck::half_t;
 using IndexDataType = int32_t;
 
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using OutLayout = ck::tensor_layout::convolution::NDHWC;
+
 constexpr ck::index_t InOutRank  = 5;
 constexpr ck::index_t WindowRank = 3;
 #if 0
@@ -44,33 +47,41 @@ struct SimpleDeviceMem
 
 int main(int argc, char* argv[])
 {
-    ck::index_t N               = 2;
-    ck::index_t C               = 32;
-    ck::index_t Z               = 2;
-    ck::index_t Y               = 2;
-    ck::index_t X               = 2;
-    ck::index_t Di              = 30;
-    ck::index_t Hi              = 30;
-    ck::index_t Wi              = 30;
-    ck::index_t window_stride_d = 2;
-    ck::index_t window_stride_h = 2;
-    ck::index_t window_stride_w = 2;
-    ck::index_t in_left_pad_d   = 1;
-    ck::index_t in_left_pad_h   = 1;
-    ck::index_t in_left_pad_w   = 1;
-    ck::index_t in_right_pad_d  = 1;
-    ck::index_t in_right_pad_h  = 1;
-    ck::index_t in_right_pad_w  = 1;
+    ck::index_t N                 = 2;
+    ck::index_t C                 = 32;
+    ck::index_t Z                 = 2;
+    ck::index_t Y                 = 2;
+    ck::index_t X                 = 2;
+    ck::index_t Di                = 30;
+    ck::index_t Hi                = 30;
+    ck::index_t Wi                = 30;
+    ck::index_t window_stride_d   = 2;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_d = 1;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_d     = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_d    = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
 
-    ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
-    ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
-    ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+    const ck::index_t Zs = (Z - 1) * window_dilation_d + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    ck::index_t Do       = (Di + in_left_pad_d + in_right_pad_d - Zs) / window_stride_d + 1;
+    ck::index_t Ho       = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    ck::index_t Wo       = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
 
     // Pool API only support the order of NCDHW
     std::vector<ck::index_t> in_length              = {N, C, Di, Hi, Wi};
     std::vector<ck::index_t> out_length             = {N, C, Do, Ho, Wo};
     std::vector<ck::index_t> window_spatial_lengths = {Z, Y, X};
-    std::vector<ck::index_t> window_strides   = {window_stride_d, window_stride_h, window_stride_w};
+    std::vector<ck::index_t> window_strides = {window_stride_d, window_stride_h, window_stride_w};
+    std::vector<ck::index_t> window_dilations{
+        window_dilation_d, window_dilation_h, window_dilation_w};
     std::vector<ck::index_t> input_left_pads  = {in_left_pad_d, in_left_pad_h, in_left_pad_w};
     std::vector<ck::index_t> input_right_pads = {in_right_pad_d, in_right_pad_h, in_right_pad_w};
 
@@ -90,6 +101,8 @@ int main(int argc, char* argv[])
                                                                  InDataType,
                                                                  OutDataType,
                                                                  IndexDataType,
+                                                                 InLayout,
+                                                                 OutLayout,
                                                                  ReduceOpId,
                                                                  OutputIndex>;
 
@@ -122,6 +135,7 @@ int main(int argc, char* argv[])
             out_tensor_stride,
             out_tensor_stride,
             window_strides,
+            window_dilations,
             input_left_pads,
             input_right_pads,
             {2, 3, 4});
@@ -181,6 +195,7 @@ int main(int argc, char* argv[])
             out_tensor_stride,
             out_tensor_stride,
             window_strides,
+            window_dilations,
             input_left_pads,
             input_right_pads,
             {2, 3, 4});
diff --git a/client_example/19_pool_fwd/max_pool2d_fwd.cpp b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
index c776dc12da..84b818a60f 100644
--- a/client_example/19_pool_fwd/max_pool2d_fwd.cpp
+++ b/client_example/19_pool_fwd/max_pool2d_fwd.cpp
@@ -10,14 +10,18 @@
 #include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
 
 using InDataType    = ck::half_t;
 using OutDataType   = ck::half_t;
 using IndexDataType = int32_t;
 
-constexpr ck::index_t InOutRank  = 4;
-constexpr ck::index_t WindowRank = 2;
+// We use pool3d to implement pool2d in this example
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using OutLayout = ck::tensor_layout::convolution::NDHWC;
+
+constexpr ck::index_t InOutRank  = 5;
+constexpr ck::index_t WindowRank = 3;
 #if 1
 constexpr auto ReduceOpId  = ck::ReduceTensorOp::MAX;
 constexpr bool OutputIndex = true;
@@ -42,31 +46,66 @@ struct SimpleDeviceMem
     void* p_mem_;
 };
 
+void TransformPool2dparamToPool3d(std::vector<ck::index_t>& input_lengths,
+                                  std::vector<ck::index_t>& window_lengths,
+                                  std::vector<ck::index_t>& output_lengths,
+                                  std::vector<ck::index_t>& input_stride,
+                                  std::vector<ck::index_t>& output_stride,
+                                  std::vector<ck::index_t>& indices_stride,
+                                  std::vector<ck::index_t>& window_strides,
+                                  std::vector<ck::index_t>& window_dilations,
+                                  std::vector<ck::index_t>& input_left_pads,
+                                  std::vector<ck::index_t>& input_right_pads,
+                                  std::vector<ck::index_t>& pooling_dims)
+{
+    // NCHW to NCDHW
+    input_lengths.insert(input_lengths.begin() + 2, 1);
+    output_lengths.insert(output_lengths.begin() + 2, 1);
+    input_stride.insert(input_stride.begin() + 2, 0);
+    output_stride.insert(output_stride.begin() + 2, 0);
+    indices_stride.insert(indices_stride.begin() + 2, 0);
+
+    // YX to ZYX
+    window_lengths.insert(window_lengths.begin(), 1);
+    window_strides.insert(window_strides.begin(), 0);
+    window_dilations.insert(window_dilations.begin(), 0);
+    input_left_pads.insert(input_left_pads.begin(), 0);
+    input_right_pads.insert(input_right_pads.begin(), 0);
+
+    pooling_dims = {2, 3, 4};
+}
+
 int main(int argc, char* argv[])
 {
-    ck::index_t N               = 2;
-    ck::index_t C               = 32;
-    ck::index_t Y               = 2;
-    ck::index_t X               = 2;
-    ck::index_t Hi              = 30;
-    ck::index_t Wi              = 30;
-    ck::index_t window_stride_h = 2;
-    ck::index_t window_stride_w = 2;
-    ck::index_t in_left_pad_h   = 1;
-    ck::index_t in_left_pad_w   = 1;
-    ck::index_t in_right_pad_h  = 1;
-    ck::index_t in_right_pad_w  = 1;
+    ck::index_t N                 = 2;
+    ck::index_t C                 = 32;
+    ck::index_t Y                 = 2;
+    ck::index_t X                 = 2;
+    ck::index_t Hi                = 30;
+    ck::index_t Wi                = 30;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
 
-    ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
-    ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    ck::index_t Ho       = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    ck::index_t Wo       = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
 
     // Pool API only support the order of NCHW
     std::vector<ck::index_t> in_length              = {N, C, Hi, Wi};
     std::vector<ck::index_t> out_length             = {N, C, Ho, Wo};
     std::vector<ck::index_t> window_spatial_lengths = {Y, X};
     std::vector<ck::index_t> window_strides         = {window_stride_h, window_stride_w};
+    std::vector<ck::index_t> window_dilations       = {window_dilation_h, window_dilation_w};
     std::vector<ck::index_t> input_left_pads        = {in_left_pad_h, in_left_pad_w};
     std::vector<ck::index_t> input_right_pads       = {in_right_pad_h, in_right_pad_w};
+    std::vector<ck::index_t> pooling_dims           = {2, 3};
 
     std::size_t in_tensor_size  = N * C * Hi * Wi;
     std::size_t out_tensor_size = N * C * Ho * Wo;
@@ -75,6 +114,18 @@ int main(int argc, char* argv[])
     std::vector<ck::index_t> in_tensor_stride  = {C * Hi * Wi, 1, Wi * C, C};
     std::vector<ck::index_t> out_tensor_stride = {C * Ho * Wo, 1, Wo * C, C};
 
+    TransformPool2dparamToPool3d(in_length,
+                                 window_spatial_lengths,
+                                 out_length,
+                                 in_tensor_stride,
+                                 out_tensor_stride,
+                                 out_tensor_stride,
+                                 window_strides,
+                                 window_dilations,
+                                 input_left_pads,
+                                 input_right_pads,
+                                 pooling_dims);
+
     SimpleDeviceMem in_device_buf(sizeof(InDataType) * in_tensor_size);
     SimpleDeviceMem out_device_buf(sizeof(OutDataType) * out_tensor_size);
     SimpleDeviceMem out_indices_device_buf(sizeof(IndexDataType) * out_tensor_size);
@@ -84,6 +135,8 @@ int main(int argc, char* argv[])
                                                                  InDataType,
                                                                  OutDataType,
                                                                  IndexDataType,
+                                                                 InLayout,
+                                                                 OutLayout,
                                                                  ReduceOpId,
                                                                  OutputIndex>;
 
@@ -116,9 +169,10 @@ int main(int argc, char* argv[])
             out_tensor_stride,
             out_tensor_stride,
             window_strides,
+            window_dilations,
             input_left_pads,
             input_right_pads,
-            {2, 3});
+            pooling_dims);
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
@@ -175,9 +229,10 @@ int main(int argc, char* argv[])
             out_tensor_stride,
             out_tensor_stride,
             window_strides,
+            window_dilations,
             input_left_pads,
             input_right_pads,
-            {2, 3});
+            pooling_dims);
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_common.hpp b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
index 1157ccd387..3ce08fd2af 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_common.hpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_common.hpp
@@ -39,31 +39,35 @@ bool pool_test(bool do_verification,
                ck::index_t Wi,
                ck::index_t window_stride_h,
                ck::index_t window_stride_w,
+               ck::index_t window_dilation_h,
+               ck::index_t window_dilation_w,
                ck::index_t in_left_pad_h,
                ck::index_t in_left_pad_w,
                ck::index_t in_right_pad_h,
                ck::index_t in_right_pad_w)
 {
     using DevicePoolFwdInstance =
-        ck::tensor_operation::device::DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<
-            InDataType,      // InDataType
-            OutDataType,     // OutDataType
-            IndexDataType,   // IndexDataType
-            ComputeDataType, // ComputeDataType
-            ReduceOpId,
-            OutputIndex,
-            64, // BlockSize
-            64, // ReduceMThreadClusterSize
-            1,  // ReduceKThreadClusterSize
-            4,  // ReduceMThreadSliceSize
-            1,  // ReduceKThreadSliceSize
-            4>; // InSrcOutDstVectorSize
+        ck::tensor_operation::device::DevicePool2dFwd_NHWC_NHWC<InDataType,
+                                                                OutDataType,
+                                                                IndexDataType,
+                                                                ComputeDataType,
+                                                                ReduceOpId,
+                                                                OutputIndex,
+                                                                64, // BlockSize
+                                                                64, // ReduceMThreadClusterSize
+                                                                1,  // ReduceKThreadClusterSize
+                                                                4,  // ReduceMThreadSliceSize
+                                                                1,  // ReduceKThreadSliceSize
+                                                                1>; // InSrcOutDstVectorSize
 
-    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
-    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
 
     const std::vector<ck::index_t> window_spatial_lengths{Y, X};
     const std::vector<ck::index_t> window_strides{window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> window_dilations{window_dilation_h, window_dilation_w};
     const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
     const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
 
@@ -123,6 +127,7 @@ bool pool_test(bool do_verification,
         {C * Ho * Wo, 1, Wo * C, C},
         {C * Ho * Wo, 1, Wo * C, C},
         window_strides,
+        window_dilations,
         input_left_pads,
         input_right_pads,
         {2, 3});
@@ -144,8 +149,8 @@ bool pool_test(bool do_verification,
 
     float gb_per_sec = num_btype / 1.E6 / ave_time;
 
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
-              << std::endl;
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+              << " GB / s " << std::endl;
 
     bool pass = true;
 
@@ -169,6 +174,7 @@ bool pool_test(bool do_verification,
                                                              out_indices_n_c_ho_wo_host,
                                                              window_spatial_lengths,
                                                              window_strides,
+                                                             window_dilations,
                                                              input_left_pads,
                                                              input_right_pads);
 
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
index daf8540d49..d767e92248 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp16.cpp
@@ -34,18 +34,20 @@ int main(int argc, char* argv[])
     bool time_kernel;
 
     // Pool shape
-    ck::index_t N               = 128;
-    ck::index_t C               = 192;
-    ck::index_t Y               = 3;
-    ck::index_t X               = 3;
-    ck::index_t Hi              = 71;
-    ck::index_t Wi              = 71;
-    ck::index_t window_stride_h = 2;
-    ck::index_t window_stride_w = 2;
-    ck::index_t in_left_pad_h   = 1;
-    ck::index_t in_left_pad_w   = 1;
-    ck::index_t in_right_pad_h  = 1;
-    ck::index_t in_right_pad_w  = 1;
+    ck::index_t N                 = 128;
+    ck::index_t C                 = 192;
+    ck::index_t Y                 = 3;
+    ck::index_t X                 = 3;
+    ck::index_t Hi                = 71;
+    ck::index_t Wi                = 71;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
 
     if(argc == 1)
     {
@@ -59,31 +61,33 @@ int main(int argc, char* argv[])
         init_method     = std::stoi(argv[2]);
         time_kernel     = static_cast<bool>(std::stoi(argv[3]));
     }
-    else if(argc == 16)
+    else if(argc == 18)
     {
         do_verification = std::stoi(argv[1]);
         init_method     = std::stoi(argv[2]);
         time_kernel     = static_cast<bool>(std::stoi(argv[3]));
 
-        N               = std::stoi(argv[4]);
-        C               = std::stoi(argv[5]);
-        Y               = std::stoi(argv[6]);
-        X               = std::stoi(argv[7]);
-        Hi              = std::stoi(argv[8]);
-        Wi              = std::stoi(argv[9]);
-        window_stride_h = std::stoi(argv[10]);
-        window_stride_w = std::stoi(argv[11]);
-        in_left_pad_h   = std::stoi(argv[12]);
-        in_left_pad_w   = std::stoi(argv[13]);
-        in_right_pad_h  = std::stoi(argv[14]);
-        in_right_pad_w  = std::stoi(argv[15]);
+        N                 = std::stoi(argv[4]);
+        C                 = std::stoi(argv[5]);
+        Y                 = std::stoi(argv[6]);
+        X                 = std::stoi(argv[7]);
+        Hi                = std::stoi(argv[8]);
+        Wi                = std::stoi(argv[9]);
+        window_stride_h   = std::stoi(argv[10]);
+        window_stride_w   = std::stoi(argv[11]);
+        window_dilation_h = std::stoi(argv[12]);
+        window_dilation_w = std::stoi(argv[13]);
+        in_left_pad_h     = std::stoi(argv[14]);
+        in_left_pad_w     = std::stoi(argv[15]);
+        in_right_pad_h    = std::stoi(argv[16]);
+        in_right_pad_w    = std::stoi(argv[17]);
     }
     else
     {
         printf("arg1: verification (0=no, 1=yes)\n");
         printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
         printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 15: N, C, Y, X, Hi, Wi, Sy, Sx, LeftPy, LeftPx, RightPy, "
+        printf("arg4 to 15: N, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
                "RightPx\n");
         exit(0);
     }
@@ -107,6 +111,8 @@ int main(int argc, char* argv[])
                                        Wi,
                                        window_stride_h,
                                        window_stride_w,
+                                       window_dilation_h,
+                                       window_dilation_w,
                                        in_left_pad_h,
                                        in_left_pad_w,
                                        in_right_pad_h,
diff --git a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
index 323e3f61f8..2621500ef1 100644
--- a/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
+++ b/example/13_pool2d_fwd/pool2d_fwd_fp32.cpp
@@ -34,18 +34,20 @@ int main(int argc, char* argv[])
     bool time_kernel;
 
     // Pool shape
-    ck::index_t N               = 128;
-    ck::index_t C               = 192;
-    ck::index_t Y               = 3;
-    ck::index_t X               = 3;
-    ck::index_t Hi              = 71;
-    ck::index_t Wi              = 71;
-    ck::index_t window_stride_h = 2;
-    ck::index_t window_stride_w = 2;
-    ck::index_t in_left_pad_h   = 1;
-    ck::index_t in_left_pad_w   = 1;
-    ck::index_t in_right_pad_h  = 1;
-    ck::index_t in_right_pad_w  = 1;
+    ck::index_t N                 = 128;
+    ck::index_t C                 = 192;
+    ck::index_t Y                 = 3;
+    ck::index_t X                 = 3;
+    ck::index_t Hi                = 71;
+    ck::index_t Wi                = 71;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
 
     if(argc == 1)
     {
@@ -59,31 +61,33 @@ int main(int argc, char* argv[])
         init_method     = std::stoi(argv[2]);
         time_kernel     = static_cast<bool>(std::stoi(argv[3]));
     }
-    else if(argc == 16)
+    else if(argc == 18)
     {
         do_verification = std::stoi(argv[1]);
         init_method     = std::stoi(argv[2]);
         time_kernel     = static_cast<bool>(std::stoi(argv[3]));
 
-        N               = std::stoi(argv[4]);
-        C               = std::stoi(argv[5]);
-        Y               = std::stoi(argv[6]);
-        X               = std::stoi(argv[7]);
-        Hi              = std::stoi(argv[8]);
-        Wi              = std::stoi(argv[9]);
-        window_stride_h = std::stoi(argv[10]);
-        window_stride_w = std::stoi(argv[11]);
-        in_left_pad_h   = std::stoi(argv[12]);
-        in_left_pad_w   = std::stoi(argv[13]);
-        in_right_pad_h  = std::stoi(argv[14]);
-        in_right_pad_w  = std::stoi(argv[15]);
+        N                 = std::stoi(argv[4]);
+        C                 = std::stoi(argv[5]);
+        Y                 = std::stoi(argv[6]);
+        X                 = std::stoi(argv[7]);
+        Hi                = std::stoi(argv[8]);
+        Wi                = std::stoi(argv[9]);
+        window_stride_h   = std::stoi(argv[10]);
+        window_stride_w   = std::stoi(argv[11]);
+        window_dilation_h = std::stoi(argv[12]);
+        window_dilation_w = std::stoi(argv[13]);
+        in_left_pad_h     = std::stoi(argv[14]);
+        in_left_pad_w     = std::stoi(argv[15]);
+        in_right_pad_h    = std::stoi(argv[16]);
+        in_right_pad_w    = std::stoi(argv[17]);
     }
     else
     {
         printf("arg1: verification (0=no, 1=yes)\n");
         printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
         printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 15: N, C, Y, X, Hi, Wi, Sy, Sx, LeftPy, LeftPx, RightPy, "
+        printf("arg4 to 15: N, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
                "RightPx\n");
         exit(0);
     }
@@ -107,6 +111,8 @@ int main(int argc, char* argv[])
                                        Wi,
                                        window_stride_h,
                                        window_stride_w,
+                                       window_dilation_h,
+                                       window_dilation_w,
                                        in_left_pad_h,
                                        in_left_pad_w,
                                        in_right_pad_h,
diff --git a/example/48_pool3d_fwd/pool3d_fwd_common.hpp b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
index 565bb94e47..39032fa123 100644
--- a/example/48_pool3d_fwd/pool3d_fwd_common.hpp
+++ b/example/48_pool3d_fwd/pool3d_fwd_common.hpp
@@ -18,7 +18,45 @@
 #include "ck/library/utility/literals.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
 
-template <typename InDataType,
+template <typename TensorLayout>
+std::vector<ck::index_t> f_tensor_strides_ncdhw(ck::index_t N_,
+                                                ck::index_t C_,
+                                                ck::index_t D,
+                                                ck::index_t H,
+                                                ck::index_t W,
+                                                TensorLayout layout)
+{
+    using namespace ck::literals;
+    (void)N_;
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+        return {C_ * D * H * W, D * H * W, H * W, W, 1_uz};
+    else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+        return {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_};
+};
+
+template <typename TensorLayout>
+HostTensorDescriptor f_host_tensor_descriptor(std::size_t N_,
+                                              std::size_t C_,
+                                              std::size_t D,
+                                              std::size_t H,
+                                              std::size_t W,
+                                              TensorLayout layout)
+{
+    using namespace ck::literals;
+
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
+    {
+        return HostTensorDescriptor({N_, C_, D, H, W}, {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
+    }
+    else if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+    {
+        return HostTensorDescriptor({N_, C_, D, H, W},
+                                    {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+    }
+};
+
+template <typename DevicePoolFwdInstance,
+          typename InDataType,
           typename OutDataType,
           typename ComputeDataType,
           typename IndexDataType,
@@ -40,6 +78,9 @@ bool pool3d_test(bool do_verification,
                  ck::index_t window_stride_d,
                  ck::index_t window_stride_h,
                  ck::index_t window_stride_w,
+                 ck::index_t window_dilation_d,
+                 ck::index_t window_dilation_h,
+                 ck::index_t window_dilation_w,
                  ck::index_t in_left_pad_d,
                  ck::index_t in_left_pad_h,
                  ck::index_t in_left_pad_w,
@@ -47,53 +88,21 @@ bool pool3d_test(bool do_verification,
                  ck::index_t in_right_pad_h,
                  ck::index_t in_right_pad_w)
 {
-    using DevicePoolFwdInstance =
-        ck::tensor_operation::device::DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<
-            InDataType,      // InDataType
-            OutDataType,     // OutDataType
-            IndexDataType,   // IndexDataType
-            ComputeDataType, // ComputeDataType
-            ReduceOpId,
-            OutputIndex,
-            64, // BlockSize
-            64, // ReduceMThreadClusterSize
-            1,  // ReduceKThreadClusterSize
-            4,  // ReduceMThreadSliceSize
-            1,  // ReduceKThreadSliceSize
-            4>; // InSrcOutDstVectorSize
-
-    const ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Z) / window_stride_d + 1;
-    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
-    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+    const ck::index_t Zs = (Z - 1) * window_dilation_d + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    const ck::index_t Do = (Di + in_left_pad_d + in_right_pad_d - Zs) / window_stride_d + 1;
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
 
     const std::vector<ck::index_t> window_spatial_lengths{Z, Y, X};
     const std::vector<ck::index_t> window_strides{
         window_stride_d, window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> window_dilations{
+        window_dilation_d, window_dilation_h, window_dilation_w};
     const std::vector<ck::index_t> input_left_pads{in_left_pad_d, in_left_pad_h, in_left_pad_w};
     const std::vector<ck::index_t> input_right_pads{in_right_pad_d, in_right_pad_h, in_right_pad_w};
 
-    // tensor layout
-    auto f_host_tensor_descriptor = [](std::size_t N_,
-                                       std::size_t C_,
-                                       std::size_t D,
-                                       std::size_t H,
-                                       std::size_t W,
-                                       auto layout) {
-        using namespace ck::literals;
-
-        if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NCDHW>::value)
-        {
-            return HostTensorDescriptor({N_, C_, D, H, W},
-                                        {C_ * D * H * W, D * H * W, H * W, W, 1_uz});
-        }
-        else if constexpr(ck::is_same<decltype(layout),
-                                      ck::tensor_layout::convolution::NDHWC>::value)
-        {
-            return HostTensorDescriptor({N_, C_, D, H, W},
-                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
-        }
-    };
-
     Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi, InLayout{}));
     Tensor<OutDataType> out_n_c_do_ho_wo_host(
         f_host_tensor_descriptor(N, C, Do, Ho, Wo, OutLayout{}));
@@ -126,10 +135,11 @@ bool pool3d_test(bool do_verification,
         {N, C, Di, Hi, Wi},
         {Z, Y, X},
         {N, C, Do, Ho, Wo},
-        {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C},
-        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
-        {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
+        f_tensor_strides_ncdhw(N, C, Di, Hi, Wi, InLayout{}),
+        f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, OutLayout{}),
+        f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, OutLayout{}),
         window_strides,
+        window_dilations,
         input_left_pads,
         input_right_pads,
         {2, 3, 4});
@@ -165,6 +175,7 @@ bool pool3d_test(bool do_verification,
                                                              out_indices_n_c_do_ho_wo_host,
                                                              window_spatial_lengths,
                                                              window_strides,
+                                                             window_dilations,
                                                              input_left_pads,
                                                              input_right_pads);
 
diff --git a/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
index 9afb51201d..b9ac61033d 100644
--- a/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
+++ b/example/48_pool3d_fwd/pool3d_fwd_fp16.cpp
@@ -27,31 +27,49 @@ static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
 static constexpr bool OutputIndex  = false;
 static constexpr bool PropagateNan = false;
 
+using DevicePoolFwdInstance =
+    ck::tensor_operation::device::DevicePool3dFwd_NDHWC_NDHWC<InDataType,
+                                                              OutDataType,
+                                                              IndexDataType,
+                                                              ComputeDataType,
+                                                              ReduceOpId,
+                                                              OutputIndex,
+                                                              64, // BlockSize
+                                                              64, // ReduceMThreadClusterSize
+                                                              1,  // ReduceKThreadClusterSize
+                                                              1,  // ReduceMThreadSliceSize
+                                                              1,  // ReduceKThreadSliceSize
+                                                              1>; // InSrcOutDstVectorSize
+
 int main()
 {
     bool do_verification = true;
     bool time_kernel     = false;
 
     // Pool shape
-    ck::index_t N               = 2;
-    ck::index_t C               = 32;
-    ck::index_t Z               = 2;
-    ck::index_t Y               = 2;
-    ck::index_t X               = 2;
-    ck::index_t Di              = 30;
-    ck::index_t Hi              = 30;
-    ck::index_t Wi              = 30;
-    ck::index_t window_stride_d = 2;
-    ck::index_t window_stride_h = 2;
-    ck::index_t window_stride_w = 2;
-    ck::index_t in_left_pad_d   = 1;
-    ck::index_t in_left_pad_h   = 1;
-    ck::index_t in_left_pad_w   = 1;
-    ck::index_t in_right_pad_d  = 1;
-    ck::index_t in_right_pad_h  = 1;
-    ck::index_t in_right_pad_w  = 1;
+    ck::index_t N                 = 2;
+    ck::index_t C                 = 32;
+    ck::index_t Z                 = 2;
+    ck::index_t Y                 = 2;
+    ck::index_t X                 = 2;
+    ck::index_t Di                = 30;
+    ck::index_t Hi                = 30;
+    ck::index_t Wi                = 30;
+    ck::index_t window_stride_d   = 2;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_d = 1;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_d     = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_d    = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
 
-    bool pass = pool3d_test<InDataType,
+    bool pass = pool3d_test<DevicePoolFwdInstance,
+                            InDataType,
                             OutDataType,
                             ComputeDataType,
                             IndexDataType,
@@ -72,6 +90,9 @@ int main()
                                          window_stride_d,
                                          window_stride_h,
                                          window_stride_w,
+                                         window_dilation_d,
+                                         window_dilation_h,
+                                         window_dilation_w,
                                          in_left_pad_d,
                                          in_left_pad_h,
                                          in_left_pad_w,
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
index 08a8009c6c..a6c8f3b884 100644
--- a/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_bf16.cpp
@@ -24,18 +24,20 @@ int main()
     bool time_kernel     = false;
 
     // Pool shape
-    ck::index_t N               = 1;
-    ck::index_t C               = 1;
-    ck::index_t Y               = 3;
-    ck::index_t X               = 3;
-    ck::index_t Hi              = 32;
-    ck::index_t Wi              = 32;
-    ck::index_t window_stride_h = 1;
-    ck::index_t window_stride_w = 1;
-    ck::index_t in_left_pad_h   = 0;
-    ck::index_t in_left_pad_w   = 0;
-    ck::index_t in_right_pad_h  = 0;
-    ck::index_t in_right_pad_w  = 0;
+    ck::index_t N                 = 1;
+    ck::index_t C                 = 1;
+    ck::index_t Y                 = 3;
+    ck::index_t X                 = 3;
+    ck::index_t Hi                = 32;
+    ck::index_t Wi                = 32;
+    ck::index_t window_stride_h   = 1;
+    ck::index_t window_stride_w   = 1;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 0;
+    ck::index_t in_left_pad_w     = 0;
+    ck::index_t in_right_pad_h    = 0;
+    ck::index_t in_right_pad_w    = 0;
 
     bool pass = maxpool_bwd_test<InDataType,
                                  OutDataType,
@@ -53,6 +55,8 @@ int main()
                                                Wi,
                                                window_stride_h,
                                                window_stride_w,
+                                               window_dilation_h,
+                                               window_dilation_w,
                                                in_left_pad_h,
                                                in_left_pad_w,
                                                in_right_pad_h,
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
index 045793cc2a..d730067940 100644
--- a/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
@@ -36,6 +36,8 @@ bool maxpool_bwd_test(bool do_verification,
                       ck::index_t Wi,
                       ck::index_t window_stride_h,
                       ck::index_t window_stride_w,
+                      ck::index_t window_dilation_h,
+                      ck::index_t window_dilation_w,
                       ck::index_t in_left_pad_h,
                       ck::index_t in_left_pad_w,
                       ck::index_t in_right_pad_h,
@@ -44,28 +46,30 @@ bool maxpool_bwd_test(bool do_verification,
     using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
     using DevicePoolFwdInstance =
-        ck::tensor_operation::device::DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<
-            InDataType,      // InDataType
-            OutDataType,     // OutDataType
-            IndexDataType,   // IndexDataType
-            ComputeDataType, // ComputeDataType
-            ck::ReduceTensorOp::MAX,
-            true, // OutputIndex
-            64,   // BlockSize
-            64,   // ReduceMThreadClusterSize
-            1,    // ReduceKThreadClusterSize
-            4,    // ReduceMThreadSliceSize
-            1,    // ReduceKThreadSliceSize
-            1>;   // InSrcOutDstVectorSize
+        ck::tensor_operation::device::DevicePool2dFwd_NHWC_NHWC<InDataType,      // InDataType
+                                                                OutDataType,     // OutDataType
+                                                                IndexDataType,   // IndexDataType
+                                                                ComputeDataType, // ComputeDataType
+                                                                ck::ReduceTensorOp::MAX,
+                                                                true,
+                                                                64, // BlockSize
+                                                                64, // ReduceMThreadClusterSize
+                                                                1,  // ReduceKThreadClusterSize
+                                                                4,  // ReduceMThreadSliceSize
+                                                                1,  // ReduceKThreadSliceSize
+                                                                1>; // InSrcOutDstVectorSize
 
     using DeviceMaxPoolBwdInstance = ck::tensor_operation::device::
         DeviceIndexPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 4>;
 
-    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Y) / window_stride_h + 1;
-    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - X) / window_stride_w + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    const ck::index_t Ho = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    const ck::index_t Wo = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
 
     const std::vector<ck::index_t> window_spatial_lengths{Y, X};
     const std::vector<ck::index_t> window_strides{window_stride_h, window_stride_w};
+    const std::vector<ck::index_t> window_dilations{window_dilation_h, window_dilation_w};
     const std::vector<ck::index_t> input_left_pads{in_left_pad_h, in_left_pad_w};
     const std::vector<ck::index_t> input_right_pads{in_right_pad_h, in_right_pad_w};
 
@@ -128,6 +132,7 @@ bool maxpool_bwd_test(bool do_verification,
         {C * Ho * Wo, 1, Wo * C, C},
         {C * Ho * Wo, 1, Wo * C, C},
         window_strides,
+        window_dilations,
         input_left_pads,
         input_right_pads,
         {2, 3});
@@ -191,6 +196,7 @@ bool maxpool_bwd_test(bool do_verification,
                                                                      indices_n_c_ho_wo_host,
                                                                      window_spatial_lengths,
                                                                      window_strides,
+                                                                     window_dilations,
                                                                      input_left_pads,
                                                                      input_right_pads);
         ref_pooling_fwd_invoker.Run(ref_pooling_fwd_argument);
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
index c6b0b2c860..a4f982d855 100644
--- a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp16.cpp
@@ -24,18 +24,20 @@ int main()
     bool time_kernel     = false;
 
     // Pool shape
-    ck::index_t N               = 1;
-    ck::index_t C               = 1;
-    ck::index_t Y               = 3;
-    ck::index_t X               = 3;
-    ck::index_t Hi              = 32;
-    ck::index_t Wi              = 32;
-    ck::index_t window_stride_h = 1;
-    ck::index_t window_stride_w = 1;
-    ck::index_t in_left_pad_h   = 0;
-    ck::index_t in_left_pad_w   = 0;
-    ck::index_t in_right_pad_h  = 0;
-    ck::index_t in_right_pad_w  = 0;
+    ck::index_t N                 = 1;
+    ck::index_t C                 = 1;
+    ck::index_t Y                 = 3;
+    ck::index_t X                 = 3;
+    ck::index_t Hi                = 32;
+    ck::index_t Wi                = 32;
+    ck::index_t window_stride_h   = 1;
+    ck::index_t window_stride_w   = 1;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 0;
+    ck::index_t in_left_pad_w     = 0;
+    ck::index_t in_right_pad_h    = 0;
+    ck::index_t in_right_pad_w    = 0;
 
     bool pass = maxpool_bwd_test<InDataType,
                                  OutDataType,
@@ -53,6 +55,8 @@ int main()
                                                Wi,
                                                window_stride_h,
                                                window_stride_w,
+                                               window_dilation_h,
+                                               window_dilation_w,
                                                in_left_pad_h,
                                                in_left_pad_w,
                                                in_right_pad_h,
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
index c79b84c489..c14928a936 100644
--- a/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_fp32.cpp
@@ -24,18 +24,20 @@ int main()
     bool time_kernel     = false;
 
     // Pool shape
-    ck::index_t N               = 1;
-    ck::index_t C               = 1;
-    ck::index_t Y               = 2;
-    ck::index_t X               = 2;
-    ck::index_t Hi              = 32;
-    ck::index_t Wi              = 32;
-    ck::index_t window_stride_h = 2;
-    ck::index_t window_stride_w = 2;
-    ck::index_t in_left_pad_h   = 0;
-    ck::index_t in_left_pad_w   = 0;
-    ck::index_t in_right_pad_h  = 0;
-    ck::index_t in_right_pad_w  = 0;
+    ck::index_t N                 = 1;
+    ck::index_t C                 = 1;
+    ck::index_t Y                 = 2;
+    ck::index_t X                 = 2;
+    ck::index_t Hi                = 32;
+    ck::index_t Wi                = 32;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 0;
+    ck::index_t in_left_pad_w     = 0;
+    ck::index_t in_right_pad_h    = 0;
+    ck::index_t in_right_pad_w    = 0;
 
     bool pass = maxpool_bwd_test<InDataType,
                                  OutDataType,
@@ -53,6 +55,8 @@ int main()
                                                Wi,
                                                window_stride_h,
                                                window_stride_w,
+                                               window_dilation_h,
+                                               window_dilation_w,
                                                in_left_pad_h,
                                                in_left_pad_w,
                                                in_right_pad_h,
diff --git a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
index 8b227fdfbe..62071c43a1 100644
--- a/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_pool_fwd.hpp
@@ -17,6 +17,8 @@ template <index_t InOutRank,
           typename InDataType,
           typename OutDataType,
           typename IndexDataType,
+          typename InLayout,
+          typename OutLayout,
           ReduceTensorOp ReduceOpId,
           bool OutputIndex>
 struct DevicePoolFwd : public BaseOperator
@@ -25,13 +27,14 @@ struct DevicePoolFwd : public BaseOperator
     MakeArgumentPointer(const void* p_in_dev,
                         void* p_out_dev,
                         void* p_out_indices_dev,
-                        std::vector<ck::index_t> input_lengths,
-                        std::vector<ck::index_t> window_lengths,
-                        std::vector<ck::index_t> output_lengths,
-                        std::vector<ck::index_t> input_stride,
-                        std::vector<ck::index_t> output_stride,
-                        std::vector<ck::index_t> indices_stride,
-                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> input_n_c_wis_lengths,
+                        std::vector<ck::index_t> window_xs_lengths,
+                        std::vector<ck::index_t> output_n_c_wos_lengths,
+                        std::vector<ck::index_t> input_n_c_wis_stride,
+                        std::vector<ck::index_t> output_n_c_wis_stride,
+                        std::vector<ck::index_t> indices_n_c_wis_stride,
+                        std::vector<ck::index_t> window_xs_strides,
+                        std::vector<ck::index_t> window_xs_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
                         std::vector<ck::index_t> pooling_dims) = 0;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
index 3f27c629dd..c94c568c49 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp
@@ -3,16 +3,7 @@
 
 #pragma once
 
-#include <iostream>
-#include <sstream>
-
-#include "ck/tensor_description/tensor_descriptor.hpp"
-#include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
-#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
-#include "ck/host_utility/device_prop.hpp"
-#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
 
 namespace ck {
 namespace tensor_operation {
@@ -30,255 +21,32 @@ template <typename InDataType,
           ck::index_t ReduceMThreadSliceSize,
           ck::index_t ReduceKThreadSliceSize,
           ck::index_t InSrcOutDstVectorSize>
-struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C
-    : public DevicePoolFwd<4, 2, InDataType, OutDataType, IndexDataType, ReduceOpId, OutputIndex>
+struct DevicePool2dFwd_NHWC_NHWC : public DevicePool3dFwd_NDHWC_NDHWC<InDataType,
+                                                                      OutDataType,
+                                                                      IndexDataType,
+                                                                      ComputeDataType,
+                                                                      ReduceOpId,
+                                                                      OutputIndex,
+                                                                      BlockSize,
+                                                                      ReduceMThreadClusterSize,
+                                                                      ReduceKThreadClusterSize,
+                                                                      ReduceMThreadSliceSize,
+                                                                      ReduceKThreadSliceSize,
+                                                                      InSrcOutDstVectorSize>
 {
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-    static constexpr auto I4 = Number<4>{};
-    static constexpr auto I5 = Number<5>{};
-
-    static constexpr index_t InOutRank  = 4;
-    static constexpr index_t WindowRank = 2;
-
-    using ReduceOperation = typename reduce_binary_operator<ReduceOpId>::opType;
-
-    using InElementwiseOperation =
-        typename reduce_unary_operator<ReduceOpId, true, true>::InElementwiseOperation;
-
-    using AccElementwiseOperation =
-        typename reduce_unary_operator<ReduceOpId, true, true>::AccElementwiseOperation;
-
-    static constexpr index_t InSrcOutDstVectorDim =
-        0; // for NHWC, the dim C is the vector Dim for both input and output in memory, which is
-           // not reduced.
-
-    static constexpr ck::index_t ReduceM_BlockTileSize =
-        ReduceMThreadClusterSize * ReduceMThreadSliceSize;
-    static constexpr ck::index_t ReduceK_BlockTileSize =
-        ReduceKThreadClusterSize * ReduceKThreadSliceSize;
-
-    static auto MakeABGridDescriptor_A_M_K_B_M(ck::index_t N,
-                                               ck::index_t C,
-                                               std::vector<ck::index_t> input_spatial_lengths,
-                                               std::vector<ck::index_t> window_spatial_lengths,
-                                               std::vector<ck::index_t> output_spatial_lengths,
-                                               std::vector<ck::index_t> window_strides,
-                                               std::vector<ck::index_t> input_left_pads,
-                                               std::vector<ck::index_t> input_right_pads)
-    {
-        const index_t Hi = input_spatial_lengths[0];
-        const index_t Wi = input_spatial_lengths[1];
-
-        const index_t Ho = output_spatial_lengths[0];
-        const index_t Wo = output_spatial_lengths[1];
-
-        const index_t Y = window_spatial_lengths[0];
-        const index_t X = window_spatial_lengths[1];
-
-        const index_t ConvStrideH = window_strides[0];
-        const index_t ConvStrideW = window_strides[1];
-
-        const index_t InLeftPadH = input_left_pads[0];
-        const index_t InLeftPadW = input_left_pads[1];
-
-        const index_t InRightPadH = input_right_pads[0];
-        const index_t InRightPadW = input_right_pads[1];
-
-        const index_t ReduceMRaw = N * Ho * Wo * C;
-        const index_t ReduceMPad =
-            math::integer_least_multiple(ReduceMRaw, ReduceM_BlockTileSize) - ReduceMRaw;
-
-        const index_t ReduceKRaw = Y * X;
-        const index_t ReduceKPad =
-            math::integer_least_multiple(ReduceKRaw, ReduceK_BlockTileSize) - ReduceKRaw;
-
-        // A[ReduceM, ReduceK]
-        const auto in_grid_desc_n_hi_wi_c =
-            make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
-
-        const auto in_grid_desc_n_hip_wip_c = transform_tensor_descriptor(
-            in_grid_desc_n_hi_wi_c,
-            make_tuple(make_pass_through_transform(N),
-                       make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                       make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                       make_pass_through_transform(C)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-        const auto in_grid_desc_n_y_ho_x_wo_c = transform_tensor_descriptor(
-            in_grid_desc_n_hip_wip_c,
-            make_tuple(make_pass_through_transform(N),
-                       make_embed_transform(make_tuple(Y, Ho), make_tuple(I1, ConvStrideH)),
-                       make_embed_transform(make_tuple(X, Wo), make_tuple(I1, ConvStrideW)),
-                       make_pass_through_transform(C)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-        const auto in_grid_desc_reducemraw_reducekraw =
-            transform_tensor_descriptor(in_grid_desc_n_y_ho_x_wo_c,
-                                        make_tuple(make_merge_transform(make_tuple(N, Ho, Wo, C)),
-                                                   make_merge_transform(make_tuple(Y, X))),
-                                        make_tuple(Sequence<0, 2, 4, 5>{}, Sequence<1, 3>{}),
-                                        make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        const auto in_grid_desc_reducem_reducek = transform_tensor_descriptor(
-            in_grid_desc_reducemraw_reducekraw,
-            make_tuple(make_right_pad_transform(ReduceMRaw, ReduceMPad),
-                       make_right_pad_transform(ReduceKRaw, ReduceKPad)),
-            make_tuple(Sequence<0>{}, Sequence<1>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        // B[ReduceM]
-        const auto out_grid_desc_reducemraw =
-            make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo * C));
-
-        const auto out_grid_desc_reducem = transform_tensor_descriptor(
-            out_grid_desc_reducemraw,
-            make_tuple(make_right_pad_transform(ReduceMRaw, ReduceMPad)),
-            make_tuple(Sequence<0>{}),
-            make_tuple(Sequence<0>{}));
-
-        return make_tuple(in_grid_desc_reducem_reducek, out_grid_desc_reducem);
-    }
-
-    using ABGridDescs   = decltype(MakeABGridDescriptor_A_M_K_B_M(1, 1, {}, {}, {}, {}, {}, {}));
-    using AGridDesc_M_K = remove_cvref_t<decltype(ABGridDescs{}[I0])>;
-    using BGridDesc_M   = remove_cvref_t<decltype(ABGridDescs{}[I1])>;
-
-    // TODO
-    struct Argument : public BaseArgument
-    {
-        Argument(const InDataType* p_in_dev,
-                 OutDataType* p_out_dev,
-                 IndexDataType* p_out_indices_dev,
-                 ck::index_t N,
-                 ck::index_t C,
-                 std::vector<ck::index_t>& input_spatial_lengths,
-                 std::vector<ck::index_t>& window_spatial_lengths,
-                 std::vector<ck::index_t>& output_spatial_lengths,
-                 std::vector<ck::index_t>& window_strides,
-                 std::vector<ck::index_t>& input_left_pads,
-                 std::vector<ck::index_t>& input_right_pads)
-            : p_in_dev_{p_in_dev},
-              p_out_dev_{p_out_dev},
-              p_out_indices_dev_{p_out_indices_dev},
-              a_grid_desc_m_k_{},
-              b_grid_desc_m_{}
-        {
-            const auto descs = MakeABGridDescriptor_A_M_K_B_M(N,
-                                                              C,
-                                                              input_spatial_lengths,
-                                                              window_spatial_lengths,
-                                                              output_spatial_lengths,
-                                                              window_strides,
-                                                              input_left_pads,
-                                                              input_right_pads);
-
-            a_grid_desc_m_k_ = descs[I0];
-            b_grid_desc_m_   = descs[I1];
-
-            invariant_lowest_length_ = C;
-            reduce_lowest_length_    = window_spatial_lengths[1];
-
-            int32_t reduceLength = window_spatial_lengths[0] * window_spatial_lengths[1];
-
-            std::tie(in_element_op_, acc_element_op_) =
-                reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(reduceLength);
-        }
-
-        const InDataType* p_in_dev_;
-        OutDataType* p_out_dev_;
-        IndexDataType* p_out_indices_dev_;
-        AGridDesc_M_K a_grid_desc_m_k_;
-        BGridDesc_M b_grid_desc_m_;
-        InElementwiseOperation in_element_op_;
-        AccElementwiseOperation acc_element_op_;
-
-        // for checking vector load/store
-        ck::index_t invariant_lowest_length_;
-        ck::index_t reduce_lowest_length_;
-    };
-
-    struct Invoker : public BaseInvoker
-    {
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
-        {
-            using gridwise_reduce =
-                GridwiseReduction_mk_to_m_threadwise<InDataType,
+    using DevicePool3D = DevicePool3dFwd_NDHWC_NDHWC<InDataType,
                                                      OutDataType,
-                                                     ComputeDataType,
                                                      IndexDataType,
-                                                     AGridDesc_M_K,
-                                                     BGridDesc_M,
-                                                     ReduceOperation,
-                                                     InElementwiseOperation,
-                                                     AccElementwiseOperation,
-                                                     InMemoryDataOperationEnum::Set,
-                                                     false, // propagate_nan
+                                                     ComputeDataType,
+                                                     ReduceOpId,
+                                                     OutputIndex,
                                                      BlockSize,
+                                                     ReduceMThreadClusterSize,
+                                                     ReduceKThreadClusterSize,
                                                      ReduceMThreadSliceSize,
                                                      ReduceKThreadSliceSize,
-                                                     InSrcOutDstVectorDim,
-                                                     InSrcOutDstVectorSize,
                                                      InSrcOutDstVectorSize>;
 
-            const auto kernel =
-                kernel_reduce_threadwise<gridwise_reduce,
-                                         OutputIndex,
-                                         true,  // pooling need to return global index
-                                         false, // don't have index input
-                                         InDataType,
-                                         OutDataType,
-                                         ComputeDataType,
-                                         IndexDataType,
-                                         AGridDesc_M_K,
-                                         BGridDesc_M,
-                                         InElementwiseOperation,
-                                         AccElementwiseOperation>;
-
-            ck::index_t ReduceM = arg.a_grid_desc_m_k_.GetLength(I0);
-
-            const index_t grid_size = (ReduceM / ReduceM_BlockTileSize);
-
-            return launch_and_time_kernel(stream_config,
-                                          kernel,
-                                          dim3(grid_size),
-                                          dim3(BlockSize),
-                                          0,
-                                          arg.a_grid_desc_m_k_,
-                                          arg.b_grid_desc_m_,
-                                          arg.in_element_op_,
-                                          arg.acc_element_op_,
-                                          float(1),
-                                          arg.p_in_dev_,
-                                          nullptr,
-                                          float(0),
-                                          arg.p_out_dev_,
-                                          arg.p_out_indices_dev_);
-        }
-
-        float Run(const BaseArgument* p_arg,
-                  const StreamConfig& stream_config = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
-        }
-    };
-
-    bool IsSupportedArgument(const BaseArgument* p_arg) override
-    {
-        const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
-
-        if(pArg->invariant_lowest_length_ % InSrcOutDstVectorSize != 0)
-        {
-            return (false);
-        }
-
-        return (true);
-    }
-
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in_dev,
                         void* p_out_dev,
@@ -286,62 +54,57 @@ struct DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C
                         std::vector<ck::index_t> input_lengths,
                         std::vector<ck::index_t> window_lengths,
                         std::vector<ck::index_t> output_lengths,
-                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
-                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
-                        std::vector<ck::index_t>, // Suppose tensor layout = NHWC
+                        std::vector<ck::index_t> input_stride,
+                        std::vector<ck::index_t> output_stride,
+                        std::vector<ck::index_t> indices_stride,
                         std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> window_dilations,
                         std::vector<ck::index_t> input_left_pads,
                         std::vector<ck::index_t> input_right_pads,
                         std::vector<ck::index_t> pooling_dims) override
     {
+        static constexpr index_t InOutRank  = 4;
+        static constexpr index_t WindowRank = 2;
+
         if(input_lengths.size() != InOutRank || window_lengths.size() != WindowRank ||
            input_lengths.size() != InOutRank || window_strides.size() != WindowRank ||
-           input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+           window_dilations.size() != WindowRank || input_left_pads.size() != WindowRank ||
+           input_right_pads.size() != WindowRank)
             throw std::runtime_error("dimension is incorrect");
 
         if(pooling_dims != std::vector<ck::index_t>{2, 3})
             throw std::runtime_error("pooling_dims only support {2, 3} in pool2d so far");
 
-        index_t N  = input_lengths[0];
-        index_t C  = input_lengths[1];
-        index_t Hi = input_lengths[2];
-        index_t Wi = input_lengths[3];
-        index_t Ho = output_lengths[2];
-        index_t Wo = output_lengths[3];
+        // NCHW to NCDHW
+        input_lengths.insert(input_lengths.begin() + 2, 1);
+        output_lengths.insert(output_lengths.begin() + 2, 1);
+        input_stride.insert(input_stride.begin() + 2, 0);
+        output_stride.insert(output_stride.begin() + 2, 0);
+        indices_stride.insert(indices_stride.begin() + 2, 0);
 
-        std::vector<ck::index_t> input_spatial_lengths  = {Hi, Wi};
-        std::vector<ck::index_t> output_spatial_lengths = {Ho, Wo};
+        // YX to ZYX
+        window_lengths.insert(window_lengths.begin(), 1);
+        window_strides.insert(window_strides.begin(), 0);
+        window_dilations.insert(window_dilations.begin(), 0);
+        input_left_pads.insert(input_left_pads.begin(), 0);
+        input_right_pads.insert(input_right_pads.begin(), 0);
 
-        return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_dev),
-                                          static_cast<OutDataType*>(p_out_dev),
-                                          static_cast<IndexDataType*>(p_out_indices_dev),
-                                          N,
-                                          C,
-                                          input_spatial_lengths,
-                                          window_lengths,
-                                          output_spatial_lengths,
-                                          window_strides,
-                                          input_left_pads,
-                                          input_right_pads);
-    }
+        pooling_dims = {2, 3, 4};
 
-    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<" << BlockSize << ",";
-        str << "M_C" << ReduceMThreadClusterSize << "_S" << ReduceMThreadSliceSize << ",";
-        str << "K_C" << ReduceKThreadClusterSize << "_S" << ReduceKThreadSliceSize << ",";
-        str <<"InSrcOutDstVectorSize_" << InSrcOutDstVectorSize << ">";
-        // clang-format on
-
-        return str.str();
+        return DevicePool3D::MakeArgumentPointer(p_in_dev,
+                                                 p_out_dev,
+                                                 p_out_indices_dev,
+                                                 input_lengths,
+                                                 window_lengths,
+                                                 output_lengths,
+                                                 input_stride,
+                                                 output_stride,
+                                                 indices_stride,
+                                                 window_strides,
+                                                 window_dilations,
+                                                 input_left_pads,
+                                                 input_right_pads,
+                                                 pooling_dims);
     }
 };
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
index 0ab6c24758..384805a0a3 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp
@@ -8,8 +8,10 @@
 
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/reduction_operator_mapping.hpp"
 #include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_reduce_common.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_2d_reduction_threadwise.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
@@ -30,8 +32,15 @@ template <typename InDataType,
           ck::index_t MThreadSliceSize,
           ck::index_t KThreadSliceSize,
           ck::index_t InSrcOutDstVectorSize>
-struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
-    : public DevicePoolFwd<5, 3, InDataType, OutDataType, IndexDataType, ReduceOpId, OutputIndex>
+struct DevicePool3dFwd_NDHWC_NDHWC : public DevicePoolFwd<5,
+                                                          3,
+                                                          InDataType,
+                                                          OutDataType,
+                                                          IndexDataType,
+                                                          tensor_layout::convolution::NDHWC,
+                                                          tensor_layout::convolution::NDHWC,
+                                                          ReduceOpId,
+                                                          OutputIndex>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
@@ -51,45 +60,48 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
     using AccElementwiseOperation =
         typename reduce_unary_operator<ReduceOpId, true, true>::AccElementwiseOperation;
 
-    // for NDHWC, the dim C is the vector Dim for both input and output in memory, which is not
-    // reduced.
-    static constexpr index_t InSrcOutDstVectorDim = 0;
-
     static constexpr ck::index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
     static constexpr ck::index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
 
-    static auto MakeABGridDescriptor_A_M_K_B_M(ck::index_t N,
-                                               ck::index_t C,
-                                               std::vector<ck::index_t> input_spatial_lengths,
-                                               std::vector<ck::index_t> window_spatial_lengths,
-                                               std::vector<ck::index_t> output_spatial_lengths,
-                                               std::vector<ck::index_t> window_strides,
-                                               std::vector<ck::index_t> input_left_pads,
-                                               std::vector<ck::index_t> input_right_pads)
+    static auto MakeABGridDescriptor_A_M_K_B_M(std::vector<ck::index_t> input_ncdhw_lengths,
+                                               std::vector<ck::index_t> output_ncdhw_lengths,
+                                               std::vector<ck::index_t> input_ncdhw_stride,
+                                               std::vector<ck::index_t> output_ncdhw_stride,
+                                               std::vector<ck::index_t> window_spatial_zyx_lengths,
+                                               std::vector<ck::index_t> window_zyx_strides,
+                                               std::vector<ck::index_t> window_zyx_dilations,
+                                               std::vector<ck::index_t> input_left_dhw_pads,
+                                               std::vector<ck::index_t> input_right_dhw_pads)
     {
-        const index_t Di = input_spatial_lengths[0];
-        const index_t Hi = input_spatial_lengths[1];
-        const index_t Wi = input_spatial_lengths[2];
+        const index_t N  = input_ncdhw_lengths[0];
+        const index_t C  = input_ncdhw_lengths[1];
+        const index_t Di = input_ncdhw_lengths[2];
+        const index_t Hi = input_ncdhw_lengths[3];
+        const index_t Wi = input_ncdhw_lengths[4];
 
-        const index_t Do = output_spatial_lengths[0];
-        const index_t Ho = output_spatial_lengths[1];
-        const index_t Wo = output_spatial_lengths[2];
+        const index_t Do = output_ncdhw_lengths[2];
+        const index_t Ho = output_ncdhw_lengths[3];
+        const index_t Wo = output_ncdhw_lengths[4];
 
-        const index_t Z = window_spatial_lengths[0];
-        const index_t Y = window_spatial_lengths[1];
-        const index_t X = window_spatial_lengths[2];
+        const index_t Z = window_spatial_zyx_lengths[0];
+        const index_t Y = window_spatial_zyx_lengths[1];
+        const index_t X = window_spatial_zyx_lengths[2];
 
-        const index_t ConvStrideD = window_strides[0];
-        const index_t ConvStrideH = window_strides[1];
-        const index_t ConvStrideW = window_strides[2];
+        const index_t WindowStrideD = window_zyx_strides[0];
+        const index_t WindowStrideH = window_zyx_strides[1];
+        const index_t WindowStrideW = window_zyx_strides[2];
 
-        const index_t InLeftPadD = input_left_pads[0];
-        const index_t InLeftPadH = input_left_pads[1];
-        const index_t InLeftPadW = input_left_pads[2];
+        const index_t WindowDilationD = window_zyx_dilations[0];
+        const index_t WindowDilationH = window_zyx_dilations[1];
+        const index_t WindowDilationW = window_zyx_dilations[2];
 
-        const index_t InRightPadD = input_right_pads[0];
-        const index_t InRightPadH = input_right_pads[1];
-        const index_t InRightPadW = input_right_pads[2];
+        const index_t InLeftPadD = input_left_dhw_pads[0];
+        const index_t InLeftPadH = input_left_dhw_pads[1];
+        const index_t InLeftPadW = input_left_dhw_pads[2];
+
+        const index_t InRightPadD = input_right_dhw_pads[0];
+        const index_t InRightPadH = input_right_dhw_pads[1];
+        const index_t InRightPadW = input_right_dhw_pads[2];
 
         const index_t MRaw = N * Do * Ho * Wo * C;
         const index_t MPad = math::integer_least_multiple(MRaw, M_BlockTileSize) - MRaw;
@@ -98,8 +110,15 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
         const index_t KPad = math::integer_least_multiple(KRaw, K_BlockTileSize) - KRaw;
 
         // A[ReduceM, ReduceK]
-        const auto in_grid_desc_n_di_hi_wi_c =
-            make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
+        const index_t Ni_stride = input_ncdhw_stride[0];
+        const index_t Ci_stride = input_ncdhw_stride[1];
+        const index_t Di_stride = input_ncdhw_stride[2];
+        const index_t Hi_stride = input_ncdhw_stride[3];
+        const index_t Wi_stride = input_ncdhw_stride[4];
+
+        const auto in_grid_desc_n_di_hi_wi_c = make_naive_tensor_descriptor(
+            make_tuple(N, Di, Hi, Wi, C),
+            make_tuple(Ni_stride, Di_stride, Hi_stride, Wi_stride, Ci_stride));
 
         const auto in_grid_desc_n_dip_hip_wip_c = transform_tensor_descriptor(
             in_grid_desc_n_di_hi_wi_c,
@@ -113,11 +132,12 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
 
         const auto in_grid_desc_n_z_do_y_ho_x_wo_c = transform_tensor_descriptor(
             in_grid_desc_n_dip_hip_wip_c,
-            make_tuple(make_pass_through_transform(N),
-                       make_embed_transform(make_tuple(Z, Do), make_tuple(I1, ConvStrideD)),
-                       make_embed_transform(make_tuple(Y, Ho), make_tuple(I1, ConvStrideH)),
-                       make_embed_transform(make_tuple(X, Wo), make_tuple(I1, ConvStrideW)),
-                       make_pass_through_transform(C)),
+            make_tuple(
+                make_pass_through_transform(N),
+                make_embed_transform(make_tuple(Z, Do), make_tuple(WindowDilationD, WindowStrideD)),
+                make_embed_transform(make_tuple(Y, Ho), make_tuple(WindowDilationH, WindowStrideH)),
+                make_embed_transform(make_tuple(X, Wo), make_tuple(WindowDilationW, WindowStrideW)),
+                make_pass_through_transform(C)),
             make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
             make_tuple(Sequence<0>{},
                        Sequence<1, 2>{},
@@ -139,8 +159,21 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
             make_tuple(Sequence<0>{}, Sequence<1>{}));
 
         // B[ReduceM]
-        const auto out_grid_desc_reducemraw =
-            make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo * C));
+        const index_t No_stride = output_ncdhw_stride[0];
+        const index_t Co_stride = output_ncdhw_stride[1];
+        const index_t Do_stride = output_ncdhw_stride[2];
+        const index_t Ho_stride = output_ncdhw_stride[3];
+        const index_t Wo_stride = output_ncdhw_stride[4];
+
+        const auto out_grid_desc_n_do_ho_wo_c = make_naive_tensor_descriptor(
+            make_tuple(N, Di, Hi, Wi, C),
+            make_tuple(No_stride, Do_stride, Ho_stride, Wo_stride, Co_stride));
+
+        const auto out_grid_desc_reducemraw = transform_tensor_descriptor(
+            out_grid_desc_n_do_ho_wo_c,
+            make_tuple(make_merge_transform(make_tuple(N, Do, Ho, Wo, C))),
+            make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+            make_tuple(Sequence<0>{}));
 
         const auto out_grid_desc_reducem =
             transform_tensor_descriptor(out_grid_desc_reducemraw,
@@ -151,7 +184,9 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
         return make_tuple(in_grid_desc_reducem_reducek, out_grid_desc_reducem);
     }
 
-    using ABGridDescs   = decltype(MakeABGridDescriptor_A_M_K_B_M(1, 1, {}, {}, {}, {}, {}, {}));
+    using ABGridDescs =
+        decltype(MakeABGridDescriptor_A_M_K_B_M({}, {}, {}, {}, {}, {}, {}, {}, {}));
+
     using AGridDesc_M_K = remove_cvref_t<decltype(ABGridDescs{}[I0])>;
     using BGridDesc_M   = remove_cvref_t<decltype(ABGridDescs{}[I1])>;
 
@@ -160,36 +195,41 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
         Argument(const InDataType* p_in_dev,
                  OutDataType* p_out_dev,
                  IndexDataType* p_out_indices_dev,
-                 ck::index_t N,
-                 ck::index_t C,
-                 std::vector<ck::index_t>& input_spatial_lengths,
-                 std::vector<ck::index_t>& window_spatial_lengths,
-                 std::vector<ck::index_t>& output_spatial_lengths,
-                 std::vector<ck::index_t>& window_strides,
-                 std::vector<ck::index_t>& input_left_pads,
-                 std::vector<ck::index_t>& input_right_pads)
+                 std::vector<ck::index_t>& input_ncdhw_lengths,
+                 std::vector<ck::index_t>& output_ncdhw_lengths,
+                 std::vector<ck::index_t>& input_ncdhw_stride,
+                 std::vector<ck::index_t>& output_ncdhw_stride,
+                 std::vector<ck::index_t>&, // indices_ncdhw_stride
+                 std::vector<ck::index_t>& window_spatial_zyx_lengths,
+                 std::vector<ck::index_t>& window_zyx_strides,
+                 std::vector<ck::index_t>& window_zyx_dilations,
+                 std::vector<ck::index_t>& input_left_dhw_pads,
+                 std::vector<ck::index_t>& input_right_dhw_pads)
             : p_in_dev_{p_in_dev},
               p_out_dev_{p_out_dev},
               p_out_indices_dev_{p_out_indices_dev},
               a_grid_desc_m_k_{},
-              b_grid_desc_m_{}
+              b_grid_desc_m_{},
+              input_ncdhw_lengths_{input_ncdhw_lengths},
+              output_ncdhw_lengths_{output_ncdhw_lengths},
+              input_ncdhw_stride_{input_ncdhw_stride},
+              output_ncdhw_stride_{output_ncdhw_stride}
         {
-            const auto descs = MakeABGridDescriptor_A_M_K_B_M(N,
-                                                              C,
-                                                              input_spatial_lengths,
-                                                              window_spatial_lengths,
-                                                              output_spatial_lengths,
-                                                              window_strides,
-                                                              input_left_pads,
-                                                              input_right_pads);
+            const auto descs = MakeABGridDescriptor_A_M_K_B_M(input_ncdhw_lengths,
+                                                              output_ncdhw_lengths,
+                                                              input_ncdhw_stride,
+                                                              output_ncdhw_stride,
+                                                              window_spatial_zyx_lengths,
+                                                              window_zyx_strides,
+                                                              window_zyx_dilations,
+                                                              input_left_dhw_pads,
+                                                              input_right_dhw_pads);
 
             a_grid_desc_m_k_ = descs[I0];
             b_grid_desc_m_   = descs[I1];
 
-            invariant_lowest_length_ = C;
-
-            int32_t reduceLength =
-                window_spatial_lengths[0] * window_spatial_lengths[1] * window_spatial_lengths[2];
+            int32_t reduceLength = window_spatial_zyx_lengths[0] * window_spatial_zyx_lengths[1] *
+                                   window_spatial_zyx_lengths[2];
 
             std::tie(in_element_op_, acc_element_op_) =
                 reduce_unary_operator<ReduceOpId, true, true>::GetElementwiseOperator(reduceLength);
@@ -200,17 +240,25 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
         IndexDataType* p_out_indices_dev_;
         AGridDesc_M_K a_grid_desc_m_k_;
         BGridDesc_M b_grid_desc_m_;
+
         InElementwiseOperation in_element_op_;
         AccElementwiseOperation acc_element_op_;
 
         // for checking vector load/store
-        ck::index_t invariant_lowest_length_;
+        std::vector<ck::index_t> input_ncdhw_lengths_;
+        std::vector<ck::index_t> output_ncdhw_lengths_;
+        std::vector<ck::index_t> input_ncdhw_stride_;
+        std::vector<ck::index_t> output_ncdhw_stride_;
     };
 
     struct Invoker : public BaseInvoker
     {
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
+            // for NDHWC, the dim C is the fastest dimension, and is not reduced.
+            // Hence, it is in M dimension for reduction kernel.
+            static constexpr index_t InSrcOutDstVectorDim = 0; // 0: M, 1: K
+
             using gridwise_reduce =
                 GridwiseReduction_mk_to_m_threadwise<InDataType,
                                                      OutDataType,
@@ -276,60 +324,66 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
     {
         const Argument* pArg = dynamic_cast<const Argument*>(p_arg);
 
-        if(pArg->invariant_lowest_length_ % InSrcOutDstVectorSize != 0)
-        {
+        // C should be fastest dimension
+        if(pArg->input_ncdhw_stride_[1] != 1)
             return false;
+
+        for(int i = 0; i < InOutRank; ++i)
+        {
+            if(pArg->input_ncdhw_stride_[i] == 1 &&
+               pArg->input_ncdhw_lengths_[i] % InSrcOutDstVectorSize != 0)
+                return false;
+
+            if(pArg->output_ncdhw_stride_[i] == 1 &&
+               pArg->output_ncdhw_lengths_[i] % InSrcOutDstVectorSize != 0)
+                return false;
         }
 
         return true;
     }
 
-    std::unique_ptr<BaseArgument>
+    virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in_dev,
                         void* p_out_dev,
                         void* p_out_indices_dev,
-                        std::vector<ck::index_t> input_lengths,
-                        std::vector<ck::index_t> window_lengths,
-                        std::vector<ck::index_t> output_lengths,
-                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
-                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
-                        std::vector<ck::index_t>, // Suppose tensor layout = NDHWC
-                        std::vector<ck::index_t> window_strides,
-                        std::vector<ck::index_t> input_left_pads,
-                        std::vector<ck::index_t> input_right_pads,
+                        std::vector<ck::index_t> input_ncdhw_lengths,
+                        std::vector<ck::index_t> window_zyx_lengths,
+                        std::vector<ck::index_t> output_ncdhw_lengths,
+                        std::vector<ck::index_t> input_ncdhw_stride,
+                        std::vector<ck::index_t> output_ncdhw_stride,
+                        std::vector<ck::index_t> indices_ncdhw_stride,
+                        std::vector<ck::index_t> window_zyx_strides,
+                        std::vector<ck::index_t> window_zyx_dilations,
+                        std::vector<ck::index_t> input_left_dhw_pads,
+                        std::vector<ck::index_t> input_right_dhw_pads,
                         std::vector<ck::index_t> pooling_dims) override
     {
-        if(input_lengths.size() != InOutRank || window_lengths.size() != WindowRank ||
-           input_lengths.size() != InOutRank || window_strides.size() != WindowRank ||
-           input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+        if(input_ncdhw_lengths.size() != InOutRank || window_zyx_lengths.size() != WindowRank ||
+           input_ncdhw_lengths.size() != InOutRank || window_zyx_strides.size() != WindowRank ||
+           window_zyx_dilations.size() != WindowRank || input_left_dhw_pads.size() != WindowRank ||
+           input_right_dhw_pads.size() != WindowRank)
             throw std::runtime_error("dimension is incorrect");
 
         if(pooling_dims != std::vector<ck::index_t>{2, 3, 4})
             throw std::runtime_error("pooling_dims only support {2, 3, 4} in pool3d so far");
 
-        index_t N  = input_lengths[0];
-        index_t C  = input_lengths[1];
-        index_t Di = input_lengths[2];
-        index_t Hi = input_lengths[3];
-        index_t Wi = input_lengths[4];
-        index_t Do = output_lengths[2];
-        index_t Ho = output_lengths[3];
-        index_t Wo = output_lengths[4];
-
-        std::vector<ck::index_t> input_spatial_lengths  = {Di, Hi, Wi};
-        std::vector<ck::index_t> output_spatial_lengths = {Do, Ho, Wo};
+        if(output_ncdhw_stride != indices_ncdhw_stride)
+            throw std::runtime_error(
+                "output_ncdhw_stride need to be equal to indices_ncdhw_stride for now");
 
         return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_dev),
                                           static_cast<OutDataType*>(p_out_dev),
                                           static_cast<IndexDataType*>(p_out_indices_dev),
-                                          N,
-                                          C,
-                                          input_spatial_lengths,
-                                          window_lengths,
-                                          output_spatial_lengths,
-                                          window_strides,
-                                          input_left_pads,
-                                          input_right_pads);
+                                          input_ncdhw_lengths,
+                                          output_ncdhw_lengths,
+                                          input_ncdhw_stride,
+                                          output_ncdhw_stride,
+                                          indices_ncdhw_stride,
+                                          window_zyx_lengths,
+                                          window_zyx_strides,
+                                          window_zyx_dilations,
+                                          input_left_dhw_pads,
+                                          input_right_dhw_pads);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
@@ -342,7 +396,7 @@ struct DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<" << BlockSize << ",";
+        str << "DevicePool3dFwd_NDHWC_NDHWC<" << BlockSize << ",";
         str << "M_C" << MThreadClusterSize << "_S" << MThreadSliceSize << ",";
         str << "K_C" << KThreadClusterSize << "_S" << KThreadSliceSize << ",";
         str <<"InSrcOutDstVectorSize_" << InSrcOutDstVectorSize << ">";
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
index 696fb5eaf1..067e0b2eb9 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
@@ -39,6 +39,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                  Tensor<IndexDataType>& out_indices,
                  const std::vector<ck::index_t>& window_spatial_lengths,
                  const std::vector<ck::index_t>& window_strides,
+                 const std::vector<ck::index_t>& window_dilations,
                  const std::vector<ck::index_t>& in_left_pads,
                  const std::vector<ck::index_t>& /*in_right_pads*/)
             : in_(in),
@@ -46,6 +47,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
               out_indices_(out_indices),
               window_spatial_lengths_(window_spatial_lengths),
               window_strides_(window_strides),
+              window_dilations_(window_dilations),
               in_left_pads_(in_left_pads),
               reduceLength_(1)
         {
@@ -58,6 +60,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
         Tensor<IndexDataType>& out_indices_;
         const std::vector<ck::index_t>& window_spatial_lengths_;
         const std::vector<ck::index_t>& window_strides_;
+        const std::vector<ck::index_t>& window_dilations_;
         const std::vector<ck::index_t>& in_left_pads_;
         int reduceLength_;
     };
@@ -85,14 +88,17 @@ struct ReferencePoolingFwd : public device::BaseOperator
 
                     for(ck::index_t z = 0; z < arg.window_spatial_lengths_[0]; ++z)
                     {
-                        ck::index_t di = do_ * arg.window_strides_[0] + z - arg.in_left_pads_[0];
+                        ck::index_t di = do_ * arg.window_strides_[0] +
+                                         z * arg.window_dilations_[0] - arg.in_left_pads_[0];
                         for(ck::index_t y = 0; y < arg.window_spatial_lengths_[1]; ++y)
                         {
-                            ck::index_t hi = ho * arg.window_strides_[1] + y - arg.in_left_pads_[1];
+                            ck::index_t hi = ho * arg.window_strides_[1] +
+                                             y * arg.window_dilations_[1] - arg.in_left_pads_[1];
                             for(ck::index_t x = 0; x < arg.window_spatial_lengths_[2]; ++x)
                             {
-                                ck::index_t wi =
-                                    wo * arg.window_strides_[2] + x - arg.in_left_pads_[2];
+                                ck::index_t wi = wo * arg.window_strides_[2] +
+                                                 x * arg.window_dilations_[2] -
+                                                 arg.in_left_pads_[2];
                                 if(di >= 0 &&
                                    di < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
                                    hi >= 0 &&
@@ -136,14 +142,17 @@ struct ReferencePoolingFwd : public device::BaseOperator
 
                     for(ck::index_t z = 0; z < arg.window_spatial_lengths_[0]; ++z)
                     {
-                        ck::index_t di = do_ * arg.window_strides_[0] + z - arg.in_left_pads_[0];
+                        ck::index_t di = do_ * arg.window_strides_[0] +
+                                         z * arg.window_dilations_[0] - arg.in_left_pads_[0];
                         for(ck::index_t y = 0; y < arg.window_spatial_lengths_[1]; ++y)
                         {
-                            ck::index_t hi = ho * arg.window_strides_[1] + y - arg.in_left_pads_[1];
+                            ck::index_t hi = ho * arg.window_strides_[1] +
+                                             y * arg.window_dilations_[1] - arg.in_left_pads_[1];
                             for(ck::index_t x = 0; x < arg.window_spatial_lengths_[2]; ++x)
                             {
-                                ck::index_t wi =
-                                    wo * arg.window_strides_[2] + x - arg.in_left_pads_[2];
+                                ck::index_t wi = wo * arg.window_strides_[2] +
+                                                 x * arg.window_dilations_[2] -
+                                                 arg.in_left_pads_[2];
                                 if(di >= 0 &&
                                    di < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
                                    hi >= 0 &&
@@ -202,10 +211,12 @@ struct ReferencePoolingFwd : public device::BaseOperator
 
                     for(ck::index_t y = 0; y < arg.window_spatial_lengths_[0]; ++y)
                     {
-                        ck::index_t hi = ho * arg.window_strides_[0] + y - arg.in_left_pads_[0];
+                        ck::index_t hi = ho * arg.window_strides_[0] +
+                                         y * arg.window_dilations_[0] - arg.in_left_pads_[0];
                         for(ck::index_t x = 0; x < arg.window_spatial_lengths_[1]; ++x)
                         {
-                            ck::index_t wi = wo * arg.window_strides_[1] + x - arg.in_left_pads_[1];
+                            ck::index_t wi = wo * arg.window_strides_[1] +
+                                             x * arg.window_dilations_[1] - arg.in_left_pads_[1];
                             if(hi >= 0 &&
                                hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
                                wi >= 0 &&
@@ -308,6 +319,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                              Tensor<IndexDataType>& out_indices,
                              const std::vector<ck::index_t>& window_spatial_lengths,
                              const std::vector<ck::index_t>& window_strides,
+                             const std::vector<ck::index_t>& window_dilations,
                              const std::vector<ck::index_t>& in_left_pads,
                              const std::vector<ck::index_t>& in_right_pads)
     {
@@ -316,6 +328,7 @@ struct ReferencePoolingFwd : public device::BaseOperator
                         out_indices,
                         window_spatial_lengths,
                         window_strides,
+                        window_dilations,
                         in_left_pads,
                         in_right_pads};
     }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
deleted file mode 100644
index ebdc5f9308..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp
+++ /dev/null
@@ -1,114 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr auto InOutRank  = 4;
-static constexpr auto WindowRank = 2;
-
-static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
-static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-#ifdef __fp16__
-// FP16
-void add_device_pool2d_fwd_nhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
-
-void add_device_pool2d_fwd_nhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
-
-// FP16 - return index
-void add_device_pool2d_fwd_nhwc_index_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
-#endif
-#ifdef __fp32__
-// FP32
-void add_device_pool2d_fwd_nhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
-
-void add_device_pool2d_fwd_nhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
-
-// FP32 - return index
-void add_device_pool2d_fwd_nhwc_index_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
-#endif
-template <typename InDataType,
-          typename OutDataType,
-          typename IndexDataType,
-          ck::ReduceTensorOp ReduceOpId,
-          bool OutputIndex>
-struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
-                                                                                  WindowRank,
-                                                                                  InDataType,
-                                                                                  OutDataType,
-                                                                                  IndexDataType,
-                                                                                  ReduceOpId,
-                                                                                  OutputIndex>>
-{
-    using DeviceOp = DevicePoolFwd<InOutRank,
-                                   WindowRank,
-                                   InDataType,
-                                   OutDataType,
-                                   IndexDataType,
-                                   ReduceOpId,
-                                   OutputIndex>;
-
-    static auto GetInstances()
-    {
-        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
-        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
-                     is_same_v<IndexDataType, I32>)
-        {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
-            {
-                add_device_pool2d_fwd_nhwc_index_f16_instances(op_ptrs);
-            }
-            else
-            {
-                add_device_pool2d_fwd_nhwc_f16_instances(op_ptrs);
-            }
-        }
-#endif
-#ifdef __fp32__
-        if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                     is_same_v<IndexDataType, I32>)
-        {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
-            {
-                add_device_pool2d_fwd_nhwc_index_f32_instances(op_ptrs);
-            }
-            else
-            {
-                add_device_pool2d_fwd_nhwc_f32_instances(op_ptrs);
-            }
-        }
-#endif
-        return op_ptrs;
-    }
-};
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
index b4b4c140ec..ae0514faa0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -25,36 +25,38 @@ static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
 #ifdef __fp16__
 // FP16
 void add_device_pool3d_fwd_ndhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, false>>>&);
 
 void add_device_pool3d_fwd_ndhwc_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, AvgOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, AvgOp, false>>>&);
 
 // FP16 - return index
 void add_device_pool3d_fwd_ndhwc_index_f16_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, MaxOp, true>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
 #endif
 #ifdef __fp32__
 // FP32
 void add_device_pool3d_fwd_ndhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, NDHWC, NDHWC, MaxOp, false>>>&);
 
 void add_device_pool3d_fwd_ndhwc_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, AvgOp, false>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, NDHWC, NDHWC, AvgOp, false>>>&);
 
 // FP32 - return index
 void add_device_pool3d_fwd_ndhwc_index_f32_instances(
-    std::vector<
-        std::unique_ptr<DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, MaxOp, true>>>&);
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, F32, F32, I32, NDHWC, NDHWC, MaxOp, true>>>&);
 #endif
 template <typename InDataType,
           typename OutDataType,
           typename IndexDataType,
+          typename InLayout,
+          typename OutLayout,
           ck::ReduceTensorOp ReduceOpId,
           bool OutputIndex>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFwd<InOutRank,
@@ -62,6 +64,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                                                                                   InDataType,
                                                                                   OutDataType,
                                                                                   IndexDataType,
+                                                                                  InLayout,
+                                                                                  OutLayout,
                                                                                   ReduceOpId,
                                                                                   OutputIndex>>
 {
@@ -70,40 +74,46 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                                    InDataType,
                                    OutDataType,
                                    IndexDataType,
+                                   InLayout,
+                                   OutLayout,
                                    ReduceOpId,
                                    OutputIndex>;
 
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
-        if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
-                     is_same_v<IndexDataType, I32>)
+        if constexpr(is_same_v<InLayout, NDHWC> && is_same_v<OutLayout, NDHWC>)
         {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+#ifdef __fp16__
+            if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
+                         is_same_v<IndexDataType, I32>)
             {
-                add_device_pool3d_fwd_ndhwc_index_f16_instances(op_ptrs);
+                if constexpr(OutputIndex && ReduceOpId == MaxOp)
+                {
+                    add_device_pool3d_fwd_ndhwc_index_f16_instances(op_ptrs);
+                }
+                else
+                {
+                    add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
+                }
             }
-            else
-            {
-                add_device_pool3d_fwd_ndhwc_f16_instances(op_ptrs);
-            }
-        }
 #endif
 #ifdef __fp32__
-        if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                     is_same_v<IndexDataType, I32>)
-        {
-            if constexpr(OutputIndex && ReduceOpId == MaxOp)
+            if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                         is_same_v<IndexDataType, I32>)
             {
-                add_device_pool3d_fwd_ndhwc_index_f32_instances(op_ptrs);
+                if constexpr(OutputIndex && ReduceOpId == MaxOp)
+                {
+                    add_device_pool3d_fwd_ndhwc_index_f32_instances(op_ptrs);
+                }
+                else
+                {
+                    add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
+                }
             }
-            else
-            {
-                add_device_pool3d_fwd_ndhwc_f32_instances(op_ptrs);
-            }
-        }
 #endif
+        }
+
         return op_ptrs;
     }
 };
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
new file mode 100644
index 0000000000..dd843426b2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
@@ -0,0 +1,10 @@
+set(DEVICE_POOL3D_FWD_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_f32_instance.cpp)
+endif()
+add_instance_library(device_pool3d_fwd_instance ${DEVICE_POOL3D_FWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
similarity index 81%
rename from library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
index 62bcad992a..4ebd50bae6 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -11,7 +11,9 @@ namespace instance {
 static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
 
 void add_device_pool3d_fwd_ndhwc_f16_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, false>>>& instances)
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, NDHWC, NDHWC, ReduceOpId, false>>>&
+        instances)
 {
     add_device_operation_instances(
         instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
similarity index 81%
rename from library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
index 47896be911..dcb19110b2 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -11,7 +11,9 @@ namespace instance {
 static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
 
 void add_device_pool3d_fwd_ndhwc_f32_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, false>>>& instances)
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, NDHWC, NDHWC, ReduceOpId, false>>>&
+        instances)
 {
     add_device_operation_instances(
         instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
similarity index 74%
rename from library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
index dbfc4acfdf..46b16bd005 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_f16_instance.cpp
@@ -11,14 +11,18 @@ namespace instance {
 static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
 
 void add_device_pool3d_fwd_ndhwc_f16_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, false>>>& instances)
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, NDHWC, NDHWC, ReduceOpId, false>>>&
+        instances)
 {
     add_device_operation_instances(
         instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F16, ReduceOpId, false>{});
 }
 
 void add_device_pool3d_fwd_ndhwc_index_f16_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, ReduceOpId, true>>>& instances)
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, F16, F16, I32, NDHWC, NDHWC, ReduceOpId, true>>>&
+        instances)
 {
     add_device_operation_instances(
         instances, device_pool3d_fwd_ndhwc_instances<F16, F16, I32, F16, ReduceOpId, true>{});
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
similarity index 74%
rename from library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
index 63b3e8df8e..b4b0e74d28 100644
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_f32_instance.cpp
@@ -11,14 +11,18 @@ namespace instance {
 static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
 
 void add_device_pool3d_fwd_ndhwc_f32_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, false>>>& instances)
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, NDHWC, NDHWC, ReduceOpId, false>>>&
+        instances)
 {
     add_device_operation_instances(
         instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
 }
 
 void add_device_pool3d_fwd_ndhwc_index_f32_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, ReduceOpId, true>>>& instances)
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, F32, F32, I32, NDHWC, NDHWC, ReduceOpId, true>>>&
+        instances)
 {
     add_device_operation_instances(
         instances, device_pool3d_fwd_ndhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp
new file mode 100644
index 0000000000..4d23ceab27
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I32   = int32_t;
+using F16   = ck::half_t;
+using F32   = float;
+using NDHWC = ck::tensor_layout::convolution::NDHWC;
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename ComputeDataType,
+          ReduceTensorOp ReduceOpId,
+          bool OutputIndex>
+using device_pool3d_fwd_ndhwc_instances =
+    // clang-format off
+    std::tuple <
+        DevicePool3dFwd_NDHWC_NDHWC<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
+        DevicePool3dFwd_NDHWC_NDHWC<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
+        DevicePool3dFwd_NDHWC_NDHWC<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
+               // clang-format on
+               >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
deleted file mode 100644
index ba2a8991bd..0000000000
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/CMakeLists.txt
+++ /dev/null
@@ -1,14 +0,0 @@
-set(DEVICE_POOL_FWD_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_POOL_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f16_instance.cpp
-                                          device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
-                                          device_max_pool2d_fwd_nhwc_f16_instance.cpp
-                                          device_max_pool3d_fwd_ndhwc_f16_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_POOL_FWD_INSTANCES device_avg_pool2d_fwd_nhwc_f32_instance.cpp
-                                          device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
-                                          device_max_pool2d_fwd_nhwc_f32_instance.cpp
-                                          device_max_pool3d_fwd_ndhwc_f32_instance.cpp)
-endif()
-add_instance_library(device_pool_fwd_instance ${DEVICE_POOL_FWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
deleted file mode 100644
index 508ad3873b..0000000000
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f16_instance.cpp
+++ /dev/null
@@ -1,23 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "pool_fwd_instance_common.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
-
-void add_device_pool2d_fwd_nhwc_f16_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, false>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F32, ReduceOpId, false>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
deleted file mode 100644
index ada96a93a2..0000000000
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_avg_pool2d_fwd_nhwc_f32_instance.cpp
+++ /dev/null
@@ -1,23 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "pool_fwd_instance_common.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
-
-void add_device_pool2d_fwd_nhwc_f32_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, false>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
deleted file mode 100644
index 35c8522d9f..0000000000
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f16_instance.cpp
+++ /dev/null
@@ -1,30 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "pool_fwd_instance_common.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
-
-void add_device_pool2d_fwd_nhwc_f16_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, false>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F16, ReduceOpId, false>{});
-}
-
-void add_device_pool2d_fwd_nhwc_index_f16_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F16, F16, I32, ReduceOpId, true>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_pool2d_fwd_nhwc_instances<F16, F16, I32, F16, ReduceOpId, true>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
deleted file mode 100644
index 75b7629f24..0000000000
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/device_max_pool2d_fwd_nhwc_f32_instance.cpp
+++ /dev/null
@@ -1,30 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "pool_fwd_instance_common.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
-
-void add_device_pool2d_fwd_nhwc_f32_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, false>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, false>{});
-}
-
-void add_device_pool2d_fwd_nhwc_index_f32_instances(
-    std::vector<std::unique_ptr<DevicePoolFwd<4, 2, F32, F32, I32, ReduceOpId, true>>>& instances)
-{
-    add_device_operation_instances(
-        instances, device_pool2d_fwd_nhwc_instances<F32, F32, I32, F32, ReduceOpId, true>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
deleted file mode 100644
index 8aa707885b..0000000000
--- a/library/src/tensor_operation_instance/gpu/pool_fwd/pool_fwd_instance_common.hpp
+++ /dev/null
@@ -1,55 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_pool3d_fwd_ndhwc_ndhwc.hpp"
-#include "ck/utility/data_type.hpp"
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using I32 = int32_t;
-using F16 = ck::half_t;
-using F32 = float;
-
-template <typename InDataType,
-          typename OutDataType,
-          typename IndexDataType,
-          typename ComputeDataType,
-          ReduceTensorOp ReduceOpId,
-          bool OutputIndex>
-using device_pool2d_fwd_nhwc_instances =
-    // clang-format off
-    std::tuple <
-        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
-        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
-        DevicePool2dFwd_Input_N_Hi_Wi_C_Output_N_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
-               // clang-format on
-               >;
-
-template <typename InDataType,
-          typename OutDataType,
-          typename IndexDataType,
-          typename ComputeDataType,
-          ReduceTensorOp ReduceOpId,
-          bool OutputIndex>
-using device_pool3d_fwd_ndhwc_instances =
-    // clang-format off
-    std::tuple <
-        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 1, 1, 1>,
-        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 2, 1, 2>,
-        DevicePool3dFwd_Input_N_Di_Hi_Wi_C_Output_N_Do_Ho_Wo_C<InDataType, OutDataType, IndexDataType, ComputeDataType, ReduceOpId, OutputIndex, 256, 256, 1, 4, 1, 4>
-               // clang-format on
-               >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/profiler/include/profiler/profile_pool2d_fwd_impl.hpp b/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
deleted file mode 100644
index 0c888db1f4..0000000000
--- a/profiler/include/profiler/profile_pool2d_fwd_impl.hpp
+++ /dev/null
@@ -1,264 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iomanip>
-
-#include "ck/ck.hpp"
-#include "ck/library/tensor_operation_instance/gpu/pool2d_fwd.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
-
-namespace ck {
-namespace profiler {
-
-template <typename InDataType,
-          typename OutDataType,
-          typename ComputeDataType,
-          typename IndexDataType,
-          ck::ReduceTensorOp ReduceOpId,
-          bool PropagateNan,
-          bool OutputIndex>
-bool profile_pool2d_fwd_impl(int do_verification,
-                             int init_method,
-                             bool do_log,
-                             bool time_kernel,
-                             std::vector<index_t> in_length, // NCHW
-                             std::vector<index_t> window_spatial_lengths,
-                             std::vector<index_t> window_strides,
-                             std::vector<index_t> input_left_pads,
-                             std::vector<index_t> input_right_pads)
-{
-    constexpr index_t InOutRank  = 4;
-    constexpr index_t WindowRank = 2;
-
-    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
-       window_strides.size() != WindowRank || input_left_pads.size() != WindowRank ||
-       input_right_pads.size() != WindowRank)
-        return false;
-
-    std::vector<index_t> out_length(InOutRank);
-
-    int N = in_length[0];
-    int C = in_length[1];
-
-    out_length[0] = N;
-    out_length[1] = C;
-
-    // Calculate Ho, Wo
-    for(int i = 2; i < InOutRank; ++i)
-    {
-        auto pad1           = input_left_pads[i - 2];
-        auto pad2           = input_right_pads[i - 2];
-        auto windows_size   = window_spatial_lengths[i - 2];
-        auto windows_stride = window_strides[i - 2];
-        out_length[i]       = (in_length[i] + pad1 + pad2 - windows_size) / windows_stride + 1;
-    }
-
-    int Hi = in_length[2];
-    int Wi = in_length[3];
-    int Ho = out_length[2];
-    int Wo = out_length[3];
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t N_, std::size_t C_, std::size_t H, std::size_t W) {
-            using namespace ck::literals;
-            return HostTensorDescriptor({N_, C_, H, W}, {C_ * H * W, 1_uz, W * C_, C_});
-        };
-
-    Tensor<InDataType> in_n_c_hi_wi(f_host_tensor_descriptor(N, C, Hi, Wi));
-    Tensor<OutDataType> out_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
-    Tensor<IndexDataType> out_indices_n_c_ho_wo_host(f_host_tensor_descriptor(N, C, Ho, Wo));
-
-    Tensor<OutDataType> out_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
-    Tensor<IndexDataType> out_indices_n_c_ho_wo_device(f_host_tensor_descriptor(N, C, Ho, Wo));
-
-    switch(init_method)
-    {
-    case 0: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{}); break;
-    case 1: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5}); break;
-    default: in_n_c_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
-    }
-
-    DeviceMem in_device_buf(sizeof(InDataType) * in_n_c_hi_wi.mDesc.GetElementSpaceSize());
-    DeviceMem out_device_buf(sizeof(OutDataType) *
-                             out_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
-    DeviceMem out_indices_device_buf(sizeof(IndexDataType) *
-                                     out_indices_n_c_ho_wo_device.mDesc.GetElementSpaceSize());
-
-    in_device_buf.ToDevice(in_n_c_hi_wi.mData.data());
-
-    // add device normalization instances
-    using DeviceOp = ck::tensor_operation::device::DevicePoolFwd<InOutRank,
-                                                                 WindowRank,
-                                                                 InDataType,
-                                                                 OutDataType,
-                                                                 IndexDataType,
-                                                                 ReduceOpId,
-                                                                 OutputIndex>;
-
-    // get device op instances
-    const auto instance_ptrs =
-        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
-            DeviceOp>::GetInstances();
-
-    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
-
-    std::string best_instance_name;
-    float best_avg_time   = std::numeric_limits<float>::max();
-    float best_gb_per_sec = 0;
-
-    if(do_verification)
-    {
-        using ReferenceInstance = ck::tensor_operation::host::ReferencePoolingFwd<InOutRank,
-                                                                                  WindowRank,
-                                                                                  InDataType,
-                                                                                  OutDataType,
-                                                                                  ComputeDataType,
-                                                                                  IndexDataType,
-                                                                                  ReduceOpId,
-                                                                                  PropagateNan,
-                                                                                  OutputIndex>;
-
-        ReferenceInstance ref;
-        auto ref_argument = ref.MakeArgument(in_n_c_hi_wi,
-                                             out_n_c_ho_wo_host,
-                                             out_indices_n_c_ho_wo_host,
-                                             window_spatial_lengths,
-                                             window_strides,
-                                             input_left_pads,
-                                             input_right_pads);
-        auto ref_invoker  = ref.MakeInvoker();
-        ref_invoker.Run(ref_argument);
-    }
-
-    int num_kernel = 0;
-
-    for(auto& inst_ptr : instance_ptrs)
-    {
-        auto argument_ptr = inst_ptr->MakeArgumentPointer(
-            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
-            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
-            static_cast<IndexDataType*>(out_indices_device_buf.GetDeviceBuffer()),
-            in_length,
-            window_spatial_lengths,
-            out_length,
-            {C * Hi * Wi, 1, Wi * C, C},
-            {C * Ho * Wo, 1, Wo * C, C},
-            {C * Ho * Wo, 1, Wo * C, C},
-            window_strides,
-            input_left_pads,
-            input_right_pads,
-            {2, 3});
-
-        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
-        {
-            ++num_kernel;
-        }
-        else
-        {
-            if(time_kernel)
-            {
-                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
-                LogRange(std::cout << "input lengths = ", in_length, ", ") << std::endl;
-            }
-
-            continue;
-        }
-
-        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
-
-        float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
-
-        std::size_t num_bytes = in_n_c_hi_wi.mDesc.GetElementSize() * sizeof(InDataType) +
-                                out_n_c_ho_wo_host.mDesc.GetElementSize() * sizeof(OutDataType);
-
-        if constexpr(OutputIndex)
-            num_bytes += out_indices_n_c_ho_wo_host.mDesc.GetElementSize() * sizeof(IndexDataType);
-
-        float gb_per_sec = num_bytes / 1.E6 / avg_time;
-
-        if(time_kernel)
-            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
-                      << inst_ptr->GetTypeString() << std::endl;
-
-        if(avg_time < best_avg_time)
-        {
-            best_instance_name = inst_ptr->GetTypeString();
-            best_avg_time      = avg_time;
-            best_gb_per_sec    = gb_per_sec;
-        }
-
-        if(do_verification)
-        {
-            out_device_buf.FromDevice(out_n_c_ho_wo_device.mData.data());
-
-            bool pass = ck::utils::check_err(out_n_c_ho_wo_device.mData,
-                                             out_n_c_ho_wo_host.mData,
-                                             "Error: Incorrect results",
-                                             1e-3,
-                                             1e-3);
-
-            if constexpr(OutputIndex)
-            {
-                out_indices_device_buf.FromDevice(out_indices_n_c_ho_wo_device.mData.data());
-
-                pass = pass && ck::utils::check_err(out_indices_n_c_ho_wo_device,
-                                                    out_indices_n_c_ho_wo_host);
-            }
-
-            if(do_log)
-            {
-                LogRangeAsType<float>(std::cout << "in_n_c_hi_wi  : ", in_n_c_hi_wi.mData, ",")
-                    << std::endl;
-                LogRangeAsType<float>(
-                    std::cout << "out_n_c_ho_wo_host  : ", out_n_c_ho_wo_host.mData, ",")
-                    << std::endl;
-                LogRangeAsType<float>(
-                    std::cout << "out_n_c_ho_wo_device  : ", out_n_c_ho_wo_device.mData, ",")
-                    << std::endl;
-
-                if constexpr(OutputIndex)
-                    LogRangeAsType<float>(std::cout << "out_indices_n_c_ho_wo_device  : ",
-                                          out_indices_n_c_ho_wo_device.mData,
-                                          ",")
-                        << std::endl;
-            }
-
-            if(!pass)
-            {
-                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
-                LogRange(std::cout << "lengths = [", in_length, ", ") << "]." << std::endl;
-                return false;
-            }
-            else
-            {
-                if(time_kernel)
-                    std::cout << "pass" << std::endl;
-            }
-        }
-    }
-
-    if(time_kernel)
-    {
-        LogRange(std::cout << "length = ", in_length, ",") << std::endl;
-        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
-                  << best_instance_name << std::endl;
-    }
-
-    if(num_kernel == 0)
-    {
-        std::cout << "Error: No kernel is applicable" << std::endl;
-        return false;
-    }
-
-    return true;
-}
-
-} // namespace profiler
-} // namespace ck
diff --git a/profiler/include/profiler/profile_pool3d_fwd_impl.hpp b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
index 41b57fd853..02fde48d6e 100644
--- a/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_pool3d_fwd_impl.hpp
@@ -21,6 +21,8 @@ template <typename InDataType,
           typename OutDataType,
           typename ComputeDataType,
           typename IndexDataType,
+          typename InLayout,
+          typename OutLayout,
           ck::ReduceTensorOp ReduceOpId,
           bool PropagateNan,
           bool OutputIndex>
@@ -31,6 +33,7 @@ bool profile_pool3d_fwd_impl(int do_verification,
                              std::vector<index_t> in_length, // NCDHW
                              std::vector<index_t> window_spatial_lengths,
                              std::vector<index_t> window_strides,
+                             std::vector<index_t> window_dilations,
                              std::vector<index_t> input_left_pads,
                              std::vector<index_t> input_right_pads)
 {
@@ -38,8 +41,8 @@ bool profile_pool3d_fwd_impl(int do_verification,
     constexpr index_t WindowRank = 3;
 
     if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
-       window_strides.size() != WindowRank || input_left_pads.size() != WindowRank ||
-       input_right_pads.size() != WindowRank)
+       window_strides.size() != WindowRank || window_dilations.size() != WindowRank ||
+       input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
         return false;
 
     std::vector<index_t> out_length(InOutRank);
@@ -53,11 +56,13 @@ bool profile_pool3d_fwd_impl(int do_verification,
     // Calculate Do, Ho, Wo
     for(int i = 2; i < InOutRank; ++i)
     {
-        auto pad1           = input_left_pads[i - 2];
-        auto pad2           = input_right_pads[i - 2];
-        auto windows_size   = window_spatial_lengths[i - 2];
-        auto windows_stride = window_strides[i - 2];
-        out_length[i]       = (in_length[i] + pad1 + pad2 - windows_size) / windows_stride + 1;
+        auto pad1             = input_left_pads[i - 2];
+        auto pad2             = input_right_pads[i - 2];
+        auto windows_size     = window_spatial_lengths[i - 2];
+        auto windows_stride   = window_strides[i - 2];
+        auto windows_dilation = window_dilations[i - 2];
+        auto eff              = (windows_size - 1) * windows_dilation + 1;
+        out_length[i]         = (in_length[i] + pad1 + pad2 - eff) / windows_stride + 1;
     }
 
     int Di = in_length[2];
@@ -104,6 +109,8 @@ bool profile_pool3d_fwd_impl(int do_verification,
                                                                  InDataType,
                                                                  OutDataType,
                                                                  IndexDataType,
+                                                                 InLayout,
+                                                                 OutLayout,
                                                                  ReduceOpId,
                                                                  OutputIndex>;
 
@@ -136,6 +143,7 @@ bool profile_pool3d_fwd_impl(int do_verification,
                                              out_indices_n_c_do_ho_wo_host,
                                              window_spatial_lengths,
                                              window_strides,
+                                             window_dilations,
                                              input_left_pads,
                                              input_right_pads);
         auto ref_invoker  = ref.MakeInvoker();
@@ -157,6 +165,7 @@ bool profile_pool3d_fwd_impl(int do_verification,
             {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
             {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C},
             window_strides,
+            window_dilations,
             input_left_pads,
             input_right_pads,
             {2, 3, 4});
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index a4ea5c399c..f98747c93a 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -17,7 +17,6 @@ set(PROFILER_SOURCES
     profile_reduce.cpp
     profile_groupnorm.cpp
     profile_layernorm.cpp
-    profile_avg_pool2d_fwd.cpp
     profile_max_pool3d_fwd.cpp
     profile_softmax.cpp
     profile_batchnorm_fwd.cpp
@@ -74,7 +73,7 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool_fwd_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool3d_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 if(DL_KERNELS)
diff --git a/profiler/src/profile_avg_pool2d_fwd.cpp b/profiler/src/profile_avg_pool2d_fwd.cpp
deleted file mode 100644
index c67897c044..0000000000
--- a/profiler/src/profile_avg_pool2d_fwd.cpp
+++ /dev/null
@@ -1,141 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <vector>
-#include <unordered_map>
-
-#include "profiler/data_type_enum.hpp"
-#include "profiler/profile_pool2d_fwd_impl.hpp"
-#include "profiler_operation_registry.hpp"
-
-using ck::index_t;
-
-struct avgPoolFwdArgParser
-{
-    std::unordered_map<std::string, std::vector<int>> long_opts = {
-        {"length", {}}, {"wsize", {}}, {"wstride", {}}, {"pad1", {}}, {"pad2", {}}};
-
-    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
-    {
-        if(std::string("--") + key == argv[i])
-        {
-            int pos = i;
-            while(++i < argc && argv[i][0] != '-') {}
-            int end = i;
-            for(int j = pos + 1; j < end; j++)
-            {
-                long_opts[key].push_back(std::stoi(argv[j]));
-            }
-            return true;
-        }
-        return false;
-    }
-
-    void operator()(int argc, char* argv[])
-    {
-        for(auto& kv : long_opts)
-        {
-            for(int i = 1; i < argc; i++)
-            {
-                if(parse_opt(argc, argv, kv.first, i))
-                    break;
-            }
-        }
-    }
-};
-
-void print_help_avg_pool2d_fwd()
-{
-    std::cout << "arg1: data type (0: fp16; 1: fp32)\n"
-              << "arg2: verification (0: no; 1: yes)\n"
-              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
-              << "arg4: print tensor value (0: no; 1: yes)\n"
-              << "arg5: time kernel (0=no, 1=yes)\n"
-              << "--length: input tensor length for NDHW(e.g, --length 2 32 30 30) \n"
-              << "--wsize: window size for YX (e.g, --wsize 2 2) \n"
-              << "--wstride: window stride for HW (e.g, --wstride 2 2) \n"
-              << "--pad1: left side of padding in HW (e.g, --pad1 1 1) \n"
-              << "--pad2: right side of padding in HW (e.g, --pad2 1 1) \n"
-              << "eg: ckProfiler avg_pool2d_fwd 0 1 2 0 1 0 --length 2 32 30 30 --wsize 2 2 "
-                 "--wstride 2 2 --pad1 1 1 --pad2 1 1"
-              << std::endl;
-}
-
-int profile_avg_pool2d_fwd(int argc, char* argv[])
-{
-    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
-    bool do_verification       = true;
-    int init_method            = 0;
-    bool do_log                = false;
-    bool time_kernel           = true;
-
-    std::vector<index_t> in_length = {2, 32, 30, 30};
-    std::vector<index_t> wsize     = {2, 2};
-    std::vector<index_t> wstride   = {2, 2};
-    std::vector<index_t> pad1      = {1, 1};
-    std::vector<index_t> pad2      = {1, 1};
-
-    if(argc != 2 && argc != 25)
-    {
-        print_help_avg_pool2d_fwd();
-        return 0;
-    }
-    else if(argc == 25)
-    {
-        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
-        do_verification = std::stoi(argv[3]);
-        init_method     = std::stoi(argv[4]);
-        do_log          = std::stoi(argv[5]);
-        time_kernel     = std::stoi(argv[6]);
-
-        // parse the long options
-        avgPoolFwdArgParser arg_parser;
-        arg_parser(argc, argv);
-        in_length = arg_parser.long_opts["length"];
-        wsize     = arg_parser.long_opts["wsize"];
-        wstride   = arg_parser.long_opts["wstride"];
-        pad1      = arg_parser.long_opts["pad1"];
-        pad2      = arg_parser.long_opts["pad2"];
-    }
-
-    using F16                 = ck::half_t;
-    using F32                 = float;
-    using I32                 = int32_t;
-    constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
-
-    if(data_type == ck::DataTypeEnum::Half)
-    {
-        ck::profiler::profile_pool2d_fwd_impl<F16, F16, F32, I32, ReduceOpId, false, false>(
-            do_verification,
-            init_method,
-            do_log,
-            time_kernel,
-            in_length,
-            wsize,
-            wstride,
-            pad1,
-            pad2);
-    }
-    else if(data_type == ck::DataTypeEnum::Float)
-    {
-        ck::profiler::profile_pool2d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, false>(
-            do_verification,
-            init_method,
-            do_log,
-            time_kernel,
-            in_length,
-            wsize,
-            wstride,
-            pad1,
-            pad2);
-    }
-    else
-    {
-        throw std::runtime_error("not implemented yet");
-    }
-
-    return 0;
-}
-
-REGISTER_PROFILER_OPERATION("avg_pool2d_fwd", "avg_pool2d fwd", profile_avg_pool2d_fwd);
diff --git a/profiler/src/profile_max_pool3d_fwd.cpp b/profiler/src/profile_max_pool3d_fwd.cpp
index cf6db2cfc9..da7ea9af4b 100644
--- a/profiler/src/profile_max_pool3d_fwd.cpp
+++ b/profiler/src/profile_max_pool3d_fwd.cpp
@@ -13,8 +13,12 @@ using ck::index_t;
 
 struct maxPoolFwdArgParser
 {
-    std::unordered_map<std::string, std::vector<int>> long_opts = {
-        {"length", {}}, {"wsize", {}}, {"wstride", {}}, {"pad1", {}}, {"pad2", {}}};
+    std::unordered_map<std::string, std::vector<int>> long_opts = {{"length", {}},
+                                                                   {"wsize", {}},
+                                                                   {"wstride", {}},
+                                                                   {"wdilation", {}},
+                                                                   {"pad1", {}},
+                                                                   {"pad2", {}}};
 
     bool parse_opt(int argc, char* argv[], const std::string& key, int i)
     {
@@ -56,10 +60,11 @@ void print_help_max_pool3d_fwd()
               << "--length: input tensor length for NCDHW(e.g, --length 2 32 30 30 30) \n"
               << "--wsize: window size for ZYX (e.g, --wsize 2 2 2) \n"
               << "--wstride: window stride for DHW (e.g, --wstride 2 2 2) \n"
+              << "--wdilation: window dilation for DHW (e.g, --wdilation 1 1 1) \n"
               << "--pad1: left side of padding in DHW (e.g, --pad1 1 1 1) \n"
               << "--pad2: right side of padding in DHW (e.g, --pad2 1 1 1) \n"
               << "eg: ckProfiler max_pool3d_fwd 0 1 2 0 1 0 --length 2 32 30 30 30 --wsize 2 2 2 "
-                 "--wstride 2 2 2 --pad1 1 1 1 --pad2 1 1 1"
+                 "--wstride 2 2 2 --wdilation 1 1 1 --pad1 1 1 1 --pad2 1 1 1"
               << std::endl;
 }
 
@@ -75,15 +80,16 @@ int profile_max_pool3d_fwd(int argc, char* argv[])
     std::vector<index_t> in_length = {2, 32, 30, 30, 30};
     std::vector<index_t> wsize     = {2, 2, 2};
     std::vector<index_t> wstride   = {2, 2, 2};
+    std::vector<index_t> wdilation = {1, 1, 1};
     std::vector<index_t> pad1      = {1, 1, 1};
     std::vector<index_t> pad2      = {1, 1, 1};
 
-    if(argc != 2 && argc != 30)
+    if(argc != 2 && argc != 34)
     {
         print_help_max_pool3d_fwd();
         return 0;
     }
-    else if(argc == 30)
+    else if(argc == 34)
     {
         data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
         do_verification = std::stoi(argv[3]);
@@ -98,64 +104,79 @@ int profile_max_pool3d_fwd(int argc, char* argv[])
         in_length = arg_parser.long_opts["length"];
         wsize     = arg_parser.long_opts["wsize"];
         wstride   = arg_parser.long_opts["wstride"];
+        wdilation = arg_parser.long_opts["wdilation"];
         pad1      = arg_parser.long_opts["pad1"];
         pad2      = arg_parser.long_opts["pad2"];
     }
 
-    using F16                 = ck::half_t;
-    using F32                 = float;
-    using I32                 = int32_t;
+    using F16   = ck::half_t;
+    using F32   = float;
+    using I32   = int32_t;
+    using NDHWC = ck::tensor_layout::convolution::NDHWC;
+
+#if 1
     constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+#else
+    constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+#endif
 
     if(data_type == ck::DataTypeEnum::Half)
     {
         if(return_index)
-            ck::profiler::profile_pool3d_fwd_impl<F16, F16, F16, I32, ReduceOpId, false, true>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                in_length,
-                wsize,
-                wstride,
-                pad1,
-                pad2);
+            ck::profiler::
+                profile_pool3d_fwd_impl<F16, F16, F16, I32, NDHWC, NDHWC, ReduceOpId, false, true>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    in_length,
+                    wsize,
+                    wstride,
+                    wdilation,
+                    pad1,
+                    pad2);
         else
-            ck::profiler::profile_pool3d_fwd_impl<F16, F16, F16, I32, ReduceOpId, false, false>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                in_length,
-                wsize,
-                wstride,
-                pad1,
-                pad2);
+            ck::profiler::
+                profile_pool3d_fwd_impl<F16, F16, F16, I32, NDHWC, NDHWC, ReduceOpId, false, false>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    in_length,
+                    wsize,
+                    wstride,
+                    wdilation,
+                    pad1,
+                    pad2);
     }
     else if(data_type == ck::DataTypeEnum::Float)
     {
         if(return_index)
-            ck::profiler::profile_pool3d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, true>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                in_length,
-                wsize,
-                wstride,
-                pad1,
-                pad2);
+            ck::profiler::
+                profile_pool3d_fwd_impl<F32, F32, F32, I32, NDHWC, NDHWC, ReduceOpId, false, true>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    in_length,
+                    wsize,
+                    wstride,
+                    wdilation,
+                    pad1,
+                    pad2);
         else
-            ck::profiler::profile_pool3d_fwd_impl<F32, F32, F32, I32, ReduceOpId, false, false>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                in_length,
-                wsize,
-                wstride,
-                pad1,
-                pad2);
+            ck::profiler::
+                profile_pool3d_fwd_impl<F32, F32, F32, I32, NDHWC, NDHWC, ReduceOpId, false, false>(
+                    do_verification,
+                    init_method,
+                    do_log,
+                    time_kernel,
+                    in_length,
+                    wsize,
+                    wstride,
+                    wdilation,
+                    pad1,
+                    pad2);
     }
     else
     {
diff --git a/test/pool_fwd/CMakeLists.txt b/test/pool_fwd/CMakeLists.txt
index 6f59b95f6f..e421a79fde 100644
--- a/test/pool_fwd/CMakeLists.txt
+++ b/test/pool_fwd/CMakeLists.txt
@@ -1,16 +1,10 @@
 add_custom_target(test_pool_fwd)
 
-add_gtest_executable(test_avg_pool2d_fwd test_avg_pool2d_fwd.cpp)
 add_gtest_executable(test_avg_pool3d_fwd test_avg_pool3d_fwd.cpp)
-add_gtest_executable(test_max_pool2d_fwd test_max_pool2d_fwd.cpp)
 add_gtest_executable(test_max_pool3d_fwd test_max_pool3d_fwd.cpp)
 
-target_link_libraries(test_avg_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
-target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
-target_link_libraries(test_max_pool2d_fwd PRIVATE utility device_pool_fwd_instance)
-target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool_fwd_instance)
+target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
+target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
 
-add_dependencies(test_pool_fwd test_avg_pool2d_fwd)
 add_dependencies(test_pool_fwd test_avg_pool3d_fwd)
-add_dependencies(test_pool_fwd test_max_pool2d_fwd)
 add_dependencies(test_pool_fwd test_max_pool3d_fwd)
diff --git a/test/pool_fwd/test_avg_pool2d_fwd.cpp b/test/pool_fwd/test_avg_pool2d_fwd.cpp
deleted file mode 100644
index 31f4a166f2..0000000000
--- a/test/pool_fwd/test_avg_pool2d_fwd.cpp
+++ /dev/null
@@ -1,60 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "gtest/gtest.h"
-#include "profiler/profile_pool2d_fwd_impl.hpp"
-#include "test_pool_fwd_common.hpp"
-
-template <typename Tuple>
-class TestAvgPool2dFwd : public ::testing::Test
-{
-    protected:
-    using InDataType      = std::tuple_element_t<0, Tuple>;
-    using OutDataType     = std::tuple_element_t<1, Tuple>;
-    using ComputeDataType = std::tuple_element_t<2, Tuple>;
-    using IndexDataType   = std::tuple_element_t<3, Tuple>;
-
-    std::vector<PoolingParam> params;
-
-    void Run()
-    {
-        for(auto param : params)
-        {
-            bool success =
-                ck::profiler::profile_pool2d_fwd_impl<InDataType,
-                                                      OutDataType,
-                                                      ComputeDataType,
-                                                      IndexDataType,
-                                                      ck::ReduceTensorOp::AVG,
-                                                      false,
-                                                      false>(true,
-                                                             2,
-                                                             false,
-                                                             false,
-                                                             param.length_,
-                                                             param.window_spatial_lengths_,
-                                                             param.window_strides_,
-                                                             param.input_left_pads_,
-                                                             param.input_right_pads_);
-            EXPECT_TRUE(success);
-        }
-    }
-};
-
-#ifdef __fp16__
-using KernelTypes =
-    ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
-#else
-using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
-#endif
-
-TYPED_TEST_SUITE(TestAvgPool2dFwd, KernelTypes);
-TYPED_TEST(TestAvgPool2dFwd, Test_Pool)
-{
-    // length, window_length, window_stride, left_pad, right_pad
-    this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
-                    {{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
-                    {{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
-
-    this->Run();
-}
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool_fwd/test_avg_pool3d_fwd.cpp
index a99a3c698e..81849ffa6a 100644
--- a/test/pool_fwd/test_avg_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_avg_pool3d_fwd.cpp
@@ -25,6 +25,8 @@ class TestAvgPool3dFwd : public ::testing::Test
                                                       OutDataType,
                                                       ComputeDataType,
                                                       IndexDataType,
+                                                      ck::tensor_layout::convolution::NDHWC,
+                                                      ck::tensor_layout::convolution::NDHWC,
                                                       ck::ReduceTensorOp::AVG,
                                                       false,
                                                       false>(true,
@@ -34,6 +36,7 @@ class TestAvgPool3dFwd : public ::testing::Test
                                                              param.length_,
                                                              param.window_spatial_lengths_,
                                                              param.window_strides_,
+                                                             param.window_dilations_,
                                                              param.input_left_pads_,
                                                              param.input_right_pads_);
             EXPECT_TRUE(success);
@@ -49,10 +52,11 @@ using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
 TYPED_TEST_SUITE(TestAvgPool3dFwd, KernelTypes);
 TYPED_TEST(TestAvgPool3dFwd, Test_Pool)
 {
-    // length, window_length, window_stride, left_pad, right_pad
-    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
-                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
-                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
+    // length, window_length, window_stride, window_dilation, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {4, 4, 4}, {4, 4, 4}, {2, 2, 2}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}};
 
     this->Run();
 }
diff --git a/test/pool_fwd/test_max_pool2d_fwd.cpp b/test/pool_fwd/test_max_pool2d_fwd.cpp
deleted file mode 100644
index c9050848d2..0000000000
--- a/test/pool_fwd/test_max_pool2d_fwd.cpp
+++ /dev/null
@@ -1,77 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "gtest/gtest.h"
-#include "profiler/profile_pool2d_fwd_impl.hpp"
-#include "test_pool_fwd_common.hpp"
-
-template <typename Tuple>
-class TestMaxPool2dFwd : public ::testing::Test
-{
-    protected:
-    using InDataType      = std::tuple_element_t<0, Tuple>;
-    using OutDataType     = std::tuple_element_t<1, Tuple>;
-    using ComputeDataType = std::tuple_element_t<2, Tuple>;
-    using IndexDataType   = std::tuple_element_t<3, Tuple>;
-
-    std::vector<PoolingParam> params;
-
-    void Run()
-    {
-        for(auto param : params)
-        {
-            // max pool
-            bool success =
-                ck::profiler::profile_pool2d_fwd_impl<InDataType,
-                                                      OutDataType,
-                                                      ComputeDataType,
-                                                      IndexDataType,
-                                                      ck::ReduceTensorOp::MAX,
-                                                      false,
-                                                      false>(true,
-                                                             2,
-                                                             false,
-                                                             false,
-                                                             param.length_,
-                                                             param.window_spatial_lengths_,
-                                                             param.window_strides_,
-                                                             param.input_left_pads_,
-                                                             param.input_right_pads_);
-            EXPECT_TRUE(success);
-
-            // max pool + index
-            success = ck::profiler::profile_pool2d_fwd_impl<InDataType,
-                                                            OutDataType,
-                                                            ComputeDataType,
-                                                            IndexDataType,
-                                                            ck::ReduceTensorOp::MAX,
-                                                            false,
-                                                            true>(true,
-                                                                  2,
-                                                                  false,
-                                                                  false,
-                                                                  param.length_,
-                                                                  param.window_spatial_lengths_,
-                                                                  param.window_strides_,
-                                                                  param.input_left_pads_,
-                                                                  param.input_right_pads_);
-            EXPECT_TRUE(success);
-        }
-    }
-};
-#ifdef __fp16__
-using KernelTypes =
-    ::testing::Types<std::tuple<F16, F16, F16, I32>, std::tuple<F32, F32, F32, I32>>;
-#else
-using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
-#endif
-TYPED_TEST_SUITE(TestMaxPool2dFwd, KernelTypes);
-TYPED_TEST(TestMaxPool2dFwd, Test_Pool)
-{
-    // length, window_length, window_stride, left_pad, right_pad
-    this->params = {{{1, 1, 1, 1}, {1, 1}, {1, 1}, {0, 0}, {0, 0}},
-                    {{2, 16, 64, 64}, {64, 64}, {1, 1}, {0, 0}, {0, 0}},
-                    {{2, 32, 30, 30}, {2, 2}, {2, 2}, {1, 1}, {1, 1}}};
-
-    this->Run();
-}
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool_fwd/test_max_pool3d_fwd.cpp
index 7c0aada473..cd79c1ad8e 100644
--- a/test/pool_fwd/test_max_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_max_pool3d_fwd.cpp
@@ -26,6 +26,8 @@ class TestMaxPool3dFwd : public ::testing::Test
                                                       OutDataType,
                                                       ComputeDataType,
                                                       IndexDataType,
+                                                      ck::tensor_layout::convolution::NDHWC,
+                                                      ck::tensor_layout::convolution::NDHWC,
                                                       ck::ReduceTensorOp::MAX,
                                                       false,
                                                       false>(true,
@@ -35,6 +37,7 @@ class TestMaxPool3dFwd : public ::testing::Test
                                                              param.length_,
                                                              param.window_spatial_lengths_,
                                                              param.window_strides_,
+                                                             param.window_dilations_,
                                                              param.input_left_pads_,
                                                              param.input_right_pads_);
             EXPECT_TRUE(success);
@@ -44,6 +47,8 @@ class TestMaxPool3dFwd : public ::testing::Test
                                                             OutDataType,
                                                             ComputeDataType,
                                                             IndexDataType,
+                                                            ck::tensor_layout::convolution::NDHWC,
+                                                            ck::tensor_layout::convolution::NDHWC,
                                                             ck::ReduceTensorOp::MAX,
                                                             false,
                                                             true>(true,
@@ -53,6 +58,7 @@ class TestMaxPool3dFwd : public ::testing::Test
                                                                   param.length_,
                                                                   param.window_spatial_lengths_,
                                                                   param.window_strides_,
+                                                                  param.window_dilations_,
                                                                   param.input_left_pads_,
                                                                   param.input_right_pads_);
             EXPECT_TRUE(success);
@@ -70,10 +76,11 @@ using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, I32>>;
 TYPED_TEST_SUITE(TestMaxPool3dFwd, KernelTypes);
 TYPED_TEST(TestMaxPool3dFwd, Test_Pool)
 {
-    // length, window_length, window_stride, left_pad, right_pad
-    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
-                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
-                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}}};
+    // length, window_length, window_stride, window_dilation, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {64, 64, 64}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {4, 4, 4}, {4, 4, 4}, {2, 2, 2}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}};
 
     this->Run();
 }
diff --git a/test/pool_fwd/test_pool_fwd_common.hpp b/test/pool_fwd/test_pool_fwd_common.hpp
index f018635170..6bfcb47bad 100644
--- a/test/pool_fwd/test_pool_fwd_common.hpp
+++ b/test/pool_fwd/test_pool_fwd_common.hpp
@@ -14,11 +14,13 @@ struct PoolingParam
     PoolingParam(const std::vector<index_t>& length,
                  const std::vector<index_t>& window_spatial_lengths,
                  const std::vector<index_t>& window_strides,
+                 const std::vector<index_t>& window_dilations,
                  const std::vector<index_t>& input_left_pads,
                  const std::vector<index_t>& input_right_pads)
         : length_(length),
           window_spatial_lengths_(window_spatial_lengths),
           window_strides_(window_strides),
+          window_dilations_(window_dilations),
           input_left_pads_(input_left_pads),
           input_right_pads_(input_right_pads)
     {
@@ -26,6 +28,7 @@ struct PoolingParam
     std::vector<index_t> length_;
     std::vector<index_t> window_spatial_lengths_;
     std::vector<index_t> window_strides_;
+    std::vector<index_t> window_dilations_;
     std::vector<index_t> input_left_pads_;
     std::vector<index_t> input_right_pads_;
 };

From d4c84256f790d210ea5c952cb18ad13542f6c698 Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Mon, 14 Aug 2023 22:46:27 +0200
Subject: [PATCH 092/194] Implement DPP8 based GEMM for Navi21 (#826)

---
 example/01_gemm/CMakeLists.txt                |   2 +
 example/01_gemm/gemm_dl_dpp8_fp16.cpp         |  37 ++
 include/ck/ck.hpp                             |   3 +
 .../gpu/block/blockwise_gemm_dl_dpp8.hpp      | 370 ++++++++++++++++++
 .../gpu/block/blockwise_gemm_dl_v2r3.hpp      |   2 +-
 .../gpu/device/gemm_dl_algorithm.hpp          |  18 +
 .../gpu/device/impl/device_gemm_dl.hpp        |  34 +-
 .../gpu/device/impl/device_gemm_dl_dpp8.hpp   | 133 +++++++
 .../gpu/grid/gridwise_gemm_dl_v1r3.hpp        |  77 +++-
 .../thread/threadwise_contraction_dl_dpp8.hpp | 136 +++++++
 include/ck/utility/amd_gemm_dpp.hpp           |  22 ++
 include/ck/utility/inner_product_dpp8.hpp     | 142 +++++++
 .../tensor_operation_instance/gpu/gemm.hpp    |  24 ++
 .../gpu/gemm/CMakeLists.txt                   |   4 +
 ..._dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp |  61 +++
 ..._dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp |  61 +++
 ..._dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp |  60 +++
 ..._dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp |  61 +++
 18 files changed, 1217 insertions(+), 30 deletions(-)
 create mode 100644 example/01_gemm/gemm_dl_dpp8_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp
 create mode 100644 include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp
 create mode 100644 include/ck/utility/amd_gemm_dpp.hpp
 create mode 100644 include/ck/utility/inner_product_dpp8.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index f8b886ebbf..a5933262a5 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -6,6 +6,8 @@ if(DL_KERNELS)
   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
     add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
     add_dependencies(example_gemm_dl example_gemm_dl_fp16)
+    add_example_executable(example_gemm_dl_dpp8_fp16 gemm_dl_dpp8_fp16.cpp)
+    add_dependencies(example_gemm_dl example_gemm_dl_dpp8_fp16)
   endif()
   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
     add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
diff --git a/example/01_gemm/gemm_dl_dpp8_fp16.cpp b/example/01_gemm/gemm_dl_dpp8_fp16.cpp
new file mode 100644
index 0000000000..ea0ba39076
--- /dev/null
+++ b/example/01_gemm/gemm_dl_dpp8_fp16.cpp
@@ -0,0 +1,37 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
+
+using ADataType   = ck::half_t;
+using BDataType   = ck::half_t;
+using CDataType   = ck::half_t;
+using AccDataType = float;
+
+using ALayout = Col;
+using BLayout = Row;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDlDpp8
+// ######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
+// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise| Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
+// ######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
+// ######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   128,   128,    16,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,               5,                  4>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 950121feb3..0e7e690421 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -125,6 +125,9 @@
 // `s_nop`s to avoid hazard
 #define CK_USE_AMD_V_DOT_INLINE_ASM 0
 
+// inner product using V_DOT with DPP8 modifiers
+#define CK_USE_AMD_V_DOT_DPP8_INLINE_ASM 1
+
 // block synchronization only s_wait lgkmcnt(0), not vmcnt(0)
 #define CK_EXPERIMENTAL_BLOCK_SYNC_LDS_WITHOUT_SYNC_VMEM 1
 
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp
new file mode 100644
index 0000000000..e527509f57
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp
@@ -0,0 +1,370 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/amd_gemm_dpp.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_adaptor.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp"
+
+namespace ck {
+
+/**
+ * DPP8 version of blockwise GEMM algorithm. It uses DPP8 instruction modifier to limit
+ * the data loaded from LDS to registers.
+ *
+ * The algorithm groups threads into groups of size `dpp8::lane_group_size` and splits the matrix C
+ * between them in such a way that threads from the same group need the same chunk of either
+ * matrix A (or B, respectively). Without the usage of DPP8, each thread would need to load the
+ * whole chunk from LDS to its own register space.
+ * Usage of DPP8 modifiers allow each thread to load less data, exactly `1 / dpp8::lane_group_size`
+ * of the chunk, and then share that data with other threads from the same lane group.
+ *
+ * Assumptions coming from the usage of DPP8:
+ *   1. `BM10BN10ThreadClusterBM10Xs[1] == dpp8::lane_group_size` or
+ *      `BM10BN10ThreadClusterBN10Xs[1] == dpp8::lane_group_size` -
+ *        - it makes consecutive `dpp8::lane_group_size` threads use the same chunk of either
+ *          matrix A or B;
+ *        - based on these values we determine which matrix to share.
+ *   2. `BM1PerThreadBM11 % dpp8::lane_group_size == 0` (if sharing A) or
+ *      `BN1PerThreadBN11 % dpp8::lane_group_size == 0` (if sharing B) -
+ *        - we have to make sure that the data to split is divisible by the number of
+ *          threads in the group.
+ *
+ * General algorithm:
+ * C[BM0, BM1, BN0, BN1] += transpose(A[K, BM0, BM1]) * B[K, BN0, BN1]
+ * A and B are visible to the whole block, C is distributed among each thread
+ * Assume:
+ *   1. A:
+ *     1. ABlockDesc_BK0_BM_BK1 is known at compile-time
+ *     2. ABlockBuffer is DynamicBuffer
+ *   2. B:
+ *     1. BBlockDesc_BK0_BN_BK1 is known at compile-time
+ *     2. BBlockBuffer is DynamicBuffer
+ *   3. C:
+ *     1. CThreadDesc_BM0_BM11_BN0_BN11 is known at compile-time
+ *     2. CThreadBuffer is StaticBuffer
+ *   4. BM10BN10ThreadClusterBM10Xs::Size() = BM10BN10ThreadClusterBN10Xs::Size() == 2
+ */
+template <index_t BlockSize,
+          typename FloatA,
+          typename FloatB,
+          typename FloatC,
+          typename ABlockDesc_BK0_BM_BK1,
+          typename BBlockDesc_BK0_BN_BK1,
+          index_t BM1PerThreadBM11,
+          index_t BN1PerThreadBN11,
+          index_t BK0PerThread,
+          typename BM10BN10ThreadClusterBM10Xs, // Sequence<BM10BN10ThreadClusterBM100,
+                                                //          BM10BN10ThreadClusterBM101, ...>
+          typename BM10BN10ThreadClusterBN10Xs, // Sequence<BM10BN10ThreadClusterBN100,
+                                                //          BM10BN10ThreadClusterBN101, ...>
+          index_t AThreadCopyScalarPerVector_BM11,
+          index_t BThreadCopyScalarPerVector_BN11,
+          typename enable_if<ABlockDesc_BK0_BM_BK1::IsKnownAtCompileTime() &&
+                                 BBlockDesc_BK0_BN_BK1::IsKnownAtCompileTime(),
+                             bool>::type = false>
+struct BlockwiseGemmDlDpp8_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_loop_BM0_BN0
+{
+    using AIndex = MultiIndex<4>;
+    using BIndex = MultiIndex<4>;
+    using CIndex = MultiIndex<4>;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+    static constexpr index_t BK0 = ABlockDesc_BK0_BM_BK1{}.GetLength(I0);
+    static constexpr index_t BK1 = ABlockDesc_BK0_BM_BK1{}.GetLength(I2);
+    static constexpr index_t BM  = ABlockDesc_BK0_BM_BK1{}.GetLength(I1);
+    static constexpr index_t BN  = BBlockDesc_BK0_BN_BK1{}.GetLength(I1);
+
+    static constexpr index_t BM100 = BM10BN10ThreadClusterBM10Xs{}[I0];
+    static constexpr index_t BN100 = BM10BN10ThreadClusterBN10Xs{}[I0];
+
+    static constexpr index_t BM101 = BM10BN10ThreadClusterBM10Xs{}[I1];
+    static constexpr index_t BN101 = BM10BN10ThreadClusterBN10Xs{}[I1];
+
+    static constexpr index_t BM11 = BM1PerThreadBM11;
+    static constexpr index_t BN11 = BN1PerThreadBN11;
+
+    static constexpr index_t BM1 = BM100 * BM101 * BM11;
+    static constexpr index_t BN1 = BN100 * BN101 * BN11;
+
+    static constexpr index_t BM0 = BM / BM1;
+    static constexpr index_t BN0 = BN / BN1;
+
+    // We assume that either `BM101` or `BN101` is equal to `dpp8::lane_group_size`. It makes all
+    // threads in a lane group need the same chunk of B or A matrices and we can share them using
+    // DPP.
+    static_assert(BM101 == dpp8::lane_group_size || BN101 == dpp8::lane_group_size);
+    static constexpr bool ShareB = BM101 == dpp8::lane_group_size ? true : false;
+    static constexpr bool ShareA = !ShareB;
+
+    // If DPP shares A (B, respectively), lane group gets `BM1PerThreadBM11` (`BN1PerThreadBN11`,
+    // respectively) elements, so we split them between threads in lane group so each thread loads
+    // less data from LDS.
+    static constexpr index_t BM1PerThread =
+        ShareA ? BM1PerThreadBM11 / dpp8::lane_group_size : BM1PerThreadBM11;
+    static constexpr index_t BN1PerThread =
+        ShareB ? BN1PerThreadBN11 / dpp8::lane_group_size : BN1PerThreadBN11;
+
+    __host__ __device__ static constexpr auto
+    MakeABlockDescriptor_BK0_BM0_BM1_BK1(const ABlockDesc_BK0_BM_BK1& a_block_desc_bk0_bm_bk1)
+    {
+        const auto a_block_bk0_bm0_bm1_bk1 = transform_tensor_descriptor(
+            a_block_desc_bk0_bm_bk1,
+            make_tuple(make_pass_through_transform(Number<BK0>{}),
+                       make_unmerge_transform(make_tuple(Number<BM0>{}, Number<BM1>{})),
+                       make_pass_through_transform(Number<BK1>{})),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}));
+
+        return a_block_bk0_bm0_bm1_bk1;
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeBBlockDescriptor_BK0_BN0_BN1_BK1(const BBlockDesc_BK0_BN_BK1& b_block_desc_bk0_bn_bk1)
+    {
+        const auto b_block_desc_bk0_bn0_bn1_bk1 = transform_tensor_descriptor(
+            b_block_desc_bk0_bn_bk1,
+            make_tuple(make_pass_through_transform(Number<BK0>{}),
+                       make_unmerge_transform(make_tuple(Number<BN0>{}, Number<BN1>{})),
+                       make_pass_through_transform(Number<BK1>{})),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}));
+
+        return b_block_desc_bk0_bn0_bn1_bk1;
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeCBlockAdaptor_BM0_BM100_BM101_BM11_BN0_BN100_BN101_BN11_To_BM_BN()
+    {
+        // upper: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
+        // lower: [BM, BN]
+        constexpr auto c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m_n =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_unmerge_transform(make_tuple(
+                               Number<BM0>{}, Number<BM100>{}, Number<BM101>{}, Number<BM11>{})),
+                           make_unmerge_transform(make_tuple(
+                               Number<BN0>{}, Number<BN100>{}, Number<BN101>{}, Number<BN11>{}))),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 2, 3>{}, Sequence<4, 5, 6, 7>{}));
+
+        return c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m_n;
+    }
+
+    __host__ __device__ static constexpr auto
+    MakeCBlockAdaptor_BM0_BM100_BM101_BM11_BN0_BN100_BN101_BN11_To_BM0_BM1_BN0_BN1()
+    {
+        // upper: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
+        // lower: [BM0, BM1, BN0, BN1]
+        constexpr auto c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m0_m1_n0_n1 =
+            make_single_stage_tensor_adaptor(
+                make_tuple(make_pass_through_transform(Number<BM0>{}),
+                           make_unmerge_transform(
+                               make_tuple(Number<BM100>{}, Number<BM101>{}, Number<BM11>{})),
+                           make_pass_through_transform(Number<BN0>{}),
+                           make_unmerge_transform(
+                               make_tuple(Number<BN100>{}, Number<BN101>{}, Number<BN11>{}))),
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1, 2, 3>{}, Sequence<4>{}, Sequence<5, 6, 7>{}));
+
+        return c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m0_m1_n0_n1;
+    }
+
+    __host__ __device__ static constexpr auto GetCThreadTensorLengths_BM0_BM1_BN0_BN1()
+    {
+        return Sequence<BM0, BM11, BN0, BN11>{};
+    }
+
+    static constexpr auto a_block_desc_bk0_bm0_bm1_bk1_ =
+        MakeABlockDescriptor_BK0_BM0_BM1_BK1(ABlockDesc_BK0_BM_BK1{});
+
+    static constexpr auto b_block_desc_bk0_bn0_bn1_bk1_ =
+        MakeBBlockDescriptor_BK0_BN0_BN1_BK1(BBlockDesc_BK0_BN_BK1{});
+
+    public:
+    __device__ BlockwiseGemmDlDpp8_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_loop_BM0_BN0()
+        : c_thread_origin_data_idx_{CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(
+              get_thread_local_1d_id())},
+          a_thread_copy_{CalculateAThreadOriginOnBlock_BK0_BM0_BM1_BK1()},
+          b_thread_copy_{CalculateBThreadOriginOnBlock_BK0_BN0_BN1_BK1()}
+    {
+        static_assert(ABlockDesc_BK0_BM_BK1::IsKnownAtCompileTime() &&
+                          BBlockDesc_BK0_BN_BK1::IsKnownAtCompileTime(),
+                      "wrong! Desc should be known at compile-time");
+
+        static_assert(BM % BM1 == 0 && BN % BN1 == 0, "wrong!");
+
+        static_assert(ABlockDesc_BK0_BM_BK1{}.GetLength(I0) ==
+                          BBlockDesc_BK0_BN_BK1{}.GetLength(I0),
+                      "wrong! K dimension not consistent");
+
+        static_assert(BM10BN10ThreadClusterBM10Xs::Size() == 2 &&
+                          BM10BN10ThreadClusterBN10Xs::Size() == 2,
+                      "wrong!");
+    }
+
+    __device__ static CIndex CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(index_t thread_id)
+    {
+        // lower: [BM0, BM1, BN0, BN1]
+        // upper: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
+        constexpr auto adaptor0 =
+            MakeCBlockAdaptor_BM0_BM100_BM101_BM11_BN0_BN100_BN101_BN11_To_BM0_BM1_BN0_BN1();
+
+        // lower: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
+        // upper: [Tid, BM0, BM11, BN0, BN11]
+        constexpr auto adaptor1 = make_single_stage_tensor_adaptor(
+            make_tuple(make_merge_transform(make_tuple(BM100, BN100, BM101, BN101)),
+                       make_pass_through_transform(BM0),
+                       make_pass_through_transform(BM11),
+                       make_pass_through_transform(BN0),
+                       make_pass_through_transform(BN11)),
+            make_tuple(
+                Sequence<1, 5, 2, 6>{}, Sequence<0>{}, Sequence<3>{}, Sequence<4>{}, Sequence<7>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+        constexpr auto adaptor = chain_tensor_adaptors(adaptor0, adaptor1);
+
+        return adaptor.CalculateBottomIndex(make_multi_index(thread_id, 0, 0, 0, 0));
+    }
+
+    __device__ AIndex CalculateAThreadOriginOnBlock_BK0_BM0_BM1_BK1()
+    {
+        const auto offsetBM0 = c_thread_origin_data_idx_[I0];
+        // If sharing matrix A, we need a separate BM1 offset for each thread in lane group.
+        const auto offsetBM1 = ShareA ? c_thread_origin_data_idx_[I1] +
+                                            dpp8::get_thread_idx_in_lane_group() * BM1PerThread
+                                      : c_thread_origin_data_idx_[I1];
+        return make_tuple(0, offsetBM0, offsetBM1, 0);
+    }
+
+    __device__ BIndex CalculateBThreadOriginOnBlock_BK0_BN0_BN1_BK1()
+    {
+        const auto offsetBN0 = c_thread_origin_data_idx_[I2];
+        // If sharing matrix B, we need a separate BN1 offset for each thread in lane group.
+        const auto offsetBN1 = ShareB ? c_thread_origin_data_idx_[I3] +
+                                            dpp8::get_thread_idx_in_lane_group() * BN1PerThread
+                                      : c_thread_origin_data_idx_[I3];
+        return make_tuple(0, offsetBN0, offsetBN1, 0);
+    }
+
+    template <typename CThreadDesc_BM0_BM11_BN0_BN11,
+              typename ABlockBuffer,
+              typename BBlockBuffer,
+              typename CThreadBuffer>
+    __device__ void Run(const CThreadDesc_BM0_BM11_BN0_BN11&,
+                        const ABlockBuffer& a_block_buf,
+                        const BBlockBuffer& b_block_buf,
+                        CThreadBuffer& c_thread_buf) const
+    {
+        static_assert(CThreadDesc_BM0_BM11_BN0_BN11::IsKnownAtCompileTime(),
+                      "wrong! Desc should be known at compile-time");
+
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatA>(
+            a_thread_desc_bk0_bm0_bm1_bk1_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
+            b_thread_desc_bk0_bn0_bn1_bk1_.GetElementSpaceSize());
+
+        constexpr auto threadwise_contraction =
+            ThreadwiseContractionDlDpp8_A_TK0_TM0_TM1_TK1_B_TK0_TN0_TN1_TK1_C_TM0_TM1_TN0_TN1<
+                FloatA,
+                FloatB,
+                FloatC,
+                decltype(a_thread_desc_bk0_bm0_bm1_bk1_),
+                decltype(b_thread_desc_bk0_bn0_bn1_bk1_),
+                CThreadDesc_BM0_BM11_BN0_BN11,
+                Sequence<BK0PerThread, BK1>,
+                Sequence<1, BM1PerThreadBM11>,
+                Sequence<1, BN1PerThreadBN11>,
+                ShareA>{};
+
+        static_for<0, BN0, 1>{}([&](auto bn0) {
+            static_for<0, BM0, 1>{}([&](auto bm0) {
+                a_thread_copy_.Run(a_block_desc_bk0_bm0_bm1_bk1_,
+                                   make_tuple(I0, bm0, I0, I0),
+                                   a_block_buf,
+                                   a_thread_desc_bk0_bm0_bm1_bk1_,
+                                   make_tuple(I0, I0, I0, I0),
+                                   a_thread_buf);
+
+                b_thread_copy_.Run(b_block_desc_bk0_bn0_bn1_bk1_,
+                                   make_tuple(I0, bn0, I0, I0),
+                                   b_block_buf,
+                                   b_thread_desc_bk0_bn0_bn1_bk1_,
+                                   make_tuple(I0, I0, I0, I0),
+                                   b_thread_buf);
+
+                threadwise_contraction.Run(a_thread_buf,
+                                           make_tuple(I0, I0, I0, I0),
+                                           b_thread_buf,
+                                           make_tuple(I0, I0, I0, I0),
+                                           c_thread_buf,
+                                           make_tuple(bm0, I0, bn0, I0));
+
+                static_for<BK0PerThread, BK0, BK0PerThread>{}([&](auto bk0) {
+                    a_thread_copy_.Run(a_block_desc_bk0_bm0_bm1_bk1_,
+                                       make_tuple(bk0, bm0, I0, I0),
+                                       a_block_buf,
+                                       a_thread_desc_bk0_bm0_bm1_bk1_,
+                                       make_tuple(I0, I0, I0, I0),
+                                       a_thread_buf);
+
+                    b_thread_copy_.Run(b_block_desc_bk0_bn0_bn1_bk1_,
+                                       make_tuple(bk0, bn0, I0, I0),
+                                       b_block_buf,
+                                       b_thread_desc_bk0_bn0_bn1_bk1_,
+                                       make_tuple(I0, I0, I0, I0),
+                                       b_thread_buf);
+
+                    threadwise_contraction.Run(a_thread_buf,
+                                               make_tuple(I0, I0, I0, I0),
+                                               b_thread_buf,
+                                               make_tuple(I0, I0, I0, I0),
+                                               c_thread_buf,
+                                               make_tuple(bm0, I0, bn0, I0));
+                });
+            });
+        });
+    }
+
+    private:
+    // A[BK0, BM0, BM1, BK1]
+    static constexpr auto a_thread_desc_bk0_bm0_bm1_bk1_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<BK0PerThread>{}, Number<BM0>{}, Number<BM1PerThread>{}, Number<BK1>{}));
+
+    // B[BK0, BN0, BN1, BK1]
+    static constexpr auto b_thread_desc_bk0_bn0_bn1_bk1_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<BK0PerThread>{}, Number<BN0>{}, Number<BN1PerThread>{}, Number<BK1>{}));
+
+    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4r1<
+        FloatA,
+        FloatA,
+        decltype(a_block_desc_bk0_bm0_bm1_bk1_),
+        decltype(a_thread_desc_bk0_bm0_bm1_bk1_),
+        Sequence<BK0PerThread, 1, BM1PerThread, BK1>, // SliceLengths
+        Sequence<0, 1, 2, 3>,                         // DimAccessOrder
+        Sequence<1, 1, BM1PerThread, BK1>,            // SrcVectorTensorLengths
+        Sequence<0, 1, 2, 3>>;                        // SrcVectorTensorContiguousDimOrder
+
+    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4r1<
+        FloatB,
+        FloatB,
+        decltype(b_block_desc_bk0_bn0_bn1_bk1_),
+        decltype(b_thread_desc_bk0_bn0_bn1_bk1_),
+        Sequence<BK0PerThread, 1, BN1PerThread, BK1>, // SliceLengths
+        Sequence<0, 1, 2, 3>,                         // DimAccessOrder
+        Sequence<1, 1, BN1PerThread, BK1>,            // SrcVectorTensorLengths
+        Sequence<0, 1, 2, 3>>;                        // SrcVectorTensorContiguousDimOrder
+
+    CIndex c_thread_origin_data_idx_;
+
+    AThreadCopy a_thread_copy_;
+    BThreadCopy b_thread_copy_;
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
index b3caa3214a..f23404a1d7 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp
@@ -11,7 +11,7 @@
 namespace ck {
 
 // C[BM0, BM1, BN0, BN1] += transpose(A[K, BM0, BM1]) * B[K, BN0, BN1]
-// A and B are visable to the whole block, C is distributed among each thread
+// A and B are visible to the whole block, C is distributed among each thread
 // Assume:
 //   1. A:
 //     1. ABlockDesc_BK0_BM_BK1 is known at compile-time
diff --git a/include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp b/include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp
new file mode 100644
index 0000000000..16ca582b89
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp
@@ -0,0 +1,18 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+enum struct GemmDlAlgorithm
+{
+    Default, // Uses DOT vector instructions
+    Dpp8,    // Uses DOT vector instructions with DPP8 SEL modifier to reduce data loads from LDS
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
index 13e9f96910..eedf384cd9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
@@ -11,6 +11,7 @@
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
 #include "ck/host_utility/device_prop.hpp"
@@ -59,6 +60,7 @@ template <
     typename CThreadTransferSrcDstAccessOrder,
     index_t CThreadTransferSrcDstVectorDim,
     index_t CThreadTransferDstScalarPerVector,
+    GemmDlAlgorithm GemmDlAlg = GemmDlAlgorithm::Default,
     enable_if_t<
         is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
             is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
@@ -236,7 +238,8 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                      BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
                                      CThreadTransferSrcDstAccessOrder,
                                      CThreadTransferSrcDstVectorDim,
-                                     CThreadTransferDstScalarPerVector>;
+                                     CThreadTransferDstScalarPerVector,
+                                     GemmDlAlg>;
 
     using AGridDesc_K0_M0_M1_K1 =
         decltype(GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
@@ -372,7 +375,8 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         true,
-                                        true>;
+                                        true,
+                                        GemmDlAlg>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -398,7 +402,8 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         true,
-                                        false>;
+                                        false,
+                                        GemmDlAlg>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -424,7 +429,8 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         false,
-                                        true>;
+                                        true,
+                                        GemmDlAlg>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -450,7 +456,8 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         false,
-                                        false>;
+                                        false,
+                                        GemmDlAlg>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -485,6 +492,16 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if constexpr(GemmDlAlg == GemmDlAlgorithm::Dpp8)
+        {
+            if(ck::get_device_name() == "gfx1030")
+            {
+                return GridwiseGemm::CheckValidity(
+                    arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
+            }
+            return false;
+        }
+
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
            ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
            ck::get_device_name() == "gfx1102")
@@ -492,10 +509,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
         }
-        else
-        {
-            return false;
-        }
+        return false;
     }
 
     // polymorphic
@@ -572,7 +586,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
     }
 
     // polymorphic
-    std::string GetTypeString() const override
+    virtual std::string GetTypeString() const override
     {
         auto str = std::stringstream();
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp
new file mode 100644
index 0000000000..336ce31e82
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <
+    typename ADataType,
+    typename BDataType,
+    typename CDataType,
+    typename AccDataType,
+    typename ALayout,
+    typename BLayout,
+    typename CLayout,
+    typename AElementwiseOperation,
+    typename BElementwiseOperation,
+    typename CElementwiseOperation,
+    GemmSpecialization GemmSpec,
+    index_t BlockSize,
+    index_t MPerBlock,
+    index_t NPerBlock,
+    index_t K0PerBlock,
+    index_t K1,
+    index_t M1PerThread,
+    index_t N1PerThread,
+    index_t KPerThread,
+    typename M1N1ThreadClusterM1Xs,
+    typename M1N1ThreadClusterN1Xs,
+    typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+    typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+    typename ABlockTransferThreadClusterArrangeOrder,
+    typename ABlockTransferSrcAccessOrder,
+    typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+    typename ABlockTransferSrcVectorTensorContiguousDimOrder,
+    typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+    typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+    typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+    typename BBlockTransferThreadClusterArrangeOrder,
+    typename BBlockTransferSrcAccessOrder,
+    typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+    typename BBlockTransferSrcVectorTensorContiguousDimOrder,
+    typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+    typename CThreadTransferSrcDstAccessOrder,
+    index_t CThreadTransferSrcDstVectorDim,
+    index_t CThreadTransferDstScalarPerVector,
+    enable_if_t<
+        is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+            is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
+            is_same_v<CElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
+        bool> = false>
+struct DeviceGemmDlDpp8 : public DeviceGemmDl<ADataType,
+                                              BDataType,
+                                              CDataType,
+                                              AccDataType,
+                                              ALayout,
+                                              BLayout,
+                                              CLayout,
+                                              AElementwiseOperation,
+                                              BElementwiseOperation,
+                                              CElementwiseOperation,
+                                              GemmSpec,
+                                              BlockSize,
+                                              MPerBlock,
+                                              NPerBlock,
+                                              K0PerBlock,
+                                              K1,
+                                              M1PerThread,
+                                              N1PerThread,
+                                              KPerThread,
+                                              M1N1ThreadClusterM1Xs,
+                                              M1N1ThreadClusterN1Xs,
+                                              ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
+                                              ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
+                                              ABlockTransferThreadClusterArrangeOrder,
+                                              ABlockTransferSrcAccessOrder,
+                                              ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
+                                              ABlockTransferSrcVectorTensorContiguousDimOrder,
+                                              ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
+                                              BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
+                                              BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
+                                              BBlockTransferThreadClusterArrangeOrder,
+                                              BBlockTransferSrcAccessOrder,
+                                              BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
+                                              BBlockTransferSrcVectorTensorContiguousDimOrder,
+                                              BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
+                                              CThreadTransferSrcDstAccessOrder,
+                                              CThreadTransferSrcDstVectorDim,
+                                              CThreadTransferDstScalarPerVector,
+                                              GemmDlAlgorithm::Dpp8>
+
+{
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGemmDlDpp8"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << M1PerThread << ", "
+            << N1PerThread << ", "
+            << KPerThread
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
index d46aea5e22..1d1bb6ed2d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
@@ -7,9 +7,11 @@
 #include "ck/tensor_description/multi_index_transform_helper.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp"
@@ -17,6 +19,8 @@
 
 namespace ck {
 
+using GemmDlAlgorithm = tensor_operation::device::GemmDlAlgorithm;
+
 template <typename GridwiseGemm,
           typename FloatAB,
           typename FloatC,
@@ -25,7 +29,8 @@ template <typename GridwiseGemm,
           typename CGridDesc_M0_M10_M11_N0_N10_N11,
           typename Block2CTileMap,
           bool HasMainKBlockLoop,
-          bool HasDoubleTailKBlockLoop>
+          bool HasDoubleTailKBlockLoop,
+          GemmDlAlgorithm GemmDlAlg = GemmDlAlgorithm::Default>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
@@ -38,6 +43,13 @@ __global__ void
                             const CGridDesc_M0_M10_M11_N0_N10_N11 c_grid_desc_m0_m10_m11_n0_n10_n11,
                             const Block2CTileMap block_2_ctile_map)
 {
+// DPP8 is currently only supported on gfx1030
+#if !defined(__gfx1030__)
+    if(GemmDlAlg == GemmDlAlgorithm::Dpp8)
+    {
+        return;
+    }
+#endif
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 
@@ -88,7 +100,8 @@ template <index_t BlockSize,
           typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
           typename CThreadTransferSrcDstAccessOrder,
           index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector>
+          index_t CThreadTransferDstScalarPerVector,
+          GemmDlAlgorithm GemmDlAlg = GemmDlAlgorithm::Default>
 struct GridwiseGemmDl_km_kn_mn_v1r3
 {
     static constexpr auto I0 = Number<0>{};
@@ -244,6 +257,45 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
             c_grid_desc_m_n);
     }
 
+    template <typename ABlockDesc_BK0_BM_BK1, typename BBlockDesc_BK0_BN_BK1>
+    __host__ __device__ static constexpr auto GetBlockwiseGemm()
+    {
+        if constexpr(GemmDlAlg == GemmDlAlgorithm::Dpp8)
+        {
+            return BlockwiseGemmDlDpp8_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_loop_BM0_BN0<
+                BlockSize,
+                FloatAB,
+                FloatAB,
+                FloatAcc,
+                ABlockDesc_BK0_BM_BK1,
+                BBlockDesc_BK0_BN_BK1,
+                M1PerThreadM111,
+                N1PerThreadN111,
+                KPerThread,
+                M11N11ThreadClusterM110Xs,
+                M11N11ThreadClusterN110Xs,
+                M1PerThreadM111,
+                N1PerThreadN111>{};
+        }
+        else
+        {
+            return BlockwiseGemmDl_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
+                BlockSize,
+                FloatAB,
+                FloatAB,
+                FloatAcc,
+                ABlockDesc_BK0_BM_BK1,
+                BBlockDesc_BK0_BN_BK1,
+                M1PerThreadM111,
+                N1PerThreadN111,
+                KPerThread,
+                M11N11ThreadClusterM110Xs,
+                M11N11ThreadClusterN110Xs,
+                M1PerThreadM111,
+                N1PerThreadN111>{};
+        }
+    }
+
     using AGridDesc_K0_M0_M1_K1 = decltype(MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
     using BGridDesc_K0_N0_N1_K1 = decltype(MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
     using CGridDesc_M0_M10_M11_N0_N10_N11 =
@@ -274,7 +326,7 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
         const auto c_m0_n0_block_cluster_idx =
             block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
 
-        // HACK: this force index data into SGPR
+        // HACK: this forces index data into SGPR
         const index_t im0 = __builtin_amdgcn_readfirstlane(c_m0_n0_block_cluster_idx[I0]);
         const index_t in0 = __builtin_amdgcn_readfirstlane(c_m0_n0_block_cluster_idx[I1]);
 
@@ -372,20 +424,7 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         const auto blockwise_gemm =
-            BlockwiseGemmDl_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
-                BlockSize,
-                FloatAB,
-                FloatAB,
-                FloatAcc,
-                decltype(a_k0_m_k1_block_desc),
-                decltype(b_k0_n_k1_block_desc),
-                M1PerThreadM111,
-                N1PerThreadN111,
-                KPerThread,
-                M11N11ThreadClusterM110Xs,
-                M11N11ThreadClusterN110Xs,
-                M1PerThreadM111,
-                N1PerThreadN111>{};
+            GetBlockwiseGemm<decltype(a_k0_m_k1_block_desc), decltype(b_k0_n_k1_block_desc)>();
 
         constexpr auto c_m10_m11_n10_n11_thread_tensor_lengths =
             decltype(blockwise_gemm)::GetCThreadTensorLengths_BM0_BM1_BN0_BN1();
@@ -472,7 +511,7 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
                 b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_k0_n0_n1_k1,
                                                     b_block_slice_copy_step);
 
-                // LDS doubel buffer: load next data from device mem
+                // LDS double buffer: load next data from device mem
                 a_blockwise_copy.RunRead(a_grid_desc_k0_m0_m1_k1, a_global_buf);
                 b_blockwise_copy.RunRead(b_grid_desc_k0_n0_n1_k1, b_global_buf);
 
@@ -992,7 +1031,7 @@ struct GridwiseGemmDl_bkm_bkn_mn_v1r3
                 b_blockwise_copy.MoveSrcSliceWindow(b_grid_desc_b_k0_n0_n1_k1,
                                                     b_block_slice_copy_step);
 
-                // LDS doubel buffer: load next data from device mem
+                // LDS double buffer: load next data from device mem
                 a_blockwise_copy.RunRead(a_grid_desc_b_k0_m0_m1_k1, a_global_buf);
                 b_blockwise_copy.RunRead(b_grid_desc_b_k0_n0_n1_k1, b_global_buf);
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp
new file mode 100644
index 0000000000..d0d214381d
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp
@@ -0,0 +1,136 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/amd_gemm_dpp.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/utility/inner_product_dpp8.hpp"
+#include "ck/utility/math.hpp"
+
+namespace ck {
+
+/**
+ * Threadwise contraction using dot instructions with DPP8 modifier.
+ *
+ * Assumptions:
+ *   1. `AThreadDesc_TK0_TM0_TM1_TK1`, `BThreadDesc_TK0_TN0_TN1_TK1`, `CThreadDesc_TM0_TM1_TN0_TN1`
+ *      are known at compile-time;
+ *   2. `AOriginIdx`, `BOriginIdx`, `COriginIdx` are known at compile-time;
+ *   3. `TM0` is equal to 1 and `TN0` is equal to 1;
+ *   4. When `ShareA` is set (unset, respectively), `TM1` (`TN1`, respectively) is divisible by
+ *      the size of the lane group (`dpp8::lane_group_size`).
+ */
+template <typename FloatA,
+          typename FloatB,
+          typename FloatC,
+          typename AThreadDesc_TK0_TM0_TM1_TK1,
+          typename BThreadDesc_TK0_TN0_TN1_TK1,
+          typename CThreadDesc_TM0_TM1_TN0_TN1,
+          typename TKLengths,
+          typename TMLengths,
+          typename TNLengths,
+          bool ShareA,
+          typename enable_if<AThreadDesc_TK0_TM0_TM1_TK1::IsKnownAtCompileTime() &&
+                                 BThreadDesc_TK0_TN0_TN1_TK1::IsKnownAtCompileTime() &&
+                                 CThreadDesc_TM0_TM1_TN0_TN1::IsKnownAtCompileTime(),
+                             bool>::type = false>
+struct ThreadwiseContractionDlDpp8_A_TK0_TM0_TM1_TK1_B_TK0_TN0_TN1_TK1_C_TM0_TM1_TN0_TN1
+{
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    static constexpr index_t TK0 = TKLengths{}[I0];
+    static constexpr index_t TK1 = TKLengths{}[I1];
+    static constexpr index_t TM0 = TMLengths{}[I0];
+    static constexpr index_t TM1 = TMLengths{}[I1];
+    static constexpr index_t TN0 = TNLengths{}[I0];
+    static constexpr index_t TN1 = TNLengths{}[I1];
+
+    static_assert(TM0 == 1 && TN0 == 1);
+
+    static_assert((ShareA && TM1 % dpp8::lane_group_size == 0) ||
+                  (!ShareA && TN1 % dpp8::lane_group_size == 0));
+    static constexpr index_t shared_elems_per_lane =
+        ShareA ? TM1 / dpp8::lane_group_size : TN1 / dpp8::lane_group_size;
+
+    __device__ constexpr ThreadwiseContractionDlDpp8_A_TK0_TM0_TM1_TK1_B_TK0_TN0_TN1_TK1_C_TM0_TM1_TN0_TN1()
+    {
+        static_assert(AThreadDesc_TK0_TM0_TM1_TK1::IsKnownAtCompileTime() &&
+                          BThreadDesc_TK0_TN0_TN1_TK1::IsKnownAtCompileTime() &&
+                          CThreadDesc_TM0_TM1_TN0_TN1::IsKnownAtCompileTime(),
+                      "wrong! Desc should be known at compile-time");
+
+        static_assert(TKLengths::Size() == 2 && TMLengths::Size() == 2 && TNLengths::Size() == 2,
+                      "wrong!");
+    }
+
+    template <typename ABuffer,
+              typename AOriginIdx,
+              typename BBuffer,
+              typename BOriginIdx,
+              typename CBuffer,
+              typename COriginIdx>
+    __device__ static void Run(const ABuffer& a_buf,
+                               AOriginIdx,
+                               const BBuffer& b_buf,
+                               BOriginIdx,
+                               CBuffer& c_buf,
+                               COriginIdx)
+    {
+        static_assert(is_known_at_compile_time<remove_cvref_t<AOriginIdx>>::value &&
+                          is_known_at_compile_time<remove_cvref_t<BOriginIdx>>::value &&
+                          is_known_at_compile_time<remove_cvref_t<COriginIdx>>::value,
+                      "wrong! AOriginIdx, BOriginIdx, COringinIdx should be known at compile-time");
+
+        static_assert(
+            is_same<remove_cvref_t<typename ABuffer::type>, remove_cvref_t<FloatA>>::value &&
+            is_same<remove_cvref_t<typename BBuffer::type>, remove_cvref_t<FloatB>>::value &&
+            is_same<remove_cvref_t<typename CBuffer::type>, remove_cvref_t<FloatC>>::value &&
+            "wrong! inconsistent type");
+
+        constexpr auto a_origin_idx = to_multi_index(AOriginIdx{});
+        constexpr auto b_origin_idx = to_multi_index(BOriginIdx{});
+        constexpr auto c_origin_idx = to_multi_index(COriginIdx{});
+
+        static_for<0, TK0, 1>{}([&](auto tk0) {
+            static_for<0, TM1, 1>{}([&](auto tm1) {
+                static_for<0, TN1, 1>{}([&](auto tn1) {
+                    vector_type<FloatA, TK1> a_vec;
+                    vector_type<FloatB, TK1> b_vec;
+
+                    static_for<0, TK1, 1>{}([&](auto tk1) {
+                        constexpr index_t local_tm1 = ShareA ? tm1 % shared_elems_per_lane : tm1;
+                        constexpr index_t a_offset  = AThreadDesc_TK0_TM0_TM1_TK1{}.CalculateOffset(
+                            a_origin_idx + make_multi_index(tk0, 0, local_tm1, tk1));
+
+                        constexpr index_t local_tn1 = ShareA ? tn1 : tn1 % shared_elems_per_lane;
+                        constexpr index_t b_offset  = BThreadDesc_TK0_TN0_TN1_TK1{}.CalculateOffset(
+                            b_origin_idx + make_multi_index(tk0, 0, local_tn1, tk1));
+
+                        a_vec.template AsType<FloatA>()(tk1) = a_buf[Number<a_offset>{}];
+                        b_vec.template AsType<FloatB>()(tk1) = b_buf[Number<b_offset>{}];
+                    });
+
+                    using a_vector_t = typename vector_type<FloatA, TK1>::type;
+                    using b_vector_t = typename vector_type<FloatB, TK1>::type;
+
+                    constexpr index_t c_offset = CThreadDesc_TM0_TM1_TN0_TN1{}.CalculateOffset(
+                        c_origin_idx + make_multi_index(0, tm1, 0, tn1));
+
+                    constexpr int src_lane =
+                        ShareA ? (tm1 / shared_elems_per_lane) % dpp8::lane_group_size
+                               : (tn1 / shared_elems_per_lane) % dpp8::lane_group_size;
+
+                    dpp8::inner_product_dpp<a_vector_t, b_vector_t, FloatC, src_lane, ShareA>(
+                        a_vec.template AsType<a_vector_t>()[I0],
+                        b_vec.template AsType<b_vector_t>()[I0],
+                        c_buf(Number<c_offset>{}));
+                });
+            });
+        });
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/utility/amd_gemm_dpp.hpp b/include/ck/utility/amd_gemm_dpp.hpp
new file mode 100644
index 0000000000..8d6c7eede9
--- /dev/null
+++ b/include/ck/utility/amd_gemm_dpp.hpp
@@ -0,0 +1,22 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/utility/math.hpp"
+#include "ck/utility/amd_gemm_dpp.hpp"
+
+namespace ck {
+
+namespace dpp8 {
+
+/// Number of lanes that can share data using DPP8 modifiers.
+constexpr index_t lane_group_size = 8;
+
+__device__ index_t get_lane_group_local_idx() { return threadIdx.x / lane_group_size; }
+__device__ index_t get_thread_idx_in_lane_group() { return threadIdx.x % lane_group_size; }
+
+} // namespace dpp8
+
+} // namespace ck
diff --git a/include/ck/utility/inner_product_dpp8.hpp b/include/ck/utility/inner_product_dpp8.hpp
new file mode 100644
index 0000000000..ccd7a4e628
--- /dev/null
+++ b/include/ck/utility/inner_product_dpp8.hpp
@@ -0,0 +1,142 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+#include "amd_gemm_dpp.hpp"
+#include "data_type.hpp"
+#include "type_convert.hpp"
+
+namespace ck {
+
+namespace dpp8 {
+
+template <int SrcLaneIdx>
+__device__ void inline_v_dot2c_dpp8_instr(const half2_t& a, const half2_t& b, float& c);
+
+// clang-format off
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<0>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[0, 0, 0, 0, 0, 0, 0, 0]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<1>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[1, 1, 1, 1, 1, 1, 1, 1]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<2>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[2, 2, 2, 2, 2, 2, 2, 2]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<3>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[3, 3, 3, 3, 3, 3, 3, 3]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<4>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[4, 4, 4, 4, 4, 4, 4, 4]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<5>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[5, 5, 5, 5, 5, 5, 5, 5]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<6>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[6, 6, 6, 6, 6, 6, 6, 6]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+template <>
+__device__ void inline_v_dot2c_dpp8_instr<7>(const half2_t& a, const half2_t& b, float& c){
+    asm volatile("\n v_dot2c_f32_f16_dpp %0, %1, %2 dpp8:[7, 7, 7, 7, 7, 7, 7, 7]" : "=v"(c) : "v"(a), "v"(b), "0"(c));
+}
+// clang-format on
+
+/**
+ * Dot product of two vectors using `v_dot` instruction with DPP8 submitted as inline assembly.
+ */
+template <int SrcLaneIdx, bool ShareA>
+__device__ void inline_v_dot2c_dpp8(const half2_t& a, const half2_t& b, float& c)
+{
+    static_assert(SrcLaneIdx >= 0 && SrcLaneIdx < dpp8::lane_group_size,
+                  "DPP8 src broadcast lane out of range <0, 7>.");
+    if constexpr(ShareA)
+    {
+        inline_v_dot2c_dpp8_instr<SrcLaneIdx>(a, b, c);
+    }
+    else
+    {
+        inline_v_dot2c_dpp8_instr<SrcLaneIdx>(b, a, c);
+    }
+}
+
+/**
+ * DPP8 instrinsics expects to get an integer mask, hardcoding integers for specific broadcast
+ * patters.
+ */
+constexpr std::array<int, dpp8::lane_group_size> IntrinsicMaskDpp8 = {
+    0,        // 0, 0, 0, 0, 0, 0, 0, 0
+    2396745,  // 1, 1, 1, 1, 1, 1, 1, 1
+    4793490,  // 2, 2, 2, 2, 2, 2, 2, 2
+    7190235,  // 3, 3, 3, 3, 3, 3, 3, 3
+    9586980,  // 4, 4, 4, 4, 4, 4, 4, 4
+    11983725, // 5, 5, 5, 5, 5, 5, 5, 5
+    14380470, // 6, 6, 6, 6, 6, 6, 6, 6
+    16777215, // 7, 7, 7, 7, 7, 7, 7, 7
+};
+
+/**
+ * Returns DPP8 sel modifier as an integer required for the intrinsic instruction.
+ */
+template <int SrcLaneIdx>
+constexpr int get_dpp_sel_mask_broadcast()
+{
+    static_assert(SrcLaneIdx >= 0 && SrcLaneIdx < dpp8::lane_group_size,
+                  "DPP8 src broadcast lane out of range <0, 7>.");
+    return IntrinsicMaskDpp8[SrcLaneIdx];
+}
+
+template <int SrcLaneIdx>
+__device__ void intrinsic_fdot2_impl(const half2_t& a, const half2_t& b, float& c)
+{
+    constexpr int sel_mask = get_dpp_sel_mask_broadcast<SrcLaneIdx>();
+    const half2_t val_from_other_lane =
+        bit_cast<half2_t>(__builtin_amdgcn_mov_dpp8(bit_cast<int>(a), sel_mask));
+    c = __builtin_amdgcn_fdot2(val_from_other_lane, b, c, false);
+}
+
+/**
+ * Dot product of two vectors using `v_dot` instruction with DPP8 submitted using intrinsics.
+ */
+template <int SrcLaneIdx, bool ShareA>
+__device__ void intrinsic_fdot2(const half2_t& a, const half2_t& b, float& c)
+{
+    if constexpr(ShareA)
+    {
+        intrinsic_fdot2_impl<SrcLaneIdx>(a, b, c);
+    }
+    else
+    {
+        intrinsic_fdot2_impl<SrcLaneIdx>(b, a, c);
+    }
+}
+
+/**
+ * Dot product of two input vectors `a`, `b` using `v_dot` instructions with DPP modifier.
+ *
+ * DPP modifier allows us to share one of the vectors between lanes in a lane group.
+ * When `ShareA` is set, instruction uses vector `a` from lane `SrcLaneIdx` from the same
+ * lane group (8 lanes per lane group in DPP8). When `ShareA` is not set, vector `b` is shared.
+ * Note that all the threads in a lane group uses the same vector - broadcast pattern.
+ *
+ * `SrcLaneIdx` must be in range from 0 to 7.
+ */
+template <typename TA, typename TB, typename TC, int SrcLaneIdx, bool ShareA>
+__device__ void inner_product_dpp(const TA& a, const TB& b, TC& c)
+{
+#if CK_USE_AMD_V_DOT_DPP8_INLINE_ASM
+    inline_v_dot2c_dpp8<SrcLaneIdx, ShareA>(a, b, c);
+#else
+    intrinsic_fdot2<SrcLaneIdx, ShareA>(a, b, c);
+#endif
+}
+
+} // namespace dpp8
+
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 538ca758f7..06197d9f46 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -23,6 +23,11 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -33,6 +38,11 @@ void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -43,6 +53,11 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -53,6 +68,11 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -354,6 +374,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
+                add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
@@ -364,6 +385,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
+                add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
@@ -375,6 +397,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
+                add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
@@ -385,6 +408,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
+                add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 266a5d4b50..5162617b18 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -31,6 +31,10 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp)
    endif()
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000..53fc74a392
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
+        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<1, 1, 4, 2>,       S<4, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,      S<16, 1, 2, 2>,       S<1, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,        S<1, 1, 2, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 8, 2>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<2, 1, 4, 2>,      S<4, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<2, 1, 4, 2>,      S<8, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,      S<4, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<1, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000..35d46b32f5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
+        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<1, 1, 4, 2>,       S<4, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 1, 2>,      S<1, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,      S<16, 1, 2, 2>,      S<1, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 8, 2>,      S<4, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<2, 1, 4, 2>,      S<4, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<2, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<2, 1, 4, 2>,      S<8, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,      S<4, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<4, 1,  64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..22e2e5f398
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,60 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
+        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<4, 1, 1, 2>,       S<1, 1,  8, 1>,  S<1, 2, 0 ,3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<16, 1, 2, 2>,       S<1, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,        S<1, 1, 2, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 8, 2>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<4, 1, 2, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<1, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<4, 1, 2, 2>,       S<4, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..49e9416c81
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
+        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
+        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
+        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<4, 1, 1, 2>,       S<1, 1,  8, 1>,  S<1, 2, 0 ,3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 1, 2>,      S<1, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<16, 1, 2, 2>,      S<1, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 8, 2>,      S<4, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<2, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<4, 1, 2, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
+        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<4, 1, 2, 2>,       S<4, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<4, 1,  64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
+    // clang-format on
+    >;
+
+void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 32fe996da0aefc5274493b857a91f217f7fe3561 Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Thu, 17 Aug 2023 17:54:11 +0200
Subject: [PATCH 093/194] Fix datatype in inner_product when V_DOT2 is disabled
 (#849)

---
 include/ck/utility/inner_product.hpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index 0087ae0f7a..b58b2b3319 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -94,8 +94,8 @@ __device__ void inner_product<half2_t, half2_t, float>(const half2_t& a, const h
     const vector_type<half_t, 2> b_vector{b};
 
     static_for<0, 2, 1>{}([&](auto i) {
-        c += type_convert<int32_t>(a_vector.AsType<half_t>()[i]) *
-             type_convert<int32_t>(b_vector.AsType<half_t>()[i]);
+        c += type_convert<float>(a_vector.AsType<half_t>()[i]) *
+             type_convert<float>(b_vector.AsType<half_t>()[i]);
     });
 #endif
 }

From d52ec01652b7d620386251db92455968d8d90bdc Mon Sep 17 00:00:00 2001
From: cloudhan <cloudhan@outlook.com>
Date: Fri, 18 Aug 2023 11:14:59 +0800
Subject: [PATCH 094/194] Use asynchronous version of hipMemset (#850)

---
 ...conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp | 5 +++--
 .../gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp   | 6 ++++--
 .../gpu/device/impl/device_gemm_xdl_streamk.hpp            | 7 +++++--
 .../impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp       | 6 ++++--
 .../device_splitk_contraction_multiple_d_xdl_cshuffle.hpp  | 5 +++--
 5 files changed, 19 insertions(+), 10 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
index 1e50e1ae89..23440e24f6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_conv2d_backward_weight_xdl_c_shuffle_nhwc_kyxc_nhwk.hpp
@@ -532,11 +532,12 @@ struct DeviceConv2dBwdWeightXdl_C_Shuffle_Input_N_Hi_Wi_C_Weight_K_Y_X_C_Output_
             float ave_time = 0;
 
             const auto Run = [&](const auto& kernel) {
-                hipGetErrorString(hipMemset(
+                hipGetErrorString(hipMemsetAsync(
                     arg.p_c_grid_,
                     0,
                     arg.c_grid_desc_mblock_mperblock_nblock_nperblock_.GetElementSpaceSize() *
-                        sizeof(CDataType)));
+                        sizeof(CDataType),
+                    stream_config.stream_id_));
 
                 ave_time =
                     launch_and_time_kernel(stream_config,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 2b8f760737..505e1846d1 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -158,8 +158,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
 
             const auto Run = [&](const auto& kernel) {
                 if(kbatch > 1)
-                    hipGetErrorString(
-                        hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(CDataType)));
+                    hipGetErrorString(hipMemsetAsync(karg.p_c_grid,
+                                                     0,
+                                                     karg.M * karg.N * sizeof(CDataType),
+                                                     stream_config.stream_id_));
 
                 ave_time = launch_and_time_kernel(
                     stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg, b2c_map);
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp
index 6fe5784c28..8de42ba9ef 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_streamk.hpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
 
@@ -147,7 +147,10 @@ struct DeviceGemmXdlStreamK : public DeviceGemmStreamK<ALayout,
             if constexpr(GridwiseGemm::Block2CTileMap::ReductionStrategy ==
                          StreamKReductionStrategy::Atomic)
             {
-                hipGetErrorString(hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(CDataType)));
+                hipGetErrorString(hipMemsetAsync(karg.p_c_grid,
+                                                 0,
+                                                 karg.M * karg.N * sizeof(CDataType),
+                                                 stream_config.stream_id_));
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
                                                   grid_dims,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 4f4413b78f..ba66d8358c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -421,8 +421,10 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                     for(const auto& trans_arg : arg.gemm_kernel_args_)
                     {
                         const auto& karg = trans_arg.karg_;
-                        hip_check_error(
-                            hipMemset(karg.p_c_grid, 0, karg.M * karg.N * sizeof(EDataType)));
+                        hip_check_error(hipMemsetAsync(karg.p_c_grid,
+                                                       0,
+                                                       karg.M * karg.N * sizeof(EDataType),
+                                                       stream_config.stream_id_));
                     }
                 }
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index 7ba2d96a9c..a52968bcef 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -886,11 +886,12 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
                     typename GridwiseGemmAtomicAdd::DefaultBlock2ETileMap,
                     has_main_loop>;
 
-                hipGetErrorString(hipMemset(
+                hipGetErrorString(hipMemsetAsync(
                     arg.p_e_grid_,
                     0,
                     arg.e_grid_desc_mblock_mperblock_nblock_nperblock_.GetElementSpaceSize() *
-                        sizeof(EDataType)));
+                        sizeof(EDataType),
+                    stream_config.stream_id_));
 
                 return launch_and_time_kernel(stream_config,
                                               kernel,

From eac50708d9893645ae6e0a590bb4706debe0a6b7 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Tue, 22 Aug 2023 09:34:49 -0500
Subject: [PATCH 095/194] Add instances/ckProfiler/client example for fp8/fp16
 mixed precision Gemm (#853)

* Add ComputeType arg to splitk device and gridwise ops

* Update for gridwise op compatibility

* Update bf16 and int8 splitk gemm examples with ComputeType

* Add instances

* Update ckProfiler for mixed precision cases

* Add a mixed precision splitK gemm client example

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
---
 client_example/20_splitk_gemm/CMakeLists.txt  |   2 +
 .../20_splitk_gemm/splitK_gemm_fp16_f8.cpp    | 225 ++++++++++++++++++
 .../35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp  |  11 +-
 .../35_splitK_gemm/splitK_gemm_xdl_int8.cpp   |  11 +-
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |   9 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp |   3 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  34 +--
 .../device_operation_instance_factory.hpp     |   1 +
 .../gpu/gemm_splitk.hpp                       |  88 +++++++
 .../gpu/gemm_splitk/CMakeLists.txt            |  38 ++-
 ...dl_splitk_f16_f8_f16_km_kn_mn_instance.cpp |  62 +++++
 ...dl_splitk_f16_f8_f16_km_nk_mn_instance.cpp |  62 +++++
 ...dl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp |  71 ++++++
 ...dl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp |  67 ++++++
 ...dl_splitk_f8_f16_f16_km_kn_mn_instance.cpp |  62 +++++
 ...dl_splitk_f8_f16_f16_km_nk_mn_instance.cpp |  62 +++++
 ...dl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp |  71 ++++++
 ...dl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp |  67 ++++++
 .../profiler/profile_gemm_splitk_impl.hpp     |  15 +-
 profiler/src/profile_gemm_splitk.cpp          |  37 ++-
 20 files changed, 956 insertions(+), 42 deletions(-)
 create mode 100644 client_example/20_splitk_gemm/CMakeLists.txt
 create mode 100644 client_example/20_splitk_gemm/splitK_gemm_fp16_f8.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp

diff --git a/client_example/20_splitk_gemm/CMakeLists.txt b/client_example/20_splitk_gemm/CMakeLists.txt
new file mode 100644
index 0000000000..a60bada473
--- /dev/null
+++ b/client_example/20_splitk_gemm/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_executable(client_splitK_gemm splitK_gemm_fp16_f8.cpp)
+target_link_libraries(client_splitK_gemm PRIVATE composable_kernel::device_operations)
diff --git a/client_example/20_splitk_gemm/splitK_gemm_fp16_f8.cpp b/client_example/20_splitk_gemm/splitK_gemm_fp16_f8.cpp
new file mode 100644
index 0000000000..94a57cd029
--- /dev/null
+++ b/client_example/20_splitk_gemm/splitK_gemm_fp16_f8.cpp
@@ -0,0 +1,225 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_splitk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp"
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+using ADataType = F8;
+using BDataType = F16;
+using CDataType = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideC = 4096;
+
+    ck::index_t KBatch = 1;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 8)
+    {
+        M = std::stoi(argv[1]);
+        N = std::stoi(argv[2]);
+        K = std::stoi(argv[3]);
+
+        StrideA = std::stoi(argv[4]);
+        StrideB = std::stoi(argv[5]);
+        StrideC = std::stoi(argv[6]);
+
+        KBatch = std::stoi(argv[7]);
+    }
+    else
+    {
+        printf("arg1 to 7: M, N, K, StrideA, StrideB, StrideC, KBatch\n");
+        exit(0);
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    SimpleDeviceMem a_device_buf(sizeof(ADataType) * f_matrix_space_size(M, K, StrideA, ALayout{}));
+    SimpleDeviceMem b_device_buf(sizeof(BDataType) * f_matrix_space_size(K, N, StrideB, BLayout{}));
+    SimpleDeviceMem c_device_buf(sizeof(CDataType) * f_matrix_space_size(M, N, StrideC, CLayout{}));
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<
+        ALayout,
+        BLayout,
+        CLayout,
+        ADataType,
+        BDataType,
+        CDataType,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough,
+        ck::tensor_operation::element_wise::PassThrough>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op = AElementOp{};
+    const auto b_element_op = BElementOp{};
+    const auto c_element_op = CElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                        b_device_buf.GetDeviceBuffer(),
+                                                        c_device_buf.GetDeviceBuffer(),
+                                                        M,
+                                                        N,
+                                                        K,
+                                                        StrideA,
+                                                        StrideB,
+                                                        StrideC,
+                                                        a_element_op,
+                                                        b_element_op,
+                                                        c_element_op,
+                                                        KBatch);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+
+            std::size_t num_btype =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(a_device_buf.GetDeviceBuffer(),
+                                                        b_device_buf.GetDeviceBuffer(),
+                                                        c_device_buf.GetDeviceBuffer(),
+                                                        M,
+                                                        N,
+                                                        K,
+                                                        StrideA,
+                                                        StrideB,
+                                                        StrideC,
+                                                        a_element_op,
+                                                        b_element_op,
+                                                        c_element_op,
+                                                        KBatch);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
index 1dc21a6c23..fdf49a31b7 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
@@ -33,6 +33,7 @@ using ADataType   = BF16;
 using BDataType   = BF16;
 using AccDataType = F32;
 using CDataType   = F32;
+using ComputeType = BF16;
 
 using ALayout = Row;
 using BLayout = Col;
@@ -46,11 +47,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 
 using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
     // clang-format off
-//######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer|  K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|
-//######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise| Spacialization|  Size| Block| Block| Block|    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
-//######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
-//######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
-        < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   256,   128,     4,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               4>;
+//######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer|  K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|      Compute|
+//######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise| Spacialization|  Size| Block| Block| Block|    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|         Type|
+//######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|             |
+//######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |             | 
+        < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   256,   128,     4,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               4, ComputeType>;
 // clang-format on
 
 #include "run_splitK_gemm_example.inc"
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
index 0fc7a5cc29..6b0c1aa02d 100644
--- a/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
+++ b/example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp
@@ -30,6 +30,7 @@ using ADataType   = int8_t;
 using BDataType   = int8_t;
 using AccDataType = int32_t;
 using CDataType   = int32_t;
+using ComputeType = int8_t;
 
 using ALayout = Row;
 using BLayout = Col;
@@ -43,11 +44,11 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 
 using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmXdlSplitKCShuffle
     // clang-format off
-//######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer|  K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|
-//######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise| Spacialization|  Size| Block| Block| Block|    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
-//######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
-//######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
-        < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   256,   128,     4,  16,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,             16,             16,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,             16,             16,      true,           1,           1,                   S<1, 32, 1, 8>,               4>;
+//######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer|  K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|     Compute|
+//######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise| Spacialization|  Size| Block| Block| Block|    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|        Type|
+//######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|            |
+//######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |            |
+        < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   256,   128,     4,  16,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,             16,             16,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,             16,             16,      true,           1,           1,                   S<1, 32, 1, 8>,               4, ComputeType>;
 // clang-format on
 
 #include "run_splitK_gemm_example.inc"
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 505e1846d1..3b0e937e22 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -57,7 +57,8 @@ template <typename ADataType,
           index_t CShuffleMRepeatPerShuffle,
           index_t CShuffleNRepeatPerShuffle,
           typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-          index_t CBlockTransferScalarPerVector_NWaveNPerXDL>
+          index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
+          typename ComputeType = CDataType>
 struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                              BLayout,
                                                              CLayout,
@@ -80,7 +81,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
 
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
         AccDataType,
         CDataType,
         ALayout,
@@ -120,7 +122,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         CBlockTransferScalarPerVector_NWaveNPerXDL,
         CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
         LoopSched,
-        PipelineVer>;
+        PipelineVer,
+        ComputeType>;
 
     using Argument              = typename GridwiseGemm::Argument;
     using DefaultBlock2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index ba66d8358c..d9950c79ac 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -142,7 +142,8 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
         AccDataType,
         EDataType,
         ALayout,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 4cd70cc902..58f76e5368 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -45,7 +45,8 @@ __global__ void
 }
 
 template <index_t BlockSize,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatAcc,
           typename FloatC,
           typename ALayout,
@@ -85,7 +86,8 @@ template <index_t BlockSize,
           index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
           typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           LoopScheduler LoopSched     = make_default_loop_scheduler(),
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          PipelineVersion PipelineVer = PipelineVersion::v1,
+          typename ComputeType        = FloatC>
 struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 {
     static constexpr auto I0 = Number<0>{};
@@ -113,8 +115,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
     struct Argument : public ck::tensor_operation::device::BaseArgument
     {
-        const FloatAB* p_a_grid;
-        const FloatAB* p_b_grid;
+        const FloatA* p_a_grid;
+        const FloatB* p_b_grid;
         FloatC* p_c_grid;
         index_t M;
         index_t N;
@@ -128,8 +130,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         index_t K0;
         index_t k_batch;
 
-        Argument(const FloatAB* p_a_grid_,
-                 const FloatAB* p_b_grid_,
+        Argument(const FloatA* p_a_grid_,
+                 const FloatB* p_b_grid_,
                  FloatC* p_c_grid_,
                  index_t M_,
                  index_t N_,
@@ -365,7 +367,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         constexpr auto c_block_size =
             GetCBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock().GetElementSpaceSize();
 
-        return math::max((a_block_space_size + b_block_space_size) * sizeof(FloatAB),
+        return math::max((a_block_space_size + b_block_space_size) * sizeof(ComputeType),
                          c_block_size * sizeof(FloatC));
     }
 
@@ -577,8 +579,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                                void* __restrict__ p_shared_block,
                                const Block2CTileMap& block_2_ctile_map)
     {
-        const FloatAB* p_a_grid          = karg.p_a_grid;
-        const FloatAB* p_b_grid          = karg.p_b_grid;
+        const FloatA* p_a_grid           = karg.p_a_grid;
+        const FloatB* p_b_grid           = karg.p_b_grid;
         FloatC* p_c_grid                 = karg.p_c_grid;
         const auto a_b_k0_m_k1_grid_desc = MakeAGridDescriptor_KBatch_K0_M_K1(
             karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
@@ -698,8 +700,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                                                 Sequence<1, K0PerBlock, MPerBlock, K1>,
                                                 ABlockTransferThreadClusterLengths_K0_M_K1,
                                                 ABlockTransferThreadClusterArrangeOrder,
-                                                FloatAB,
-                                                FloatAB,
+                                                FloatA,
+                                                ComputeType,
                                                 decltype(a_b_k0_m_k1_grid_desc),
                                                 decltype(a_b_k0_m_k1_block_desc),
                                                 ABlockTransferSrcAccessOrder,
@@ -728,8 +730,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                                                 Sequence<1, K0PerBlock, NPerBlock, K1>,
                                                 BBlockTransferThreadClusterLengths_K0_N_K1,
                                                 BBlockTransferThreadClusterArrangeOrder,
-                                                FloatAB,
-                                                FloatAB,
+                                                FloatB,
+                                                ComputeType,
                                                 decltype(b_b_k0_n_k1_grid_desc),
                                                 decltype(b_b_k0_n_k1_block_desc),
                                                 BBlockTransferSrcAccessOrder,
@@ -759,7 +761,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            FloatAB,
+            ComputeType,
             FloatAcc,
             decltype(a_k0_m_k1_block_desc),
             decltype(b_k0_n_k1_block_desc),
@@ -776,8 +778,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         constexpr auto a_block_space_size =
             math::integer_least_multiple(a_k0_m_k1_block_desc.GetElementSpaceSize(), max_lds_align);
 
-        FloatAB* p_a_block = static_cast<FloatAB*>(p_shared_block);
-        FloatAB* p_b_block = static_cast<FloatAB*>(p_shared_block) + a_block_space_size;
+        ComputeType* p_a_block = static_cast<ComputeType*>(p_shared_block);
+        ComputeType* p_b_block = static_cast<ComputeType*>(p_shared_block) + a_block_space_size;
 
         constexpr auto a_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
         constexpr auto b_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index 851d9e497f..67c7d659e2 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -17,6 +17,7 @@ namespace instance {
 using F64  = double;
 using F32  = float;
 using F16  = ck::half_t;
+using F8   = ck::f8_t;
 using BF16 = ck::bhalf_t;
 using I8   = int8_t;
 using I32  = int32_t;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index 5348751511..dbd2b8f656 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -57,6 +57,46 @@ void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemmSplitK<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 template <typename ADataType,
           typename BDataType,
           typename CDataType,
@@ -136,6 +176,54 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+        else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
+                          is_same_v<CDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instances(op_ptrs);
+            }
+        }
+        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
+                          is_same_v<CDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances(op_ptrs);
+            }
+        }
 
         return op_ptrs;
     }
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
index 6b33622746..89dfa8f2ed 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
@@ -1,10 +1,28 @@
-add_instance_library(device_gemm_splitk_instance
-   device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
-   device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
-   device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
-   device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
-   device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
-   device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
-   device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
-)
+set(GEMM_SPLITK_INSTANCES)
+
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp)
+endif()
+
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp)
+endif()
+
+if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp)
+  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp)
+endif()
+
+add_instance_library(device_gemm_splitk_instance ${GEMM_SPLITK_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000..86a86c1035
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,     F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F8,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000..c423b493eb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..c0b6ec67bb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..fa38640ae2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000..face3f456d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000..727bd221a7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..6f378fd281
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..70f4bdc0ec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<   F8,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Col, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index 6ffa31678f..47a02b41ce 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -191,7 +191,20 @@ bool profile_gemm_splitk_impl(int do_verification,
             {
                 c_device_buf.FromDevice(c_m_n_device_result.mData.data());
 
-                pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+                // set softer tolerances for fp8
+                if constexpr(is_same_v<ADataType, f8_t> || is_same_v<BDataType, f8_t> ||
+                             is_same_v<CDataType, f8_t>)
+                {
+                    std::string msg = "Error: Incorrect results!";
+                    double rtol     = 1e-1;
+                    double atol     = 1e-1;
+                    pass            = pass & ck::utils::check_err(
+                                      c_m_n_device_result, c_m_n_host_result, msg, rtol, atol);
+                }
+                else
+                {
+                    pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+                }
 
                 if(do_log)
                 {
diff --git a/profiler/src/profile_gemm_splitk.cpp b/profiler/src/profile_gemm_splitk.cpp
index cc2da73cb7..617e0b9cd4 100644
--- a/profiler/src/profile_gemm_splitk.cpp
+++ b/profiler/src/profile_gemm_splitk.cpp
@@ -23,6 +23,8 @@ enum struct GemmDataType
     F16_F16_F16,    // 1
     BF16_BF16_BF16, // 2
     INT8_INT8_INT8, // 3
+    F8_F16_F16,     // 4
+    F16_F8_F16,     // 5
 };
 
 #define OP_NAME "gemm_splitk"
@@ -33,7 +35,7 @@ int profile_gemm_splitk(int argc, char* argv[])
     if(argc != 15)
     {
         printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
-        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n");
+        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8; 4: f8@f16; 5: f16@f8)\n");
         printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
         printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
         printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
@@ -65,6 +67,7 @@ int profile_gemm_splitk(int argc, char* argv[])
 
     using F32 = float;
     using F16 = ck::half_t;
+    using F8  = ck::f8_t;
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -143,6 +146,38 @@ int profile_gemm_splitk(int argc, char* argv[])
     {
         return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{});
     }
+    else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F8{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F8{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F16{}, F8{}, F32{}, F16{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F16{}, F8{}, F32{}, F16{}, Row{}, Col{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Col{}, Row{});
+    }
     else
     {
         std::cout << "this data_type & layout is not implemented" << std::endl;

From 595d23be14d3a23a0df1aae8350497f47ba04137 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 22 Aug 2023 18:25:41 +0200
Subject: [PATCH 096/194] Fix transform and instances for grouped conv bwd data
 (#848)

* Fix transform and instances for grouped conv bwd data

* Add instances for small K and small C

* Remove workaround after fix

* Fix interface tests
---
 include/ck/ck.hpp                             |   3 -
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |  13 +-
 .../threadwise_tensor_slice_transfer_v3r1.hpp |   3 +
 .../transform_conv_bwd_data_to_gemm_v1.hpp    | 200 ++++++++++--------
 ...ice_grouped_conv_bwd_data_xdl_instance.hpp | 129 +++++------
 .../test_contraction_interface.cpp            |   2 +-
 .../test_grouped_convnd_bwd_data.cpp          |   9 +
 test/grouped_gemm/test_grouped_gemm_util.hpp  |   4 +-
 8 files changed, 203 insertions(+), 160 deletions(-)

diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 0e7e690421..734fe549db 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -200,9 +200,6 @@
 // workaround: compiler issue on gfx908
 #define CK_WORKAROUND_SWDEV_388832 1
 
-// workaround: Grouped Conv2d_bwd_data fails for already implemented instance
-#define CK_WORKAROUND_GITHUB_ISSUE_824 1
-
 // flag to enable (1) or disable (0) the debugging output in some kernels
 #define DEBUG_LOG 0
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index ec1cc53991..5437379018 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -266,12 +266,13 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                           (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
                       "Invalid tuning param!");
 
-        const auto M = a_grid_desc_m_k.GetLength(I0);
-        const auto N = b_grid_desc_n_k.GetLength(I0);
-        const auto K = a_grid_desc_m_k.GetLength(I1);
+        const auto M  = a_grid_desc_m_k.GetLength(I0);
+        const auto N  = b_grid_desc_n_k.GetLength(I0);
+        const auto AK = a_grid_desc_m_k.GetLength(I1);
+        const auto BK = b_grid_desc_n_k.GetLength(I1);
 
         // check consistency of desc
-        if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1)))
+        if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1) && AK == BK))
         {
             return false;
         }
@@ -289,13 +290,13 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         }
 
         // check tile size
-        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && AK % KPerBlock == 0))
         {
             return false;
         }
 
         // check gridwise gemm pipeline
-        const auto num_k_loop = K / KPerBlock;
+        const auto num_k_loop = AK / KPerBlock;
 
         if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
index 6665d765f8..32ea8ae39c 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
@@ -129,6 +129,9 @@ struct ThreadwiseTensorSliceTransfer_v3r1
 
         constexpr auto src_access_lengths = SliceLengths{} / src_scalar_per_access;
 
+        static_assert(SliceLengths::At(SrcVectorDim) % SrcScalarPerVector == 0,
+                      "SliceLengths[SrcVectorDim] must be divisible by SrcScalarPerVector");
+
         constexpr auto src_dim_access_order = SrcDimAccessOrder{};
 
         constexpr auto ordered_src_access_lengths =
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
index 98a46ef450..b1b203b43f 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
@@ -236,8 +236,6 @@ struct TransformConvBwdDataToGemm_v1
         const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
         const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
 
-        const index_t AK0 = K / AK1;
-
         // n_do_ho_wo_k for 3d or n_ho_wo_k for 2d
         const auto out_grid_desc =
             make_out_grid_desc<NDimSpatial, ALayout, ConvBwdDataSpecialization>(
@@ -247,6 +245,8 @@ struct TransformConvBwdDataToGemm_v1
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
                          Filter1x1Stride1Pad0)
         {
+            const index_t AK0 = math::integer_divide_ceil(K, AK1);
+
             // A: output tensor
             const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
                 out_grid_desc,
@@ -308,6 +308,9 @@ struct TransformConvBwdDataToGemm_v1
             const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
             const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
+            const index_t AK0 =
+                math::integer_divide_ceil(ZDotSlice * YDotSlice * XDotSlice * K, AK1);
+
             if constexpr(NDimSpatial == 2)
             {
                 // A: output tensor
@@ -332,7 +335,7 @@ struct TransformConvBwdDataToGemm_v1
                     make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
                     make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
 
-                const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc =
+                const auto out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k_grid_desc =
                     transform_tensor_descriptor(
                         out_n_ydot_htilde_xdot_wtilde_k_grid_desc,
                         make_tuple(make_pass_through_transform(N),
@@ -340,7 +343,7 @@ struct TransformConvBwdDataToGemm_v1
                                    make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
                                    make_slice_transform(XDot, I0, XDotSlice),
                                    make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                                   make_unmerge_transform(make_tuple(AK0, AK1))),
+                                   make_pass_through_transform(K)),
                         make_tuple(Sequence<0>{},
                                    Sequence<1>{},
                                    Sequence<2>{},
@@ -352,21 +355,28 @@ struct TransformConvBwdDataToGemm_v1
                                    Sequence<2>{},
                                    Sequence<3>{},
                                    Sequence<4>{},
-                                   Sequence<5, 6>{}));
+                                   Sequence<5>{}));
 
-                const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
-                    out_n_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc,
-                    make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, AK0)),
-                               make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice)),
-                               make_pass_through_transform(AK1)),
-                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}, Sequence<6>{}),
-                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+                const auto out_gemmk_gemmmraw_grid_desc = transform_tensor_descriptor(
+                    out_n_ydotslice_htildeslice_xdotslice_wtildeslice_k_grid_desc,
+                    make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K)),
+                               make_merge_transform(make_tuple(N, HTildeSlice, WTildeSlice))),
+                    make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-                const auto out_gemmak0_gemmm_gemmak1_grid_desc =
+                const auto out_gemmk_gemmm_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
-                        out_gemmak0_gemmmraw_gemmak1_grid_desc,
-                        make_tuple(AK0, GemmMPerBlock, AK1),
-                        Sequence<false, DoPadGemmM, false>{});
+                        out_gemmk_gemmmraw_grid_desc,
+                        make_tuple(AK1, GemmMPerBlock),
+                        Sequence<true, DoPadGemmM>{});
+
+                const auto out_gemmak0_gemmm_gemmak1_grid_desc = transform_tensor_descriptor(
+                    out_gemmk_gemmm_padded_grid_desc,
+                    make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
+                               make_pass_through_transform(
+                                   out_gemmk_gemmm_padded_grid_desc.GetLength(I1))),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
 
                 return out_gemmak0_gemmm_gemmak1_grid_desc;
             }
@@ -411,7 +421,7 @@ struct TransformConvBwdDataToGemm_v1
                                    Sequence<7>{}));
 
                 const auto
-                    out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc =
+                    out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_k_grid_desc =
                         transform_tensor_descriptor(
                             out_n_zdot_dtilde_ydot_htilde_xdot_wtilde_k_grid_desc,
                             make_tuple(make_pass_through_transform(N),
@@ -421,7 +431,7 @@ struct TransformConvBwdDataToGemm_v1
                                        make_slice_transform(HTilde, IHTildeSliceBegin, HTildeSlice),
                                        make_slice_transform(XDot, I0, XDotSlice),
                                        make_slice_transform(WTilde, IWTildeSliceBegin, WTildeSlice),
-                                       make_unmerge_transform(make_tuple(AK0, AK1))),
+                                       make_pass_through_transform(K)),
                             make_tuple(Sequence<0>{},
                                        Sequence<1>{},
                                        Sequence<2>{},
@@ -437,22 +447,29 @@ struct TransformConvBwdDataToGemm_v1
                                        Sequence<4>{},
                                        Sequence<5>{},
                                        Sequence<6>{},
-                                       Sequence<7, 8>{}));
+                                       Sequence<7>{}));
 
-                const auto out_gemmak0_gemmmraw_gemmak1_grid_desc = transform_tensor_descriptor(
-                    out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_ak0_ak1_grid_desc,
+                const auto out_gemmk_gemmmraw_grid_desc = transform_tensor_descriptor(
+                    out_n_zdotslice_dtildeslice_ydotslice_htildeslice_xdotslice_wtildeslice_k_grid_desc,
                     make_tuple(
-                        make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice, AK0)),
-                        make_merge_transform(make_tuple(N, DTildeSlice, HTildeSlice, WTildeSlice)),
-                        make_pass_through_transform(AK1)),
-                    make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}, Sequence<8>{}),
-                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+                        make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice, K)),
+                        make_merge_transform(make_tuple(N, DTildeSlice, HTildeSlice, WTildeSlice))),
+                    make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-                const auto out_gemmak0_gemmm_gemmak1_grid_desc =
+                const auto out_gemmk_gemmm_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
-                        out_gemmak0_gemmmraw_gemmak1_grid_desc,
-                        make_tuple(AK0, GemmMPerBlock, AK1),
-                        Sequence<false, DoPadGemmM, false>{});
+                        out_gemmk_gemmmraw_grid_desc,
+                        make_tuple(AK1, GemmMPerBlock),
+                        Sequence<true, DoPadGemmM>{});
+
+                const auto out_gemmak0_gemmm_gemmak1_grid_desc = transform_tensor_descriptor(
+                    out_gemmk_gemmm_padded_grid_desc,
+                    make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
+                               make_pass_through_transform(
+                                   out_gemmk_gemmm_padded_grid_desc.GetLength(I1))),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
 
                 return out_gemmak0_gemmm_gemmak1_grid_desc;
             }
@@ -505,8 +522,6 @@ struct TransformConvBwdDataToGemm_v1
         const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
         const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
 
-        const index_t BK0 = K / BK1;
-
         // assume packed
         // k_y_x_c for 2d or k_z_y_x_c for 3d
         const auto wei_grid_desc = make_wei_grid_desc<BLayout>(K, Z, Y, X, C);
@@ -515,6 +530,8 @@ struct TransformConvBwdDataToGemm_v1
                      ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::
                          Filter1x1Stride1Pad0)
         {
+            const index_t BK0 = math::integer_divide_ceil(K, BK1);
+
             // B: weight tensor
             const auto wei_gemmbk0_gemmnraw_gemmbk1_grid_desc =
                 transform_tensor_descriptor(make_naive_tensor_descriptor_packed(make_tuple(K, C)),
@@ -551,6 +568,9 @@ struct TransformConvBwdDataToGemm_v1
             const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
             const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
+            const index_t BK0 =
+                math::integer_divide_ceil(ZDotSlice * YDotSlice * XDotSlice * K, BK1);
+
             // B weight tensor
             if constexpr(NDimSpatial == 2)
             {
@@ -566,43 +586,47 @@ struct TransformConvBwdDataToGemm_v1
                     make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
                     make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
 
-                const auto wei_bk0_bk1_ydotslice_xdotslice_c_grid_desc =
-                    transform_tensor_descriptor(
-                        wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                        make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
-                                   make_slice_transform(YDot, I0, YDotSlice),
-                                   make_slice_transform(XDot, I0, XDotSlice),
-                                   make_freeze_transform(i_ytilde),
-                                   make_freeze_transform(i_xtilde),
-                                   make_pass_through_transform(C)),
-                        make_tuple(Sequence<0>{},
-                                   Sequence<1>{},
-                                   Sequence<3>{},
-                                   Sequence<2>{},
-                                   Sequence<4>{},
-                                   Sequence<5>{}),
-                        make_tuple(Sequence<0, 1>{},
-                                   Sequence<2>{},
-                                   Sequence<3>{},
-                                   Sequence<>{},
-                                   Sequence<>{},
-                                   Sequence<4>{}));
+                const auto wei_k_ydotslice_xdotslice_c_grid_desc = transform_tensor_descriptor(
+                    wei_k_ydot_ytilde_xdot_xtilde_c_grid_desc,
+                    make_tuple(make_pass_through_transform(K),
+                               make_slice_transform(YDot, I0, YDotSlice),
+                               make_slice_transform(XDot, I0, XDotSlice),
+                               make_freeze_transform(i_ytilde),
+                               make_freeze_transform(i_xtilde),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<0>{},
+                               Sequence<1>{},
+                               Sequence<3>{},
+                               Sequence<2>{},
+                               Sequence<4>{},
+                               Sequence<5>{}),
+                    make_tuple(Sequence<0>{},
+                               Sequence<1>{},
+                               Sequence<2>{},
+                               Sequence<>{},
+                               Sequence<>{},
+                               Sequence<3>{}));
 
-                const auto wei_gemmbk0_gemmnraw_gemmbk1_grid_desc = transform_tensor_descriptor(
-                    wei_bk0_bk1_ydotslice_xdotslice_c_grid_desc,
-                    make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, BK0)),
-                               make_pass_through_transform(C),
-                               make_pass_through_transform(BK1)),
-                    make_tuple(Sequence<2, 3, 0>{}, Sequence<4>{}, Sequence<1>{}),
-                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+                const auto wei_gemmk_gemmnraw_grid_desc = transform_tensor_descriptor(
+                    wei_k_ydotslice_xdotslice_c_grid_desc,
+                    make_tuple(make_merge_transform(make_tuple(YDotSlice, XDotSlice, K)),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<1, 2, 0>{}, Sequence<3>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-                const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc =
+                const auto wei_gemmk_gemmn_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
-                        wei_gemmbk0_gemmnraw_gemmbk1_grid_desc,
-                        make_tuple(wei_gemmbk0_gemmnraw_gemmbk1_grid_desc.GetLength(I0),
-                                   GemmNPerBlock,
-                                   BK1),
-                        Sequence<false, DoPadGemmN, false>{});
+                        wei_gemmk_gemmnraw_grid_desc,
+                        make_tuple(BK1, GemmNPerBlock),
+                        Sequence<true, DoPadGemmN>{});
+
+                const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc = transform_tensor_descriptor(
+                    wei_gemmk_gemmn_padded_grid_desc,
+                    make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                               make_pass_through_transform(
+                                   wei_gemmk_gemmn_padded_grid_desc.GetLength(I1))),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
 
                 return wei_gemmbk0_gemmn_gemmbk1_grid_desc;
             }
@@ -631,10 +655,10 @@ struct TransformConvBwdDataToGemm_v1
                                    Sequence<5, 6>{},
                                    Sequence<7>{}));
 
-                const auto wei_bk0_bk1_zdotslice_ydotslice_xdotslice_c_grid_desc =
+                const auto wei_gemmk_zdotslice_ydotslice_xdotslice_c_grid_desc =
                     transform_tensor_descriptor(
                         wei_k_zdot_ztilde_ydot_ytilde_xdot_xtilde_c_grid_desc,
-                        make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                        make_tuple(make_pass_through_transform(K),
                                    make_slice_transform(ZDot, I0, ZDotSlice),
                                    make_slice_transform(YDot, I0, YDotSlice),
                                    make_slice_transform(XDot, I0, XDotSlice),
@@ -650,33 +674,37 @@ struct TransformConvBwdDataToGemm_v1
                                    Sequence<4>{},
                                    Sequence<6>{},
                                    Sequence<7>{}),
-                        make_tuple(Sequence<0, 1>{},
+                        make_tuple(Sequence<0>{},
+                                   Sequence<1>{},
                                    Sequence<2>{},
                                    Sequence<3>{},
-                                   Sequence<4>{},
                                    Sequence<>{},
                                    Sequence<>{},
                                    Sequence<>{},
-                                   Sequence<5>{}));
+                                   Sequence<4>{}));
 
-                const auto wei_gemmbk0_gemmnraw_gemmbk1_grid_desc = transform_tensor_descriptor(
-                    wei_bk0_bk1_zdotslice_ydotslice_xdotslice_c_grid_desc,
-                    make_tuple(
-                        make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice, BK0)),
-                        make_pass_through_transform(C),
-                        make_pass_through_transform(BK1)),
-                    make_tuple(Sequence<2, 3, 4, 0>{}, Sequence<5>{}, Sequence<1>{}),
-                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+                const auto wei_gemmk_gemmnraw_grid_desc = transform_tensor_descriptor(
+                    wei_gemmk_zdotslice_ydotslice_xdotslice_c_grid_desc,
+                    make_tuple(make_merge_transform(make_tuple(ZDotSlice, YDotSlice, XDotSlice, K)),
+                               make_pass_through_transform(C)),
+                    make_tuple(Sequence<1, 2, 3, 0>{}, Sequence<4>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-                const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc =
+                const auto wei_gemmk_gemm_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
-                        wei_gemmbk0_gemmnraw_gemmbk1_grid_desc,
-                        make_tuple(wei_gemmbk0_gemmnraw_gemmbk1_grid_desc.GetLength(I0),
-                                   GemmNPerBlock,
-                                   BK1),
-                        Sequence<false, DoPadGemmN, false>{});
+                        wei_gemmk_gemmnraw_grid_desc,
+                        make_tuple(BK1, GemmNPerBlock),
+                        Sequence<true, DoPadGemmN>{});
 
-                return wei_gemmbk0_gemmn_gemmbk1_grid_desc;
+                const auto wei_gemmbk0_gemm_gemmbk1_grid_desc = transform_tensor_descriptor(
+                    wei_gemmk_gemm_padded_grid_desc,
+                    make_tuple(
+                        make_unmerge_transform(make_tuple(BK0, BK1)),
+                        make_pass_through_transform(wei_gemmk_gemm_padded_grid_desc.GetLength(I1))),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}),
+                    make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+                return wei_gemmbk0_gemm_gemmbk1_grid_desc;
             }
             else
             {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
index d05e8e3a87..26acf4f5f7 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -47,24 +47,26 @@ using device_grouped_conv_bwd_data_xdl_f16_instances =
         // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
-        
-#ifndef CK_WORKAROUND_GITHUB_ISSUE_824
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial,   GNHWK,   GKYXC, Empty_Tuple,   GNHWC,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,  ConvBwdDataDefault,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        // TODO: After enable, add instance for small conv.K and conv.C
-#endif
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 16, 1, 4>,                1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 32, 1, 4>,                1>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F16, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
         // clang-format on
         >;
 
@@ -75,33 +77,34 @@ template <index_t NDimSpatial,
           typename DsLayout,
           typename ELayout,
           ConvolutionBackwardDataSpecialization ConvSpec>
-using device_grouped_conv_bwd_data_xdl_bf16_instances =
-    std::tuple<
-        // clang-format off
-        // ##############################################|       NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
-        // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
-        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
-        // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   BF16,   BF16,     F32,      BF16, Empty_Tuple,   BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>
-        
-#ifndef CK_WORKAROUND_GITHUB_ISSUE_824
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-            DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
-            // TODO: After enable, add instance for small conv.K and conv.C
-#endif
-        // clang-format on
-        >;
+using device_grouped_conv_bwd_data_xdl_bf16_instances = std::tuple<
+    // clang-format off
+        // ##############################################|          NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+        // ##############################################|       Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+        // ##############################################|              |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+        // ##############################################|              |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 16, 1, 4>,                1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,         1,            1,            1,     S<1, 32, 1, 4>,                1>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 32, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,   S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 8>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  BF16,  BF16,     F32,     BF16, Empty_Tuple,  BF16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,    S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
+    // clang-format on
+    >;
 
 // f32_f32_f32_f32
 template <index_t NDimSpatial,
@@ -117,24 +120,26 @@ using device_grouped_conv_bwd_data_xdl_f32_instances =
          // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
          // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
          // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
-         DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,  F32,  F32,      F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,            1,            1,     S<1, 16, 1, 4>,                1>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,            1,            1,     S<1, 32, 1, 4>,                1>,
 
-#ifndef CK_WORKAROUND_GITHUB_ISSUE_824
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
-        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1<NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>,
-        // TODO: After enable, add instance for small conv.K and conv.C
-#endif
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8>
         // clang-format on
         >;
 
diff --git a/test/contraction/test_contraction_interface.cpp b/test/contraction/test_contraction_interface.cpp
index c9e720c597..12a307f5da 100644
--- a/test/contraction/test_contraction_interface.cpp
+++ b/test/contraction/test_contraction_interface.cpp
@@ -38,7 +38,7 @@ class ContractionInstanceWrapper
         //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
         //#####################################|        |        |        |      |      |        |         |               |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
         //#####################################|        |        |        |      |      |        |         |               |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
-        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,         Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
+        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,         Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
     // clang-format on
 
     bool isSupported(std::vector<ck::index_t>& ADims,
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
index c2aab749d3..bdad37fe14 100644
--- a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
@@ -87,6 +87,9 @@ TYPED_TEST(TestGroupedConvndBwdData2d, Test2D)
         {2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
     this->conv_params.push_back(
         {2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back({2, 1, 1, 1, 32, {8, 8}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 1, 64, 3, {8, 8}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 1, 1, 1, {8, 8}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     this->template Run<2>();
 }
 
@@ -99,5 +102,11 @@ TYPED_TEST(TestGroupedConvndBwdData3d, Test3D)
         {3, 2, 2, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
     this->conv_params.push_back(
         {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
     this->template Run<3>();
 }
diff --git a/test/grouped_gemm/test_grouped_gemm_util.hpp b/test/grouped_gemm/test_grouped_gemm_util.hpp
index b61118b512..04b31dcc91 100644
--- a/test/grouped_gemm/test_grouped_gemm_util.hpp
+++ b/test/grouped_gemm/test_grouped_gemm_util.hpp
@@ -147,14 +147,14 @@ struct DeviceGroupedGemmSplitkInstanceWrapper
             32,
             4,
             2,
-            S<1, 4, 32, 1>,
+            S<1, 4, 16, 1>,
             ABlockTransferThreadClusterArrageOrder,
             ABlockTransferSrcAccessOrder,
             ABlockTransferSrcVectorDim::value,
             ABlockTransferSrcScalarPerVector,
             ABlockTransferDstScalarPerVector_K1::value,
             ABlockLdsAddExtraM::value,
-            S<1, 4, 32, 1>,
+            S<1, 4, 16, 1>,
             BBlockTransferThreadClusterArrageOrder,
             BBlockTransferSrcAccessOrder,
             BBlockTransferSrcVectorDim::value,

From ca3115e7e8e3fbd64fe4ef3c19c84c20fa0c80a9 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 22 Aug 2023 18:54:34 -0500
Subject: [PATCH 097/194] Ck profiler splitk (#857)

* updated regular gemm

* update ckProfiler

* fixed gtests

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |   5 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp |   5 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  48 ----
 ...l_splitk_f16_f16_f16_mk_kn_mn_instance.cpp |  54 ++--
 ...l_splitk_f16_f16_f16_mk_nk_mn_instance.cpp |  42 ++-
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |  62 +++--
 .../profiler/profile_gemm_splitk_impl.hpp     | 133 ++++++----
 .../profiler/profile_grouped_gemm_impl.hpp    | 246 ++++++++++--------
 .../test_grouped_gemm_interface.cpp           |   2 +-
 9 files changed, 330 insertions(+), 267 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index 3b0e937e22..e7365daed2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -58,7 +58,9 @@ template <typename ADataType,
           index_t CShuffleNRepeatPerShuffle,
           typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CBlockTransferScalarPerVector_NWaveNPerXDL,
-          typename ComputeType = CDataType>
+          typename ComputeType        = CDataType,
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+
 struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                              BLayout,
                                                              CLayout,
@@ -77,7 +79,6 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
     // TODO: should be exposed as Tparams.
     static constexpr index_t NumGemmKPrefetchStage = 1;
     static constexpr LoopScheduler LoopSched       = make_default_loop_scheduler();
-    static constexpr PipelineVersion PipelineVer   = PipelineVersion::v1;
 
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2<
         BlockSize,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index d9950c79ac..41445aeaf0 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -114,7 +114,8 @@ template <typename ALayout,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CDEBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
           // Current implementation does not support multiple D fusions.
           enable_if_t<AK1 == BK1 && is_same_v<DsLayout, ck::Tuple<>> &&
                           is_same_v<DsDataType, ck::Tuple<>>,
@@ -183,7 +184,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         CDEBlockTransferScalarPerVector_NPerBlock,
         CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
         LoopSched,
-        PipelineVersion::v2>;
+        PipelineVer>;
 
     using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
     using Block2ETileMapKSplit =
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 58f76e5368..0c94e28e9b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -789,53 +789,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
             p_b_block, b_k0_n_k1_block_desc.GetElementSpaceSize());
 
-#if 0
-        // preload data into LDS
-        {
-            a_blockwise_copy.RunRead(a_b_k0_m_k1_grid_desc, a_grid_buf);
-            b_blockwise_copy.RunRead(b_b_k0_n_k1_grid_desc, b_grid_buf);
-
-            a_blockwise_copy.RunWrite(a_b_k0_m_k1_block_desc, a_block_buf);
-            b_blockwise_copy.RunWrite(b_b_k0_n_k1_block_desc, b_block_buf);
-        }
-
-        // Initialize C
-        c_thread_buf.Clear();
-
-        // main body
-        if constexpr(HasMainKBlockLoop)
-        {
-            index_t k0_block_data_begin = 0;
-
-            do
-            {
-                a_blockwise_copy.MoveSrcSliceWindow(a_b_k0_m_k1_grid_desc, a_block_slice_copy_step);
-                b_blockwise_copy.MoveSrcSliceWindow(b_b_k0_n_k1_grid_desc, b_block_slice_copy_step);
-
-                a_blockwise_copy.RunRead(a_b_k0_m_k1_grid_desc, a_grid_buf);
-
-                block_sync_lds();
-
-                b_blockwise_copy.RunRead(b_b_k0_n_k1_grid_desc, b_grid_buf);
-
-                blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
-
-                block_sync_lds();
-
-                a_blockwise_copy.RunWrite(a_b_k0_m_k1_block_desc, a_block_buf);
-                b_blockwise_copy.RunWrite(b_b_k0_n_k1_block_desc, b_block_buf);
-
-                k0_block_data_begin += K0PerBlock;
-            } while(k0_block_data_begin < (karg.K0 - K0PerBlock));
-        }
-
-        // tail
-        {
-            block_sync_lds();
-
-            blockwise_gemm.Run(a_block_buf, b_block_buf, c_thread_buf);
-        }
-#else
         // gridwise GEMM pipeline
         const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
             (a_b_k0_m_k1_grid_desc.GetLength(I1) * a_b_k0_m_k1_grid_desc.GetLength(I3)) /
@@ -858,7 +811,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
                                                                blockwise_gemm,
                                                                c_thread_buf,
                                                                num_k_block_main_loop);
-#endif
 
         // output: register to global memory
         {
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index 30a2bf36d1..67a10d623f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -36,22 +36,44 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>
+        //PipelineVersion::v1
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    32,    32,     4,  8,   32,   32,    1,    1,  S<1, 2, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    16,    32,     4,  8,   16,   16,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               4,  F16, PipelineVersion::v1>,
+
+        //PipelineVersion::v2
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>
+    // clang-format on
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index 4b2a2dbdc7..28e6e15861 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -35,19 +35,35 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>,
-        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8>
+        //PipelineVersion::v1
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v1>,
+
+        //PipelineVersion::v2
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,  F16, PipelineVersion::v2>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index 83b31b07cf..e2877f2a5c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -41,26 +41,48 @@ using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instanc
      // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
      // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
         
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
+    // clang-format on
     // clang-format on
     >;
 
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index 47a02b41ce..fb68bb8811 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -94,7 +94,6 @@ bool profile_gemm_splitk_impl(int do_verification,
 
     a_device_buf.ToDevice(a_m_k.mData.data());
     b_device_buf.ToDevice(b_k_n.mData.data());
-    c_device_buf.SetZero();
 
     using DeviceOp = ck::tensor_operation::device::DeviceGemmSplitK<ALayout,
                                                                     BLayout,
@@ -136,60 +135,84 @@ bool profile_gemm_splitk_impl(int do_verification,
     float best_ave_time   = 0;
     float best_tflops     = 0;
     float best_gb_per_sec = 0;
+    float best_kbatch     = 0;
 
     // profile device GEMM instances
     for(auto& op_ptr : op_ptrs)
     {
-        auto argument_ptr =
-            op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
-                                        static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
-                                        static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
-                                        M,
-                                        N,
-                                        K,
-                                        StrideA,
-                                        StrideB,
-                                        StrideC,
-                                        a_element_op,
-                                        b_element_op,
-                                        c_element_op,
-                                        KBatch);
+        std::vector<int> kbatch_list = {1,  2,  4,  8,  12, 16,  20,  24,  32,  36,  40, 60,
+                                        64, 72, 80, 88, 96, 128, 144, 160, 176, 192, 256};
 
-        auto invoker_ptr = op_ptr->MakeInvokerPointer();
-
-        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        if(KBatch > 0)
         {
-            // re-init C to zero before profiling next kernel
-            c_device_buf.SetZero();
+            kbatch_list = {KBatch};
+        }
 
-            std::string op_name = op_ptr->GetTypeString();
+        for(std::size_t i = 0; i < kbatch_list.size(); i++)
+        {
+            auto kbatch_curr = kbatch_list[i];
 
-            float ave_time =
-                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+            auto argument_ptr =
+                op_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
+                                            static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
+                                            static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
+                                            M,
+                                            N,
+                                            K,
+                                            StrideA,
+                                            StrideB,
+                                            StrideC,
+                                            a_element_op,
+                                            b_element_op,
+                                            c_element_op,
+                                            kbatch_curr);
 
-            std::size_t flop = std::size_t(2) * M * N * K;
+            auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
-            std::size_t num_btype =
-                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(CDataType) * M * N;
-
-            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-            float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
-                      << gb_per_sec << " GB/s, " << op_name << std::endl;
-
-            if(tflops > best_tflops)
+            if(op_ptr->IsSupportedArgument(argument_ptr.get()))
             {
-                best_op_name    = op_name;
-                best_tflops     = tflops;
-                best_ave_time   = ave_time;
-                best_gb_per_sec = gb_per_sec;
-            }
 
-            if(do_verification)
-            {
-                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
+                // re-init C to zero before profiling next kernel
+                c_device_buf.SetZero();
+
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+
+                if(do_verification)
+                {
+                    c_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+                    pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+
+                    if(do_log)
+                    {
+                        LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
+                        LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                        LogRangeAsType<float>(
+                            std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
+                            << std::endl;
+                        LogRangeAsType<float>(
+                            std::cout << "c_device: ", c_m_n_device_result.mData, ",")
+                            << std::endl;
+                    }
+                }
+
+                std::string op_name = op_ptr->GetTypeString();
+
+                float ave_time =
+                    invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+                std::size_t flop = std::size_t(2) * M * N * K;
+
+                std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                                        sizeof(CDataType) * M * N;
+
+                float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+                float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+                std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
+                          << " TFlops, " << gb_per_sec << " GB/s, " << op_name << ", KBatch "
+                          << kbatch_curr << std::endl;
 
                 // set softer tolerances for fp8
                 if constexpr(is_same_v<ADataType, f8_t> || is_same_v<BDataType, f8_t> ||
@@ -206,20 +229,20 @@ bool profile_gemm_splitk_impl(int do_verification,
                     pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
                 }
 
-                if(do_log)
+                if(tflops > best_tflops)
                 {
-                    LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
-                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
-                    LogRangeAsType<float>(std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
-                        << std::endl;
-                    LogRangeAsType<float>(std::cout << "c_device: ", c_m_n_device_result.mData, ",")
-                        << std::endl;
+                    best_op_name    = op_name;
+                    best_tflops     = tflops;
+                    best_ave_time   = ave_time;
+                    best_gb_per_sec = gb_per_sec;
+                    best_kbatch     = kbatch_curr;
                 }
             }
-        }
-        else
-        {
-            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+            else
+            {
+                std::cout << op_ptr->GetTypeString() << " does not support this problem"
+                          << std::endl;
+            }
         }
     }
 
@@ -259,7 +282,7 @@ bool profile_gemm_splitk_impl(int do_verification,
     }
 
     std::cout << " M = " << M << " N = " << N << " K = " << K << " StrideA = " << StrideA
-              << " StrideB = " << StrideB << " StrideC = " << StrideC << " KBatch = " << KBatch
+              << " StrideB = " << StrideB << " StrideC = " << StrideC << " KBatch = " << best_kbatch
               << " : " << best_ave_time << " ms, " << best_tflops << " TFlops, " << best_gb_per_sec
               << " GB/s, " << best_op_name << std::endl;
 
diff --git a/profiler/include/profiler/profile_grouped_gemm_impl.hpp b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
index 09a651d77c..fe7a397606 100644
--- a/profiler/include/profiler/profile_grouped_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
@@ -70,6 +70,7 @@ bool profile_grouped_gemm_impl(int do_verification,
 
     std::vector<Tensor<ADataType>> a_m_k;
     std::vector<Tensor<BDataType>> b_k_n;
+    std::vector<Tensor<CDataType>> c_m_n_host_results;
     std::vector<Tensor<CDataType>> c_m_n_device_results;
 
     for(std::size_t i = 0; i < group_count; i++)
@@ -81,6 +82,9 @@ bool profile_grouped_gemm_impl(int do_verification,
 
         c_m_n_device_results.push_back(
             Tensor<CDataType>(f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{})));
+
+        c_m_n_host_results.push_back(
+            Tensor<CDataType>(f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{})));
 #if DEBUG_LOG
         std::cout << "group: " << i << " a_m_k[" << i << "]:" << a_m_k[i].mDesc << ", b_k_n[" << i
                   << "]:" << b_k_n[i].mDesc << ", c_m_n_device_results[" << i
@@ -137,7 +141,6 @@ bool profile_grouped_gemm_impl(int do_verification,
 
         a_device_buf[i]->ToDevice(a_m_k[i].mData.data());
         b_device_buf[i]->ToDevice(b_k_n[i].mData.data());
-        c_device_buf[i]->SetZero();
 
         gemm_descs.push_back({Ms[i], Ns[i], Ks[i], StrideAs[i], StrideBs[i], StrideCs[i], {}});
 
@@ -170,9 +173,36 @@ bool profile_grouped_gemm_impl(int do_verification,
     float best_ave_time   = 0;
     float best_tflops     = 0;
     float best_gb_per_sec = 0;
+    float best_kbatch     = 0;
 
     auto p_ds = std::vector<std::array<const void*, 0>>{};
 
+    if(do_verification)
+    {
+        for(std::size_t i = 0; i < gemm_descs.size(); i++)
+        {
+            using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                    BDataType,
+                                                                                    CDataType,
+                                                                                    AccDataType,
+                                                                                    AElementOp,
+                                                                                    BElementOp,
+                                                                                    CElementOp>;
+
+            auto ref_gemm    = ReferenceGemmInstance{};
+            auto ref_invoker = ref_gemm.MakeInvoker();
+
+            auto ref_argument = ref_gemm.MakeArgument(a_m_k[i],
+                                                      b_k_n[i],
+                                                      c_m_n_host_results[i],
+                                                      a_element_op,
+                                                      b_element_op,
+                                                      c_element_op);
+
+            ref_invoker.Run(ref_argument);
+        }
+    }
+
     // profile device GEMM instances
     for(auto& gemm_ptr : op_ptrs)
     {
@@ -193,139 +223,135 @@ bool profile_grouped_gemm_impl(int do_verification,
         gemm_ptr->SetWorkSpacePointer(argument_ptr.get(), gemm_desc_workspace.GetDeviceBuffer());
         std::string gemm_name = gemm_ptr->GetTypeString();
 
-        if(kbatch > 1)
-        {
-            using DeviceOpSplitK =
-                ck::tensor_operation::device::DeviceGroupedGemmSplitK<ALayout,
-                                                                      BLayout,
-                                                                      ck::Tuple<>,
-                                                                      CLayout,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      ck::Tuple<>,
-                                                                      CDataType,
-                                                                      AElementOp,
-                                                                      BElementOp,
-                                                                      CElementOp>;
+        using DeviceOpSplitK = ck::tensor_operation::device::DeviceGroupedGemmSplitK<ALayout,
+                                                                                     BLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     CLayout,
+                                                                                     ADataType,
+                                                                                     BDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     CDataType,
+                                                                                     AElementOp,
+                                                                                     BElementOp,
+                                                                                     CElementOp>;
 
-            if(dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get()) != nullptr)
-            {
-                dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get())
-                    ->SetKBatchSize(argument_ptr.get(), kbatch);
-            }
+        // skip non-splitk grouped_gemm
+        if(dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get()) == nullptr)
+        {
+            continue;
         }
 
-        if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
+        std::vector<int> kbatch_list = {1, 2, 4, 8, 12, 16, 20, 24, 32, 48, 64};
+
+        if(kbatch > 0)
+        {
+            kbatch_list = {kbatch};
+        }
+
+        for(std::size_t j = 0; j < kbatch_list.size(); j++)
         {
 
-            float ave_time =
-                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+            auto kbatch_curr = kbatch_list[j];
 
-            if(time_kernel)
+            dynamic_cast<DeviceOpSplitK*>(gemm_ptr.get())
+                ->SetKBatchSize(argument_ptr.get(), kbatch_curr);
+
+            if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
             {
-                std::size_t flop = 0, num_btype = 0;
                 for(std::size_t i = 0; i < gemm_descs.size(); i++)
-                {
-                    flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
-
-                    num_btype += sizeof(ADataType) * Ms[i] * Ks[i] +
-                                 sizeof(BDataType) * Ks[i] * Ns[i] +
-                                 sizeof(CDataType) * Ms[i] * Ns[i];
-                }
-
-                float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-                float gb_per_sec = num_btype / 1.E6 / ave_time;
-                std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
-                          << " TFlops, " << gb_per_sec << " GB/s, " << gemm_name << std::endl;
-
-                if(tflops > best_tflops)
-                {
-                    best_gemm_name  = gemm_name;
-                    best_tflops     = tflops;
-                    best_ave_time   = ave_time;
-                    best_gb_per_sec = gb_per_sec;
-                }
-            }
-
-            if(do_verification)
-            {
-                bool instance_pass = true;
-                for(std::size_t i = 0; i < gemm_descs.size(); i++)
-                {
-
-                    c_device_buf[i]->FromDevice(c_m_n_device_results[i].mData.data());
                     c_device_buf[i]->SetZero();
 
-                    Tensor<CDataType> c_m_n_host_result(
-                        f_host_tensor_descriptor(Ms[i], Ns[i], StrideCs[i], CLayout{}));
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
 
-                    using ReferenceGemmInstance =
-                        ck::tensor_operation::host::ReferenceGemm<ADataType,
-                                                                  BDataType,
-                                                                  CDataType,
-                                                                  AccDataType,
-                                                                  AElementOp,
-                                                                  BElementOp,
-                                                                  CElementOp>;
-
-                    auto ref_gemm    = ReferenceGemmInstance{};
-                    auto ref_invoker = ref_gemm.MakeInvoker();
-
-                    auto ref_argument = ref_gemm.MakeArgument(a_m_k[i],
-                                                              b_k_n[i],
-                                                              c_m_n_host_result,
-                                                              a_element_op,
-                                                              b_element_op,
-                                                              c_element_op);
-
-                    ref_invoker.Run(ref_argument);
-                    if(std::is_same_v<CDataType, ck::half_t> && kbatch > 1)
+                if(do_verification)
+                {
+                    bool instance_pass = true;
+                    for(std::size_t i = 0; i < gemm_descs.size(); i++)
                     {
-                        instance_pass =
-                            instance_pass && ck::utils::check_err(c_m_n_device_results[i],
-                                                                  c_m_n_host_result,
-                                                                  "Error: Incorrect results!",
-                                                                  0.06);
-                    }
-                    else
-                    {
-                        instance_pass =
-                            instance_pass &&
-                            ck::utils::check_err(c_m_n_device_results[i], c_m_n_host_result);
+
+                        c_device_buf[i]->FromDevice(c_m_n_device_results[i].mData.data());
+
+                        if(std::is_same_v<CDataType, ck::half_t> && kbatch_curr > 1)
+                        {
+                            instance_pass =
+                                instance_pass && ck::utils::check_err(c_m_n_device_results[i],
+                                                                      c_m_n_host_results[i],
+                                                                      "Error: Incorrect results!",
+                                                                      0.06);
+                        }
+                        else
+                        {
+                            instance_pass =
+                                instance_pass && ck::utils::check_err(c_m_n_device_results[i],
+                                                                      c_m_n_host_results[i]);
+                        }
+
+                        if(do_log)
+                        {
+                            LogRangeAsType<float>(std::cout << "a : ", a_m_k[i].mData, ",")
+                                << std::endl;
+                            LogRangeAsType<float>(std::cout << "b: ", b_k_n[i].mData, ",")
+                                << std::endl;
+                            LogRangeAsType<float>(
+                                std::cout << "c_device: ", c_m_n_device_results[i].mData, ",")
+                                << std::endl;
+                            LogRangeAsType<float>(
+                                std::cout << "c_host  : ", c_m_n_host_results[i].mData, ",")
+                                << std::endl;
+                        }
                     }
 
-                    if(do_log)
-                    {
-                        LogRangeAsType<float>(std::cout << "a : ", a_m_k[i].mData, ",")
-                            << std::endl;
-                        LogRangeAsType<float>(std::cout << "b: ", b_k_n[i].mData, ",") << std::endl;
-                        LogRangeAsType<float>(
-                            std::cout << "c_device: ", c_m_n_device_results[i].mData, ",")
-                            << std::endl;
-                        LogRangeAsType<float>(
-                            std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
-                            << std::endl;
-                    }
+                    std::cout << "Instance: " << gemm_name << " verification "
+                              << (instance_pass ? "SUCCEED" : "FAILED") << std::endl;
+
+                    pass = pass && instance_pass;
                 }
 
-                std::cout << "Instance: " << gemm_name << " verification "
-                          << (instance_pass ? "SUCCEED" : "FAILED") << std::endl;
+                float ave_time =
+                    invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
 
-                pass = pass && instance_pass;
+                if(time_kernel)
+                {
+                    std::size_t flop = 0, num_btype = 0;
+                    for(std::size_t i = 0; i < gemm_descs.size(); i++)
+                    {
+                        flop += std::size_t(2) * Ms[i] * Ns[i] * Ks[i];
+
+                        num_btype += sizeof(ADataType) * Ms[i] * Ks[i] +
+                                     sizeof(BDataType) * Ks[i] * Ns[i] +
+                                     sizeof(CDataType) * Ms[i] * Ns[i];
+                    }
+
+                    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+                    float gb_per_sec = num_btype / 1.E6 / ave_time;
+                    std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops
+                              << " TFlops, " << gb_per_sec << " GB/s, " << gemm_name << ", KBatch "
+                              << kbatch_curr << std::endl;
+
+                    if(tflops > best_tflops)
+                    {
+                        best_gemm_name  = gemm_name;
+                        best_tflops     = tflops;
+                        best_ave_time   = ave_time;
+                        best_gb_per_sec = gb_per_sec;
+                        best_kbatch     = kbatch_curr;
+                    }
+                }
+            }
+            else
+            {
+                std::cout << "Instance: " << gemm_name << ", does not support this GEMM problem"
+                          << std::endl;
             }
-        }
-        else
-        {
-            std::cout << "Instance: " << gemm_name << ", does not support this GEMM problem"
-                      << std::endl;
         }
     }
 
     if(time_kernel)
     {
         std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
-                  << best_gb_per_sec << " GB/s, " << best_gemm_name << std::endl;
+                  << best_gb_per_sec << " GB/s, " << best_gemm_name << ", KBatch = " << best_kbatch
+                  << std::endl;
     }
 
     return pass;
diff --git a/test/grouped_gemm/test_grouped_gemm_interface.cpp b/test/grouped_gemm/test_grouped_gemm_interface.cpp
index ffa8840fc7..6ff3a787e7 100644
--- a/test/grouped_gemm/test_grouped_gemm_interface.cpp
+++ b/test/grouped_gemm/test_grouped_gemm_interface.cpp
@@ -108,7 +108,7 @@ TEST_F(TestGGemmSplitKInterface_MKNKMN, KLoops)
 
     // kloops % 2
     Ks = std::vector<int>{256, 512, 320, 768};
-    EXPECT_FALSE(
+    EXPECT_TRUE(
         DefaultGGemmInstance{}.IsSupported(Ms, Ns, Ks, StrideAs, StrideBs, StrideCs, kbatch));
 
     // Not all gemms have same value for main_k0_block_loop!

From 8ebea3a56e68b6b638c78f3644b53f3c20e245f2 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 23 Aug 2023 09:18:10 -0500
Subject: [PATCH 098/194] add generic instances (#858)

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 ...l_splitk_f16_f16_f16_mk_kn_mn_instance.cpp | 14 +++++++-
 ...l_splitk_f16_f16_f16_mk_nk_mn_instance.cpp | 13 +++++++
 ..._streamk_f16_f16_f16_mk_kn_mn_instance.cpp | 12 +++++++
 .../profiler/profile_gemm_streamk_impl.hpp    | 36 ++++++++++---------
 4 files changed, 57 insertions(+), 18 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index 67a10d623f..218d6e0c2a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -29,6 +29,17 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 // static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
+using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_generic_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               2>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    32,    32,     4,  8,   32,   32,    1,    1,  S<1, 2, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               2>
+    // clang-format on
+    >;
+
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
@@ -74,7 +85,6 @@ using device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,  F16, PipelineVersion::v2>,
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,  F16, PipelineVersion::v2>
     // clang-format on
-    // clang-format on
     >;
 
 void add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
@@ -82,6 +92,8 @@ void add_device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemmSplitK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index 28e6e15861..b87f7ff309 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -28,6 +28,17 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
+using device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_generic_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              1,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               2,  F16, PipelineVersion::v1>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               2,  F16, PipelineVersion::v1>
+    // clang-format on
+    >;
+
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // clang-format off
@@ -72,6 +83,8 @@ void add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemmSplitK<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
index 7e058a895d..60e5d8f5fa 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_streamk/device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -29,6 +29,16 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 // static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 // static constexpr auto GemmMNPadding =
 // ck::tensor_operation::device::GemmSpecialization::MNPadding;
+using device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_generic_instances = std::tuple<
+    // clang-format off
+        //##################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer|
+        //##################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
+        //##################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##################|     |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               2>,
+        DeviceGemmXdlStreamK<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,    32,    64,     4,  8,   32,   32,    1,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,                   S<1, 16, 1, 8>,               2>
+    // clang-format on
+    >;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
@@ -61,6 +71,8 @@ void add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemmStreamK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(
+        instances, device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances{});
 }
diff --git a/profiler/include/profiler/profile_gemm_streamk_impl.hpp b/profiler/include/profiler/profile_gemm_streamk_impl.hpp
index b5d24367d4..71b54c1f47 100644
--- a/profiler/include/profiler/profile_gemm_streamk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_streamk_impl.hpp
@@ -170,6 +170,25 @@ bool profile_gemm_streamk_impl(int do_verification,
             // re-init C to zero before profiling next kernel
             c_device_buf.SetZero();
 
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+
+            if(do_verification)
+            {
+                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
+
+                pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "c_device: ", c_m_n_device_result.mData, ",")
+                        << std::endl;
+                }
+            }
+
             std::string op_name = op_ptr->GetTypeString();
 
             float ave_time =
@@ -194,23 +213,6 @@ bool profile_gemm_streamk_impl(int do_verification,
                 best_ave_time   = ave_time;
                 best_gb_per_sec = gb_per_sec;
             }
-
-            if(do_verification)
-            {
-                c_device_buf.FromDevice(c_m_n_device_result.mData.data());
-
-                pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
-
-                if(do_log)
-                {
-                    LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
-                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
-                    LogRangeAsType<float>(std::cout << "c_host  : ", c_m_n_host_result.mData, ",")
-                        << std::endl;
-                    LogRangeAsType<float>(std::cout << "c_device: ", c_m_n_device_result.mData, ",")
-                        << std::endl;
-                }
-            }
         }
         else
         {

From 7c71dc7e70cfc9b866268337872fc8bd9f8328ca Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 23 Aug 2023 09:52:11 -0700
Subject: [PATCH 099/194] use correct data types in cmake conditions for splitk
 gemm example (#862)

---
 example/35_splitK_gemm/CMakeLists.txt | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index fa66a5a520..251a9b93c5 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -3,15 +3,15 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_splitK_gemm_xdl)
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
     add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
     add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
    endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
     add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
     add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
    endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
+   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
     add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
     add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bfp16)
    endif()

From 350d64f351701af68e58c64e0e97efe07f7bf126 Mon Sep 17 00:00:00 2001
From: Qianfeng <qianfeng.zhang@amd.com>
Date: Thu, 24 Aug 2023 01:13:07 +0800
Subject: [PATCH 100/194] Add workspace setting up for batchnorm bwd/fwd client
 examples (#860)

---
 client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp | 6 ++++++
 client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp | 6 ++++++
 2 files changed, 12 insertions(+)

diff --git a/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp b/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
index c0140f71c1..1ed36e0f50 100644
--- a/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_bwd_nhwc.cpp
@@ -191,6 +191,12 @@ int main(int argc, char* argv[])
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+
+            SimpleDeviceMem workspace(workspace_sz);
+
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
             invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
         }
 
diff --git a/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp b/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
index 3653733436..f9af011c84 100644
--- a/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
+++ b/client_example/13_batchnorm/batchnorm_fwd_nhwc.cpp
@@ -187,6 +187,12 @@ int main(int argc, char* argv[])
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+
+            SimpleDeviceMem workspace(workspace_sz);
+
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
             invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
         }
 

From c8a8385fdd9ff1d16fcd9fee81a44145c06552a6 Mon Sep 17 00:00:00 2001
From: Jun Liu <Liu.Jun@amd.com>
Date: Wed, 23 Aug 2023 11:36:17 -0700
Subject: [PATCH 101/194] [HotFix] add config and version files to pass on
 build info (#856)

* experiment with config file

* experiment with version.h config

* add more info to version.h

* minor updates

* minor updates

* fix case where DTYPE is not used

* large amount of files but minor changes

* remove white space

* minor changes to add more MACROs

* fix cmakedefine01

* fix issue with CK internal conflict

* fix define and define value

* fix clang-format

* fix formatting issue

* experiment with cmake

* clang format v12 to be consistent with miopen

* avoid clang-format for config file
---
 .pre-commit-config.yaml                       |   2 +-
 CMakeLists.txt                                | 113 +++++++++++-------
 Dockerfile                                    |   2 +-
 Jenkinsfile                                   |   2 +-
 client_example/CMakeLists.txt                 |  53 +++++---
 include/ck/ck.hpp                             |   2 +
 include/ck/config.h.in                        | 102 ++++++++++++++++
 .../multi_index_transform.hpp                 |  18 +--
 .../gpu/block/blockwise_softmax.hpp           |   4 +-
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |  20 ++--
 .../device_batched_gemm_e_permute_xdl.hpp     |  15 ++-
 .../impl/device_batched_gemm_multi_d_xdl.hpp  |  20 ++--
 ...ultiple_d_gemm_multiple_d_xdl_cshuffle.hpp |  15 ++-
 ...ce_contraction_multiple_d_xdl_cshuffle.hpp |  20 ++--
 ...gemm_multiple_d_layernorm_xdl_cshuffle.hpp |  10 +-
 ...emm_multiple_d_multiple_r_xdl_cshuffle.hpp |  10 +-
 .../device_gemm_multiple_d_xdl_cshuffle.hpp   |  20 ++--
 .../device_gemm_xdl_waveletmodel_cshuffle.hpp |  10 +-
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |  20 ++--
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  10 +-
 ..._conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp |   4 +-
 ...ice_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp |   4 +-
 ...fwd_multiple_d_multiple_r_xdl_cshuffle.hpp |  14 ++-
 ...uped_conv_fwd_multiple_d_wmma_cshuffle.hpp |   4 +-
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |  24 ++--
 ...gemm_softmax_gemm_permute_xdl_cshuffle.hpp |  12 +-
 .../device/impl/device_grouped_gemm_xdl.hpp   |  20 ++--
 ...tk_contraction_multiple_d_xdl_cshuffle.hpp |  10 +-
 .../gridwise_multiblock_batchnorm_forward.hpp |   4 +-
 ...e_second_half_batchnorm_backward_final.hpp |   4 +-
 ..._half_batchnorm_forward_final_obsolete.hpp |   4 +-
 ...cond_half_multiblock_reduce_first_half.hpp |   4 +-
 ...iple_d_welford_first_half_xdl_cshuffle.hpp |  24 ++--
 ...wise_batched_gemm_gemm_xdl_cshuffle_v1.hpp |   9 +-
 ...iple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp |  15 ++-
 ...ultiple_d_softmax_gemm_xdl_cshuffle_v1.hpp |  16 +--
 ...ched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp |   9 +-
 ...e_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp |  19 +--
 ...emm_multiple_d_multiple_r_xdl_cshuffle.hpp |   9 +-
 ...gridwise_gemm_multiple_d_wmma_cshuffle.hpp |  14 ++-
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |   4 +-
 .../gridwise_gemm_reduce_xdl_cshuffle_v1.hpp  |   9 +-
 ...e_gemm_split_k_multiple_d_xdl_cshuffle.hpp |  10 +-
 .../gpu/grid/gridwise_gemm_wmma.hpp           |   9 +-
 .../grid/gridwise_gemm_xdl_cshuffle_v1.hpp    |   4 +-
 ...ridwise_gemm_xdl_layernorm_cshuffle_v1.hpp |   9 +-
 ...ridwise_gemm_xdl_waveletmodel_cshuffle.hpp |   5 +-
 .../grid/gridwise_gemm_xdlops_bwd_weight.hpp  |   4 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |   4 +-
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |   4 +-
 .../gpu/grid/gridwise_gemm_xdlops_v3r1.hpp    |  14 +--
 .../gpu/grid/gridwise_gemm_xdlops_v3r2.hpp    |  21 ++--
 .../gpu/grid/gridwise_gemm_xdlops_v3r3.hpp    |  28 ++---
 ...se_sparse_embeddings_forward_layernorm.hpp |   4 +-
 .../gridwise_normalization_splitk_1st.hpp     |   6 +-
 include/ck/version.h.in                       |  40 +++++++
 .../gpu/batched_gemm.hpp                      |  16 +--
 .../gpu/batched_gemm_add_relu_gemm_add.hpp    |   2 +-
 .../gpu/batched_gemm_bias_permute.hpp         |   2 +-
 ...batched_gemm_bias_softmax_gemm_permute.hpp |   8 +-
 .../gpu/batched_gemm_gemm.hpp                 |   2 +-
 .../gpu/batched_gemm_multi_d.hpp              |   8 +-
 .../gpu/batched_gemm_softmax_gemm.hpp         |   2 +-
 .../gpu/batched_gemm_softmax_gemm_permute.hpp |   8 +-
 .../gpu/contraction_bilinear.hpp              |   8 +-
 .../gpu/contraction_scale.hpp                 |   8 +-
 .../gpu/convolution_backward_data.hpp         |  48 ++++----
 .../gpu/convolution_forward.hpp               |  14 +--
 .../gpu/elementwise_normalization.hpp         |   2 +-
 .../tensor_operation_instance/gpu/gemm.hpp    |  22 ++--
 .../gpu/gemm_add_relu_add_layernorm.hpp       |   2 +-
 .../gpu/gemm_bilinear.hpp                     |   2 +-
 .../gpu/gemm_fastgelu.hpp                     |   2 +-
 .../gpu/gemm_streamk.hpp                      |   2 +-
 .../gpu/grouped_gemm.hpp                      |   2 +-
 .../gpu/normalization.hpp                     |   8 +-
 .../gpu/pool3d_fwd.hpp                        |   8 +-
 .../gpu/quantization/gemm_quantization.hpp    |   2 +-
 ...n_bias_forward_perchannel_quantization.hpp |   2 +-
 ...ion_bias_forward_perlayer_quantization.hpp |   2 +-
 ...lution_forward_perchannel_quantization.hpp |   2 +-
 ...volution_forward_perlayer_quantization.hpp |   2 +-
 .../device_reduce_instance_blockwise.hpp      |   8 +-
 ..._reduce_instance_multiblock_atomic_add.hpp |   8 +-
 .../device_reduce_instance_threadwise.hpp     |   4 +-
 .../tensor_operation_instance/gpu/softmax.hpp |   4 +-
 library/include/ck/library/utility/fill.hpp   |   7 +-
 ..._data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp |   2 +-
 ..._data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp |   2 +-
 ...d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp |   2 +-
 ...2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp |   2 +-
 ...2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp |   2 +-
 ...d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp |   2 +-
 ...ice_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp |   2 +-
 ...l_i8_i8_i8_km_kn_mn_irregular_instance.cpp |   2 +-
 ...ice_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp |   2 +-
 ...l_i8_i8_i8_km_nk_mn_irregular_instance.cpp |   2 +-
 ...ice_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp |   2 +-
 ...l_i8_i8_i8_mk_kn_mn_irregular_instance.cpp |   2 +-
 ...ice_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp |   2 +-
 ...l_i8_i8_i8_mk_nk_mn_irregular_instance.cpp |   2 +-
 ...l_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp |   2 +-
 ...l_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp |   2 +-
 ...l_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp |   2 +-
 ...l_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp |   2 +-
 profiler/src/profile_batched_gemm_multi_d.cpp |   4 +-
 profiler/src/profile_conv_bwd_data.cpp        |   8 +-
 profiler/src/profile_gemm.cpp                 |  12 +-
 profiler/src/profile_grouped_gemm.cpp         |   2 +-
 script/clang-format-overwrite.sh              |   4 +-
 script/install_precommit.sh                   |   2 +-
 .../test_batched_gemm_multi_d.cpp             |   2 +-
 test/pool_fwd/test_avg_pool3d_fwd.cpp         |   2 +-
 test/pool_fwd/test_max_pool3d_fwd.cpp         |   2 +-
 test/softmax/test_softmax_rank3.cpp           |   4 +-
 test/softmax/test_softmax_rank4.cpp           |   4 +-
 116 files changed, 709 insertions(+), 456 deletions(-)
 create mode 100644 include/ck/config.h.in
 create mode 100644 include/ck/version.h.in

diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 3e1e8c0169..d6700ae05b 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -3,7 +3,7 @@ repos:
     hooks:
     -   id: clang-format
         name: clang-format
-        entry: clang-format-10 -i --style=file
+        entry: clang-format-12 -i --style=file
         language: system
         types_or: [c++, inc]
     -   id: copyright-year-checker
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 6eba503942..fd03c9b207 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,43 +1,61 @@
 cmake_minimum_required(VERSION 3.14)
 
+set(version 1.1.0)
 # Check support for CUDA/HIP in Cmake
-project(composable_kernel)
+project(composable_kernel VERSION ${version})
 
 list(APPEND CMAKE_MODULE_PATH "${PROJECT_SOURCE_DIR}/cmake")
 
 if (DTYPES)
-        add_definitions(-DDTYPES)
-        if (DTYPES MATCHES "int8")
-                add_definitions(-D__int8__)
-        endif()
-        if (DTYPES MATCHES "fp8")
-                add_definitions(-D__fp8__)
-        endif()
-        if (DTYPES MATCHES "fp16")
-                add_definitions(-D__fp16__)
-        endif()
-        if (DTYPES MATCHES "fp32")
-                add_definitions(-D__fp32__)
-        endif()
-        if (DTYPES MATCHES "fp64")
-                add_definitions(-D__fp64__)
-        endif()
-        if (DTYPES MATCHES "bf16")
-                add_definitions(-D__bf16__)
-        endif()
-        message("DTYPES macro set to ${DTYPES}")
+    add_definitions(-DDTYPES)
+    if (DTYPES MATCHES "int8")
+        add_definitions(-DCK_ENABLE_INT8)
+        set(CK_ENABLE_INT8 "ON")
+    endif()
+    if (DTYPES MATCHES "fp8")
+        add_definitions(-DCK_ENABLE_FP8)
+        set(CK_ENABLE_FP8 "ON")
+    endif()
+    if (DTYPES MATCHES "fp16")
+        add_definitions(-DCK_ENABLE_FP16)
+        set(CK_ENABLE_FP16 "ON")
+    endif()
+    if (DTYPES MATCHES "fp32")
+        add_definitions(-DCK_ENABLE_FP32)
+        set(CK_ENABLE_FP32 "ON")
+    endif()
+    if (DTYPES MATCHES "fp64")
+        add_definitions(-DCK_ENABLE_FP64)
+        set(CK_ENABLE_FP64 "ON")
+    endif()
+    if (DTYPES MATCHES "bf16")
+        add_definitions(-DCK_ENABLE_BF16)
+        set(CK_ENABLE_BF16 "ON")
+    endif()
+    message("DTYPES macro set to ${DTYPES}")
 else()
-        add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
+    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
+    set(CK_ENABLE_ALL_DTYPES "ON")
 endif()
 
 if(DL_KERNELS)
     add_definitions(-DDL_KERNELS)
+    set(CK_ENABLE_DL_KERNELS "ON")
 endif()
 
 if(INSTANCES_ONLY)
     add_definitions(-DINSTANCES_ONLY)
+    set(CK_ENABLE_INSTANCES_ONLY "ON")
 endif()
 
+# CK config file to record supported datatypes, etc.
+configure_file("${PROJECT_SOURCE_DIR}/include/ck/config.h.in" "${PROJECT_BINARY_DIR}/include/ck/config.h")
+
+# CK version file to record release version as well as git commit hash
+find_package(Git REQUIRED)
+execute_process(COMMAND "${GIT_EXECUTABLE}" rev-parse HEAD OUTPUT_VARIABLE COMMIT_ID OUTPUT_STRIP_TRAILING_WHITESPACE)
+configure_file("${PROJECT_SOURCE_DIR}/include/ck/version.h.in" "${PROJECT_BINARY_DIR}/include/ck/version.h")
+
 enable_testing()
 
 set(ROCM_SYMLINK_LIBS OFF)
@@ -50,8 +68,10 @@ include(ROCMInstallSymlinks)
 include(ROCMCreatePackage)
 include(CheckCXXCompilerFlag)
 include(ROCMCheckTargetIds)
-rocm_setup_version(VERSION 0.2.0)
 include(TargetFlags)
+
+rocm_setup_version(VERSION ${version})
+
 list(APPEND CMAKE_PREFIX_PATH ${CMAKE_INSTALL_PREFIX} ${CMAKE_INSTALL_PREFIX}/llvm ${CMAKE_INSTALL_PREFIX}/hip /opt/rocm /opt/rocm/llvm /opt/rocm/hip)
 
 message("GPU_TARGETS= ${GPU_TARGETS}")
@@ -315,13 +335,14 @@ set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/lib)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/bin)
 
+# set CK project include directories
 include_directories(BEFORE
+    ${PROJECT_BINARY_DIR}/include
     ${PROJECT_SOURCE_DIR}/include
     ${PROJECT_SOURCE_DIR}/library/include
     ${HIP_INCLUDE_DIRS}
 )
 
-
 SET(BUILD_DEV ON CACHE BOOL "BUILD_DEV")
 if(BUILD_DEV)
     add_compile_options(-Werror)
@@ -341,35 +362,35 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
     file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
     set(add_inst 0)
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8")
-            #message("fp8 instance found!")
-            set(add_inst 1)
+        #message("fp8 instance found!")
+        set(add_inst 1)
     endif()
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
-            #message("fp16 instance found!")
-            set(add_inst 1)
+        #message("fp16 instance found!")
+        set(add_inst 1)
     endif()
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32")
-            #message("fp32 instance found!")
-            set(add_inst 1)
+        #message("fp32 instance found!")
+        set(add_inst 1)
     endif()
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64")
-            #message("fp64 instance found!")
-            set(add_inst 1)
+        #message("fp64 instance found!")
+        set(add_inst 1)
     endif()
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16")
-            #message("bf16 instance found!")
-            set(add_inst 1)
+        #message("bf16 instance found!")
+        set(add_inst 1)
     endif()
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8")
-            #message("int8 instance found!")
-            set(add_inst 1)
+        #message("int8 instance found!")
+        set(add_inst 1)
     endif()
     if(NOT "${cmake_instance}" MATCHES "DTYPES")
-            #message("instance should be built for all types!")
-            set(add_inst 1)
+        #message("instance should be built for all types!")
+        set(add_inst 1)
     endif()
     if(add_inst EQUAL 1 OR NOT DEFINED DTYPES)
-      list(APPEND CK_DEVICE_INSTANCES device_${subdir_path}_instance)
+        list(APPEND CK_DEVICE_INSTANCES device_${subdir_path}_instance)
     endif()
 ENDIF()
 ENDFOREACH()
@@ -409,7 +430,6 @@ endif()
 #Create an interface target for the include only files and call it "composablekernels"
 include(CMakePackageConfigHelpers)
 
-set(version 1.0.0)
 write_basic_package_version_file(
     "${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfigVersion.cmake"
     VERSION "${version}"
@@ -417,9 +437,9 @@ write_basic_package_version_file(
 )
 
 configure_package_config_file(${CMAKE_CURRENT_SOURCE_DIR}/Config.cmake.in
-        "${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfig.cmake"
-        INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
-        NO_CHECK_REQUIRED_COMPONENTS_MACRO
+    "${CMAKE_CURRENT_BINARY_DIR}/composable_kernelConfig.cmake"
+    INSTALL_DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
+    NO_CHECK_REQUIRED_COMPONENTS_MACRO
 )
 
 rocm_install(FILES
@@ -428,6 +448,13 @@ rocm_install(FILES
     DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
 )
 
+# Install CK version and configuration files
+install(FILES
+    ${PROJECT_BINARY_DIR}/include/ck/version.h
+    ${PROJECT_BINARY_DIR}/include/ck/config.h
+    DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck/
+)
+
 set(CPACK_RESOURCE_FILE_LICENSE "${CMAKE_CURRENT_SOURCE_DIR}/LICENSE")
 set(CPACK_RPM_PACKAGE_LICENSE "MIT")
 
diff --git a/Dockerfile b/Dockerfile
index b9c815f527..e479268f48 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -63,7 +63,7 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     nano \
     zlib1g-dev \
     openssh-server \
-    clang-format-10 \
+    clang-format-12 \
     kmod && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
diff --git a/Jenkinsfile b/Jenkinsfile
index 3fce4dabec..668d9a613b 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -689,7 +689,7 @@ pipeline {
                                 -o -iname \'*.cpp.in\' \
                                 -o -iname \'*.cl\' \
                                 | grep -v 'build/' \
-                                | xargs -n 1 -P 1 -I{} -t sh -c \'clang-format-10 -style=file {} | diff - {}\'"
+                                | xargs -n 1 -P 1 -I{} -t sh -c \'clang-format-12 -style=file {} | diff - {}\'"
                     }
                     steps{
                         buildHipClangJobAndReboot(setup_cmd: "", build_cmd: "", execute_cmd: execute_cmd, no_reboot:true)
diff --git a/client_example/CMakeLists.txt b/client_example/CMakeLists.txt
index 526abf5e77..eb793b3cbd 100644
--- a/client_example/CMakeLists.txt
+++ b/client_example/CMakeLists.txt
@@ -3,31 +3,52 @@ project(ck_app)
 add_compile_options(-std=c++17)
 
 if (DTYPES)
-        add_definitions(-DDTYPES)
-        if (DTYPES MATCHES "int8")
-                add_definitions(-D__int8__)
+    add_definitions(-DDTYPES)
+    if (DTYPES MATCHES "int8")
+        add_definitions(-DCK_ENABLE_INT8)
+        if(NOT DEFINED ${CK_ENABLE_INT8})
+            set(CK_ENABLE_INT8 "ON")
         endif()
-        if (DTYPES MATCHES "fp8")
-                add_definitions(-D__fp8__)
+    endif()
+    if (DTYPES MATCHES "fp8")
+        add_definitions(-DCK_ENABLE_FP8)
+        if(NOT DEFINED ${CK_ENABLE_FP8})
+            set(CK_ENABLE_FP8 "ON")
         endif()
-        if (DTYPES MATCHES "fp16")
-                add_definitions(-D__fp16__)
+    endif()
+    if (DTYPES MATCHES "fp16")
+        add_definitions(-DCK_ENABLE_FP16)
+        if(NOT DEFINED ${CK_ENABLE_FP16})
+            set(CK_ENABLE_FP16 "ON")
         endif()
-        if (DTYPES MATCHES "fp32")
-                add_definitions(-D__fp32__)
+    endif()
+    if (DTYPES MATCHES "fp32")
+        add_definitions(-DCK_ENABLE_FP32)
+        if(NOT DEFINED ${CK_ENABLE_FP32})
+            set(CK_ENABLE_FP32 "ON")
         endif()
-        if (DTYPES MATCHES "fp64")
-                add_definitions(-D__fp64__)
+    endif()
+    if (DTYPES MATCHES "fp64")
+        add_definitions(-DCK_ENABLE_FP64)
+        if(NOT DEFINED ${CK_ENABLE_FP64})
+            set(CK_ENABLE_FP64 "ON")
         endif()
-        if (DTYPES MATCHES "bf16")
-                add_definitions(-D__bf16__)
+    endif()
+    if (DTYPES MATCHES "bf16")
+        add_definitions(-DCK_ENABLE_BF16)
+        if(NOT DEFINED ${CK_ENABLE_BF16})
+            set(CK_ENABLE_BF16 "ON")
         endif()
-        message("DTYPES macro set to ${DTYPES}")
+    endif()
+    message("DTYPES macro set to ${DTYPES}")
 else()
-        add_definitions(-D__int8__ -D__fp8__ -D__fp16__ -D__fp32__ -D__fp64__ -D__bf16__)
+    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
+    if(NOT DEFINED ${CK_ENABLE_ALL_DTYPES})
+        set(CK_ENABLE_ALL_DTYPES "ON")
+    endif()
 endif()
 
-find_package(composable_kernel 1.0.0 COMPONENTS device_operations)
+find_package(composable_kernel COMPONENTS device_operations)
 find_package(hip REQUIRED PATHS /opt/rocm)
 message(STATUS "Build with HIP ${hip_VERSION}")
 
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 734fe549db..069ff7fc74 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -3,6 +3,8 @@
 
 #pragma once
 
+#include "ck/config.h"
+
 #ifndef CK_DONT_USE_HIP_RUNTIME_HEADERS
 #include "hip/hip_runtime.h"
 #include "hip/hip_fp16.h"
diff --git a/include/ck/config.h.in b/include/ck/config.h.in
new file mode 100644
index 0000000000..13dc5da5d1
--- /dev/null
+++ b/include/ck/config.h.in
@@ -0,0 +1,102 @@
+/*******************************************************************************
+ *
+ * MIT License
+ *
+ * Copyright (c) 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ *******************************************************************************/
+#ifndef CK_CONFIG_H_IN
+#define CK_CONFIG_H_IN
+
+// clang-format off
+//
+// DataType supports in the current CK build
+//
+#ifndef DTYPES
+#cmakedefine DTYPES "@DTYPES@"
+#endif
+// if DTYPES is not defined, enable all datatypes in headerfiles
+#ifndef CK_ENABLE_ALL_DTYPES
+#cmakedefine CK_ENABLE_ALL_DTYPES @CK_ENABLE_ALL_DTYPES@
+#if defined(CK_ENABLE_ALL_DTYPES)
+#ifndef CK_ENABLE_INT8
+#define CK_ENABLE_INT8 "ON"
+#endif
+#ifndef CK_ENABLE_FP8
+#define CK_ENABLE_FP8 "ON"
+#endif
+#ifndef CK_ENABLE_FP16
+#define CK_ENABLE_FP16 "ON"
+#endif
+#ifndef CK_ENABLE_BF16
+#define CK_ENABLE_BF16 "ON"
+#endif
+#ifndef CK_ENABLE_FP32
+#define CK_ENABLE_FP32 "ON"
+#endif
+#ifndef CK_ENABLE_FP64
+#define CK_ENABLE_FP64 "ON"
+#endif
+#endif
+#endif
+// if DTYPES are selectively enabled
+#ifndef CK_ENABLE_INT8
+#cmakedefine CK_ENABLE_INT8 @CK_ENABLE_INT8@
+#endif
+
+#ifndef CK_ENABLE_FP8
+#cmakedefine CK_ENABLE_FP8 @CK_ENABLE_FP8@
+#endif
+
+#ifndef CK_ENABLE_FP16
+#cmakedefine CK_ENABLE_FP16 @CK_ENABLE_FP16@
+#endif
+
+#ifndef CK_ENABLE_BF16
+#cmakedefine CK_ENABLE_BF16 @CK_ENABLE_BF16@
+#endif
+
+#ifndef CK_ENABLE_FP32
+#cmakedefine CK_ENABLE_FP32 @CK_ENABLE_FP32@
+#endif
+
+#ifndef CK_ENABLE_FP64
+#cmakedefine CK_ENABLE_FP64 @CK_ENABLE_FP64@
+#endif
+
+//
+// Legacy DL kernel supports in the current CK build
+// by default DL kernels are turned OFF
+//
+#ifndef CK_ENABLE_DL_KERNELS
+#cmakedefine CK_ENABLE_DL_KERNELS @CK_ENABLE_DL_KERNELS@
+#endif
+
+//
+// Instances supports in the current CK build
+//
+#ifndef CK_ENABLE_INSTANCES_ONLY
+#cmakedefine CK_ENABLE_INSTANCES_ONLY @CK_ENABLE_INSTANCES_ONLY@
+#endif
+
+// clang-format on
+
+#endif // CK_CONFIG_H_IN
diff --git a/include/ck/tensor_description/multi_index_transform.hpp b/include/ck/tensor_description/multi_index_transform.hpp
index 6854226dd4..ae3139ce78 100644
--- a/include/ck/tensor_description/multi_index_transform.hpp
+++ b/include/ck/tensor_description/multi_index_transform.hpp
@@ -1042,13 +1042,13 @@ struct Merge_v2_magic_division
     using UpLengths =
         decltype(make_tuple(container_reduce(LowLengths{}, math::multiplies{}, Number<1>{})));
 
-    using LowLengthsMagicDivisorMultipiler = decltype(
-        generate_tuple(lambda_merge_generate_MagicDivision_calculate_magic_multiplier<LowLengths>{},
-                       Number<NDimLow>{}));
+    using LowLengthsMagicDivisorMultipiler = decltype(generate_tuple(
+        lambda_merge_generate_MagicDivision_calculate_magic_multiplier<LowLengths>{},
+        Number<NDimLow>{}));
 
-    using LowLengthsMagicDivisorShift = decltype(
-        generate_tuple(lambda_merge_generate_MagicDivision_calculate_magic_shift<LowLengths>{},
-                       Number<NDimLow>{}));
+    using LowLengthsMagicDivisorShift = decltype(generate_tuple(
+        lambda_merge_generate_MagicDivision_calculate_magic_shift<LowLengths>{},
+        Number<NDimLow>{}));
 
     LowLengths low_lengths_;
     LowLengthsMagicDivisorMultipiler low_lengths_magic_divisor_multiplier_;
@@ -1201,9 +1201,9 @@ struct Merge_v2r2_magic_division
         lambda_merge_generate_MagicDivision_calculate_magic_multiplier<LowLengthsScan>{},
         Number<NDimLow>{}));
 
-    using LowLengthsScanMagicDivisorShift = decltype(
-        generate_tuple(lambda_merge_generate_MagicDivision_calculate_magic_shift<LowLengthsScan>{},
-                       Number<NDimLow>{}));
+    using LowLengthsScanMagicDivisorShift = decltype(generate_tuple(
+        lambda_merge_generate_MagicDivision_calculate_magic_shift<LowLengthsScan>{},
+        Number<NDimLow>{}));
 
     LowLengths low_lengths_;
     LowLengthsScan low_lengths_scan_;
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp b/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
index 82bcff6947..2fb7242708 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_softmax.hpp
@@ -35,8 +35,8 @@ struct BlockwiseSoftmax
     static constexpr index_t MRepeat = ThreadSliceDesc_M_K{}.GetLength(I0);
     static constexpr index_t KRepeat = ThreadSliceDesc_M_K{}.GetLength(I1);
 
-    using ThreadSliceDesc_M = decltype(
-        make_naive_tensor_descriptor_packed(make_tuple(ThreadSliceDesc_M_K{}.GetLength(I0))));
+    using ThreadSliceDesc_M = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(ThreadSliceDesc_M_K{}.GetLength(I0))));
 
     using ThreadwiseMaxReduce = typename conditional<
         IgnoreNaN,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index 3270325069..fe97117181 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -588,14 +588,18 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
         LoopSched>;
 
     // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     // block-to-e-tile map
     using Block2ETileMap =
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
index 8f46e0c498..d6b6405bb7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -378,13 +378,16 @@ struct DeviceBatchedGemmEPermuteXdl : public DeviceBatchedGemmEPermute<ALayout,
         CDEBlockTransferScalarPerVector_NPerBlock,
         LoopSched>;
 
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
 
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}));
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}));
     using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;
 
     // Argument
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index bd87126f99..3df2ee38f2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -368,14 +368,18 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
         LoopSched>;
 
     // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     // block-to-e-tile map
     using Block2ETileMap =
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
index dfc84ccf67..9455ec48b3 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle.hpp
@@ -510,12 +510,15 @@ struct DeviceBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
         CDE1ShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched>;
 
-    using A0GridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultA0GridDescriptor_AK0_M_AK1(A0GridDesc_M_K{}))>;
-    using B0GridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultB0GridDescriptor_BK0_N_BK1(B0GridDesc_N_K{}))>;
-    using B1GridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultB1GridDescriptor_BK0_N_BK1(B1GridDesc_N_K{}))>;
+    using A0GridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultA0GridDescriptor_AK0_M_AK1(
+            A0GridDesc_M_K{}))>;
+    using B0GridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultB0GridDescriptor_BK0_N_BK1(
+            B0GridDesc_N_K{}))>;
+    using B1GridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultB1GridDescriptor_BK0_N_BK1(
+            B1GridDesc_N_K{}))>;
 
     // Argument
     struct Argument : public BaseArgument
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index db263b41d6..deb938156d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -355,14 +355,18 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
         LoopSched>;
 
     // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     // block-to-e-tile map
     using Block2ETileMap =
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
index b086151e77..b0efa9d4e4 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_layernorm_xdl_cshuffle.hpp
@@ -364,11 +364,13 @@ struct DeviceGemmMultipleDLayernorm_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
     // We have to separate mean var descriptor for gemm and layernorm bacause of different grid
     // layout(different padding)
-    using GemmMeanVarGridDesc_M_NBlock = decltype(
-        MakeMeanVarDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1, 1));
+    using GemmMeanVarGridDesc_M_NBlock =
+        decltype(MakeMeanVarDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1,
+                                                                                                1));
 
-    using GemmCountGridDesc_M_NBlock = decltype(
-        MakeCountDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1, 1));
+    using GemmCountGridDesc_M_NBlock =
+        decltype(MakeCountDescriptor_M_N<Sequence<true, false>, GemmMPerBlock, GemmNPerBlock>(1,
+                                                                                              1));
 
     using LayernormMeanVarGridDesc_M_NBlock =
         decltype(MakeMeanVarDescriptor_M_N<Sequence<true, true>,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index 1f3ea43c7b..916f29a904 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -337,10 +337,12 @@ struct DeviceGemmMultipleDMultipleR_Xdl_CShuffle
         RThreadTransferDstScalarPerVector_MPerBlock,
         LoopSched>;
 
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
 
     using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index d5b07e213a..9a75add9ce 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -288,14 +288,18 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
         PipelineVer>;
 
     // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     // block-to-e-tile map
     using Block2ETileMap =
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
index a086e1a4bd..1b34e2dba2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -248,10 +248,12 @@ struct DeviceGemm_Xdl_WaveletModel_CShuffle : public DeviceGemm<ALayout,
         CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
         CShuffleBlockTransferScalarPerVector_NPerBlock>;
 
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
 
     using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 39a6dd7ab2..0b1bf0f179 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -400,14 +400,18 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
         LoopSched>;
 
     // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     struct GroupedContractionBlock2ETileMap
     {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index a9f69ab4e6..8a0a445370 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -422,10 +422,12 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
     using AGridDesc_M_K = decltype(transform_k0_m_k1_to_m_k(AGridDesc_AK0_M_AK1{}));
     using BGridDesc_N_K = decltype(transform_k0_m_k1_to_m_k(BGridDesc_BK0_N_BK1{}));
 
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}));
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}));
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}));
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}));
 
     // block-to-e-tile map
     using Block2ETileMap =
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index ea958a4eb5..8b22bd2090 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -381,8 +381,8 @@ struct DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK
     }
 
     // desc for problem definition
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        MakeAGridDescriptor_AK0_M_AK1<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(MakeAGridDescriptor_AK0_M_AK1<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
     using BGridDesc_BK0_N_BK1 =
         remove_cvref_t<decltype(MakeBGridDescriptor_BK0_N_BK1<BLayout>({}, {}))>;
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
index 9080936589..f18fbcfe4b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -320,8 +320,8 @@ struct DeviceGroupedConvFwdDl_NHWC_KYXC_NHWK : public DeviceGroupedConvFwd<NDimS
     }
 
     // desc for problem definition
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        MakeAGridDescriptor_AK0_M_AK1<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(MakeAGridDescriptor_AK0_M_AK1<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
     using BGridDesc_BK0_N_BK1 =
         remove_cvref_t<decltype(MakeBGridDescriptor_BK0_N_BK1<BLayout>({}, {}))>;
     using CGridDesc_M_N = remove_cvref_t<decltype(MakeCGridDescriptor_M_N<CLayout>({}, {}))>;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
index af8e0e0279..caa18b709c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -446,8 +446,8 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
         return GetPaddedRGridDescriptor(r_grid_desc_mraw, NHoWo);
     }
 
-    using AGridDesc_M_K = remove_cvref_t<decltype(
-        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_M_K = remove_cvref_t<decltype(MakeAGridDescriptor_M_K<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
     using BGridDesc_N_K = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
     using EGridDesc_M_N = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<DELayout>({}, {}))>;
     using RGridDesc_M   = remove_cvref_t<decltype(MakeRGridDescriptor_M<RLayout>({}, {}))>;
@@ -507,10 +507,12 @@ struct DeviceGroupedConvFwdMultipleDMultipleR_Xdl_CShuffle
         RThreadTransferDstScalarPerVector_MPerBlock,
         LoopSched>;
 
-    using AGridDesc_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
 
     using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
index 526108b874..7c726cd851 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -245,8 +245,8 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
     }
 
     // desc for problem definition
-    using AGridDesc_M_K  = remove_cvref_t<decltype(
-        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_M_K  = remove_cvref_t<decltype(MakeAGridDescriptor_M_K<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
     using BGridDesc_N_K  = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
     using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index a9294f0344..c80598c4e3 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -361,8 +361,8 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
     }
 
     // desc for problem definition
-    using AGridDesc_M_K  = remove_cvref_t<decltype(
-        MakeAGridDescriptor_M_K<ALayout>({}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using AGridDesc_M_K  = remove_cvref_t<decltype(MakeAGridDescriptor_M_K<ALayout>(
+        {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
     using BGridDesc_N_K  = remove_cvref_t<decltype(MakeBGridDescriptor_N_K<BLayout>({}, {}))>;
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
     using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
@@ -412,14 +412,18 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
         LoopSched>;
 
     // desc for blockwise copy
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     // block-to-e-tile map
     using Block2ETileMap =
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
index a7ff7f3d7e..ddd762af1c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_softmax_gemm_permute_xdl_cshuffle.hpp
@@ -735,12 +735,12 @@ struct DeviceGroupedGemmSoftmaxGemmPermute_Xdl_CShuffle
             }
 
             // Check vector load/store requirement
-            const auto a_stride_lowest = ABlockTransferSrcVectorDim == 2
-                                             ? device_arg.a_mz_kz_strides_[1]
-                                             : device_arg.a_mz_kz_strides_[0];
-            const auto b_stride_lowest = BBlockTransferSrcVectorDim == 2
-                                             ? device_arg.b_nz_kz_strides_[1]
-                                             : device_arg.b_nz_kz_strides_[0];
+            const auto a_stride_lowest  = ABlockTransferSrcVectorDim == 2
+                                              ? device_arg.a_mz_kz_strides_[1]
+                                              : device_arg.a_mz_kz_strides_[0];
+            const auto b_stride_lowest  = BBlockTransferSrcVectorDim == 2
+                                              ? device_arg.b_nz_kz_strides_[1]
+                                              : device_arg.b_nz_kz_strides_[0];
             const auto b1_stride_lowest = B1BlockTransferSrcVectorDim == 2
                                               ? device_arg.b1_nz_kz_strides_[1]
                                               : device_arg.b1_nz_kz_strides_[0];
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index 9d3fe77fe2..db89bee96c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -272,14 +272,18 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
         CDEBlockTransferScalarPerVector_NPerBlock,
         LoopSched>;
 
-    using AGridDesc_AK0_M_AK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
-    using BGridDesc_BK0_N_BK1                          = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using AGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AK0_M_AK1(
+            AGridDesc_M_K{}))>;
+    using BGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BK0_N_BK1(
+            BGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     struct GroupedGemmBlock2ETileMap
     {
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
index a52968bcef..cfab7d29c9 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_splitk_contraction_multiple_d_xdl_cshuffle.hpp
@@ -617,10 +617,12 @@ struct DeviceSplitKContractionMultipleD_Xdl_CShuffle
         CDEBlockTransferScalarPerVector_NPerBlock,
         LoopSched>;
 
-    using AGridDesc_AKB_AK0_M_AK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultAGridDescriptor_AKB_AK0_M_AK1(AGridDesc_M_K{}, 1))>;
-    using BGridDesc_BKB_BK0_N_BK1 = remove_cvref_t<decltype(
-        GridwiseGemm::MakeDefaultBGridDescriptor_BKB_BK0_N_BK1(BGridDesc_N_K{}, 1))>;
+    using AGridDesc_AKB_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultAGridDescriptor_AKB_AK0_M_AK1(
+            AGridDesc_M_K{}, 1))>;
+    using BGridDesc_BKB_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeDefaultBGridDescriptor_BKB_BK0_N_BK1(
+            BGridDesc_N_K{}, 1))>;
 
     using Block2ETileMap = typename GridwiseGemm::DefaultBlock2ETileMap;
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
index f73fe282db..47573107cf 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_batchnorm_forward.hpp
@@ -136,8 +136,8 @@ struct GridwiseMultiblockBatchNormForward
     using ThreadReduceDstDesc_M =
         decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
 
-    using ThreadReduceSrcDesc_M_1 = decltype(
-        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
+    using ThreadReduceSrcDesc_M_1 = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
 
     using ThreadwiseWelford1 =
         ThreadwiseWelford<AccDataType, ThreadReduceSrcDesc_M_K, ThreadReduceDstDesc_M>;
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
index e73e7e6817..4e182ec29d 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_reduce_second_half_batchnorm_backward_final.hpp
@@ -118,8 +118,8 @@ struct GridwiseReduceSecondHalfBatchNormBackwardFinal
     static constexpr auto thread_cluster_desc =
         make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
 
-    using ThreadReduceSrcDesc_M_1 = decltype(
-        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
+    using ThreadReduceSrcDesc_M_1 = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
     using ThreadReduceDstDesc_M =
         decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
index dd514567d2..672be91a79 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_batchnorm_forward_final_obsolete.hpp
@@ -121,8 +121,8 @@ struct GridwiseWelfordSecondHalfBatchNormForwardFinal
     static constexpr auto thread_cluster_desc =
         make_cluster_descriptor(ThreadClusterLengths_M_K{}, ThreadClusterArrangeOrder{});
 
-    using ThreadReduceSrcDesc_M_1 = decltype(
-        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
+    using ThreadReduceSrcDesc_M_1 = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
     using ThreadReduceDstDesc_M =
         decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
 
diff --git a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
index 6fe78edb34..2d5dc90bfb 100644
--- a/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
+++ b/include/ck/tensor_operation/gpu/grid/batchnorm_multiblock/gridwise_multiblock_welford_second_half_multiblock_reduce_first_half.hpp
@@ -115,8 +115,8 @@ struct GridwiseWelfordSecondHalfReduceFirstHalf
 
     using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<KThreadSliceSize>{})));
-    using ThreadReduceSrcDesc_M_1 = decltype(
-        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
+    using ThreadReduceSrcDesc_M_1 = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MThreadSliceSize>{}, Number<1>{})));
     using ThreadReduceDstDesc_M =
         decltype(make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{})));
 
diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
index 523e7f7c53..b25f136a37 100644
--- a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
@@ -101,8 +101,8 @@ struct GridwiseGemmMultipleDWelfordFirstHalf_xdl_cshuffle
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
@@ -346,14 +346,18 @@ struct GridwiseGemmMultipleDWelfordFirstHalf_xdl_cshuffle
         remove_cvref_t<decltype(MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
     using DefaultBGridDesc_BK0_N_BK1 =
         remove_cvref_t<decltype(MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
-    using MeanVarGridDescriptor_MBlock_MPerBlock_NBlock      = remove_cvref_t<decltype(
-        MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(MeanVarGridDesc_M_NBlock{}))>;
-    using CountGridDescriptor_MBlock_MPerBlock_NBlock        = remove_cvref_t<decltype(
-        MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(CountGridDesc_M_NBlock{}))>;
-    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
+    using MeanVarGridDescriptor_MBlock_MPerBlock_NBlock =
+        remove_cvref_t<decltype(MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(
+            MeanVarGridDesc_M_NBlock{}))>;
+    using CountGridDescriptor_MBlock_MPerBlock_NBlock =
+        remove_cvref_t<decltype(MakeMeanVarCountGridDescriptor_MBlock_MPerBlock_NBlock(
+            CountGridDesc_M_NBlock{}))>;
+    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
 
     using DefaultBlock2ETileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2ETileMap(EGridDesc_M_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
index a8a1f803fc..c2f47bd444 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
@@ -102,8 +102,8 @@ struct GridwiseBatchedGemmGemm_Xdl_CShuffle
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     template <typename ABlockDesc_AK0_M_AK1>
     __host__ __device__ static constexpr auto
@@ -286,8 +286,9 @@ struct GridwiseBatchedGemmGemm_Xdl_CShuffle
             c_grid_desc_m_n);
     }
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}))>;
 
     using DefaultBlock2CTileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
index 2c40defddd..d2920570e4 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
@@ -446,14 +446,17 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
             e1_grid_desc_m_n);
     }
 
-    using E1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeE1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(E1GridDesc_M_N{}))>;
+    using E1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeE1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            E1GridDesc_M_N{}))>;
 
-    using D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5 = remove_cvref_t<decltype(
-        MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(D0sGridDesc_M_N{}))>;
+    using D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5 =
+        remove_cvref_t<decltype(MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(
+            D0sGridDesc_M_N{}))>;
 
-    using D1sGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeD1sGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(D1sGridDesc_M_N{}))>;
+    using D1sGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeD1sGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            D1sGridDesc_M_N{}))>;
 
     using DefaultBlock2E1TileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2E1TileMap(E1GridDesc_M_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
index 135b9da6a0..18cfeebcf3 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -114,8 +114,8 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     template <typename ABlockDesc_AK0_M_AK1>
     __host__ __device__ static constexpr auto
@@ -368,12 +368,14 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
             Number<NumD0Tensor>{});
     }
 
-    using D0sGridPointer                                  = decltype(MakeD0sGridPointer());
-    using D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5 = remove_cvref_t<decltype(
-        MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(D0sGridDesc_M_N{}))>;
+    using D0sGridPointer = decltype(MakeD0sGridPointer());
+    using D0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5 =
+        remove_cvref_t<decltype(MakeD0sGridDescriptor_M0_N0_M1_N1_M2_N2_M3_N3_N4_N5(
+            D0sGridDesc_M_N{}))>;
 
-    using C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeC1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(C1GridDesc_M_N{}))>;
+    using C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeC1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            C1GridDesc_M_N{}))>;
 
     using DefaultBlock2CTileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(C1GridDesc_M_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
index 9eb2bf8aa8..f4b82badf1 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -113,8 +113,8 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     template <typename ABlockDesc_AK0_M_AK1>
     __host__ __device__ static constexpr auto
@@ -300,8 +300,9 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
             c_grid_desc_m_n);
     }
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}))>;
 
     using DefaultBlock2CTileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
index 533559adf1..3ced4b9ad6 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
@@ -191,8 +191,8 @@ struct GridwiseGemmBiasAddReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
@@ -346,14 +346,17 @@ struct GridwiseGemmBiasAddReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             c_grid_desc_m_n);
     }
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}))>;
 
-    using C0GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(C0GridDesc_M_N{}))>;
+    using C0GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            C0GridDesc_M_N{}))>;
 
-    using C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(C1GridDesc_M_N{}))>;
+    using C1GridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            C1GridDesc_M_N{}))>;
 
     using ReduceGridDescriptor_MBlock_MPerBlock =
         remove_cvref_t<decltype(MakeReduceGridDescriptor_MBlock_MPerBlock(ReduceGridDesc_M{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index 7289a20da0..d710fc1894 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -92,8 +92,8 @@ struct GridwiseGemmMultipleDMultipleR_k0mk1_k0nk1_mn_xdl_cshuffle_v1
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
@@ -300,8 +300,9 @@ struct GridwiseGemmMultipleDMultipleR_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         remove_cvref_t<decltype(MakeDefaultAGridDescriptor_AK0_M_AK1(AGridDesc_M_K{}))>;
     using DefaultBGridDesc_BK0_N_BK1 =
         remove_cvref_t<decltype(MakeDefaultBGridDescriptor_BK0_N_BK1(BGridDesc_N_K{}))>;
-    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     // Support 2 dimension in the future. Not only M
     using RGridDescriptor_MBlock_MPerBlock =
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
index 119c1ea596..98ade85a30 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
@@ -346,8 +346,8 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
     {
@@ -565,10 +565,12 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
             e_grid_desc_m_n);
     }
 
-    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
-    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
     using DefaultBlock2CTileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(EGridDesc_M_N{}, 1, 1))>;
     using DsGridPointer = decltype(MakeDsGridPointer());
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index 5437379018..3b8a5ec8f5 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -89,8 +89,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     // denorm test fix, required to work around fp16 mfma issue
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
index 3d1fbb73bd..99e410f688 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
@@ -164,8 +164,8 @@ struct GridwiseGemmReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
@@ -318,8 +318,9 @@ struct GridwiseGemmReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             c_grid_desc_m_n);
     }
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}))>;
 
     using ReduceGridDescriptor_MBlock_MPerBlock =
         remove_cvref_t<decltype(MakeReduceGridDescriptor_MBlock_MPerBlock(ReduceGridDesc_M{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
index e7577bdcbd..18cf80041b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
@@ -375,10 +375,12 @@ struct GridwiseGemmSplitKMultipleD_xdl_cshuffle
         remove_cvref_t<decltype(MakeDefaultAGridDescriptor_AKB_AK0_M_AK1(AGridDesc_M_K{}, 1))>;
     using DefaultBGridDesc_BK0_N_BK1 =
         remove_cvref_t<decltype(MakeDefaultBGridDescriptor_BKB_BK0_N_BK1(BGridDesc_N_K{}, 1))>;
-    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock  = remove_cvref_t<decltype(
-        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
-    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(DsGridDesc_M_N{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
+    using DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
 
     using DefaultBlock2ETileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2ETileMap(EGridDesc_M_N{}, 1))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
index 740dde5c65..d8b31311b1 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp
@@ -138,8 +138,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
     {
@@ -308,8 +308,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
             c_grid_desc_m_n);
     }
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}))>;
     using DefaultBlock2CTileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2CTileMap(CGridDesc_M_N{}, 1, 1))>;
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index 29cbf14a57..9c09f3a539 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -491,8 +491,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
     };
 
     // FIXME: pass GridwiseGemmPipe as a template arguement into GridwiseGemm
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
index 8675350094..0404d88ab8 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
@@ -173,8 +173,8 @@ struct GridwiseGemmLayernorm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
@@ -345,8 +345,9 @@ struct GridwiseGemmLayernorm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             c_grid_desc_m_n);
     }
 
-    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(CGridDesc_M_N{}))>;
+    using CGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeCGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}))>;
 
     using C0GridDescriptor_NBlock_NPerBlock =
         remove_cvref_t<decltype(MakeC0GridDescriptor_NBlock_NPerBlock(C0GridDesc_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
index 31c59d14ed..bbd01a238e 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -330,8 +330,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_waveletmodel_cshuffle
         return e_grid_desc_mblock_mperblock_nblock_nperblock;
     }
 
-    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<decltype(
-        MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(EGridDesc_M_N{}))>;
+    using EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
 
     using DefaultBlock2ETileMap =
         remove_cvref_t<decltype(MakeDefaultBlock2ETileMap(EGridDesc_M_N{}))>;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index 0fc18bb923..0920a17fc7 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -259,8 +259,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     // denorm test fix, required to work around fp16 mfma issue
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index 3d1b13f998..31aee70db2 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -247,8 +247,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         FloatC* p_c_grid;
     };
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     // denorm test fix, required to work around fp16 mfma issue
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 0c94e28e9b..371281dfe2 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -110,8 +110,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
     struct Argument : public ck::tensor_operation::device::BaseArgument
     {
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
index d090ba54d8..8d7bd9a8d1 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
@@ -139,8 +139,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r1
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
     {
@@ -315,8 +315,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r1
             c_grid_desc_m_n);
     }
     using CGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl =
-        remove_cvref_t<decltype(
-            MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
+        remove_cvref_t<
+            decltype(MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
                 CGridDesc_M_N{}))>;
 
     using DefaultBlock2CTileMap =
@@ -634,10 +634,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r1
                 Sequence<0, 1, 2, 3, 4, 5>, // typename ThreadClusterArrangeOrder,
                 FloatCShuffle,              // typename SrcData,
                 FloatC,                     // typename DstData,
-                decltype(
-                    c_block_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
-                decltype(
-                    c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c_block_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
                 Sequence<0, 1, 2, 3, 4, 5>,                 // typename DimAccessOrder,
                 5,                                          // index_t VectorDim,
                 CBlockTransferScalarPerVector_NWaveNPerXdl, // index_t ScalarPerVector,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
index 2c09e80fdc..79202cb5cf 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
@@ -142,8 +142,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r2
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
     {
@@ -323,13 +323,13 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r2
     }
 
     using CGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl =
-        remove_cvref_t<decltype(
-            MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
+        remove_cvref_t<
+            decltype(MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
                 CGridDesc_M_N{}))>;
 
     using C0GridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl =
-        remove_cvref_t<decltype(
-            MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
+        remove_cvref_t<
+            decltype(MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
                 C0GridDesc_M_N{}))>;
 
     using DefaultBlock2CTileMap =
@@ -654,12 +654,9 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r2
                 FloatC,                     // typename Src0Data,
                 FloatC,                     // typename Src1Data,
                 FloatC,                     // typename DstData,
-                decltype(
-                    c_block_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
-                decltype(
-                    c0_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
-                decltype(
-                    c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c_block_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c0_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
                 Sequence<0, 1, 2, 3, 4, 5>,                 // typename DimAccessOrder,
                 5,                                          // index_t VectorDim,
                 CBlockTransferScalarPerVector_NWaveNPerXdl, // index_t ScalarPerVector,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
index c1bc6a8fec..0d461b4fbf 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
@@ -151,8 +151,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r3
 
     using ThisThreadBlock = ThisThreadBlock<BlockSize>;
 
-    using GridwiseGemmPipe = remove_cvref_t<decltype(
-        GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_K0PerBlock_MPerBlock_K1()
     {
@@ -331,18 +331,18 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r3
             c_grid_desc_m_n);
     }
     using CGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl =
-        remove_cvref_t<decltype(
-            MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
+        remove_cvref_t<
+            decltype(MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
                 CGridDesc_M_N{}))>;
 
     using C0GridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl =
-        remove_cvref_t<decltype(
-            MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
+        remove_cvref_t<
+            decltype(MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
                 C0GridDesc_M_N{}))>;
 
     using C1GridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl =
-        remove_cvref_t<decltype(
-            MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
+        remove_cvref_t<
+            decltype(MakeCGridDescriptor_MBlock_MXdlPerWave_MWaveMPerXdl_NBlock_NXdlPerWave_NWaveNPerXdl(
                 C1GridDesc_M_N{}))>;
 
     using DefaultBlock2CTileMap =
@@ -674,14 +674,10 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r3
                 FloatC,                     // typename Src1Data,
                 FloatC,                     // typename Src2Data,
                 FloatC,                     // typename DstData,
-                decltype(
-                    c_block_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
-                decltype(
-                    c0_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
-                decltype(
-                    c1_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
-                decltype(
-                    c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c_block_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c0_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c1_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
+                decltype(c_grid_desc_mblock_mxdlperwave_mwavemperxdl_nblock_nxdlperwave_nwavenperxdl),
                 Sequence<0, 1, 2, 3, 4, 5>,                 // typename DimAccessOrder,
                 5,                                          // index_t VectorDim,
                 CBlockTransferScalarPerVector_NWaveNPerXdl, // index_t ScalarPerVector,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
index ee68660a06..287b4e5421 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_sparse_embeddings_forward_layernorm.hpp
@@ -78,8 +78,8 @@ struct GridwiseSparseEmbeddingsForwardLayernorm
     using ThreadwiseWolfordDesc2D = decltype(make_naive_tensor_descriptor_packed(make_tuple(
         Number<DimSubBlocks * DimThreadSize>{}, Number<RowSubBlocks * RowVectorSize>{})));
 
-    using ThreadwiseWolfordDescReduce = decltype(
-        make_naive_tensor_descriptor_packed(make_tuple(Number<DimSubBlocks * DimThreadSize>{})));
+    using ThreadwiseWolfordDescReduce = decltype(make_naive_tensor_descriptor_packed(
+        make_tuple(Number<DimSubBlocks * DimThreadSize>{})));
 
     using ThreadwiseWelford =
         ThreadwiseWelford<AccDataType, ThreadwiseWolfordDesc2D, ThreadwiseWolfordDescReduce>;
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
index 0fb961eb7a..80e9a84f96 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_1st.hpp
@@ -87,9 +87,9 @@ struct GridwiseNormalizationSplitK1st
             int left_kPerBlock  = math::integer_divide_ceil(k, kGridSize);
             int kRightmostBlock = kRaw - left_kPerBlock * (kGridSize - 1);
             int kPerThread      = kRightmostBlock < K_BlockTileSize
-                                 ? 0
-                                 : KThreadSliceSize * (kRightmostBlock / K_BlockTileSize);
-            int kPerBlockTail = kRightmostBlock - kPerThread * KThreadClusterSize;
+                                      ? 0
+                                      : KThreadSliceSize * (kRightmostBlock / K_BlockTileSize);
+            int kPerBlockTail   = kRightmostBlock - kPerThread * KThreadClusterSize;
 
             if(kPerBlockTail > 0)
             {
diff --git a/include/ck/version.h.in b/include/ck/version.h.in
new file mode 100644
index 0000000000..0d6a6512fb
--- /dev/null
+++ b/include/ck/version.h.in
@@ -0,0 +1,40 @@
+/*******************************************************************************
+ *
+ * MIT License
+ *
+ * Copyright (c) 2023 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ *******************************************************************************/
+
+/* the configured version and settings for miopen- Composable Kernel */
+
+#ifndef CK_VERSION_H_
+#define CK_VERSION_H_
+
+// clang-format off
+#define CK_VERSION @CMAKE_PROJECT_VERSION@
+#define CK_VERSION_MAJOR @CMAKE_PROJECT_VERSION_MAJOR@
+#define CK_VERSION_MINOR @CMAKE_PROJECT_VERSION_MINOR@
+#define CK_VERSION_PATCH @CMAKE_PROJECT_VERSION_PATCH@
+#define CK_COMMIT_ID @COMMIT_ID@
+// clang-format on
+
+#endif
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
index a89358538e..8f15e80794 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
@@ -37,7 +37,7 @@ void add_device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instances(
         DeviceBatchedGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -58,7 +58,7 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances(
         DeviceBatchedGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -79,7 +79,7 @@ void add_device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instances(
         DeviceBatchedGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemm<Col,
                                                   Row,
@@ -154,7 +154,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<CDataType, float>)
         {
@@ -180,7 +180,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
             }
         }
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<CDataType, half_t>)
         {
@@ -206,7 +206,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
             }
         }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
         if constexpr(is_same_v<ADataType, bhalf_t> && is_same_v<BDataType, bhalf_t> &&
                      is_same_v<CDataType, bhalf_t>)
         {
@@ -232,7 +232,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
             }
         }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                      is_same_v<CDataType, int8_t>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
index 7f326879f3..462b7f1102 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add.hpp
@@ -14,7 +14,7 @@
 
 using CDE0ElementOp = ck::tensor_operation::element_wise::AddRelu;
 using CDE1ElementOp = ck::tensor_operation::element_wise::Add;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
index 0b2a0c729b..b471268765 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_permute.hpp
@@ -13,7 +13,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
index bfe221af7b..161085e4d0 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_bias_softmax_gemm_permute.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -59,7 +59,7 @@ void add_device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_batched_gemm_bias_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -148,7 +148,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<B0DataType, half_t> &&
                      is_same_v<B1DataType, half_t> && is_same_v<CDataType, half_t> &&
                      Acc0BiasDataType::Size() == 1 &&
@@ -166,7 +166,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
         else if constexpr(is_same_v<ADataType, BF16> && is_same_v<B0DataType, BF16> &&
                           is_same_v<B1DataType, BF16> && is_same_v<CDataType, BF16> &&
                           Acc0BiasDataType::Size() == 1 &&
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
index e57a4206d5..77ad36b97b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_gemm.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance(
     std::vector<std::unique_ptr<DeviceBatchedGemmGemm<Row,
                                                       Col,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
index 4cb81e8770..c9535e828b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_multi_d.hpp
@@ -19,7 +19,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
                                                         Row,
@@ -124,7 +124,7 @@ void add_device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instan
                                                         PassThrough,
                                                         PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instances(
     std::vector<std::unique_ptr<DeviceBatchedGemmMultiD<Col,
                                                         Row,
@@ -263,7 +263,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<EDataType, half_t>)
         {
@@ -297,7 +297,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceBatche
             }
         }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<EDataType, int8_t>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
index c6df0d2441..88bdaea23a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
index c1d7bf9002..1814cfabb4 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -59,7 +59,7 @@ void add_device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_g
                                                             MaskingSpecialization::MaskDisabled>>>&
         instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_batched_gemm_masking_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instances(
     std::vector<std::unique_ptr<
         DeviceBatchedGemmSoftmaxGemmPermute<2,
@@ -148,7 +148,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<B0DataType, half_t> &&
                      is_same_v<B1DataType, half_t> && is_same_v<CDataType, half_t>)
         {
@@ -164,7 +164,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
         else if constexpr(is_same_v<ADataType, BF16> && is_same_v<B0DataType, BF16> &&
                           is_same_v<B1DataType, BF16> && is_same_v<CDataType, BF16>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index da391feef3..bd3af891ec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,7 +66,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            PassThrough,
                                                            Bilinear>>>& instances);
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
 // double
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -150,7 +150,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<DDataType, float> && is_same_v<EDataType, float>)
         {
@@ -167,7 +167,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
             }
         }
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<DDataType, double> && is_same_v<EDataType, double>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index 6b11523392..4dd735d062 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,7 +66,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            PassThrough,
                                                            Scale>>>& instances);
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
 // double
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -149,7 +149,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<EDataType, float>)
         {
@@ -166,7 +166,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
             }
         }
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<EDataType, double>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index 8ec9832896..571ff0b672 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 // conv1d backward data
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
@@ -30,19 +30,19 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvBwdData<1, NWC, KXC, NWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<1,
                                                   NWC,
@@ -55,7 +55,7 @@ void add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_int8_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 // conv2d backward data
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -69,7 +69,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -82,7 +82,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -95,7 +95,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -109,7 +109,7 @@ void add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough>>>& instances);
 #endif
 #ifdef DL_KERNELS
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 // conv2d dl
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
@@ -123,7 +123,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -136,7 +136,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<2,
                                                   NHWC,
@@ -150,7 +150,7 @@ void add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instances(
                                                   PassThrough>>>& instances);
 #endif
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 // conv3d backward data
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
@@ -164,7 +164,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -177,7 +177,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -190,7 +190,7 @@ void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(
                                                   PassThrough,
                                                   PassThrough>>>& instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvBwdData<3,
                                                   NDHWC,
@@ -245,21 +245,21 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(op_ptrs);
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f16_instances(op_ptrs);
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
@@ -278,7 +278,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
                 add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
 #endif
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
@@ -288,14 +288,14 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
 #endif
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
@@ -314,21 +314,21 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(op_ptrs);
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f16_instances(op_ptrs);
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                          is_same_v<OutDataType, int8_t>)
             {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
index af4c2aee66..ad2da3364f 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
@@ -18,7 +18,7 @@ namespace device {
 namespace instance {
 
 // conv2d forward
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(
     std::vector<std::unique_ptr<
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -28,7 +28,7 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvFwd<2,
                                               NHWC,
@@ -41,13 +41,13 @@ void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(
                                               PassThrough,
                                               PassThrough>>>& instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(
     std::vector<std::unique_ptr<
         DeviceConvFwd<2, NHWC, KYXC, NHWK, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceConvFwd<2,
                                               NHWC,
@@ -103,7 +103,7 @@ struct DeviceOperationInstanceFactory<
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
             }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
             {
@@ -111,7 +111,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
             }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&
                               is_same_v<OutDataType, ck::bhalf_t>)
@@ -119,7 +119,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
index b863243c1b..931110267a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/elementwise_normalization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 06197d9f46..9310fd433b 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -17,7 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-#if defined(__fp16__) && defined(DL_KERNELS)
+#if defined(CK_ENABLE_FP16) && defined(DL_KERNELS)
 void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -78,7 +78,7 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#if defined(__fp32__) && defined(DL_KERNELS)
+#if defined(CK_ENABLE_FP32) && defined(DL_KERNELS)
 void add_device_gemm_dl_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -99,7 +99,7 @@ void add_device_gemm_dl_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#if defined(__int8__) && defined(DL_KERNELS)
+#if defined(CK_ENABLE_INT8) && defined(DL_KERNELS)
 void add_device_gemm_dl_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -140,7 +140,7 @@ void add_device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instances(
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 void add_device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
@@ -161,7 +161,7 @@ void add_device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, int8_t, int8_t, int8_t, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -208,7 +208,7 @@ void add_device_gemm_xdl_f16_f16_f16_mk_nk_mn_instances(
         instances);
 
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
@@ -229,7 +229,7 @@ void add_device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, BF16, BF16, BF16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 void add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -270,7 +270,7 @@ void add_device_gemm_xdl_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
-#ifdef __fp64__
+#ifdef CK_ENABLE_FP64
 void add_device_gemm_xdl_f64_f64_f64_km_kn_mn_instances(
     std::vector<std::unique_ptr<
 
@@ -363,7 +363,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instances(op_ptrs);
             }
         }
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -414,7 +414,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
         else if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
                           is_same_v<CDataType, ck::bhalf_t>)
         {
@@ -440,7 +440,7 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if constexpr(is_same_v<ADataType, int8_t> && is_same_v<BDataType, int8_t> &&
                           is_same_v<CDataType, int8_t>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
index 4b11a9b262..dd8ecae62c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm.hpp
@@ -9,7 +9,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_layernorm.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index 09dc40e3ec..9d8eaada22 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
index c30ba4a1f5..f6612fc259 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_fastgelu.hpp
@@ -10,7 +10,7 @@
 #include "ck/ck.hpp"
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
index 963a18cb44..2df378b0c6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_streamk.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_xdl_streamk_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmStreamK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
index 0cd2a262ec..070c7e5b17 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm.hpp
@@ -10,7 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
index 21d3b88799..8e90a7ea98 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 // FP16
 void add_device_normalization_rank_2_1_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 2, 1>>>&);
@@ -27,7 +27,7 @@ void add_device_normalization_rank_4_3_f16_instances(
 void add_device_normalization_rank_5_3_f16_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 5, 3>>>&);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // FP32
 void add_device_normalization_rank_2_1_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, PassThrough, 2, 1>>>&);
@@ -66,7 +66,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
                      is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
         {
@@ -84,7 +84,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
             }
         }
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
                      is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
         {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
index ae0514faa0..f520c7eded 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -22,7 +22,7 @@ static constexpr auto WindowRank = 3;
 
 static constexpr auto MaxOp = ck::ReduceTensorOp::MAX;
 static constexpr auto AvgOp = ck::ReduceTensorOp::AVG;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 // FP16
 void add_device_pool3d_fwd_ndhwc_f16_instances(
     std::vector<std::unique_ptr<
@@ -37,7 +37,7 @@ void add_device_pool3d_fwd_ndhwc_index_f16_instances(
     std::vector<std::unique_ptr<
         DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 // FP32
 void add_device_pool3d_fwd_ndhwc_f32_instances(
     std::vector<std::unique_ptr<
@@ -84,7 +84,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
         if constexpr(is_same_v<InLayout, NDHWC> && is_same_v<OutLayout, NDHWC>)
         {
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, F16> && is_same_v<OutDataType, F16> &&
                          is_same_v<IndexDataType, I32>)
             {
@@ -98,7 +98,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 }
             }
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
                          is_same_v<IndexDataType, I32>)
             {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
index 04d184f981..19600a90f8 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/gemm_quantization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
index 733ac516fb..d9f1cdceb9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perchannel_quantization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
index e487c5fab9..9f3919e394 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_bias_forward_perlayer_quantization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
index a883b2b935..1845cb8a95 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perchannel_quantization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
index 0a71571ba7..c2f41de783 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/quantization/grouped_convolution_forward_perlayer_quantization.hpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
index c9c1475f18..f6fddc74fb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise.hpp
@@ -89,13 +89,13 @@ void add_device_reduce_instance_blockwise(
 {
     static_for<0, std::tuple_size<reduce_configuration_1_instances_blockwise>::value, 1>{}(
         [&](auto i) {
-            using cfg1 = remove_cvref_t<decltype(
-                std::get<i.value>(reduce_configuration_1_instances_blockwise{}))>;
+            using cfg1 = remove_cvref_t<decltype(std::get<i.value>(
+                reduce_configuration_1_instances_blockwise{}))>;
 
             static_for<0, std::tuple_size<reduce_configuration_2_instances_blockwise>::value, 1>{}(
                 [&](auto j) {
-                    using cfg2 = remove_cvref_t<decltype(
-                        std::get<j.value>(reduce_configuration_2_instances_blockwise{}))>;
+                    using cfg2 = remove_cvref_t<decltype(std::get<j.value>(
+                        reduce_configuration_2_instances_blockwise{}))>;
 
                     using ReduceOpInstance =
                         DeviceReduceMultiBlock<InDataType,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
index bb45295ed8..5a65a2ff88 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add.hpp
@@ -90,14 +90,14 @@ void add_device_reduce_instance_multiblock_atomic_add(
     static_for<0,
                std::tuple_size<reduce_configuration_1_instances_multiblock_atomic_add>::value,
                1>{}([&](auto i) {
-        using cfg1 = remove_cvref_t<decltype(
-            std::get<i.value>(reduce_configuration_1_instances_multiblock_atomic_add{}))>;
+        using cfg1 = remove_cvref_t<decltype(std::get<i.value>(
+            reduce_configuration_1_instances_multiblock_atomic_add{}))>;
 
         static_for<0,
                    std::tuple_size<reduce_configuration_2_instances_multiblock_atomic_add>::value,
                    1>{}([&](auto j) {
-            using cfg2 = remove_cvref_t<decltype(
-                std::get<j.value>(reduce_configuration_2_instances_multiblock_atomic_add{}))>;
+            using cfg2 = remove_cvref_t<decltype(std::get<j.value>(
+                reduce_configuration_2_instances_multiblock_atomic_add{}))>;
 
             using ReduceOpInstance = DeviceReduceMultiBlock<InDataType,
                                                             AccDataType,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
index 04bab9d283..f3267c0d02 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise.hpp
@@ -77,8 +77,8 @@ void add_device_reduce_instance_threadwise(
 
     static_for<0, std::tuple_size<reduce_configuration_2_instances_threadwise>::value, 1>{}(
         [&](auto j) {
-            using cfg2 = remove_cvref_t<decltype(
-                std::get<j.value>(reduce_configuration_2_instances_threadwise{}))>;
+            using cfg2 = remove_cvref_t<decltype(std::get<j.value>(
+                reduce_configuration_2_instances_threadwise{}))>;
 
             using ReduceOpInstance = DeviceReduceThreadWise<InDataType,
                                                             AccDataType,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
index f03eb36d31..0a25f8e6ad 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/softmax.hpp
@@ -40,7 +40,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftma
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
         if constexpr(std::is_same_v<InDataType, F16> && std::is_same_v<AccDataType, F32> &&
                      std::is_same_v<OutDataType, F16>)
         {
@@ -66,7 +66,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceSoftma
             }
         }
 #endif
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
         if constexpr(std::is_same_v<InDataType, F32> && std::is_same_v<AccDataType, F32> &&
                      std::is_same_v<OutDataType, F32>)
         {
diff --git a/library/include/ck/library/utility/fill.hpp b/library/include/ck/library/utility/fill.hpp
index c01e139ea0..4e075df43b 100644
--- a/library/include/ck/library/utility/fill.hpp
+++ b/library/include/ck/library/utility/fill.hpp
@@ -102,9 +102,10 @@ struct FillMonotonicSeq
     }
 
     template <typename ForwardRange>
-    auto operator()(ForwardRange&& range) const -> std::void_t<decltype(
-        std::declval<const FillMonotonicSeq&>()(std::begin(std::forward<ForwardRange>(range)),
-                                                std::end(std::forward<ForwardRange>(range))))>
+    auto operator()(ForwardRange&& range) const
+        -> std::void_t<decltype(std::declval<const FillMonotonicSeq&>()(
+            std::begin(std::forward<ForwardRange>(range)),
+            std::end(std::forward<ForwardRange>(range))))>
     {
         (*this)(std::begin(std::forward<ForwardRange>(range)),
                 std::end(std::forward<ForwardRange>(range)));
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index ca8ceb2fa1..366d1fe160 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 4bd714980a..36610ae205 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -11,7 +11,7 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
index 8eb423f6f1..63c612523f 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
index 93d3cfe57b..0f3b9e7939 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
index 204612d877..14f9b5cd6a 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
index 25bec9f8a6..3f641cdadc 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/impl/device_conv2d_fwd_xdl_nhwc_kyxc_nhwk.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
index 8377b13ba6..05f399e218 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
index 2976c3ad1f..d1eb8edf97 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
index db4ca19a8f..8df3bb9641 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
index e6643ff085..3de5458e05 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
index 872eb28caf..8969983356 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
index 1c19cdd15c..745a4bb31d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
index 75aac31a26..2bda30f82c 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
index 87fda6cbe0..b5e8b8ecd6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
index 2c8b2e3acb..d02fb8f70b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
index e5156c4ce8..abf79262e6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
index 3aa4ea90f6..5da89c3421 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
index a5de884004..0ade7a61ce 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
@@ -8,7 +8,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/profiler/src/profile_batched_gemm_multi_d.cpp b/profiler/src/profile_batched_gemm_multi_d.cpp
index 8c0c75c1bc..7cd4636d98 100644
--- a/profiler/src/profile_batched_gemm_multi_d.cpp
+++ b/profiler/src/profile_batched_gemm_multi_d.cpp
@@ -71,7 +71,7 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
     const int BatchCount = std::stoi(argv[17]);
 
     using F16 = ck::half_t;
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
     using INT8 = int8_t;
 #endif
 
@@ -165,7 +165,7 @@ int profile_batched_gemm_multi_d(int argc, char* argv[])
     {
         return profile(F16{}, F16{}, F16{}, Col{}, Col{}, Row{});
     }
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
     else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(INT8{}, INT8{}, INT8{}, Row{}, Row{}, Row{});
diff --git a/profiler/src/profile_conv_bwd_data.cpp b/profiler/src/profile_conv_bwd_data.cpp
index f8314b5169..e08a39aeb0 100644
--- a/profiler/src/profile_conv_bwd_data.cpp
+++ b/profiler/src/profile_conv_bwd_data.cpp
@@ -77,7 +77,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
     using F32  = float;
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
     using INT8 = int8_t;
 #endif
 
@@ -140,7 +140,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
         {
             return profile(I1, NWC{}, KXC{}, NWK{}, BF16{}, BF16{}, BF16{});
         }
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if(data_type == ConvDataType::INT8_INT8_INT8)
         {
             return profile(I1, NWC{}, KXC{}, NWK{}, INT8{}, INT8{}, INT8{});
@@ -161,7 +161,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
         {
             return profile(I2, NHWC{}, KYXC{}, NHWK{}, BF16{}, BF16{}, BF16{});
         }
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if(data_type == ConvDataType::INT8_INT8_INT8)
         {
             return profile(I2, NHWC{}, KYXC{}, NHWK{}, INT8{}, INT8{}, INT8{});
@@ -182,7 +182,7 @@ int profile_conv_bwd_data(int argc, char* argv[])
         {
             return profile(I3, NDHWC{}, KZYXC{}, NDHWK{}, BF16{}, BF16{}, BF16{});
         }
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
         else if(data_type == ConvDataType::INT8_INT8_INT8)
         {
             return profile(I3, NDHWC{}, KZYXC{}, NDHWK{}, INT8{}, INT8{}, INT8{});
diff --git a/profiler/src/profile_gemm.cpp b/profiler/src/profile_gemm.cpp
index d1f5debc7d..9ca7fc4c88 100644
--- a/profiler/src/profile_gemm.cpp
+++ b/profiler/src/profile_gemm.cpp
@@ -69,10 +69,10 @@ int profile_gemm(int argc, char* argv[])
 
     using F32 = float;
     using F16 = ck::half_t;
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
     using BF16 = ck::bhalf_t;
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
     using INT8  = int8_t;
     using INT32 = int32_t;
 #endif
@@ -123,7 +123,7 @@ int profile_gemm(int argc, char* argv[])
 
     if(false)
         ;
-#ifdef __fp32__
+#ifdef CK_ENABLE_FP32
     else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, F32{}, F32{}, F32{}, F32{});
@@ -141,7 +141,7 @@ int profile_gemm(int argc, char* argv[])
         return profile(Col{}, Col{}, Row{}, F32{}, F32{}, F32{}, F32{});
     }
 #endif
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, F16{}, F16{}, F32{}, F16{});
@@ -159,7 +159,7 @@ int profile_gemm(int argc, char* argv[])
         return profile(Col{}, Col{}, Row{}, F16{}, F16{}, F32{}, F16{});
     }
 #endif
-#ifdef __bf16__
+#ifdef CK_ENABLE_BF16
     else if(data_type == GemmDataType::BF16_BF16_BF16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, BF16{}, BF16{}, F32{}, BF16{});
@@ -177,7 +177,7 @@ int profile_gemm(int argc, char* argv[])
         return profile(Col{}, Col{}, Row{}, BF16{}, BF16{}, F32{}, BF16{});
     }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
     else if(data_type == GemmDataType::INT8_INT8_INT8 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(Row{}, Row{}, Row{}, INT8{}, INT8{}, INT32{}, INT8{});
diff --git a/profiler/src/profile_grouped_gemm.cpp b/profiler/src/profile_grouped_gemm.cpp
index 1d4707a6c8..5636656ba3 100644
--- a/profiler/src/profile_grouped_gemm.cpp
+++ b/profiler/src/profile_grouped_gemm.cpp
@@ -88,7 +88,7 @@ int profile_grouped_gemm(int argc, char* argv[])
     const auto StrideBs = argToIntArray(argv[12]);
     const auto StrideCs = argToIntArray(argv[13]);
     const int kbatch    = argc == 15 ? std::stoi(argv[14]) : 1;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
     if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         ck::profiler::profile_grouped_gemm_impl<ck::half_t,
diff --git a/script/clang-format-overwrite.sh b/script/clang-format-overwrite.sh
index f9d11fcd8c..da83254f00 100755
--- a/script/clang-format-overwrite.sh
+++ b/script/clang-format-overwrite.sh
@@ -1,2 +1,2 @@
-#find . -name deps -prune -o -name build -prune -o -iname '*.h' -o -iname '*.hpp' -o -iname '*.cpp' -o -iname '*.h.in' -o -iname '*.hpp.in' -o -iname '*.cpp.in' -o -iname '*.cl' -o -iname '*.cuh' -o -iname '*.cu' -o -iname '*.inc' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-10 -i -style=file {}'
-git status --porcelain | awk '$1 != "D" && (match($2, "\\.cpp|hpp|inc")) {print $2}' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-10 -i -style=file {}'
+#find . -name deps -prune -o -name build -prune -o -iname '*.h' -o -iname '*.hpp' -o -iname '*.cpp' -o -iname '*.h.in' -o -iname '*.hpp.in' -o -iname '*.cpp.in' -o -iname '*.cl' -o -iname '*.cuh' -o -iname '*.cu' -o -iname '*.inc' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-12 -i -style=file {}'
+git status --porcelain | awk '$1 != "D" && (match($2, "\\.cpp|hpp|inc")) {print $2}' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-12 -i -style=file {}'
diff --git a/script/install_precommit.sh b/script/install_precommit.sh
index daa5cb29ae..296280bb03 100755
--- a/script/install_precommit.sh
+++ b/script/install_precommit.sh
@@ -11,7 +11,7 @@ run_and_check() {
 }
 
 echo "I: Installing tools required for pre-commit checks..."
-run_and_check apt install clang-format-10
+run_and_check apt install clang-format-12
 
 echo "I: Installing pre-commit itself..."
 run_and_check pip3 install pre-commit
diff --git a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
index 5ccae83805..6c04086e0e 100644
--- a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
+++ b/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
@@ -71,6 +71,6 @@ TYPED_TEST_SUITE(TestBatchedGemmMultiD, KernelTypes);
 #ifdef __fp16
 TYPED_TEST(TestBatchedGemmMultiD, f16) { this->template Run<F16>(); }
 #endif
-#ifdef __int8__
+#ifdef CK_ENABLE_INT8
 TYPED_TEST(TestBatchedGemmMultiD, int8) { this->template Run<int8_t>(); }
 #endif
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool_fwd/test_avg_pool3d_fwd.cpp
index 81849ffa6a..fc196a8a07 100644
--- a/test/pool_fwd/test_avg_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_avg_pool3d_fwd.cpp
@@ -43,7 +43,7 @@ class TestAvgPool3dFwd : public ::testing::Test
         }
     }
 };
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 using KernelTypes =
     ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
 #else
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool_fwd/test_max_pool3d_fwd.cpp
index cd79c1ad8e..7189f1b104 100644
--- a/test/pool_fwd/test_max_pool3d_fwd.cpp
+++ b/test/pool_fwd/test_max_pool3d_fwd.cpp
@@ -66,7 +66,7 @@ class TestMaxPool3dFwd : public ::testing::Test
     }
 };
 
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 using KernelTypes =
     ::testing::Types<std::tuple<F16, F16, F32, I32>, std::tuple<F32, F32, F32, I32>>;
 #else
diff --git a/test/softmax/test_softmax_rank3.cpp b/test/softmax/test_softmax_rank3.cpp
index f17672286f..a8b950ce6b 100644
--- a/test/softmax/test_softmax_rank3.cpp
+++ b/test/softmax/test_softmax_rank3.cpp
@@ -10,7 +10,7 @@
 
 template <ck::index_t N>
 using I = ck::Number<N>;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 using F16 = ck::half_t;
 #endif
 using F32 = float;
@@ -23,7 +23,7 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 // clang-format off
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
     std::tuple<       F16,         F32,         F16,    I<3>>,
 #endif
     std::tuple<       F32,         F32,         F32,    I<3>>
diff --git a/test/softmax/test_softmax_rank4.cpp b/test/softmax/test_softmax_rank4.cpp
index 161aebf617..cbd790ac9b 100644
--- a/test/softmax/test_softmax_rank4.cpp
+++ b/test/softmax/test_softmax_rank4.cpp
@@ -10,7 +10,7 @@
 
 template <ck::index_t N>
 using I = ck::Number<N>;
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
 using F16 = ck::half_t;
 #endif
 using F32 = float;
@@ -23,7 +23,7 @@ class TestSoftmax : public ck::TestSoftmax<Tuple>
 // clang-format off
 using KernelTypes = ::testing::Types<
     //         InDataType, AccDataType, OutDataType, Rank
-#ifdef __fp16__
+#ifdef CK_ENABLE_FP16
     std::tuple<       F16,         F32,         F16,    I<4>>,
 #endif
     std::tuple<       F32,         F32,         F32,    I<4>>

From 31ea132aa21ec37fe11735b2e0e71041b95f1911 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Mon, 28 Aug 2023 16:27:32 -0500
Subject: [PATCH 102/194] Fp16/fp8 mixed-precision Gemm with multiply+add
 fusion (#865)

* add compute_type

* add multiply_add ckProfiler

* add f8_fp16 support

* clean

* clean

* fixed lds size calc

* format

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |    6 +-
 .../device_batched_gemm_e_permute_xdl.hpp     |    7 +-
 .../impl/device_batched_gemm_multi_d_xdl.hpp  |    4 +
 ...ce_contraction_multiple_d_xdl_cshuffle.hpp |    4 +
 .../device_gemm_multiple_d_xdl_cshuffle.hpp   |   12 +-
 ...ed_contraction_multiple_d_xdl_cshuffle.hpp |    4 +
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |    2 +
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |    4 +
 .../device/impl/device_grouped_gemm_xdl.hpp   |    4 +
 .../gpu/element/element_wise_operation.hpp    |   45 +
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |   41 +-
 ...emm_split_k_multiple_d_xdl_cshuffle_v2.hpp | 1076 +++++++++++++++++
 .../device_operation_instance_factory.hpp     |    3 +
 .../gpu/gemm_multiply_add.hpp                 |  161 +++
 .../gpu/gemm_multiply_add/CMakeLists.txt      |    7 +
 ...16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp |   83 ++
 ...16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp |   82 ++
 ...f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp |   84 ++
 ...f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp |   83 ++
 .../profile_gemm_multiply_add_impl.hpp        |  242 ++++
 profiler/src/CMakeLists.txt                   |    2 +
 profiler/src/profile_gemm_multiply_add.cpp    |  153 +++
 22 files changed, 2082 insertions(+), 27 deletions(-)
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
 create mode 100644 profiler/include/profiler/profile_gemm_multiply_add_impl.hpp
 create mode 100644 profiler/src/profile_gemm_multiply_add.cpp

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
index fe97117181..32c45bc57e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_contraction_multiple_d_xdl_cshuffle.hpp
@@ -543,9 +543,13 @@ struct DeviceBatchedContractionMultipleD_Xdl_CShuffle
         EGridDesc_G_M_N e_grid_desc_g_m_n_;
     };
 
+    using ComputeDataType = ADataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
index d6b6405bb7..ba22cf0bf8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_e_permute_xdl.hpp
@@ -331,8 +331,13 @@ struct DeviceBatchedGemmEPermuteXdl : public DeviceBatchedGemmEPermute<ALayout,
         EGridDesc_G0_G1_M_N e_grid_desc_g0_g1_m_n_;
     };
 
+    using ComputeDataType = ADataType;
+
+    // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         ck::Tuple<>, // DsDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
index 3df2ee38f2..545d7e576f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multi_d_xdl.hpp
@@ -324,8 +324,12 @@ struct DeviceBatchedGemmMultiD_Xdl : public DeviceBatchedGemmMultiD<ALayout,
         index_t BatchStrideE_;
     };
 
+    using ComputeDataType = ADataType;
+
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index deb938156d..5ae06836fa 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -310,9 +310,13 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({{}}, {{}}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N({}, {}));
 
+    using ComputeDataType = ADataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index 9a75add9ce..c90c28f5a8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -20,7 +20,8 @@
 namespace ck {
 
 template <typename GridwiseGemm,
-          typename ABDataType,
+          typename ADataType,
+          typename BDataType,
           typename DsPointer,
           typename EDataType,
           typename AElementwiseOperation,
@@ -36,8 +37,8 @@ __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_multiple_d_xdl_cshuffle(const ABDataType* __restrict__ p_a_grid,
-                                            const ABDataType* __restrict__ p_b_grid,
+        kernel_gemm_multiple_d_xdl_cshuffle(const ADataType* __restrict__ p_a_grid,
+                                            const BDataType* __restrict__ p_b_grid,
                                             DsPointer p_ds_grid,
                                             EDataType* __restrict__ p_e_grid,
                                             const AElementwiseOperation a_element_op,
@@ -242,9 +243,13 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
 
+    using ComputeDataType = EDataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
@@ -442,6 +447,7 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
                 const auto kernel = kernel_gemm_multiple_d_xdl_cshuffle<
                     GridwiseGemm,
                     ADataType, // TODO: distiguish A/B datatype
+                    BDataType, // TODO: distiguish A/B datatype
                     typename GridwiseGemm::DsGridPointer,
                     EDataType,
                     AElementwiseOperation,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
index 0b1bf0f179..4823f5d489 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_contraction_multiple_d_xdl_cshuffle.hpp
@@ -355,9 +355,13 @@ struct DeviceGroupedContractionMultipleD_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({{}}, {{}}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N({}, {}));
 
+    using ComputeDataType = ADataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 8a0a445370..7f6fb908fa 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -355,6 +355,8 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
+        ABDataType, // TODO: distinguish A/B datatype
+        ABDataType, // TODO: distinguish A/B datatype
         ABDataType, // TODO: distinguish A/B datatype
         AccDataType,
         CShuffleDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index c80598c4e3..bcef5c3b62 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -367,9 +367,13 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
     using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
 
+    using ComputeDataType = ADataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
index db89bee96c..9290a31554 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl.hpp
@@ -228,9 +228,13 @@ struct DeviceGroupedGemm_Xdl : public DeviceGroupedGemm<ALayout,
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
 
+    using ComputeDataType = ADataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeDataType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
index 3fdb391a02..9f5ed6adea 100644
--- a/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/element_wise_operation.hpp
@@ -195,6 +195,51 @@ struct AddMultiply
     }
 };
 
+// C = A * B
+// E = C x D0 + D1
+struct MultiplyAdd
+{
+    template <typename E, typename C, typename D0, typename D1>
+    __host__ __device__ void operator()(E& e, const C& c, const D0& d0, const D1& d1) const;
+
+    template <>
+    __host__ __device__ void operator()<half_t, half_t, half_t, half_t>(half_t& e,
+                                                                        const half_t& c,
+                                                                        const half_t& d0,
+                                                                        const half_t& d1) const
+    {
+        const half_t y = (c * d0) + d1;
+        e              = y;
+    }
+    template <>
+    __host__ __device__ void operator()<half_t, float, half_t, half_t>(half_t& e,
+                                                                       const float& c,
+                                                                       const half_t& d0,
+                                                                       const half_t& d1) const
+    {
+        const half_t y = type_convert<half_t>(c) * d0 + d1;
+        e              = y;
+    }
+    template <>
+    __host__ __device__ void operator()<float, float, half_t, half_t>(float& e,
+                                                                      const float& c,
+                                                                      const half_t& d0,
+                                                                      const half_t& d1) const
+    {
+        const float y = c * d0 + d1;
+        e             = y;
+    }
+    template <>
+    __host__ __device__ void operator()<half_t, float, float, float>(half_t& e,
+                                                                     const float& c,
+                                                                     const float& d0,
+                                                                     const float& d1) const
+    {
+        const float y = c * d0 + d1;
+        e             = y;
+    }
+};
+
 // E = FastGelu(C + D0 + D1)
 struct AddAddFastGelu
 {
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index 3b8a5ec8f5..8f9e0915d2 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -26,7 +26,9 @@ namespace ck {
 //   E = cde_op(C, D0, D1, ...)
 // Assume:
 //   D0, D1, ... and E have the same layout
-template <typename ABDataType, // FIXME: don't assume A/B have same datatype
+template <typename ADataType,
+          typename BDataType,
+          typename ComputeDataType_,
           typename AccDataType,
           typename CShuffleDataType,
           typename DsDataType,
@@ -92,15 +94,11 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
     using GridwiseGemmPipe = remove_cvref_t<
         decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
-    // denorm test fix, required to work around fp16 mfma issue
-    // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
-    // when mfma if fixed, remove this section and update
-    // ABDataTypeAdjusted -> ABDataType throughout this file
 #if CK_WORKAROUND_DENORM_FIX
-    using ABDataTypeAdjusted =
-        conditional_t<is_same_v<ABDataType, ck::half_t>, ck::bhalf_t, ABDataType>;
+    using ComputeDataType =
+        conditional_t<is_same_v<ComputeDataType_, ck::half_t>, ck::bhalf_t, ComputeDataType_>;
 #else
-    using ABDataTypeAdjusted = ABDataType;
+    using ComputeDataType = ComputeDataType_;
 #endif
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
@@ -170,7 +168,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
             c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
 
         return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
-                             sizeof(ABDataType),
+                             sizeof(ComputeDataType),
                          c_block_size * sizeof(CShuffleDataType));
     }
 
@@ -313,8 +311,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         // check tensor size: cannot be larger than 2GB each
         constexpr long_index_t TwoGB = (long_index_t{1} << 31);
 
-        if(!(a_grid_desc_m_k.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
-             b_grid_desc_n_k.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+        if(!(a_grid_desc_m_k.GetElementSpaceSize() * sizeof(ADataType) <= TwoGB &&
+             b_grid_desc_n_k.GetElementSpaceSize() * sizeof(BDataType) <= TwoGB &&
              e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
         {
             return false;
@@ -338,8 +336,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
               typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
               typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
               typename Block2ETileMap>
-    __device__ static void Run(const ABDataType* __restrict__ p_a_grid,
-                               const ABDataType* __restrict__ p_b_grid,
+    __device__ static void Run(const ADataType* __restrict__ p_a_grid,
+                               const BDataType* __restrict__ p_b_grid,
                                DsGridPointer p_ds_grid,
                                EDataType* __restrict__ p_e_grid,
                                void* __restrict__ p_shared,
@@ -408,8 +406,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                                                 Sequence<AK0PerBlock, MPerBlock, AK1>,
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
-                                                ABDataType,
-                                                ABDataTypeAdjusted,
+                                                ADataType,
+                                                ComputeDataType,
                                                 decltype(a_grid_desc_ak0_m_ak1),
                                                 decltype(a_block_desc_ak0_m_ak1),
                                                 ABlockTransferSrcAccessOrder,
@@ -439,8 +437,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                                                 Sequence<BK0PerBlock, NPerBlock, BK1>,
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
-                                                ABDataType,
-                                                ABDataTypeAdjusted,
+                                                BDataType,
+                                                ComputeDataType,
                                                 decltype(b_grid_desc_bk0_n_bk1),
                                                 decltype(b_block_desc_bk0_n_bk1),
                                                 BBlockTransferSrcAccessOrder,
@@ -470,11 +468,11 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         // sanity check
         constexpr index_t KPack =
             math::max(math::lcm(AK1, BK1),
-                      MfmaSelector<ABDataTypeAdjusted, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+                      MfmaSelector<ComputeDataType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ABDataTypeAdjusted,
+            ComputeDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -492,11 +490,10 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ABDataTypeAdjusted*>(p_shared),
-            a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+            static_cast<ComputeDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ABDataTypeAdjusted*>(p_shared) + a_block_space_size_aligned,
+            static_cast<ComputeDataType*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp
new file mode 100644
index 0000000000..caf8f040f4
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle_v2.hpp
@@ -0,0 +1,1076 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+
+namespace ck {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <typename ADataType, // FIXME: don't assume A/B have same datatype
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t MPerXdl,
+          index_t NPerXdl,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_KBatch_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_KBatch_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched,
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    using GemmSpecialization = ck::tensor_operation::device::GemmSpecialization;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto AK1         = Number<AK1Value>{};
+    static constexpr auto BK1         = Number<BK1Value>{};
+    static constexpr auto AK0PerBlock = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0PerBlock = Number<KPerBlock / BK1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_KBatch_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(I1, AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(AK0PerBlock * Number<MPerBlock + ABlockLdsExtraM>{} * AK1,
+                       Number<MPerBlock + ABlockLdsExtraM>{} * AK1,
+                       AK1,
+                       I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_KBatch_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(I1, BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(BK0PerBlock * Number<NPerBlock + BBlockLdsExtraN>{} * BK1,
+                       Number<NPerBlock + BBlockLdsExtraN>{} * BK1,
+                       BK1,
+                       I1));
+    }
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    {
+        constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+        constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl>{},
+                           I1,
+                           Number<CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>{}));
+
+        return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // ck::Tuple<const D0DataType*, const D1DataType*, ...>
+    static constexpr auto MakeDsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                return static_cast<const DDataType*>(nullptr);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        // LDS allocation for C shuffle in LDS
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+        constexpr auto c_block_size =
+            c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
+
+        return math::max(a_block_space_size_aligned * sizeof(ADataType) +
+                             b_block_space_size_aligned * sizeof(BDataType),
+                         c_block_size * sizeof(CShuffleDataType));
+    }
+
+    __host__ __device__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_least_multiple(M, MPerBlock);
+    }
+
+    __host__ __device__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_least_multiple(N, NPerBlock);
+    }
+
+    __host__ __device__ static auto CalculateKPadded(index_t K, index_t K_Batch)
+    {
+        return math::integer_least_multiple(K, KPerBlock * K_Batch);
+    }
+
+    template <typename ALayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeAGridDescriptor_KBatch_AK0_M_AK1(index_t M, index_t K, index_t StrideA, index_t KBatch)
+    {
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto MPad = CalculateMPadded(M);
+        const auto KPad = CalculateKPadded(K, KBatch);
+
+        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto AK0 = KPad / (KBatch * AK1);
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, AK0, AK1)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, AK0, AK1)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    template <typename BLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeBGridDescriptor_KBatch_BK0_N_BK1(index_t K, index_t N, index_t StrideB, index_t KBatch)
+    {
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        const auto NPad = CalculateNPadded(N);
+        const auto KPad = CalculateKPadded(K, KBatch);
+
+        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
+            b_grid_desc_k_n,
+            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto BK0 = KPad / (KBatch * BK1);
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, BK0, BK1)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, BK0, BK1)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    // E desc for destination in blockwise copy
+    template <typename EGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            e_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return e_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // Ds desc for source in blockwise copy
+    template <typename DsGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const DsGridDesc_M_N& ds_grid_desc_m_n)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // return block_id to E matrix tile idx (m0, n0) mapping
+    template <typename EGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeDefaultBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, EGridDesc_M_N>(
+            e_grid_desc_m_n);
+    }
+
+    template <typename ALayout,
+              typename BLayout,
+              typename DsLayout,
+              typename ELayout,
+              GemmSpecialization GemmSpec>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const index_t M,
+                  const index_t N,
+                  const index_t K,
+                  const index_t StrideA,
+                  const index_t StrideB,
+                  const std::array<index_t, NumDTensor> StrideDs,
+                  const index_t StrideE,
+                  const index_t KBatch)
+    {
+        const auto a_grid_desc_kbatch_ak0_m_ak1 =
+            MakeAGridDescriptor_KBatch_AK0_M_AK1<ALayout, GemmSpec>(M, K, StrideA, KBatch);
+        const auto b_grid_desc_kbatch_bk0_n_bk1 =
+            MakeBGridDescriptor_KBatch_BK0_N_BK1<BLayout, GemmSpec>(K, N, StrideB, KBatch);
+
+        ignore = StrideDs;
+
+        const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+#if 0
+        // check tile size
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
+        {
+            return false;
+        }
+#endif
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = K / KPerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        // check tensor size: cannot be larger than 2GB each
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        if(!(a_grid_desc_kbatch_ak0_m_ak1.GetElementSpaceSize() * sizeof(ADataType) <= TwoGB &&
+             b_grid_desc_kbatch_bk0_n_bk1.GetElementSpaceSize() * sizeof(BDataType) <= TwoGB &&
+             e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / KPerBlock;
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    using DsGridPointer = decltype(MakeDsGridPointer());
+
+    template <typename ELayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+        const auto e_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    template <typename DsLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                             const std::array<index_t, NumDTensor>& NRaws,
+                             const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return MakeEGridDescriptor_M_N<DLayout, GemmSpec>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __device__ __host__ static constexpr auto GetMPerBlock() { return MPerBlock; }
+
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+              index_t NumDTensor_,
+              typename DsDataType_,
+              typename AGridDesc_KBatch_AK0_M_AK1,
+              typename BGridDesc_KBatch_BK0_N_BK1,
+              typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename CDEElementwiseOperation_,
+              typename Block2ETileMap>
+    __device__ static void Run(const ADataType* __restrict__ p_a_grid,
+                               const BDataType* __restrict__ p_b_grid,
+                               DsGridPointer p_ds_grid,
+                               EDataType* __restrict__ p_e_grid,
+                               void* __restrict__ p_shared,
+                               uint32_t* barrier_count_finished,
+                               const index_t KBatch,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation_& cde_element_op,
+                               const AGridDesc_KBatch_AK0_M_AK1& a_grid_desc_kbatch_ak0_m_ak1,
+                               const BGridDesc_KBatch_BK0_N_BK1& b_grid_desc_kbatch_bk0_n_bk1,
+                               const DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_kbatch_ak0_m_ak1.GetElementSpaceSize());
+
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_kbatch_bk0_n_bk1.GetElementSpaceSize());
+
+        const auto ds_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_ds_grid[i],
+                    ds_grid_desc_mblock_mperblock_nblock_nperblock[i].GetElementSpaceSize());
+            },
+            Number<NumDTensor_>{});
+
+        auto e_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_e_grid, e_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t kbatch_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
+
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * MPerBlock);
+
+        const index_t n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I2] * NPerBlock);
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_kbatch_ak0_m_ak1 =
+            GetABlockDescriptor_KBatch_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_kbatch_bk0_n_bk1 =
+            GetBBlockDescriptor_KBatch_BK0PerBlock_NPerBlock_BK1();
+
+        // A matrix blockwise copy
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                AElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<1, AK0PerBlock, MPerBlock, AK1>,
+                                                ABlockTransferThreadClusterLengths_KBatch_AK0_M_AK1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ADataType,
+                                                ComputeType,
+                                                decltype(a_grid_desc_kbatch_ak0_m_ak1),
+                                                decltype(a_block_desc_kbatch_ak0_m_ak1),
+                                                ABlockTransferSrcAccessOrder,
+                                                Sequence<2, 0, 1, 3>,
+                                                ABlockTransferSrcVectorDim,
+                                                3,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_AK1,
+                                                1,
+                                                1,
+                                                AThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                a_grid_desc_kbatch_ak0_m_ak1,
+                make_multi_index(kbatch_id, 0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_kbatch_ak0_m_ak1,
+                make_multi_index(0, 0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<1, BK0PerBlock, NPerBlock, BK1>,
+                                                BBlockTransferThreadClusterLengths_KBatch_BK0_N_BK1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                BDataType,
+                                                ComputeType,
+                                                decltype(b_grid_desc_kbatch_bk0_n_bk1),
+                                                decltype(b_block_desc_kbatch_bk0_n_bk1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<2, 0, 1, 3>,
+                                                BBlockTransferSrcVectorDim,
+                                                3,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_BK1,
+                                                1,
+                                                1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                b_grid_desc_kbatch_bk0_n_bk1,
+                make_multi_index(kbatch_id, 0, n_block_data_idx_on_grid, 0),
+                b_element_op,
+                b_block_desc_kbatch_bk0_n_bk1,
+                make_multi_index(0, 0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // GEMM definition
+        //   c_mtx += transpose(a_mtx) * b_mtx
+        //     a_mtx[K0PerBlock, MPerBlock] is in LDS
+        //     b_mtx[K0PerBlock, NPerBlock] is in LDS
+        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+        //       register
+        // sanity check
+        constexpr index_t KPack =
+            math::max(math::lcm(AK1, BK1),
+                      MfmaSelector<ComputeType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
+            BlockSize,
+            ComputeType,
+            AccDataType,
+            decltype(a_block_desc_ak0_m_ak1),
+            decltype(b_block_desc_bk0_n_bk1),
+            MPerXdl,
+            NPerXdl,
+            MXdlPerWave,
+            NXdlPerWave,
+            KPack,
+            LoopSched>();
+
+#if 1
+        if(block_work_idx[I0] == 0)
+        {
+            const index_t nThreadSize = CDEShuffleBlockTransferScalarPerVector_NPerBlock;
+            const index_t numNThreads = NPerBlock / nThreadSize;
+            const index_t numMThreads = BlockSize / numNThreads;
+            const index_t mThreadSize = MPerBlock / numMThreads;
+
+            const index_t m_tid = get_thread_local_1d_id() / numNThreads;
+            const index_t n_tid = get_thread_local_1d_id() % numNThreads;
+
+            auto c_thread_desc_mblock_mperblock_nblock_nperblock =
+                make_naive_tensor_descriptor_packed(
+                    make_tuple(I1, Number<mThreadSize>{}, I1, Number<nThreadSize>{}));
+
+            StaticBuffer<AddressSpaceEnum::Vgpr,
+                         EDataType,
+                         c_thread_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize(),
+                         true>
+                e_thread_zero_buf;
+
+            auto c_thread_copy = ThreadwiseTensorSliceTransfer_v1r3<
+                EDataType,
+                EDataType,
+                decltype(c_thread_desc_mblock_mperblock_nblock_nperblock),
+                decltype(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                ck::tensor_operation::element_wise::PassThrough,
+                Sequence<1, mThreadSize, 1, nThreadSize>,
+                Sequence<0, 1, 2, 3>,
+                3,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+                InMemoryDataOperationEnum::Set,
+                1,
+                true>{e_grid_desc_mblock_mperblock_nblock_nperblock,
+                      make_multi_index(block_work_idx[I1],
+                                       m_tid * mThreadSize,
+                                       block_work_idx[I2],
+                                       n_tid * nThreadSize),
+                      ck::tensor_operation::element_wise::PassThrough{}};
+
+            c_thread_copy.Run(c_thread_desc_mblock_mperblock_nblock_nperblock,
+                              make_tuple(I0, I0, I0, I0),
+                              e_thread_zero_buf,
+                              e_grid_desc_mblock_mperblock_nblock_nperblock,
+                              e_grid_buf);
+
+            __syncthreads();
+
+            if(threadIdx.x == 0)
+            {
+                atomicAdd(barrier_count_finished, 1);
+            }
+        }
+#endif
+
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ComputeType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ComputeType*>(p_shared) + a_block_space_size_aligned,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(0, KPerBlock / AK1, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(0, KPerBlock / BK1, 0, 0);
+
+        // gridwise GEMM pipeline
+        const auto gridwise_gemm_pipeline =
+            GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>();
+
+        const index_t num_k_block_main_loop =
+            __builtin_amdgcn_readfirstlane((a_grid_desc_kbatch_ak0_m_ak1.GetLength(I1) *
+                                            a_grid_desc_kbatch_ak0_m_ak1.GetLength(I3)) /
+                                           KPerBlock);
+
+        gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_grid_desc_kbatch_ak0_m_ak1,
+                                                               a_block_desc_kbatch_ak0_m_ak1,
+                                                               a_blockwise_copy,
+                                                               a_grid_buf,
+                                                               a_block_buf,
+                                                               a_block_slice_copy_step,
+                                                               b_grid_desc_kbatch_bk0_n_bk1,
+                                                               b_block_desc_kbatch_bk0_n_bk1,
+                                                               b_blockwise_copy,
+                                                               b_grid_buf,
+                                                               b_block_buf,
+                                                               b_block_slice_copy_step,
+                                                               blockwise_gemm,
+                                                               c_thread_buf,
+                                                               num_k_block_main_loop);
+
+        // shuffle C and write out
+        {
+            if(threadIdx.x == 0)
+            {
+                while(__atomic_load_n(barrier_count_finished, __ATOMIC_RELAXED) == 0) {}
+            }
+
+            __syncthreads();
+
+            static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
+                              NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
+                          "wrong!");
+
+            constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+            constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+            // TODO: hacky, fix it!
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+                blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            // TODO: hacky, fix it!
+            // c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp is only used to get lengths
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp =
+                blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I0);
+            constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I1);
+            constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I2);
+            constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I3);
+            constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I4);
+            constexpr auto M3 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I5);
+            constexpr auto M4 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I6);
+            constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I7);
+
+            constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+                GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<CShuffleDataType*>(p_shared),
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2 = transform_tensor_descriptor(
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMXdlPerWavePerShuffle>{}, // M0 (MXdlPerWave) per shuffle
+                        M1,                                      // M1 = MWave
+                        M2,                                      // M2 * M3 * M4 = MPerXdl
+                        M3,
+                        M4)),
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNXdlPerWavePerShuffle>{}, // N0 (NXdlPerWave) per shuffle
+                        N1,                                      // N1 = NWave
+                        N2))),                                   // N2 = NPerXdl
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(
+                    Sequence<>{}, Sequence<0, 2, 4, 5, 6>{}, Sequence<>{}, Sequence<1, 3, 7>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block =
+                blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
+                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto m_thread_data_on_block_idx =
+                m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
+                    make_multi_index(m_thread_data_on_block));
+
+            const auto n_thread_data_on_block_to_n0_n1_n2_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                    make_tuple(Sequence<0, 1, 2>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_idx =
+                n_thread_data_on_block_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                    make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   CShuffleDataType,
+                                                   decltype(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   decltype(c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMXdlPerWavePerShuffle,
+                                                            CShuffleNXdlPerWavePerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            M2,
+                                                            I1,
+                                                            M4,
+                                                            I1>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                                   7,
+                                                   1,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                    make_multi_index(0,
+                                     0,
+                                     m_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     m_thread_data_on_block_idx[I3],
+                                     m_thread_data_on_block_idx[I4],
+                                     n_thread_data_on_block_idx[I2]),
+                    ck::tensor_operation::element_wise::PassThrough{}};
+
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_desc_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_desc_mblock_mperblock_nblock_nperblock),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_desc_mblock_mperblock_nblock_nperblock[i]; },
+                    Number<NumDTensor_>{}));
+
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_buf_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_buf),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_buf[i]; },
+                    Number<NumDTensor_>{}));
+
+            // tuple of starting index of C/Ds blockwise copy
+            const auto idx_c_ds_block_begin = container_concat(
+                make_tuple(make_multi_index(0, 0, 0, 0)),
+                generate_tuple(
+                    [&](auto) {
+                        return make_multi_index(block_work_idx[I1], 0, block_work_idx[I2], 0);
+                    },
+                    Number<NumDTensor_>{}));
+
+            // space filling curve for threadwise C in VGPR before shuffle
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MXdlPerWave, NXdlPerWave, 1, 1, M2, 1, M4, 1>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                  Sequence<CShuffleMXdlPerWavePerShuffle,
+                                           CShuffleNXdlPerWavePerShuffle,
+                                           1,
+                                           1,
+                                           M2,
+                                           1,
+                                           M4,
+                                           1>>{};
+
+            // space filling curve for shuffled blockwise C/D/E
+            constexpr auto sfc_cde_block =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                                           1,
+                                           CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>>{};
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_cde_block.GetNumOfAccess(), "wrong!");
+
+            // blockwise copy C/D/E between LDS and global
+            auto cde_block_copy_lds_and_global = ThreadGroupTensorSliceTransfer_v7<
+                ThisThreadBlock,
+                decltype(container_concat(make_tuple(CShuffleDataType{}), DsDataType_{})),
+                Tuple<EDataType>,
+                decltype(c_ds_desc_refs),
+                decltype(tie(e_grid_desc_mblock_mperblock_nblock_nperblock)),
+                CDEElementwiseOperation_,
+                Sequence<static_cast<index_t>(EGlobalMemoryDataOperation)>, // FIXME: make
+                                                                            // Sequence support
+                                                                            // arbitray type
+                Sequence<1,
+                         CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                         1,
+                         CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>, // BlockSliceLengths,
+                CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                Sequence<0, 1, 2, 3>, // typename DimAccessOrder,
+                3,                    // index_t VectorDim,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+                sequence_merge_t<
+                    Sequence<true>,
+                    uniform_sequence_gen_t<NumDTensor_,
+                                           false>>, // ThreadTransferSrcResetCoordinateAfterRunFlags
+                Sequence<false>>                    // ThreadTransferDstResetCoordinateAfterRunFlags
+                {c_ds_desc_refs,
+                 idx_c_ds_block_begin,
+                 tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                 make_tuple(make_multi_index(block_work_idx[I1], 0, block_work_idx[I2], 0)),
+                 cde_element_op};
+
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // each thread write its data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each block copy its data from LDS to global
+                cde_block_copy_lds_and_global.Run(
+                    c_ds_desc_refs,
+                    c_ds_buf_refs,
+                    tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                    tie(e_grid_buf));
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto cde_lds_and_global_step =
+                        sfc_cde_block.GetForwardStep(access_id);
+
+                    // move on Ds
+                    static_for<0, NumDTensor_, 1>{}([&](auto i) {
+                        cde_block_copy_lds_and_global.MoveSrcSliceWindow(
+                            c_ds_desc_refs, i + I1, cde_lds_and_global_step);
+                    });
+
+                    // move on E
+                    cde_block_copy_lds_and_global.MoveDstSliceWindow(
+                        tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                        I0,
+                        cde_lds_and_global_step);
+                }
+            });
+
+            if(threadIdx.x == 0)
+            {
+                index_t k_id_finished_t = atomicAdd(barrier_count_finished, 1);
+
+                if(k_id_finished_t == KBatch)
+                {
+                    *barrier_count_finished = 0;
+                }
+            }
+        }
+    }
+
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+              GemmSpecialization GemmSpec,
+              typename ALayout,
+              typename BLayout,
+              typename DsLayout,
+              typename ELayout,
+              typename Block2ETileMap>
+    __device__ static void Run(const void* __restrict__ p_a_grid_,
+                               const void* __restrict__ p_b_grid_,
+                               DsGridPointer p_ds_grid,
+                               void* __restrict__ p_e_grid_,
+                               void* __restrict__ p_shared,
+                               uint32_t* barrier_count_finished,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation& cde_element_op,
+                               const index_t M,
+                               const index_t N,
+                               const index_t K,
+                               const index_t StrideA,
+                               const index_t StrideB,
+                               const std::array<index_t, NumDTensor> StrideDs,
+                               const index_t StrideE,
+                               const index_t KBatch,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        const auto p_a_grid = reinterpret_cast<const ADataType*>(p_a_grid_);
+        const auto p_b_grid = reinterpret_cast<const BDataType*>(p_b_grid_);
+        const auto p_e_grid = reinterpret_cast<EDataType*>(p_e_grid_);
+
+        using DsGridDesc_M_N =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_M_N<DsLayout, GemmSpec>({}, {}, {}))>;
+
+        DsGridDesc_M_N ds_grid_desc_m_n;
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            using DLayout = remove_cvref_t<tuple_element_t<j.value, DsLayout>>;
+
+            ds_grid_desc_m_n(j) = MakeEGridDescriptor_M_N<DLayout, GemmSpec>(M, N, StrideDs[j]);
+        });
+
+        const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+        // tensor descriptors for block/thread-wise copy
+        const auto a_grid_desc_kbatch_ak0_m_ak1 =
+            MakeAGridDescriptor_KBatch_AK0_M_AK1<ALayout, GemmSpec>(M, K, StrideA, KBatch);
+
+        const auto b_grid_desc_kbatch_bk0_n_bk1 =
+            MakeBGridDescriptor_KBatch_BK0_N_BK1<BLayout, GemmSpec>(K, N, StrideB, KBatch);
+
+        using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                DsGridDesc_M_N{}))>;
+
+        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock ds_grid_desc_mblock_mperblock_nblock_nperblock;
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            ds_grid_desc_mblock_mperblock_nblock_nperblock(j) =
+                MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[j]);
+        });
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(e_grid_desc_m_n);
+
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        const index_t kbatch_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
+
+        if(kbatch_id == KBatch - 1)
+        {
+            Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, NumDTensor, DsDataType>(
+                p_a_grid,
+                p_b_grid,
+                p_ds_grid,
+                p_e_grid,
+                p_shared,
+                barrier_count_finished,
+                KBatch,
+                a_element_op,
+                b_element_op,
+                cde_element_op,
+                a_grid_desc_kbatch_ak0_m_ak1,
+                b_grid_desc_kbatch_bk0_n_bk1,
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                block_2_etile_map);
+        }
+        else
+        {
+            Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, 0, Tuple<>>(
+                p_a_grid,
+                p_b_grid,
+                p_ds_grid,
+                p_e_grid,
+                p_shared,
+                barrier_count_finished,
+                KBatch,
+                a_element_op,
+                b_element_op,
+                ck::tensor_operation::element_wise::PassThrough{},
+                a_grid_desc_kbatch_ak0_m_ak1,
+                b_grid_desc_kbatch_bk0_n_bk1,
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                block_2_etile_map);
+        }
+    }
+};
+
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index 67c7d659e2..f874e52ddd 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -32,6 +32,8 @@ using F32_Tuple     = ck::Tuple<F32>;
 using I32_Tuple     = ck::Tuple<I32>;
 using I32_F32_Tuple = ck::Tuple<I32, F32>;
 
+using F32_F32_Tuple = ck::Tuple<F32, F32>;
+
 // GEMM layout
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -95,6 +97,7 @@ using AddFastGelu    = ck::tensor_operation::element_wise::AddFastGelu;
 using AddReluAdd     = ck::tensor_operation::element_wise::AddReluAdd;
 using FastGelu       = ck::tensor_operation::element_wise::FastGelu;
 using AddMultiply    = ck::tensor_operation::element_wise::AddMultiply;
+using MultiplyAdd    = ck::tensor_operation::element_wise::MultiplyAdd;
 using ScaleAdd       = ck::tensor_operation::element_wise::ScaleAdd;
 using Gelu           = ck::tensor_operation::element_wise::Gelu;
 using Swish          = ck::tensor_operation::element_wise::Swish;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
new file mode 100644
index 0000000000..badc06dd6f
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <vector>
+#include <memory>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>&);
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>&);
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F8,
+                                                    F32_F32_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>&);
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F8,
+                                                    F32_F32_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>&);
+
+// GEMM + Multiply + Add
+template <typename ALayout,
+          typename BLayout,
+          typename D0Layout,
+          typename D1Layout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMultipleD<
+    ALayout,
+    BLayout,
+    ck::Tuple<D0Layout, D1Layout>,
+    ELayout,
+    ADataType,
+    BDataType,
+    ck::Tuple<D0DataType, D1DataType>,
+    EDataType,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::PassThrough,
+    ck::tensor_operation::element_wise::MultiplyAdd>>
+{
+    using DeviceOp = DeviceGemmMultipleD<ALayout,
+                                         BLayout,
+                                         ck::Tuple<D0Layout, D1Layout>,
+                                         ELayout,
+                                         ADataType,
+                                         BDataType,
+                                         ck::Tuple<D0DataType, D1DataType>,
+                                         EDataType,
+                                         ck::tensor_operation::element_wise::PassThrough,
+                                         ck::tensor_operation::element_wise::PassThrough,
+                                         ck::tensor_operation::element_wise::MultiplyAdd>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<D0DataType, half_t> && is_same_v<D1DataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+        }
+
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
+                     is_same_v<D0DataType, float> && is_same_v<D1DataType, float> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<D0Layout, Row> && is_same_v<D1Layout, Row> &&
+                              is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances(
+                    op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
new file mode 100644
index 0000000000..a4f7443455
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
@@ -0,0 +1,7 @@
+add_instance_library(device_gemm_multiply_add_instance
+   device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+   device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+
+   device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
+   device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000..bb38b33405
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // M/N/K padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+        // clang-format on
+        >;
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000..4be9b51aa7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -0,0 +1,82 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16       = ck::half_t;
+using F32       = float;
+using F16_Tuple = ck::Tuple<F16, F16>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // M/N/K padding
+        // N % 8 == 0 && K % 1 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,   F16,     F32,      F16, F16_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>
+
+        // clang-format on
+        >;
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F16,
+                                                    F16_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000..7946bafc28
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -0,0 +1,84 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8        = ck::f8_t;
+using F16       = ck::half_t;
+using F32       = float;
+using F32_Tuple = ck::Tuple<F32, F32>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // M/N/K padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>
+        // clang-format on
+        >;
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F8,
+                                                    F32_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
new file mode 100644
index 0000000000..99db9f4270
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8        = ck::f8_t;
+using F16       = ck::half_t;
+using F32       = float;
+using F32_Tuple = ck::Tuple<F32, F32>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row, Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances =
+    std::tuple<
+        // clang-format off
+        // M/N/K padding
+        // N % 8 == 0 && K % 1 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  4, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>
+
+        // clang-format on
+        >;
+
+void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    F16,
+                                                    F8,
+                                                    F32_Tuple,
+                                                    F16,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    MultiplyAdd>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp b/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp
new file mode 100644
index 0000000000..5fc92d1f8d
--- /dev/null
+++ b/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp
@@ -0,0 +1,242 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename D0DataType,
+          typename D1DataType,
+          typename EDataType,
+          typename ALayout,
+          typename BLayout,
+          typename D0Layout,
+          typename D1Layout,
+          typename ELayout>
+bool profile_gemm_multiply_add_impl(int do_verification,
+                                    int init_method,
+                                    bool /*do_log*/,
+                                    bool time_kernel,
+                                    int M,
+                                    int N,
+                                    int K,
+                                    int StrideA,
+                                    int StrideB,
+                                    int StrideD0,
+                                    int StrideD1,
+                                    int StrideE)
+{
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(is_same<decltype(layout), tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<D0DataType> d0_m_n(f_host_tensor_descriptor(M, N, StrideD0, D0Layout{}));
+    Tensor<D1DataType> d1_m_n(f_host_tensor_descriptor(M, N, StrideD1, D1Layout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d0_m_n: " << d0_m_n.mDesc << std::endl;
+    std::cout << "d1_m_n: " << d1_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_device_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_2<D0DataType>{-5, 5});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_2<D1DataType>{-1, 1});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d0_m_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
+        d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
+    }
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+    using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
+
+    using AElementOp   = PassThrough;
+    using BElementOp   = PassThrough;
+    using CDEElementOp = MultiplyAdd;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceGemmMultipleD<ALayout,
+                                                          BLayout,
+                                                          ck::Tuple<D0Layout, D1Layout>,
+                                                          ELayout,
+                                                          ADataType,
+                                                          BDataType,
+                                                          ck::Tuple<D0DataType, D1DataType>,
+                                                          EDataType,
+                                                          PassThrough,
+                                                          PassThrough,
+                                                          CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    // run reference
+    if(do_verification)
+    {
+        Tensor<AccDataType> c_m_n({M, N});
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                AccDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+
+        auto ref_gemm    = ReferenceGemmInstance{};
+        auto ref_invoker = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d0_m_n(m, n), d1_m_n(m, n));
+            }
+        }
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d0_m_n_device_buf(sizeof(D0DataType) * d0_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem d1_m_n_device_buf(sizeof(D1DataType) * d1_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d0_m_n_device_buf.ToDevice(d0_m_n.mData.data());
+    d1_m_n_device_buf.ToDevice(d1_m_n.mData.data());
+
+    std::string best_op_name;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    bool pass = true;
+
+    // profile device operation instances
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            a_device_buf.GetDeviceBuffer(),
+            b_device_buf.GetDeviceBuffer(),
+            std::array<const void*, 2>{d0_m_n_device_buf.GetDeviceBuffer(),
+                                       d1_m_n_device_buf.GetDeviceBuffer()},
+            e_device_buf.GetDeviceBuffer(),
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            std::array<ck::index_t, 2>{StrideD0, StrideD1},
+            StrideE,
+            a_element_op,
+            b_element_op,
+            cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            // re-init E to zero before profiling a kernel
+            e_device_buf.SetZero();
+
+            float ave_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+            std::size_t flop = std::size_t(2) * M * N * K;
+
+            std::size_t num_btype =
+                sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+            float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+                pass = pass && ck::utils::check_err(e_m_n_device_result, e_m_n_host_result);
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index f98747c93a..b305e57598 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -5,6 +5,7 @@ set(PROFILER_SOURCES
     profile_gemm_splitk.cpp
     profile_gemm_bias_add_reduce.cpp
     profile_gemm_add_multiply.cpp
+    profile_gemm_multiply_add.cpp
     profile_gemm_reduce.cpp
     profile_batched_gemm.cpp
     profile_batched_gemm_reduce.cpp
@@ -51,6 +52,7 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE utility)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_splitk_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_multiply_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_multiply_add_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_bias_add_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_instance)
diff --git a/profiler/src/profile_gemm_multiply_add.cpp b/profiler/src/profile_gemm_multiply_add.cpp
new file mode 100644
index 0000000000..fd1f5c65c1
--- /dev/null
+++ b/profiler/src/profile_gemm_multiply_add.cpp
@@ -0,0 +1,153 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_gemm_multiply_add_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+#define OP_NAME "gemm_multiply_add"
+#define OP_DESC "GEMM+MULTIPLY+ADD"
+
+int profile_gemm_multiply_add(int argc, char* argv[])
+{
+    enum struct MatrixLayout
+    {
+        MK_KN_MN_MN_MN, // 0
+        MK_NK_MN_MN_MN, // 1
+    };
+
+    enum struct MatrixDataType
+    {
+        F16_F16_F16_F16_F16, // 0
+        F16_F8_F32_F32_F16,  // 1
+    };
+
+    if(argc != 16)
+    {
+        // clang-format off
+        printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
+        printf("arg2: data type (0: fp16; 1: fp16Afp8B)\n");
+        printf("arg3: matrix layout (0: E[m, n] = Multiply_Add((A[m, k] * B[k, n]) x D1[m, n] + D0[m, n]);\n");
+        printf("                     1: E[m, n] = Multiply_Add((A[m, k] * B[n, k]) x D1[m, n] + D0[m, n]);\n");
+        printf("arg4: verification (0: no; 1: yes)\n");
+        printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
+        printf("arg6: print tensor value (0: no; 1: yes)\n");
+        printf("arg7: time kernel (0=no, 1=yes)\n");
+        printf("arg8 to 15: M, N, K, StrideA, StrideB, StrideD0, StrideD1, StrideE\n");
+        // clang-format on
+        exit(1);
+    }
+
+    const auto data_type       = static_cast<MatrixDataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<MatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+
+    const int M = std::stoi(argv[8]);
+    const int N = std::stoi(argv[9]);
+    const int K = std::stoi(argv[10]);
+
+    const int StrideA  = std::stoi(argv[11]);
+    const int StrideB  = std::stoi(argv[12]);
+    const int StrideD0 = std::stoi(argv[13]);
+    const int StrideD1 = std::stoi(argv[14]);
+    const int StrideE  = std::stoi(argv[15]);
+
+    using F8  = ck::f8_t;
+    using F16 = ck::half_t;
+    using F32 = float;
+
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    auto profile = [&](auto a_type,
+                       auto b_type,
+                       auto acc_type,
+                       auto d0_type,
+                       auto d1_type,
+                       auto e_type,
+                       auto a_layout,
+                       auto b_layout,
+                       auto d0_layout,
+                       auto d1_layout,
+                       auto e_layout) {
+        using ADataType   = decltype(a_type);
+        using BDataType   = decltype(b_type);
+        using AccDataType = decltype(acc_type);
+        using D0DataType  = decltype(d0_type);
+        using D1DataType  = decltype(d1_type);
+        using EDataType   = decltype(e_type);
+
+        using ALayout  = decltype(a_layout);
+        using BLayout  = decltype(b_layout);
+        using D0Layout = decltype(d0_layout);
+        using D1Layout = decltype(d1_layout);
+        using ELayout  = decltype(e_layout);
+
+        const int DefaultStrideA  = ck::is_same_v<ALayout, Row> ? K : M;
+        const int DefaultStrideB  = ck::is_same_v<BLayout, Row> ? N : K;
+        const int DefaultStrideD0 = ck::is_same_v<D0Layout, Row> ? N : M;
+        const int DefaultStrideD1 = ck::is_same_v<D1Layout, Row> ? N : M;
+        const int DefaultStrideE  = ck::is_same_v<ELayout, Row> ? N : M;
+
+        bool pass = ck::profiler::profile_gemm_multiply_add_impl<ADataType,
+                                                                 BDataType,
+                                                                 AccDataType,
+                                                                 D0DataType,
+                                                                 D1DataType,
+                                                                 EDataType,
+                                                                 ALayout,
+                                                                 BLayout,
+                                                                 D0Layout,
+                                                                 D1Layout,
+                                                                 ELayout>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            M,
+            N,
+            K,
+            (StrideA < 0) ? DefaultStrideA : StrideA,
+            (StrideB < 0) ? DefaultStrideB : StrideB,
+            (StrideD0 < 0) ? DefaultStrideD0 : StrideD0,
+            (StrideD1 < 0) ? DefaultStrideD1 : StrideD1,
+            (StrideE < 0) ? DefaultStrideE : StrideE);
+
+        return pass ? 0 : 1;
+    };
+
+    if(data_type == MatrixDataType::F16_F16_F16_F16_F16 && layout == MatrixLayout::MK_KN_MN_MN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Row{}, Row{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::F16_F16_F16_F16_F16 &&
+            layout == MatrixLayout::MK_NK_MN_MN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Row{}, Col{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::F16_F8_F32_F32_F16 &&
+            layout == MatrixLayout::MK_KN_MN_MN_MN)
+    {
+        return profile(F16{}, F8{}, F32{}, F32{}, F32{}, F16{}, Row{}, Row{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::F16_F8_F32_F32_F16 &&
+            layout == MatrixLayout::MK_NK_MN_MN_MN)
+    {
+        return profile(F16{}, F8{}, F32{}, F32{}, F32{}, F16{}, Row{}, Col{}, Row{}, Row{}, Row{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
+    }
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_gemm_multiply_add);

From 38ada109ea2a400f18c74f365c6f5871cbd85765 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 29 Aug 2023 12:31:24 -0500
Subject: [PATCH 103/194] add an example of customized type convert - bfp16_rtn
 (#869)

* add an example of customized bfp16_rtn

* fixed threadwise_copy

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/01_gemm/CMakeLists.txt                |  3 ++
 example/01_gemm/gemm_xdl_bf16_rtn.cpp         | 39 +++++++++++++++++++
 .../element/unary_element_wise_operation.hpp  |  6 +++
 .../threadwise_tensor_slice_transfer_v6r1.hpp |  4 +-
 .../cpu/reference_gemm.hpp                    |  4 +-
 5 files changed, 52 insertions(+), 4 deletions(-)
 create mode 100644 example/01_gemm/gemm_xdl_bf16_rtn.cpp

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index a5933262a5..a671877f68 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -40,6 +40,9 @@ endif()
 if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
   add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
   add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
+
+  add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
 endif()
 
 if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
diff --git a/example/01_gemm/gemm_xdl_bf16_rtn.cpp b/example/01_gemm/gemm_xdl_bf16_rtn.cpp
new file mode 100644
index 0000000000..cc14dcb8eb
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_bf16_rtn.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/utility/type_convert.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::bhalf_t;
+using BDataType        = ck::bhalf_t;
+using CDataType        = ck::bhalf_t;
+using AccDataType      = float;
+using CShuffleDataType = float;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = ck::tensor_operation::element_wise::ConvertBF16RTN;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 4fb061fadb..905908a1c3 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -39,6 +39,12 @@ struct PassThrough
         y = x;
     }
 
+    template <>
+    __host__ __device__ void operator()<half_t, float>(half_t& y, const float& x) const
+    {
+        y = type_convert<half_t>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<bhalf_t, bhalf_t>(bhalf_t& y, const bhalf_t& x) const
     {
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
index 6ec9abc417..644877d393 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v6r1.hpp
@@ -104,13 +104,13 @@ struct ThreadwiseTensorSliceTransfer_v6r1
 
             // apply pointwise operation
             static_for<0, ScalarPerVector, 1>{}([&](auto i) {
-                SrcData v;
+                DstData v;
 
                 // apply element-wise operation
                 element_op_(v, src_vector_container.template AsType<SrcData>()[i]);
 
                 // apply type convert
-                dst_vector_container.template AsType<DstData>()(i) = type_convert<DstData>(v);
+                dst_vector_container.template AsType<DstData>()(i) = v;
             });
 
             const bool is_dst_valid =
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
index 9b797be925..309b4afad8 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
@@ -92,11 +92,11 @@ struct ReferenceGemm : public device::BaseOperator
                         ck::type_convert<AccDataType>(v_a) * ck::type_convert<AccDataType>(v_b);
                 }
 
-                AccDataType v_c;
+                CDataType v_c;
 
                 arg.c_element_op_(v_c, v_acc);
 
-                arg.c_m_n_(m, n) = ck::type_convert<CDataType>(v_c);
+                arg.c_m_n_(m, n) = v_c;
             };
 
             make_ParallelTensorFunctor(

From 9e86ebd62dc594909b24b694b56036e6a0520017 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 30 Aug 2023 17:33:11 +0200
Subject: [PATCH 104/194] Add number of error when fail (#868)

---
 .../include/ck/library/utility/check_err.hpp  | 24 +++++++++++++++----
 1 file changed, 20 insertions(+), 4 deletions(-)

diff --git a/library/include/ck/library/utility/check_err.hpp b/library/include/ck/library/utility/check_err.hpp
index 7f63a81a02..8a72631376 100644
--- a/library/include/ck/library/utility/check_err.hpp
+++ b/library/include/ck/library/utility/check_err.hpp
@@ -65,7 +65,11 @@ check_err(const Range& out,
     }
     if(!res)
     {
-        std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
+        const float error_percent =
+            static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
+        std::cerr << "max err: " << max_err;
+        std::cerr << ", number of errors: " << err_count;
+        std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
@@ -112,7 +116,11 @@ check_err(const Range& out,
     }
     if(!res)
     {
-        std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
+        const float error_percent =
+            static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
+        std::cerr << "max err: " << max_err;
+        std::cerr << ", number of errors: " << err_count;
+        std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
@@ -158,7 +166,11 @@ check_err(const Range& out,
     }
     if(!res)
     {
-        std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
+        const float error_percent =
+            static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
+        std::cerr << "max err: " << max_err;
+        std::cerr << ", number of errors: " << err_count;
+        std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }
@@ -209,7 +221,11 @@ check_err(const Range& out,
     }
     if(!res)
     {
-        std::cerr << "max err: " << max_err << std::endl;
+        const float error_percent =
+            static_cast<float>(err_count) / static_cast<float>(out.size()) * 100.f;
+        std::cerr << "max err: " << max_err;
+        std::cerr << ", number of errors: " << err_count;
+        std::cerr << ", " << error_percent << "% wrong values" << std::endl;
     }
     return res;
 }

From bf1912ed3d00466f89d280b2dd56ab9447216b0d Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 30 Aug 2023 20:18:38 -0700
Subject: [PATCH 105/194] fix gemm_streamk example on mi300 (#875)

---
 .../tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
index 4408b34870..70abcac0b1 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
@@ -37,7 +37,8 @@ __global__ void
                                    index_t StrideC,
                                    typename GridwiseGemm::Block2CTileMap block_mapping)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
 
     __shared__ uint8_t p_shared[shared_size];

From 866377de18775d73ae8f5094e834c37eb7a29b48 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Thu, 31 Aug 2023 21:01:50 +0800
Subject: [PATCH 106/194] MaxPool & AvgPool bwd instances, test, ckProfiler,
 client example (#861)

* Add maxpool instances

* Rename index pool to max pool.

* Add maxpool bwd bf16 instances

* Add avg pool bwd instances

* Rename avgpool and maxpool to avg_pool3d and max_pool

* Add bf16 pool fwd instances

* Add max pool bwd to ckProfiler

* Add avg pool3d bwd to ckProfiler

* Add avg pool bwd test

* Fix bug of reference pool fwd (dilation)

* Fix bug of max pool bwd  (dilation and initZero)

* Support bf16 compute data type

* Force compute type be f32. Because atomicAdd only support f32

* Add max pool bwd test

* Rename folder

* Rename pool

* Add max pool bwd client example

* Add avg pool bwd client example

* Add missing workspace

* clang format

* Rename macro

* remove useless header

* remove useless layout
---
 client_example/05_layernorm/layernorm2d.cpp   |   8 +
 .../18_groupnorm/groupnorm_swish.cpp          |   8 +
 client_example/19_pool/CMakeLists.txt         |  11 +
 client_example/19_pool/avg_pool3d_bwd.cpp     | 191 ++++++++++++
 .../avg_pool3d_fwd.cpp                        |   0
 client_example/19_pool/max_pool2d_bwd.cpp     | 280 +++++++++++++++++
 .../max_pool2d_fwd.cpp                        |   0
 client_example/19_pool_fwd/CMakeLists.txt     |   5 -
 .../49_maxpool2d_bwd/maxpool2d_bwd_common.hpp |   7 +-
 ...x_pool_bwd.hpp => device_max_pool_bwd.hpp} |   5 +-
 ..._impl.hpp => device_max_pool_bwd_impl.hpp} |  21 +-
 include/ck/utility/reduction_operator.hpp     |  61 +++-
 .../cpu/reference_maxpool_bwd.hpp             |  11 +-
 .../cpu/reference_pool_fwd.hpp                |   6 +-
 .../gpu/avg_pool3d_bwd.hpp                    |  59 ++++
 .../gpu/max_pool_bwd.hpp                      |  58 ++++
 .../gpu/pool3d_fwd.hpp                        |  33 +-
 .../gpu/avg_pool3d_bwd/CMakeLists.txt         |  11 +
 .../avg_pool3d_bwd_ndhwc_instance_common.hpp  |  59 ++++
 ...ice_avg_pool3d_bwd_ndhwc_bf16_instance.cpp |  20 ++
 ...vice_avg_pool3d_bwd_ndhwc_f16_instance.cpp |  20 ++
 ...vice_avg_pool3d_bwd_ndhwc_f32_instance.cpp |  20 ++
 .../gpu/max_pool_bwd/CMakeLists.txt           |  11 +
 .../device_max_pool_bwd_bf16_instance.cpp     |  20 ++
 .../device_max_pool_bwd_f16_instance.cpp      |  20 ++
 .../device_max_pool_bwd_f32_instance.cpp      |  20 ++
 .../max_pool_bwd_instance_common.hpp          |  35 +++
 .../gpu/pool3d_fwd/CMakeLists.txt             |   4 +
 ...ice_avg_pool3d_fwd_ndhwc_bf16_instance.cpp |  25 ++
 ...ice_max_pool3d_fwd_ndhwc_bf16_instance.cpp |  34 +++
 .../pool3d_fwd/pool_fwd_instance_common.hpp   |   1 +
 .../profiler/profile_avg_pool3d_bwd_impl.hpp  | 253 +++++++++++++++
 .../profiler/profile_max_pool3d_bwd_impl.hpp  | 288 ++++++++++++++++++
 profiler/src/CMakeLists.txt                   |   4 +
 profiler/src/profile_avg_pool3d_bwd.cpp       | 175 +++++++++++
 profiler/src/profile_max_pool3d_bwd.cpp       | 177 +++++++++++
 profiler/src/profile_max_pool3d_fwd.cpp       |  64 +++-
 test/CMakeLists.txt                           |   2 +-
 test/pool/CMakeLists.txt                      |  16 +
 test/pool/test_avg_pool3d_bwd.cpp             |  74 +++++
 .../test_avg_pool3d_fwd.cpp                   |   0
 test/pool/test_max_pool3d_bwd.cpp             |  79 +++++
 .../test_max_pool3d_fwd.cpp                   |   0
 .../test_pool_fwd_common.hpp                  |   8 +-
 test/pool_fwd/CMakeLists.txt                  |  10 -
 45 files changed, 2174 insertions(+), 40 deletions(-)
 create mode 100644 client_example/19_pool/CMakeLists.txt
 create mode 100644 client_example/19_pool/avg_pool3d_bwd.cpp
 rename client_example/{19_pool_fwd => 19_pool}/avg_pool3d_fwd.cpp (100%)
 create mode 100644 client_example/19_pool/max_pool2d_bwd.cpp
 rename client_example/{19_pool_fwd => 19_pool}/max_pool2d_fwd.cpp (100%)
 delete mode 100644 client_example/19_pool_fwd/CMakeLists.txt
 rename include/ck/tensor_operation/gpu/device/{device_index_pool_bwd.hpp => device_max_pool_bwd.hpp} (83%)
 rename include/ck/tensor_operation/gpu/device/impl/{device_index_pool_bwd_impl.hpp => device_max_pool_bwd_impl.hpp} (93%)
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/avg_pool3d_bwd.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/max_pool_bwd.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/avg_pool3d_bwd_ndhwc_instance_common.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/max_pool_bwd/max_pool_bwd_instance_common.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_bf16_instance.cpp
 create mode 100644 profiler/include/profiler/profile_avg_pool3d_bwd_impl.hpp
 create mode 100644 profiler/include/profiler/profile_max_pool3d_bwd_impl.hpp
 create mode 100644 profiler/src/profile_avg_pool3d_bwd.cpp
 create mode 100644 profiler/src/profile_max_pool3d_bwd.cpp
 create mode 100644 test/pool/CMakeLists.txt
 create mode 100644 test/pool/test_avg_pool3d_bwd.cpp
 rename test/{pool_fwd => pool}/test_avg_pool3d_fwd.cpp (100%)
 create mode 100644 test/pool/test_max_pool3d_bwd.cpp
 rename test/{pool_fwd => pool}/test_max_pool3d_fwd.cpp (100%)
 rename test/{pool_fwd => pool}/test_pool_fwd_common.hpp (88%)
 delete mode 100644 test/pool_fwd/CMakeLists.txt

diff --git a/client_example/05_layernorm/layernorm2d.cpp b/client_example/05_layernorm/layernorm2d.cpp
index 4af4d7abe8..3ee7cead7b 100644
--- a/client_example/05_layernorm/layernorm2d.cpp
+++ b/client_example/05_layernorm/layernorm2d.cpp
@@ -100,6 +100,10 @@ int main(int argc, char* argv[])
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            SimpleDeviceMem workspace(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
             float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
             std::size_t num_byte = sizeof(XDataType) * M * N + sizeof(GammaDataType) * N +
@@ -153,6 +157,10 @@ int main(int argc, char* argv[])
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            SimpleDeviceMem workspace(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
             invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
         }
 
diff --git a/client_example/18_groupnorm/groupnorm_swish.cpp b/client_example/18_groupnorm/groupnorm_swish.cpp
index e1d198d228..df0a9ceec6 100644
--- a/client_example/18_groupnorm/groupnorm_swish.cpp
+++ b/client_example/18_groupnorm/groupnorm_swish.cpp
@@ -129,6 +129,10 @@ int main(int argc, char* argv[])
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            SimpleDeviceMem workspace(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
             float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
             std::size_t num_byte =
@@ -184,6 +188,10 @@ int main(int argc, char* argv[])
 
         if(op_ptr->IsSupportedArgument(argument_ptr.get()))
         {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+            SimpleDeviceMem workspace(workspace_sz);
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
             invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
         }
 
diff --git a/client_example/19_pool/CMakeLists.txt b/client_example/19_pool/CMakeLists.txt
new file mode 100644
index 0000000000..d4e2e6d4dc
--- /dev/null
+++ b/client_example/19_pool/CMakeLists.txt
@@ -0,0 +1,11 @@
+add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp)
+target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_operations)
+
+add_executable(client_max_pool2d_bwd max_pool2d_bwd.cpp)
+target_link_libraries(client_max_pool2d_bwd PRIVATE composable_kernel::device_operations)
+
+add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp)
+target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_operations)
+
+add_executable(client_avg_pool3d_bwd avg_pool3d_bwd.cpp)
+target_link_libraries(client_avg_pool3d_bwd PRIVATE composable_kernel::device_operations)
diff --git a/client_example/19_pool/avg_pool3d_bwd.cpp b/client_example/19_pool/avg_pool3d_bwd.cpp
new file mode 100644
index 0000000000..686d1da3ad
--- /dev/null
+++ b/client_example/19_pool/avg_pool3d_bwd.cpp
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/avg_pool3d_bwd.hpp"
+
+using DOutDataType = ck::half_t;
+using DInDataType  = ck::half_t;
+
+using DOutLayout = ck::tensor_layout::convolution::NDHWC;
+using DInLayout  = ck::tensor_layout::convolution::NDHWC;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}, mMemSize_(mem_size)
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    void SetZero() const { (void)hipMemset(p_mem_, 0, mMemSize_); }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+    std::size_t mMemSize_;
+};
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N                 = 2;
+    ck::index_t C                 = 32;
+    ck::index_t Z                 = 2;
+    ck::index_t Y                 = 2;
+    ck::index_t X                 = 2;
+    ck::index_t Di                = 30;
+    ck::index_t Hi                = 30;
+    ck::index_t Wi                = 30;
+    ck::index_t window_stride_d   = 2;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_d = 1;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_d     = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_d    = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
+
+    const ck::index_t Zs = (Z - 1) * window_dilation_d + 1;
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    ck::index_t Do       = (Di + in_left_pad_d + in_right_pad_d - Zs) / window_stride_d + 1;
+    ck::index_t Ho       = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    ck::index_t Wo       = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
+
+    // Pool API only support the order of NCDHW
+    std::vector<ck::index_t> in_length              = {N, C, Di, Hi, Wi};
+    std::vector<ck::index_t> out_length             = {N, C, Do, Ho, Wo};
+    std::vector<ck::index_t> window_spatial_lengths = {Z, Y, X};
+    std::vector<ck::index_t> window_strides = {window_stride_d, window_stride_h, window_stride_w};
+    std::vector<ck::index_t> window_dilations{
+        window_dilation_d, window_dilation_h, window_dilation_w};
+    std::vector<ck::index_t> input_left_pads  = {in_left_pad_d, in_left_pad_h, in_left_pad_w};
+    std::vector<ck::index_t> input_right_pads = {in_right_pad_d, in_right_pad_h, in_right_pad_w};
+
+    std::size_t in_tensor_size  = N * C * Di * Hi * Wi;
+    std::size_t out_tensor_size = N * C * Do * Ho * Wo;
+
+    // tensor layout = NDHWC
+    std::vector<ck::index_t> in_tensor_stride  = {Di * C * Hi * Wi, 1, C * Hi * Wi, Wi * C, C};
+    std::vector<ck::index_t> out_tensor_stride = {Do * C * Ho * Wo, 1, C * Ho * Wo, Wo * C, C};
+
+    SimpleDeviceMem dout_device_buf(sizeof(DOutDataType) * out_tensor_size);
+    SimpleDeviceMem din_device_buf(sizeof(DInDataType) * in_tensor_size);
+
+    using DeviceOp = ck::tensor_operation::device::
+        DeviceAvgPoolBwd<3, DOutDataType, DInDataType, DOutLayout, DInLayout>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+            static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+            out_length,
+            in_length,
+            out_tensor_stride,
+            in_tensor_stride,
+            window_spatial_lengths,
+            window_strides,
+            window_dilations,
+            input_left_pads,
+            input_right_pads);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            din_device_buf.SetZero();
+
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes =
+                in_tensor_size * sizeof(DInDataType) + out_tensor_size * sizeof(DOutDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+            static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+            out_length,
+            in_length,
+            out_tensor_stride,
+            in_tensor_stride,
+            window_spatial_lengths,
+            window_strides,
+            window_dilations,
+            input_left_pads,
+            input_right_pads);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            din_device_buf.SetZero();
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/19_pool_fwd/avg_pool3d_fwd.cpp b/client_example/19_pool/avg_pool3d_fwd.cpp
similarity index 100%
rename from client_example/19_pool_fwd/avg_pool3d_fwd.cpp
rename to client_example/19_pool/avg_pool3d_fwd.cpp
diff --git a/client_example/19_pool/max_pool2d_bwd.cpp b/client_example/19_pool/max_pool2d_bwd.cpp
new file mode 100644
index 0000000000..53ece7425f
--- /dev/null
+++ b/client_example/19_pool/max_pool2d_bwd.cpp
@@ -0,0 +1,280 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <vector>
+#include <iostream>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_pool_fwd.hpp"
+#include "ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+#include "ck/library/tensor_operation_instance/gpu/max_pool_bwd.hpp"
+
+using InDataType    = ck::half_t;
+using OutDataType   = ck::half_t;
+using DOutDataType  = ck::half_t;
+using DInDataType   = ck::half_t;
+using IndexDataType = int32_t;
+
+// We use pool3d to implement pool2d in this example
+using InLayout  = ck::tensor_layout::convolution::NDHWC;
+using OutLayout = ck::tensor_layout::convolution::NDHWC;
+
+constexpr ck::index_t InOutRank  = 5;
+constexpr ck::index_t WindowRank = 3;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+void TransformPool2dparamToPool3d(std::vector<ck::index_t>& input_lengths,
+                                  std::vector<ck::index_t>& window_lengths,
+                                  std::vector<ck::index_t>& output_lengths,
+                                  std::vector<ck::index_t>& input_stride,
+                                  std::vector<ck::index_t>& output_stride,
+                                  std::vector<ck::index_t>& indices_stride,
+                                  std::vector<ck::index_t>& window_strides,
+                                  std::vector<ck::index_t>& window_dilations,
+                                  std::vector<ck::index_t>& input_left_pads,
+                                  std::vector<ck::index_t>& input_right_pads,
+                                  std::vector<ck::index_t>& pooling_dims)
+{
+    // NCHW to NCDHW
+    input_lengths.insert(input_lengths.begin() + 2, 1);
+    output_lengths.insert(output_lengths.begin() + 2, 1);
+    input_stride.insert(input_stride.begin() + 2, 0);
+    output_stride.insert(output_stride.begin() + 2, 0);
+    indices_stride.insert(indices_stride.begin() + 2, 0);
+
+    // YX to ZYX
+    window_lengths.insert(window_lengths.begin(), 1);
+    window_strides.insert(window_strides.begin(), 0);
+    window_dilations.insert(window_dilations.begin(), 0);
+    input_left_pads.insert(input_left_pads.begin(), 0);
+    input_right_pads.insert(input_right_pads.begin(), 0);
+
+    pooling_dims = {2, 3, 4};
+}
+
+int main(int argc, char* argv[])
+{
+    ck::index_t N                 = 2;
+    ck::index_t C                 = 32;
+    ck::index_t Y                 = 2;
+    ck::index_t X                 = 2;
+    ck::index_t Hi                = 30;
+    ck::index_t Wi                = 30;
+    ck::index_t window_stride_h   = 2;
+    ck::index_t window_stride_w   = 2;
+    ck::index_t window_dilation_h = 1;
+    ck::index_t window_dilation_w = 1;
+    ck::index_t in_left_pad_h     = 1;
+    ck::index_t in_left_pad_w     = 1;
+    ck::index_t in_right_pad_h    = 1;
+    ck::index_t in_right_pad_w    = 1;
+
+    const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
+    const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
+    ck::index_t Ho       = (Hi + in_left_pad_h + in_right_pad_h - Ys) / window_stride_h + 1;
+    ck::index_t Wo       = (Wi + in_left_pad_w + in_right_pad_w - Xs) / window_stride_w + 1;
+
+    // Pool API only support the order of NCHW
+    std::vector<ck::index_t> in_length              = {N, C, Hi, Wi};
+    std::vector<ck::index_t> out_length             = {N, C, Ho, Wo};
+    std::vector<ck::index_t> window_spatial_lengths = {Y, X};
+    std::vector<ck::index_t> window_strides         = {window_stride_h, window_stride_w};
+    std::vector<ck::index_t> window_dilations       = {window_dilation_h, window_dilation_w};
+    std::vector<ck::index_t> input_left_pads        = {in_left_pad_h, in_left_pad_w};
+    std::vector<ck::index_t> input_right_pads       = {in_right_pad_h, in_right_pad_w};
+    std::vector<ck::index_t> pooling_dims           = {2, 3};
+
+    std::size_t in_tensor_size  = N * C * Hi * Wi;
+    std::size_t out_tensor_size = N * C * Ho * Wo;
+
+    // tensor layout = NHWC
+    std::vector<ck::index_t> in_tensor_stride  = {C * Hi * Wi, 1, Wi * C, C};
+    std::vector<ck::index_t> out_tensor_stride = {C * Ho * Wo, 1, Wo * C, C};
+
+    TransformPool2dparamToPool3d(in_length,
+                                 window_spatial_lengths,
+                                 out_length,
+                                 in_tensor_stride,
+                                 out_tensor_stride,
+                                 out_tensor_stride,
+                                 window_strides,
+                                 window_dilations,
+                                 input_left_pads,
+                                 input_right_pads,
+                                 pooling_dims);
+
+    SimpleDeviceMem in_device_buf(sizeof(InDataType) * in_tensor_size);
+    SimpleDeviceMem out_device_buf(sizeof(OutDataType) * out_tensor_size);
+    SimpleDeviceMem indices_device_buf(sizeof(IndexDataType) * out_tensor_size);
+    SimpleDeviceMem dout_device_buf(sizeof(DOutDataType) * out_tensor_size);
+    SimpleDeviceMem din_device_buf(sizeof(DInDataType) * in_tensor_size);
+
+    // Generate index data from max pool forward
+    {
+        using MaxPoolFwdDeviceOp =
+            ck::tensor_operation::device::DevicePoolFwd<InOutRank,
+                                                        WindowRank,
+                                                        InDataType,
+                                                        OutDataType,
+                                                        IndexDataType,
+                                                        InLayout,
+                                                        OutLayout,
+                                                        ck::ReduceTensorOp::MAX,
+                                                        true>;
+
+        const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            MaxPoolFwdDeviceOp>::GetInstances();
+
+        auto& op_ptr = op_ptrs[0];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutDataType*>(out_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+            in_length,
+            window_spatial_lengths,
+            out_length,
+            in_tensor_stride,
+            out_tensor_stride,
+            out_tensor_stride,
+            window_strides,
+            window_dilations,
+            input_left_pads,
+            input_right_pads,
+            pooling_dims);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+    }
+
+    // Run MaxPool bwd
+    using MaxPoolBwdDeviceOp =
+        ck::tensor_operation::device::DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>;
+
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        MaxPoolBwdDeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr      = op_ptrs[i];
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(dout_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+            static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+            out_tensor_size,
+            in_tensor_size,
+            window_spatial_lengths,
+            window_strides,
+            window_dilations);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+
+            SimpleDeviceMem workspace(workspace_sz);
+
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes = in_tensor_size * sizeof(DInDataType) +
+                                    out_tensor_size * sizeof(IndexDataType) +
+                                    out_tensor_size * sizeof(DOutDataType);
+
+            float gb_per_sec = num_bytes / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << "GB / s,"
+                      << op_name << std::endl;
+
+            if(ave_time < best_ave_time)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    // run the best intance
+    if(found)
+    {
+        std::cout << "Best Perf: " << best_ave_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_op_name << std::endl;
+
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InDataType*>(dout_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+            static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+            out_tensor_size,
+            in_tensor_size,
+            window_spatial_lengths,
+            window_strides,
+            window_dilations);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            size_t workspace_sz = op_ptr->GetWorkSpaceSize(argument_ptr.get());
+
+            SimpleDeviceMem workspace(workspace_sz);
+
+            op_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace.GetDeviceBuffer());
+
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
diff --git a/client_example/19_pool_fwd/max_pool2d_fwd.cpp b/client_example/19_pool/max_pool2d_fwd.cpp
similarity index 100%
rename from client_example/19_pool_fwd/max_pool2d_fwd.cpp
rename to client_example/19_pool/max_pool2d_fwd.cpp
diff --git a/client_example/19_pool_fwd/CMakeLists.txt b/client_example/19_pool_fwd/CMakeLists.txt
deleted file mode 100644
index 13f9f73c83..0000000000
--- a/client_example/19_pool_fwd/CMakeLists.txt
+++ /dev/null
@@ -1,5 +0,0 @@
-add_executable(client_max_pool2d_fwd max_pool2d_fwd.cpp)
-target_link_libraries(client_max_pool2d_fwd PRIVATE composable_kernel::device_operations)
-
-add_executable(client_avg_pool3d_fwd avg_pool3d_fwd.cpp)
-target_link_libraries(client_avg_pool3d_fwd PRIVATE composable_kernel::device_operations)
\ No newline at end of file
diff --git a/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
index d730067940..2c1e669375 100644
--- a/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
+++ b/example/49_maxpool2d_bwd/maxpool2d_bwd_common.hpp
@@ -8,7 +8,7 @@
 #include "ck/ck.hpp"
 #include "ck/utility/reduction_enums.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_pool2d_fwd_nhwc_nhwc.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_max_pool_bwd_impl.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/check_err.hpp"
@@ -60,7 +60,7 @@ bool maxpool_bwd_test(bool do_verification,
                                                                 1>; // InSrcOutDstVectorSize
 
     using DeviceMaxPoolBwdInstance = ck::tensor_operation::device::
-        DeviceIndexPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 4>;
+        DeviceMaxPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 4>;
 
     const ck::index_t Ys = (Y - 1) * window_dilation_h + 1;
     const ck::index_t Xs = (X - 1) * window_dilation_w + 1;
@@ -155,7 +155,8 @@ bool maxpool_bwd_test(bool do_verification,
         dout_n_c_ho_wo.mDesc.GetElementSpaceSize(),
         din_n_c_hi_wi_device.mDesc.GetElementSpaceSize(),
         window_spatial_lengths,
-        window_strides);
+        window_strides,
+        window_dilations);
 
     if(!pool_bwd.IsSupportedArgument(pool_bwd_argument_ptr.get()))
     {
diff --git a/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp b/include/ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp
similarity index 83%
rename from include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
rename to include/ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp
index bf81ed9f5b..5a4a9cac1e 100644
--- a/include/ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp
@@ -13,7 +13,7 @@ namespace device {
 
 // For pooling which used indexable operation, such as MaxPool, MinPool...etc
 template <typename DOutDataType, typename IndexDataType, typename DInDataType>
-struct DeviceIndexPoolBwd : public BaseOperator
+struct DeviceMaxPoolBwd : public BaseOperator
 {
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_dout,
@@ -22,7 +22,8 @@ struct DeviceIndexPoolBwd : public BaseOperator
                         index_t dout_length,
                         index_t din_length,
                         std::vector<ck::index_t> window_lengths,
-                        std::vector<ck::index_t> window_strides) = 0;
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> window_dilations) = 0;
 
     virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
 };
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_max_pool_bwd_impl.hpp
similarity index 93%
rename from include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_max_pool_bwd_impl.hpp
index 175994d49f..e98a85defe 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_index_pool_bwd_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_max_pool_bwd_impl.hpp
@@ -8,7 +8,7 @@
 
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/device_index_pool_bwd.hpp"
+#include "ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_put_element_1d.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_elementwise_1d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
@@ -25,7 +25,7 @@ template <typename DOutDataType,
           typename IndexDataType,
           typename DInDataType,
           ck::index_t InOutVectorSize>
-struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDataType, DInDataType>
+struct DeviceMaxPoolBwdImpl : public DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>
 {
     using DInDataType_AutomicAddPreCast =
         conditional_t<is_same_v<DInDataType, float> || is_same_v<DInDataType, double>,
@@ -91,7 +91,8 @@ struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDat
                  index_t dout_length,
                  index_t din_length,
                  const std::vector<ck::index_t>& window_lengths,
-                 const std::vector<ck::index_t>& window_strides)
+                 const std::vector<ck::index_t>& window_strides,
+                 const std::vector<ck::index_t>& window_dilations)
             : p_dout_{p_dout},
               p_indices_{p_indices},
               p_din_{p_din},
@@ -102,7 +103,8 @@ struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDat
         {
             for(size_t i = 0; i < window_lengths.size(); ++i)
             {
-                windowOverlap_ |= window_lengths.at(i) > window_strides.at(i);
+                auto eff = (window_lengths.at(i) - 1) * window_dilations.at(i) + 1;
+                windowOverlap_ |= eff > window_strides.at(i);
             }
         }
 
@@ -228,6 +230,11 @@ struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDat
                 }
                 else
                 {
+                    hip_check_error(hipMemsetAsync(arg.p_din_,
+                                                   0,
+                                                   arg.din_length_raw_ * sizeof(DInDataType),
+                                                   stream_config.stream_id_));
+
                     const auto put_kernel = kernel_put_element_1d<GridwisePutElementSet,
                                                                   InOutGrid1dDesc,
                                                                   DOutDataType,
@@ -292,7 +299,8 @@ struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDat
                         index_t dout_length,
                         index_t din_length,
                         std::vector<ck::index_t> window_lengths,
-                        std::vector<ck::index_t> window_strides) override
+                        std::vector<ck::index_t> window_strides,
+                        std::vector<ck::index_t> window_dilations) override
     {
         // Assume p_dout, p_indices, p_din are packed memory space, dout_length and din_length are
         // physical size of the packed tensor
@@ -302,7 +310,8 @@ struct DeviceIndexPoolBwdImpl : public DeviceIndexPoolBwd<DOutDataType, IndexDat
                                           dout_length,
                                           din_length,
                                           window_lengths,
-                                          window_strides);
+                                          window_strides,
+                                          window_dilations);
     }
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
diff --git a/include/ck/utility/reduction_operator.hpp b/include/ck/utility/reduction_operator.hpp
index 36c25203ea..5480a98409 100644
--- a/include/ck/utility/reduction_operator.hpp
+++ b/include/ck/utility/reduction_operator.hpp
@@ -116,7 +116,15 @@ struct Max
     template <typename T>
     __host__ __device__ static constexpr T GetIdentityValue()
     {
-        return NumericLimits<T>::Lowest();
+        if constexpr(is_same_v<T, bhalf_t>)
+        {
+            float val = NumericLimits<float>::Lowest();
+            return type_convert<bhalf_t>(val);
+        }
+        else
+        {
+            return NumericLimits<T>::Lowest();
+        }
     };
 
     __host__ __device__ static constexpr bool
@@ -138,6 +146,15 @@ struct Max
             a = b;
     }
 
+    __host__ __device__ inline constexpr void operator()(bhalf_t& a, bhalf_t b) const
+    {
+        float a_ = type_convert<float>(a);
+        float b_ = type_convert<float>(b);
+
+        if(a_ < b_)
+            a = b;
+    }
+
     template <typename T>
     __host__ __device__ inline constexpr void operator()(T& a, T b, bool& changed) const
     {
@@ -152,6 +169,18 @@ struct Max
             changed = true;
         }
     }
+
+    __host__ __device__ inline constexpr void operator()(bhalf_t& a, bhalf_t b, bool& changed) const
+    {
+        float a_ = type_convert<float>(a);
+        float b_ = type_convert<float>(b);
+
+        if(a_ < b_)
+        {
+            a       = b;
+            changed = true;
+        }
+    }
 };
 
 struct Min
@@ -159,6 +188,15 @@ struct Min
     template <typename T>
     __host__ __device__ static constexpr T GetIdentityValue()
     {
+        if constexpr(is_same_v<T, bhalf_t>)
+        {
+            float val = NumericLimits<float>::Max();
+            return type_convert<bhalf_t>(val);
+        }
+        else
+        {
+            return NumericLimits<T>::Max();
+        }
         return NumericLimits<T>::Max();
     };
 
@@ -181,6 +219,15 @@ struct Min
             a = b;
     }
 
+    __host__ __device__ inline constexpr void operator()(bhalf_t& a, bhalf_t b) const
+    {
+        float a_ = type_convert<float>(a);
+        float b_ = type_convert<float>(b);
+
+        if(a_ > b_)
+            a = b;
+    }
+
     template <typename T>
     __host__ __device__ inline constexpr void operator()(T& a, T b, bool& changed) const
     {
@@ -195,6 +242,18 @@ struct Min
             changed = true;
         }
     }
+
+    __host__ __device__ inline constexpr void operator()(bhalf_t& a, bhalf_t b, bool& changed) const
+    {
+        float a_ = type_convert<float>(a);
+        float b_ = type_convert<float>(b);
+
+        if(a_ > b_)
+        {
+            a       = b;
+            changed = true;
+        }
+    }
 };
 
 struct AMax
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
index 3f1fc6165c..60c74fbf14 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp
@@ -53,7 +53,16 @@ struct ReferenceMaxPoolBwd : public device::BaseOperator
             {
                 int index = arg.indices_.mData[i];
                 if(index >= 0 && index < din_length)
-                    buf[index] += ck::type_convert<ConputeDataType>(arg.dout_.mData[i]);
+                {
+                    if constexpr(is_same_v<ConputeDataType, bhalf_t>)
+                    {
+                        float buf_val = ck::type_convert<float>(buf[index]);
+                        buf_val += ck::type_convert<float>(arg.dout_.mData[i]);
+                        buf[index] = ck::type_convert<ConputeDataType>(buf_val);
+                    }
+                    else
+                        buf[index] += ck::type_convert<ConputeDataType>(arg.dout_.mData[i]);
+                }
             }
 
             for(int i = 0; i < din_length; ++i)
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
index 067e0b2eb9..cf241ac1b1 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp
@@ -256,10 +256,12 @@ struct ReferencePoolingFwd : public device::BaseOperator
 
                     for(ck::index_t y = 0; y < arg.window_spatial_lengths_[0]; ++y)
                     {
-                        ck::index_t hi = ho * arg.window_strides_[0] + y - arg.in_left_pads_[0];
+                        ck::index_t hi = ho * arg.window_strides_[0] +
+                                         y * arg.window_dilations_[0] - arg.in_left_pads_[0];
                         for(ck::index_t x = 0; x < arg.window_spatial_lengths_[1]; ++x)
                         {
-                            ck::index_t wi = wo * arg.window_strides_[1] + x - arg.in_left_pads_[1];
+                            ck::index_t wi = wo * arg.window_strides_[1] +
+                                             x * arg.window_dilations_[1] - arg.in_left_pads_[1];
                             if(hi >= 0 &&
                                hi < static_cast<ck::index_t>(arg.in_.mDesc.GetLengths()[2]) &&
                                wi >= 0 &&
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/avg_pool3d_bwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/avg_pool3d_bwd.hpp
new file mode 100644
index 0000000000..949e1d2dd0
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/avg_pool3d_bwd.hpp
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_operation/gpu/device/device_avgpool_bwd.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+#ifdef CK_ENABLE_FP16
+void add_device_avgpool_bwd_ndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceAvgPoolBwd<3, F16, F16, NDHWC, NDHWC>>>&);
+#endif
+#ifdef CK_ENABLE_BF16
+void add_device_avgpool_bwd_ndhwc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceAvgPoolBwd<3, BF16, BF16, NDHWC, NDHWC>>>&);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_avgpool_bwd_ndhwc_f32_instances(
+    std::vector<std::unique_ptr<DeviceAvgPoolBwd<3, F32, F32, NDHWC, NDHWC>>>&);
+#endif
+template <typename DOutDataType, typename DInDataType, typename InLayout, typename OutLayout>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::
+        DeviceAvgPoolBwd<3, DOutDataType, DInDataType, InLayout, OutLayout>>
+{
+    using DeviceOp = DeviceAvgPoolBwd<3, DOutDataType, DInDataType, InLayout, OutLayout>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+        if constexpr(is_same_v<InLayout, NDHWC> && is_same_v<OutLayout, NDHWC>)
+        {
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<DOutDataType, F16> && is_same_v<DInDataType, F16>)
+                add_device_avgpool_bwd_ndhwc_f16_instances(op_ptrs);
+#endif
+#ifdef CK_ENABLE_BF16
+            else if constexpr(is_same_v<DOutDataType, BF16> && is_same_v<DInDataType, BF16>)
+                add_device_avgpool_bwd_ndhwc_bf16_instances(op_ptrs);
+#endif
+#ifdef CK_ENABLE_FP32
+            else if constexpr(is_same_v<DOutDataType, F32> && is_same_v<DInDataType, F32>)
+                add_device_avgpool_bwd_ndhwc_f32_instances(op_ptrs);
+#endif
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/max_pool_bwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/max_pool_bwd.hpp
new file mode 100644
index 0000000000..63ea4f2891
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/max_pool_bwd.hpp
@@ -0,0 +1,58 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_operation/gpu/device/device_max_pool_bwd.hpp"
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+#ifdef CK_ENABLE_FP16
+void add_device_maxpool_bwd_f16_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F16, I32, F16>>>&);
+#endif
+#ifdef CK_ENABLE_BF16
+void add_device_maxpool_bwd_bf16_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<BF16, I32, BF16>>>&);
+#endif
+#ifdef CK_ENABLE_FP32
+void add_device_maxpool_bwd_f32_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F32, I32, F32>>>&);
+#endif
+template <typename DOutDataType, typename IndexDataType, typename DInDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>>
+{
+    using DeviceOp = DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<DOutDataType, F16> && is_same_v<DInDataType, F16> &&
+                     is_same_v<IndexDataType, I32>)
+            add_device_maxpool_bwd_f16_instances(op_ptrs);
+#endif
+#ifdef CK_ENABLE_BF16
+        else if constexpr(is_same_v<DOutDataType, BF16> && is_same_v<DInDataType, BF16> &&
+                          is_same_v<IndexDataType, I32>)
+            add_device_maxpool_bwd_bf16_instances(op_ptrs);
+#endif
+#ifdef CK_ENABLE_FP32
+        else if constexpr(is_same_v<DOutDataType, F32> && is_same_v<DInDataType, F32> &&
+                          is_same_v<IndexDataType, I32>)
+            add_device_maxpool_bwd_f32_instances(op_ptrs);
+#endif
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
index f520c7eded..94ee68a409 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp
@@ -37,6 +37,21 @@ void add_device_pool3d_fwd_ndhwc_index_f16_instances(
     std::vector<std::unique_ptr<
         DevicePoolFwd<InOutRank, WindowRank, F16, F16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
 #endif
+#ifdef CK_ENABLE_BF16
+// BF16
+void add_device_pool3d_fwd_ndhwc_bf16_instances(
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, BF16, BF16, I32, NDHWC, NDHWC, MaxOp, false>>>&);
+
+void add_device_pool3d_fwd_ndhwc_bf16_instances(
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, BF16, BF16, I32, NDHWC, NDHWC, AvgOp, false>>>&);
+
+// BF16 - return index
+void add_device_pool3d_fwd_ndhwc_index_bf16_instances(
+    std::vector<std::unique_ptr<
+        DevicePoolFwd<InOutRank, WindowRank, BF16, BF16, I32, NDHWC, NDHWC, MaxOp, true>>>&);
+#endif
 #ifdef CK_ENABLE_FP32
 // FP32
 void add_device_pool3d_fwd_ndhwc_f32_instances(
@@ -98,9 +113,23 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DevicePoolFw
                 }
             }
 #endif
+#ifdef CK_ENABLE_BF16
+            else if constexpr(is_same_v<InDataType, BF16> && is_same_v<OutDataType, BF16> &&
+                              is_same_v<IndexDataType, I32>)
+            {
+                if constexpr(OutputIndex && ReduceOpId == MaxOp)
+                {
+                    add_device_pool3d_fwd_ndhwc_index_bf16_instances(op_ptrs);
+                }
+                else
+                {
+                    add_device_pool3d_fwd_ndhwc_bf16_instances(op_ptrs);
+                }
+            }
+#endif
 #ifdef CK_ENABLE_FP32
-            if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
-                         is_same_v<IndexDataType, I32>)
+            else if constexpr(is_same_v<InDataType, F32> && is_same_v<OutDataType, F32> &&
+                              is_same_v<IndexDataType, I32>)
             {
                 if constexpr(OutputIndex && ReduceOpId == MaxOp)
                 {
diff --git a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt
new file mode 100644
index 0000000000..ec079e3ba3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt
@@ -0,0 +1,11 @@
+set(DEVICE_AVGPOOL_BWD_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_f32_instance.cpp)
+endif()
+add_instance_library(device_avg_pool3d_bwd_instance ${DEVICE_AVGPOOL_BWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/avg_pool3d_bwd_ndhwc_instance_common.hpp b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/avg_pool3d_bwd_ndhwc_instance_common.hpp
new file mode 100644
index 0000000000..c989bbcd3d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/avg_pool3d_bwd_ndhwc_instance_common.hpp
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_avgpool3d_bwd_ndhwc_ndhwc.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I32   = int32_t;
+using F16   = ck::half_t;
+using BF16  = ck::bhalf_t;
+using F32   = float;
+using NDHWC = ck::tensor_layout::convolution::NDHWC;
+
+using device_avgpool_bwd_ndhwc_f16_instances =
+    // clang-format off
+    std::tuple <
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F16, F16, F32, 256, 256, 1, 1, 1, 1>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F16, F16, F32, 256, 256, 1, 2, 2, 2>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F16, F16, F32, 256, 256, 1, 4, 4, 4>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F16, F16, F32, 256, 256, 1, 8, 8, 8>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F16, F16, F32, 256, 32, 8, 8, 8, 8>
+               // clang-format on
+               >;
+
+using device_avgpool_bwd_ndhwc_bf16_instances =
+    // clang-format off
+    std::tuple <
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<BF16, BF16, F32, 256, 256, 1, 1, 1, 1>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<BF16, BF16, F32, 256, 256, 1, 2, 2, 2>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<BF16, BF16, F32, 256, 256, 1, 4, 4, 4>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<BF16, BF16, F32, 256, 256, 1, 8, 8, 8>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<BF16, BF16, F32, 256, 32, 8, 8, 8, 8>
+               // clang-format on
+               >;
+
+using device_avgpool_bwd_ndhwc_f32_instances =
+    // clang-format off
+    std::tuple <
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F32, F32, F32, 256, 256, 1, 1, 1, 1>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F32, F32, F32, 256, 256, 1, 2, 2, 2>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F32, F32, F32, 256, 256, 1, 4, 4, 4>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F32, F32, F32, 256, 256, 1, 8, 8, 8>,
+        DeviceAvgPool3dBwd_NDHWC_NDHWC<F32, F32, F32, 256, 32, 8, 8, 8, 8>
+               // clang-format on
+               >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp
new file mode 100644
index 0000000000..52a8852f30
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "avg_pool3d_bwd_ndhwc_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_avgpool_bwd_ndhwc_bf16_instances(
+    std::vector<std::unique_ptr<DeviceAvgPoolBwd<3, BF16, BF16, NDHWC, NDHWC>>>& instances)
+{
+    add_device_operation_instances(instances, device_avgpool_bwd_ndhwc_bf16_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f16_instance.cpp
new file mode 100644
index 0000000000..50de10e78e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f16_instance.cpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "avg_pool3d_bwd_ndhwc_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_avgpool_bwd_ndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceAvgPoolBwd<3, F16, F16, NDHWC, NDHWC>>>& instances)
+{
+    add_device_operation_instances(instances, device_avgpool_bwd_ndhwc_f16_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f32_instance.cpp
new file mode 100644
index 0000000000..0d4bb9a67f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/device_avg_pool3d_bwd_ndhwc_f32_instance.cpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "avg_pool3d_bwd_ndhwc_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_avgpool_bwd_ndhwc_f32_instances(
+    std::vector<std::unique_ptr<DeviceAvgPoolBwd<3, F32, F32, NDHWC, NDHWC>>>& instances)
+{
+    add_device_operation_instances(instances, device_avgpool_bwd_ndhwc_f32_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
new file mode 100644
index 0000000000..820a63480e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
@@ -0,0 +1,11 @@
+set(DEVICE_MAXPOOL_BWD_INSTANCES)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f16_instance.cpp)
+endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_bf16_instance.cpp)
+endif()
+if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f32_instance.cpp)
+endif()
+add_instance_library(device_max_pool_bwd_instance ${DEVICE_MAXPOOL_BWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_bf16_instance.cpp
new file mode 100644
index 0000000000..40628d58b2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_bf16_instance.cpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "max_pool_bwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_maxpool_bwd_bf16_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<BF16, I32, BF16>>>& instances)
+{
+    add_device_operation_instances(instances, device_maxpool_bwd_instances<BF16, I32, BF16>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f16_instance.cpp
new file mode 100644
index 0000000000..6c5cb27a7a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f16_instance.cpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "max_pool_bwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_maxpool_bwd_f16_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F16, I32, F16>>>& instances)
+{
+    add_device_operation_instances(instances, device_maxpool_bwd_instances<F16, I32, F16>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f32_instance.cpp
new file mode 100644
index 0000000000..0a8d5a7976
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/max_pool_bwd/device_max_pool_bwd_f32_instance.cpp
@@ -0,0 +1,20 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "max_pool_bwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_maxpool_bwd_f32_instances(
+    std::vector<std::unique_ptr<DeviceMaxPoolBwd<F32, I32, F32>>>& instances)
+{
+    add_device_operation_instances(instances, device_maxpool_bwd_instances<F32, I32, F32>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/max_pool_bwd/max_pool_bwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/max_pool_bwd/max_pool_bwd_instance_common.hpp
new file mode 100644
index 0000000000..0bba106ee2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/max_pool_bwd/max_pool_bwd_instance_common.hpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_max_pool_bwd_impl.hpp"
+#include "ck/utility/data_type.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I32  = int32_t;
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
+
+template <typename DOutDataType, typename IndexDataType, typename DInDataType>
+using device_maxpool_bwd_instances =
+    // clang-format off
+    std::tuple <
+        DeviceMaxPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 1>,
+        DeviceMaxPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 2>,
+        DeviceMaxPoolBwdImpl<DOutDataType, IndexDataType, DInDataType, 4>
+               // clang-format on
+               >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
index dd843426b2..63bbe75465 100644
--- a/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
@@ -3,6 +3,10 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
     list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
                                           device_max_pool3d_fwd_ndhwc_f16_instance.cpp)
 endif()
+if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+    list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp
+                                          device_max_pool3d_fwd_ndhwc_bf16_instance.cpp)
+endif()
 if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
     list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
                                           device_max_pool3d_fwd_ndhwc_f32_instance.cpp)
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp
new file mode 100644
index 0000000000..f102516992
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
+
+void add_device_pool3d_fwd_ndhwc_bf16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, BF16, BF16, I32, NDHWC, NDHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<BF16, BF16, I32, F32, ReduceOpId, false>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_bf16_instance.cpp
new file mode 100644
index 0000000000..5dc504e178
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/device_max_pool3d_fwd_ndhwc_bf16_instance.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "pool_fwd_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+static constexpr auto ReduceOpId = ck::ReduceTensorOp::MAX;
+
+void add_device_pool3d_fwd_ndhwc_bf16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, BF16, BF16, I32, NDHWC, NDHWC, ReduceOpId, false>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<BF16, BF16, I32, BF16, ReduceOpId, false>{});
+}
+
+void add_device_pool3d_fwd_ndhwc_index_bf16_instances(
+    std::vector<
+        std::unique_ptr<DevicePoolFwd<5, 3, BF16, BF16, I32, NDHWC, NDHWC, ReduceOpId, true>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_pool3d_fwd_ndhwc_instances<BF16, BF16, I32, BF16, ReduceOpId, true>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp b/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp
index 4d23ceab27..e8e7813295 100644
--- a/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/pool_fwd_instance_common.hpp
@@ -17,6 +17,7 @@ namespace instance {
 
 using I32   = int32_t;
 using F16   = ck::half_t;
+using BF16  = ck::bhalf_t;
 using F32   = float;
 using NDHWC = ck::tensor_layout::convolution::NDHWC;
 
diff --git a/profiler/include/profiler/profile_avg_pool3d_bwd_impl.hpp b/profiler/include/profiler/profile_avg_pool3d_bwd_impl.hpp
new file mode 100644
index 0000000000..e7e8f7213f
--- /dev/null
+++ b/profiler/include/profiler/profile_avg_pool3d_bwd_impl.hpp
@@ -0,0 +1,253 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+#include "ck/library/tensor_operation_instance/gpu/avg_pool3d_bwd.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_avgpool_bwd.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename TensorLayout>
+std::vector<ck::index_t> f_tensor_strides_ncdhw(ck::index_t N_,
+                                                ck::index_t C_,
+                                                ck::index_t D,
+                                                ck::index_t H,
+                                                ck::index_t W,
+                                                TensorLayout layout)
+{
+    using namespace ck::literals;
+    (void)N_;
+    if constexpr(ck::is_same<decltype(layout), ck::tensor_layout::convolution::NDHWC>::value)
+        return {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_};
+    else
+        throw std::runtime_error("not supported yet");
+};
+
+template <typename DOutDataType,
+          typename DInDataType,
+          typename ComputeDataType,
+          typename DOutLayout,
+          typename DInLayout>
+bool profile_avg_pool3d_bwd_impl(int do_verification,
+                                 int init_method,
+                                 bool do_log,
+                                 bool time_kernel,
+                                 std::vector<index_t> in_length, // NCDHW
+                                 std::vector<index_t> window_spatial_lengths,
+                                 std::vector<index_t> window_strides,
+                                 std::vector<index_t> window_dilations,
+                                 std::vector<index_t> input_left_pads,
+                                 std::vector<index_t> input_right_pads)
+{
+    constexpr index_t InOutRank  = 5;
+    constexpr index_t WindowRank = 3;
+
+    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
+       window_strides.size() != WindowRank || window_dilations.size() != WindowRank ||
+       input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+    {
+        std::cout << "Parameter is incorrect" << std::endl;
+        return false;
+    }
+
+    std::vector<index_t> out_length(InOutRank);
+
+    int N = in_length[0];
+    int C = in_length[1];
+
+    out_length[0] = N;
+    out_length[1] = C;
+
+    // Calculate Do, Ho, Wo
+    for(int i = 2; i < InOutRank; ++i)
+    {
+        auto pad1             = input_left_pads[i - 2];
+        auto pad2             = input_right_pads[i - 2];
+        auto windows_size     = window_spatial_lengths[i - 2];
+        auto windows_stride   = window_strides[i - 2];
+        auto windows_dilation = window_dilations[i - 2];
+        auto eff              = (windows_size - 1) * windows_dilation + 1;
+        out_length[i]         = (in_length[i] + pad1 + pad2 - eff) / windows_stride + 1;
+    }
+
+    int Di = in_length[2];
+    int Hi = in_length[3];
+    int Wi = in_length[4];
+    int Do = out_length[2];
+    int Ho = out_length[3];
+    int Wo = out_length[4];
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t D, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+        };
+
+    Tensor<DOutDataType> dout_n_c_do_ho_wo(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<DInDataType> din_n_c_di_hi_wi_device(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+    Tensor<DInDataType> din_n_c_di_hi_wi_host(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+
+    switch(init_method)
+    {
+    case 0: dout_n_c_do_ho_wo.GenerateTensorValue(GeneratorTensor_1<DOutDataType>{}); break;
+    case 1: dout_n_c_do_ho_wo.GenerateTensorValue(GeneratorTensor_2<DOutDataType>{-5, 5}); break;
+    default: dout_n_c_do_ho_wo.GenerateTensorValue(GeneratorTensor_3<DOutDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem dout_device_buf(sizeof(DOutDataType) * dout_n_c_do_ho_wo.mDesc.GetElementSpaceSize());
+    DeviceMem din_device_buf(sizeof(DInDataType) *
+                             din_n_c_di_hi_wi_device.mDesc.GetElementSpaceSize());
+
+    dout_device_buf.ToDevice(dout_n_c_do_ho_wo.mData.data());
+
+    using DeviceOp = ck::tensor_operation::device::
+        DeviceAvgPoolBwd<3, DOutDataType, DInDataType, DOutLayout, DInLayout>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using ReferencePoolingBwdInstance =
+            ck::tensor_operation::host::ReferenceAvgPoolBwd<3, DInDataType, DOutDataType>;
+
+        ReferencePoolingBwdInstance ref_pooling_bwd;
+        auto ref_pooling_bwd_argument = ref_pooling_bwd.MakeArgument(din_n_c_di_hi_wi_host,
+                                                                     dout_n_c_do_ho_wo,
+                                                                     window_spatial_lengths,
+                                                                     window_strides,
+                                                                     window_dilations,
+                                                                     input_left_pads,
+                                                                     input_right_pads);
+
+        auto ref_invoker = ref_pooling_bwd.MakeInvoker();
+        ref_invoker.Run(ref_pooling_bwd_argument);
+    }
+
+    int num_kernel = 0;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(
+            static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+            static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+            {N, C, Do, Ho, Wo},
+            {N, C, Di, Hi, Wi},
+            f_tensor_strides_ncdhw(N, C, Do, Ho, Wo, DOutLayout{}),
+            f_tensor_strides_ncdhw(N, C, Di, Hi, Wi, DInLayout{}),
+            window_spatial_lengths,
+            window_strides,
+            window_dilations,
+            input_left_pads,
+            input_right_pads);
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
+                LogRange(std::cout << "doutput lengths = ", out_length, ", ") << std::endl;
+            }
+
+            continue;
+        }
+
+        din_device_buf.SetZero();
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+        float avg_time   = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        std::size_t num_bytes =
+            dout_n_c_do_ho_wo.mDesc.GetElementSize() * sizeof(DOutDataType) +
+            din_n_c_di_hi_wi_device.mDesc.GetElementSize() * sizeof(DInDataType);
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            din_device_buf.FromDevice(din_n_c_di_hi_wi_device.mData.data());
+            bool pass = ck::utils::check_err(din_n_c_di_hi_wi_device.mData,
+                                             din_n_c_di_hi_wi_host.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+
+            if(do_log)
+            {
+                LogRangeAsType<float>(
+                    std::cout << "din_n_c_di_hi_wi_device: ", din_n_c_di_hi_wi_device.mData, ",")
+                    << std::endl;
+
+                LogRangeAsType<float>(
+                    std::cout << "din_n_c_di_hi_wi_host: ", din_n_c_di_hi_wi_host.mData, ",")
+                    << std::endl;
+            }
+
+            if(!pass)
+            {
+                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
+                LogRange(std::cout << "doutput lengths = [", out_length, ", ") << "]." << std::endl;
+                return false;
+            }
+            else
+            {
+                if(time_kernel)
+                    std::cout << "pass" << std::endl;
+            }
+        }
+    }
+
+    if(time_kernel)
+    {
+        LogRange(std::cout << "length = ", out_length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/include/profiler/profile_max_pool3d_bwd_impl.hpp b/profiler/include/profiler/profile_max_pool3d_bwd_impl.hpp
new file mode 100644
index 0000000000..15fb4e9034
--- /dev/null
+++ b/profiler/include/profiler/profile_max_pool3d_bwd_impl.hpp
@@ -0,0 +1,288 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/pool3d_fwd.hpp"
+#include "ck/library/tensor_operation_instance/gpu/max_pool_bwd.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_pool_fwd.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_maxpool_bwd.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <typename InDataType,
+          typename OutDataType,
+          typename IndexDataType,
+          typename DOutDataType,
+          typename DInDataType,
+          bool PropagateNan>
+bool profile_max_pool3d_bwd_impl(int do_verification,
+                                 int init_method,
+                                 bool do_log,
+                                 bool time_kernel,
+                                 std::vector<index_t> in_length, // NCDHW
+                                 std::vector<index_t> window_spatial_lengths,
+                                 std::vector<index_t> window_strides,
+                                 std::vector<index_t> window_dilations,
+                                 std::vector<index_t> input_left_pads,
+                                 std::vector<index_t> input_right_pads)
+{
+    // AtomicAdd only support f32 for now. ComputeDataType must be float32
+    using ComputeDataType = float;
+
+    constexpr index_t InOutRank  = 5;
+    constexpr index_t WindowRank = 3;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    if(in_length.size() != InOutRank || window_spatial_lengths.size() != WindowRank ||
+       window_strides.size() != WindowRank || window_dilations.size() != WindowRank ||
+       input_left_pads.size() != WindowRank || input_right_pads.size() != WindowRank)
+    {
+        std::cout << "Parameter is incorrect" << std::endl;
+        return false;
+    }
+
+    std::vector<index_t> out_length(InOutRank);
+
+    int N = in_length[0];
+    int C = in_length[1];
+
+    out_length[0] = N;
+    out_length[1] = C;
+
+    // Calculate Do, Ho, Wo
+    for(int i = 2; i < InOutRank; ++i)
+    {
+        auto pad1             = input_left_pads[i - 2];
+        auto pad2             = input_right_pads[i - 2];
+        auto windows_size     = window_spatial_lengths[i - 2];
+        auto windows_stride   = window_strides[i - 2];
+        auto windows_dilation = window_dilations[i - 2];
+        auto eff              = (windows_size - 1) * windows_dilation + 1;
+        out_length[i]         = (in_length[i] + pad1 + pad2 - eff) / windows_stride + 1;
+    }
+
+    int Di = in_length[2];
+    int Hi = in_length[3];
+    int Wi = in_length[4];
+    int Do = out_length[2];
+    int Ho = out_length[3];
+    int Wo = out_length[4];
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t N_, std::size_t C_, std::size_t D, std::size_t H, std::size_t W) {
+            using namespace ck::literals;
+
+            return HostTensorDescriptor({N_, C_, D, H, W},
+                                        {D * C_ * H * W, 1_uz, C_ * H * W, W * C_, C_});
+        };
+
+    Tensor<InDataType> in_n_c_di_hi_wi(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+    Tensor<OutDataType> out_n_c_do_ho_wo(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<IndexDataType> out_indices_n_c_do_ho_wo(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<DOutDataType> dout_n_c_do_ho_wo(f_host_tensor_descriptor(N, C, Do, Ho, Wo));
+    Tensor<DInDataType> din_n_c_di_hi_wi_host(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+
+    Tensor<DInDataType> din_n_c_di_hi_wi_device(f_host_tensor_descriptor(N, C, Di, Hi, Wi));
+
+    switch(init_method)
+    {
+    case 0:
+        in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_1<InDataType>{});
+        dout_n_c_do_ho_wo.GenerateTensorValue(GeneratorTensor_1<DOutDataType>{});
+        break;
+    case 1:
+        in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5});
+        dout_n_c_do_ho_wo.GenerateTensorValue(GeneratorTensor_2<DOutDataType>{-5, 5});
+        break;
+    default:
+        in_n_c_di_hi_wi.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+        dout_n_c_do_ho_wo.GenerateTensorValue(GeneratorTensor_3<DOutDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem indices_device_buf(sizeof(IndexDataType) *
+                                 out_indices_n_c_do_ho_wo.mDesc.GetElementSpaceSize());
+    DeviceMem dout_device_buf(sizeof(DOutDataType) * dout_n_c_do_ho_wo.mDesc.GetElementSpaceSize());
+    DeviceMem din_device_buf(sizeof(DInDataType) *
+                             din_n_c_di_hi_wi_device.mDesc.GetElementSpaceSize());
+
+    // Generate index data from forwarding
+    {
+        using ReferencePoolingFwdInstance =
+            ck::tensor_operation::host::ReferencePoolingFwd<InOutRank,
+                                                            WindowRank,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ComputeDataType,
+                                                            IndexDataType,
+                                                            ck::ReduceTensorOp::MAX,
+                                                            false,
+                                                            true>;
+
+        ReferencePoolingFwdInstance ref_pooling_fwd;
+        auto ref_pooling_fwd_argument = ref_pooling_fwd.MakeArgument(in_n_c_di_hi_wi,
+                                                                     out_n_c_do_ho_wo,
+                                                                     out_indices_n_c_do_ho_wo,
+                                                                     window_spatial_lengths,
+                                                                     window_strides,
+                                                                     window_dilations,
+                                                                     input_left_pads,
+                                                                     input_right_pads);
+        auto ref_pooling_fwd_invoker  = ref_pooling_fwd.MakeInvoker();
+        ref_pooling_fwd_invoker.Run(ref_pooling_fwd_argument);
+    }
+
+    indices_device_buf.ToDevice(out_indices_n_c_do_ho_wo.mData.data());
+    dout_device_buf.ToDevice(dout_n_c_do_ho_wo.mData.data());
+
+    using DeviceOp =
+        ck::tensor_operation::device::DeviceMaxPoolBwd<DOutDataType, IndexDataType, DInDataType>;
+
+    // get device op instances
+    const auto instance_ptrs =
+        ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+            DeviceOp>::GetInstances();
+
+    std::cout << "found " << instance_ptrs.size() << " instances" << std::endl;
+
+    std::string best_instance_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    if(do_verification)
+    {
+        using ReferencePoolingBwdInstance =
+            ck::tensor_operation::host::ReferenceMaxPoolBwd<DOutDataType,
+                                                            IndexDataType,
+                                                            ComputeDataType,
+                                                            DInDataType,
+                                                            PassThrough>;
+
+        ReferencePoolingBwdInstance ref_pooling_bwd;
+        auto ref_pooling_bwd_argument = ref_pooling_bwd.MakeArgument(
+            dout_n_c_do_ho_wo, out_indices_n_c_do_ho_wo, din_n_c_di_hi_wi_host, PassThrough{});
+        auto ref_invoker = ref_pooling_bwd.MakeInvoker();
+        ref_invoker.Run(ref_pooling_bwd_argument);
+    }
+
+    int num_kernel = 0;
+
+    for(auto& inst_ptr : instance_ptrs)
+    {
+        auto argument_ptr = inst_ptr->MakeArgumentPointer(
+            static_cast<DOutDataType*>(dout_device_buf.GetDeviceBuffer()),
+            static_cast<IndexDataType*>(indices_device_buf.GetDeviceBuffer()),
+            static_cast<DInDataType*>(din_device_buf.GetDeviceBuffer()),
+            dout_n_c_do_ho_wo.mDesc.GetElementSpaceSize(),
+            din_n_c_di_hi_wi_device.mDesc.GetElementSpaceSize(),
+            window_spatial_lengths,
+            window_strides,
+            window_dilations);
+
+        if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            ++num_kernel;
+        }
+        else
+        {
+            if(time_kernel)
+            {
+                std::cout << inst_ptr->GetTypeString() << " skipped due to unsupported argument: ";
+                LogRange(std::cout << "doutput lengths = ", out_length, ", ") << std::endl;
+            }
+
+            continue;
+        }
+
+        size_t workspace_sz = inst_ptr->GetWorkSpaceSize(argument_ptr.get());
+        DeviceMem workspace_device_buf(workspace_sz);
+        inst_ptr->SetWorkSpacePointer(argument_ptr.get(), workspace_device_buf.GetDeviceBuffer());
+
+        auto invoker_ptr = inst_ptr->MakeInvokerPointer();
+        float avg_time   = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+
+        std::size_t num_bytes =
+            dout_n_c_do_ho_wo.mDesc.GetElementSize() * sizeof(DOutDataType) +
+            out_indices_n_c_do_ho_wo.mDesc.GetElementSize() * sizeof(IndexDataType) +
+            din_n_c_di_hi_wi_device.mDesc.GetElementSize() * sizeof(DInDataType);
+
+        float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+        if(time_kernel)
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << inst_ptr->GetTypeString() << std::endl;
+
+        if(avg_time < best_avg_time)
+        {
+            best_instance_name = inst_ptr->GetTypeString();
+            best_avg_time      = avg_time;
+            best_gb_per_sec    = gb_per_sec;
+        }
+
+        if(do_verification)
+        {
+            din_device_buf.FromDevice(din_n_c_di_hi_wi_device.mData.data());
+
+            bool pass = ck::utils::check_err(din_n_c_di_hi_wi_device.mData,
+                                             din_n_c_di_hi_wi_host.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+
+            if(do_log)
+            {
+                LogRangeAsType<float>(
+                    std::cout << "out_indices_n_c_do_ho_wo: ", out_indices_n_c_do_ho_wo.mData, ",")
+                    << std::endl;
+
+                LogRangeAsType<float>(
+                    std::cout << "din_n_c_di_hi_wi_device: ", din_n_c_di_hi_wi_device.mData, ",")
+                    << std::endl;
+
+                LogRangeAsType<float>(
+                    std::cout << "din_n_c_di_hi_wi_host: ", din_n_c_di_hi_wi_host.mData, ",")
+                    << std::endl;
+            }
+
+            if(!pass)
+            {
+                std::cout << inst_ptr->GetTypeString() << " failed verification: ";
+                LogRange(std::cout << "doutput lengths = [", out_length, ", ") << "]." << std::endl;
+                return false;
+            }
+            else
+            {
+                if(time_kernel)
+                    std::cout << "pass" << std::endl;
+            }
+        }
+    }
+
+    if(time_kernel)
+    {
+        LogRange(std::cout << "length = ", out_length, ",") << std::endl;
+        std::cout << "best perf = " << best_avg_time << " ms, " << best_gb_per_sec << " GB/s, "
+                  << best_instance_name << std::endl;
+    }
+
+    if(num_kernel == 0)
+    {
+        std::cout << "Error: No kernel is applicable" << std::endl;
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index b305e57598..66755c47c5 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -19,6 +19,8 @@ set(PROFILER_SOURCES
     profile_groupnorm.cpp
     profile_layernorm.cpp
     profile_max_pool3d_fwd.cpp
+    profile_avg_pool3d_bwd.cpp
+    profile_max_pool3d_bwd.cpp
     profile_softmax.cpp
     profile_batchnorm_fwd.cpp
     profile_batchnorm_bwd.cpp
@@ -76,6 +78,8 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool3d_fwd_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_avg_pool3d_bwd_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 if(DL_KERNELS)
diff --git a/profiler/src/profile_avg_pool3d_bwd.cpp b/profiler/src/profile_avg_pool3d_bwd.cpp
new file mode 100644
index 0000000000..0ff50a5292
--- /dev/null
+++ b/profiler/src/profile_avg_pool3d_bwd.cpp
@@ -0,0 +1,175 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <unordered_map>
+
+#include "profiler/data_type_enum.hpp"
+#include "profiler/profile_avg_pool3d_bwd_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+struct maxPoolbwdArgParser
+{
+    std::unordered_map<std::string, std::vector<int>> long_opts = {{"length", {}},
+                                                                   {"wsize", {}},
+                                                                   {"wstride", {}},
+                                                                   {"wdilation", {}},
+                                                                   {"pad1", {}},
+                                                                   {"pad2", {}}};
+
+    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
+    {
+        if(std::string("--") + key == argv[i])
+        {
+            int pos = i;
+            while(++i < argc && argv[i][0] != '-') {}
+            int end = i;
+            for(int j = pos + 1; j < end; j++)
+            {
+                long_opts[key].push_back(std::stoi(argv[j]));
+            }
+            return true;
+        }
+        return false;
+    }
+
+    void operator()(int argc, char* argv[])
+    {
+        for(auto& kv : long_opts)
+        {
+            for(int i = 1; i < argc; i++)
+            {
+                if(parse_opt(argc, argv, kv.first, i))
+                    break;
+            }
+        }
+    }
+};
+
+void print_help_avg_pool3d_bwd()
+{
+    std::cout << "arg1: data type (0: fp16; 1: fp32; 5: bf16)\n"
+              << "arg2: verification (0: no; 1: yes)\n"
+              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+              << "arg4: print tensor value (0: no; 1: yes)\n"
+              << "arg5: time kernel (0=no, 1=yes)\n"
+              << "--length: input tensor length for NCDHW(e.g, --length 2 32 30 30 30) \n"
+              << "--wsize: window size for ZYX (e.g, --wsize 2 2 2) \n"
+              << "--wstride: window stride for DHW (e.g, --wstride 2 2 2) \n"
+              << "--wdilation: window dilation for DHW (e.g, --wdilation 1 1 1) \n"
+              << "--pad1: left side of padding in DHW (e.g, --pad1 1 1 1) \n"
+              << "--pad2: right side of padding in DHW (e.g, --pad2 1 1 1) \n"
+              << "eg: ckProfiler avg_pool3d_bwd 0 1 2 0 1 --length 2 32 30 30 30 --wsize 2 2 2 "
+                 "--wstride 2 2 2 --wdilation 1 1 1 --pad1 1 1 1 --pad2 1 1 1"
+              << std::endl;
+}
+
+int profile_avg_pool3d_bwd(int argc, char* argv[])
+{
+    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
+    bool do_verification       = true;
+    int init_method            = 0;
+    bool do_log                = false;
+    bool time_kernel           = true;
+
+    std::vector<index_t> in_length = {2, 32, 30, 30, 30};
+    std::vector<index_t> wsize     = {2, 2, 2};
+    std::vector<index_t> wstride   = {2, 2, 2};
+    std::vector<index_t> wdilation = {1, 1, 1};
+    std::vector<index_t> pad1      = {1, 1, 1};
+    std::vector<index_t> pad2      = {1, 1, 1};
+
+    if(argc != 2 && argc != 33)
+    {
+        print_help_avg_pool3d_bwd();
+        return 0;
+    }
+    else if(argc == 33)
+    {
+        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
+        do_verification = std::stoi(argv[3]);
+        init_method     = std::stoi(argv[4]);
+        do_log          = std::stoi(argv[5]);
+        time_kernel     = std::stoi(argv[6]);
+
+        // parse the long options
+        maxPoolbwdArgParser arg_parser;
+        arg_parser(argc, argv);
+        in_length = arg_parser.long_opts["length"];
+        wsize     = arg_parser.long_opts["wsize"];
+        wstride   = arg_parser.long_opts["wstride"];
+        wdilation = arg_parser.long_opts["wdilation"];
+        pad1      = arg_parser.long_opts["pad1"];
+        pad2      = arg_parser.long_opts["pad2"];
+    }
+
+#ifdef CK_ENABLE_FP16
+    using F16 = ck::half_t;
+#endif
+#ifdef CK_ENABLE_BF16
+    using BF16 = ck::bhalf_t;
+#endif
+#ifdef CK_ENABLE_FP32
+    using F32 = float;
+#endif
+    using NDHWC = ck::tensor_layout::convolution::NDHWC;
+
+    if(false)
+        ;
+#ifdef CK_ENABLE_FP16
+    else if(data_type == ck::DataTypeEnum::Half)
+    {
+        ck::profiler::profile_avg_pool3d_bwd_impl<F16, F16, F16, NDHWC, NDHWC>(do_verification,
+                                                                               init_method,
+                                                                               do_log,
+                                                                               time_kernel,
+                                                                               in_length,
+                                                                               wsize,
+                                                                               wstride,
+                                                                               wdilation,
+                                                                               pad1,
+                                                                               pad2);
+    }
+#endif
+#ifdef CK_ENABLE_BF16
+    else if(data_type == ck::DataTypeEnum::BFloat16)
+    {
+        ck::profiler::profile_avg_pool3d_bwd_impl<BF16, BF16, BF16, NDHWC, NDHWC>(do_verification,
+                                                                                  init_method,
+                                                                                  do_log,
+                                                                                  time_kernel,
+                                                                                  in_length,
+                                                                                  wsize,
+                                                                                  wstride,
+                                                                                  wdilation,
+                                                                                  pad1,
+                                                                                  pad2);
+    }
+#endif
+#ifdef CK_ENABLE_FP32
+    else if(data_type == ck::DataTypeEnum::Float)
+    {
+        ck::profiler::profile_avg_pool3d_bwd_impl<F32, F32, F32, NDHWC, NDHWC>(do_verification,
+                                                                               init_method,
+                                                                               do_log,
+                                                                               time_kernel,
+                                                                               in_length,
+                                                                               wsize,
+                                                                               wstride,
+                                                                               wdilation,
+                                                                               pad1,
+                                                                               pad2);
+    }
+#endif
+    else
+    {
+        throw std::runtime_error("not implemented yet");
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("avg_pool3d_bwd", "max_pool bwd", profile_avg_pool3d_bwd);
diff --git a/profiler/src/profile_max_pool3d_bwd.cpp b/profiler/src/profile_max_pool3d_bwd.cpp
new file mode 100644
index 0000000000..45a64df423
--- /dev/null
+++ b/profiler/src/profile_max_pool3d_bwd.cpp
@@ -0,0 +1,177 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <vector>
+#include <unordered_map>
+
+#include "profiler/data_type_enum.hpp"
+#include "profiler/profile_max_pool3d_bwd_impl.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "profiler_operation_registry.hpp"
+
+using ck::index_t;
+
+struct maxPoolbwdArgParser
+{
+    std::unordered_map<std::string, std::vector<int>> long_opts = {{"length", {}},
+                                                                   {"wsize", {}},
+                                                                   {"wstride", {}},
+                                                                   {"wdilation", {}},
+                                                                   {"pad1", {}},
+                                                                   {"pad2", {}}};
+
+    bool parse_opt(int argc, char* argv[], const std::string& key, int i)
+    {
+        if(std::string("--") + key == argv[i])
+        {
+            int pos = i;
+            while(++i < argc && argv[i][0] != '-') {}
+            int end = i;
+            for(int j = pos + 1; j < end; j++)
+            {
+                long_opts[key].push_back(std::stoi(argv[j]));
+            }
+            return true;
+        }
+        return false;
+    }
+
+    void operator()(int argc, char* argv[])
+    {
+        for(auto& kv : long_opts)
+        {
+            for(int i = 1; i < argc; i++)
+            {
+                if(parse_opt(argc, argv, kv.first, i))
+                    break;
+            }
+        }
+    }
+};
+
+void print_help_max_pool3d_bwd()
+{
+    std::cout << "arg1: data type (0: fp16; 1: fp32; 5: bf16)\n"
+              << "arg2: verification (0: no; 1: yes)\n"
+              << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
+              << "arg4: print tensor value (0: no; 1: yes)\n"
+              << "arg5: time kernel (0=no, 1=yes)\n"
+              << "--length: input tensor length for NCDHW(e.g, --length 2 32 30 30 30) \n"
+              << "--wsize: window size for ZYX (e.g, --wsize 2 2 2) \n"
+              << "--wstride: window stride for DHW (e.g, --wstride 2 2 2) \n"
+              << "--wdilation: window dilation for DHW (e.g, --wdilation 1 1 1) \n"
+              << "--pad1: left side of padding in DHW (e.g, --pad1 1 1 1) \n"
+              << "--pad2: right side of padding in DHW (e.g, --pad2 1 1 1) \n"
+              << "eg: ckProfiler max_pool3d_bwd 0 1 2 0 1 --length 2 32 30 30 30 --wsize 2 2 2 "
+                 "--wstride 2 2 2 --wdilation 1 1 1 --pad1 1 1 1 --pad2 1 1 1"
+              << std::endl;
+}
+
+int profile_max_pool3d_bwd(int argc, char* argv[])
+{
+    ck::DataTypeEnum data_type = ck::DataTypeEnum::Half;
+    bool do_verification       = true;
+    int init_method            = 0;
+    bool do_log                = false;
+    bool time_kernel           = true;
+
+    std::vector<index_t> in_length = {2, 32, 30, 30, 30};
+    std::vector<index_t> wsize     = {2, 2, 2};
+    std::vector<index_t> wstride   = {2, 2, 2};
+    std::vector<index_t> wdilation = {1, 1, 1};
+    std::vector<index_t> pad1      = {1, 1, 1};
+    std::vector<index_t> pad2      = {1, 1, 1};
+
+    if(argc != 2 && argc != 33)
+    {
+        print_help_max_pool3d_bwd();
+        return 0;
+    }
+    else if(argc == 33)
+    {
+        data_type       = static_cast<ck::DataTypeEnum>(std::stoi(argv[2]));
+        do_verification = std::stoi(argv[3]);
+        init_method     = std::stoi(argv[4]);
+        do_log          = std::stoi(argv[5]);
+        time_kernel     = std::stoi(argv[6]);
+
+        // parse the long options
+        maxPoolbwdArgParser arg_parser;
+        arg_parser(argc, argv);
+        in_length = arg_parser.long_opts["length"];
+        wsize     = arg_parser.long_opts["wsize"];
+        wstride   = arg_parser.long_opts["wstride"];
+        wdilation = arg_parser.long_opts["wdilation"];
+        pad1      = arg_parser.long_opts["pad1"];
+        pad2      = arg_parser.long_opts["pad2"];
+    }
+
+#ifdef CK_ENABLE_FP16
+    using F16 = ck::half_t;
+#endif
+#ifdef CK_ENABLE_BF16
+    using BF16 = ck::bhalf_t;
+#endif
+#ifdef CK_ENABLE_FP32
+    using F32 = float;
+#endif
+    using I32 = int32_t;
+
+    if(false)
+        ;
+#ifdef CK_ENABLE_FP16
+    else if(data_type == ck::DataTypeEnum::Half)
+    {
+        ck::profiler::profile_max_pool3d_bwd_impl<F16, F16, I32, F16, F16, false>(do_verification,
+                                                                                  init_method,
+                                                                                  do_log,
+                                                                                  time_kernel,
+                                                                                  in_length,
+                                                                                  wsize,
+                                                                                  wstride,
+                                                                                  wdilation,
+                                                                                  pad1,
+                                                                                  pad2);
+    }
+#endif
+#ifdef CK_ENABLE_BF16
+    else if(data_type == ck::DataTypeEnum::BFloat16)
+    {
+        ck::profiler::profile_max_pool3d_bwd_impl<BF16, BF16, I32, BF16, BF16, false>(
+            do_verification,
+            init_method,
+            do_log,
+            time_kernel,
+            in_length,
+            wsize,
+            wstride,
+            wdilation,
+            pad1,
+            pad2);
+    }
+#endif
+#ifdef CK_ENABLE_FP32
+    else if(data_type == ck::DataTypeEnum::Float)
+    {
+        ck::profiler::profile_max_pool3d_bwd_impl<F32, F32, I32, F32, F32, false>(do_verification,
+                                                                                  init_method,
+                                                                                  do_log,
+                                                                                  time_kernel,
+                                                                                  in_length,
+                                                                                  wsize,
+                                                                                  wstride,
+                                                                                  wdilation,
+                                                                                  pad1,
+                                                                                  pad2);
+    }
+#endif
+    else
+    {
+        throw std::runtime_error("not implemented yet");
+    }
+
+    return 0;
+}
+
+REGISTER_PROFILER_OPERATION("max_pool3d_bwd", "max_pool3d bwd", profile_max_pool3d_bwd);
diff --git a/profiler/src/profile_max_pool3d_fwd.cpp b/profiler/src/profile_max_pool3d_fwd.cpp
index da7ea9af4b..52fdf29fe4 100644
--- a/profiler/src/profile_max_pool3d_fwd.cpp
+++ b/profiler/src/profile_max_pool3d_fwd.cpp
@@ -51,7 +51,7 @@ struct maxPoolFwdArgParser
 
 void print_help_max_pool3d_fwd()
 {
-    std::cout << "arg1: data type (0: fp16; 1: fp32)\n"
+    std::cout << "arg1: data type (0: fp16; 1: fp32; 5: bf16)\n"
               << "arg2: verification (0: no; 1: yes)\n"
               << "arg3: initialization (0: no init; 1: integer value; 2: decimal value)\n"
               << "arg4: print tensor value (0: no; 1: yes)\n"
@@ -109,8 +109,15 @@ int profile_max_pool3d_fwd(int argc, char* argv[])
         pad2      = arg_parser.long_opts["pad2"];
     }
 
-    using F16   = ck::half_t;
-    using F32   = float;
+#ifdef CK_ENABLE_FP16
+    using F16 = ck::half_t;
+#endif
+#ifdef CK_ENABLE_BF16
+    using BF16 = ck::bhalf_t;
+#endif
+#ifdef CK_ENABLE_FP32
+    using F32 = float;
+#endif
     using I32   = int32_t;
     using NDHWC = ck::tensor_layout::convolution::NDHWC;
 
@@ -120,7 +127,10 @@ int profile_max_pool3d_fwd(int argc, char* argv[])
     constexpr auto ReduceOpId = ck::ReduceTensorOp::AVG;
 #endif
 
-    if(data_type == ck::DataTypeEnum::Half)
+    if(false)
+        ;
+#ifdef CK_ENABLE_FP16
+    else if(data_type == ck::DataTypeEnum::Half)
     {
         if(return_index)
             ck::profiler::
@@ -149,6 +159,51 @@ int profile_max_pool3d_fwd(int argc, char* argv[])
                     pad1,
                     pad2);
     }
+#endif
+#ifdef CK_ENABLE_BF16
+    else if(data_type == ck::DataTypeEnum::BFloat16)
+    {
+        if(return_index)
+            ck::profiler::profile_pool3d_fwd_impl<BF16,
+                                                  BF16,
+                                                  BF16,
+                                                  I32,
+                                                  NDHWC,
+                                                  NDHWC,
+                                                  ReduceOpId,
+                                                  false,
+                                                  true>(do_verification,
+                                                        init_method,
+                                                        do_log,
+                                                        time_kernel,
+                                                        in_length,
+                                                        wsize,
+                                                        wstride,
+                                                        wdilation,
+                                                        pad1,
+                                                        pad2);
+        else
+            ck::profiler::profile_pool3d_fwd_impl<BF16,
+                                                  BF16,
+                                                  BF16,
+                                                  I32,
+                                                  NDHWC,
+                                                  NDHWC,
+                                                  ReduceOpId,
+                                                  false,
+                                                  false>(do_verification,
+                                                         init_method,
+                                                         do_log,
+                                                         time_kernel,
+                                                         in_length,
+                                                         wsize,
+                                                         wstride,
+                                                         wdilation,
+                                                         pad1,
+                                                         pad2);
+    }
+#endif
+#ifdef CK_ENABLE_FP32
     else if(data_type == ck::DataTypeEnum::Float)
     {
         if(return_index)
@@ -178,6 +233,7 @@ int profile_max_pool3d_fwd(int argc, char* argv[])
                     pad1,
                     pad2);
     }
+#endif
     else
     {
         throw std::runtime_error("not implemented yet");
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index cd1f305313..a359c529a4 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -57,7 +57,7 @@ add_subdirectory(data_type)
 add_subdirectory(elementwise_normalization)
 add_subdirectory(batchnorm)
 add_subdirectory(contraction)
-add_subdirectory(pool_fwd)
+add_subdirectory(pool)
 add_subdirectory(batched_gemm_multi_d)
 add_subdirectory(grouped_convnd_bwd_data)
 if(GPU_TARGETS MATCHES "gfx11")
diff --git a/test/pool/CMakeLists.txt b/test/pool/CMakeLists.txt
new file mode 100644
index 0000000000..fac806897a
--- /dev/null
+++ b/test/pool/CMakeLists.txt
@@ -0,0 +1,16 @@
+add_custom_target(test_pool)
+
+add_gtest_executable(test_avg_pool3d_bwd test_avg_pool3d_bwd.cpp)
+add_gtest_executable(test_max_pool3d_bwd test_max_pool3d_bwd.cpp)
+add_gtest_executable(test_avg_pool3d_fwd test_avg_pool3d_fwd.cpp)
+add_gtest_executable(test_max_pool3d_fwd test_max_pool3d_fwd.cpp)
+
+target_link_libraries(test_avg_pool3d_bwd PRIVATE utility device_avg_pool3d_bwd_instance)
+target_link_libraries(test_max_pool3d_bwd PRIVATE utility device_max_pool_bwd_instance)
+target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
+target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
+
+add_dependencies(test_pool test_avg_pool3d_bwd)
+add_dependencies(test_pool test_max_pool3d_bwd)
+add_dependencies(test_pool test_avg_pool3d_fwd)
+add_dependencies(test_pool test_max_pool3d_fwd)
diff --git a/test/pool/test_avg_pool3d_bwd.cpp b/test/pool/test_avg_pool3d_bwd.cpp
new file mode 100644
index 0000000000..fbd03fdf45
--- /dev/null
+++ b/test/pool/test_avg_pool3d_bwd.cpp
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_avg_pool3d_bwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestAvgPool3dBwd : public ::testing::Test
+{
+    protected:
+    using DOutDataType    = std::tuple_element_t<0, Tuple>;
+    using DInDataType     = std::tuple_element_t<1, Tuple>;
+    using ComputeDataType = std::tuple_element_t<2, Tuple>;
+    using DOutLayout      = std::tuple_element_t<3, Tuple>;
+    using DInLayout       = std::tuple_element_t<4, Tuple>;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            bool success =
+                ck::profiler::profile_avg_pool3d_bwd_impl<DOutDataType,
+                                                          DInDataType,
+                                                          ComputeDataType,
+                                                          DOutLayout,
+                                                          DInLayout>(true,
+                                                                     2,
+                                                                     false,
+                                                                     false,
+                                                                     param.length_,
+                                                                     param.window_spatial_lengths_,
+                                                                     param.window_strides_,
+                                                                     param.window_dilations_,
+                                                                     param.input_left_pads_,
+                                                                     param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+#if defined(CK_ENABLE_FP16) && defined(CK_ENABLE_BF16) && defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, NDHWC, NDHWC>,
+                                     std::tuple<BF16, BF16, F32, NDHWC, NDHWC>,
+                                     std::tuple<F32, F32, F32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP16) && defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, NDHWC, NDHWC>,
+                                     std::tuple<F32, F32, F32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_BF16) && defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<BF16, BF16, F32, NDHWC, NDHWC>,
+                                     std::tuple<F32, F32, F32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP16) && defined(CK_ENABLE_BF16)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, NDHWC, NDHWC>,
+                                     std::tuple<BF16, BF16, F32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP16)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_BF16)
+using KernelTypes = ::testing::Types<std::tuple<BF16, BF16, F32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<F32, F32, F32, NDHWC, NDHWC>>;
+#endif
+
+TYPED_TEST_SUITE(TestAvgPool3dBwd, KernelTypes);
+TYPED_TEST(TestAvgPool3dBwd, Test_Pool)
+{
+    // length, window_length, window_stride, window_dilation, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {4, 4, 4}, {4, 4, 4}, {2, 2, 2}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_avg_pool3d_fwd.cpp b/test/pool/test_avg_pool3d_fwd.cpp
similarity index 100%
rename from test/pool_fwd/test_avg_pool3d_fwd.cpp
rename to test/pool/test_avg_pool3d_fwd.cpp
diff --git a/test/pool/test_max_pool3d_bwd.cpp b/test/pool/test_max_pool3d_bwd.cpp
new file mode 100644
index 0000000000..8d52bde4da
--- /dev/null
+++ b/test/pool/test_max_pool3d_bwd.cpp
@@ -0,0 +1,79 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "profiler/profile_max_pool3d_bwd_impl.hpp"
+#include "test_pool_fwd_common.hpp"
+
+template <typename Tuple>
+class TestMaxPool3dBwd : public ::testing::Test
+{
+    protected:
+    using DOutDataType  = std::tuple_element_t<0, Tuple>;
+    using DInDataType   = std::tuple_element_t<1, Tuple>;
+    using IndexDataType = std::tuple_element_t<2, Tuple>;
+
+    using InDataType  = DInDataType;
+    using OutDataType = DOutDataType;
+
+    std::vector<PoolingParam> params;
+
+    void Run()
+    {
+        for(auto param : params)
+        {
+            bool success =
+                ck::profiler::profile_max_pool3d_bwd_impl<InDataType,
+                                                          OutDataType,
+                                                          IndexDataType,
+                                                          DOutDataType,
+                                                          DInDataType,
+                                                          false>(true,
+                                                                 2,
+                                                                 false,
+                                                                 false,
+                                                                 param.length_,
+                                                                 param.window_spatial_lengths_,
+                                                                 param.window_strides_,
+                                                                 param.window_dilations_,
+                                                                 param.input_left_pads_,
+                                                                 param.input_right_pads_);
+            EXPECT_TRUE(success);
+        }
+    }
+};
+
+#if defined(CK_ENABLE_FP16) && defined(CK_ENABLE_BF16) && defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, I32, NDHWC, NDHWC>,
+                                     std::tuple<BF16, BF16, I32, NDHWC, NDHWC>,
+                                     std::tuple<F32, F32, I32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP16) && defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, I32, NDHWC, NDHWC>,
+                                     std::tuple<F32, F32, I32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_BF16) && defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<BF16, BF16, I32, NDHWC, NDHWC>,
+                                     std::tuple<F32, F32, I32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP16) && defined(CK_ENABLE_BF16)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, I32, NDHWC, NDHWC>,
+                                     std::tuple<BF16, BF16, I32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP16)
+using KernelTypes = ::testing::Types<std::tuple<F16, F16, I32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_BF16)
+using KernelTypes = ::testing::Types<std::tuple<BF16, BF16, I32, NDHWC, NDHWC>>;
+#elif defined(CK_ENABLE_FP32)
+using KernelTypes = ::testing::Types<std::tuple<F32, F32, I32, NDHWC, NDHWC>>;
+#endif
+
+TYPED_TEST_SUITE(TestMaxPool3dBwd, KernelTypes);
+TYPED_TEST(TestMaxPool3dBwd, Test_Pool)
+{
+    // length, window_length, window_stride, window_dilation, left_pad, right_pad
+    this->params = {{{1, 1, 1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 16, 64, 64, 64}, {4, 4, 4}, {4, 4, 4}, {2, 2, 2}, {0, 0, 0}, {0, 0, 0}},
+                    {{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}}};
+
+    // this->params = {{{2, 32, 30, 30, 30}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1,
+    // 1}}};
+
+    this->Run();
+}
diff --git a/test/pool_fwd/test_max_pool3d_fwd.cpp b/test/pool/test_max_pool3d_fwd.cpp
similarity index 100%
rename from test/pool_fwd/test_max_pool3d_fwd.cpp
rename to test/pool/test_max_pool3d_fwd.cpp
diff --git a/test/pool_fwd/test_pool_fwd_common.hpp b/test/pool/test_pool_fwd_common.hpp
similarity index 88%
rename from test/pool_fwd/test_pool_fwd_common.hpp
rename to test/pool/test_pool_fwd_common.hpp
index 6bfcb47bad..5917a27e56 100644
--- a/test/pool_fwd/test_pool_fwd_common.hpp
+++ b/test/pool/test_pool_fwd_common.hpp
@@ -4,10 +4,12 @@
 #include "gtest/gtest.h"
 #include "ck/ck.hpp"
 
-using F16 = ck::half_t;
-using F32 = float;
-using I32 = int32_t;
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
+using I32  = int32_t;
 using ck::index_t;
+using NDHWC = ck::tensor_layout::convolution::NDHWC;
 
 struct PoolingParam
 {
diff --git a/test/pool_fwd/CMakeLists.txt b/test/pool_fwd/CMakeLists.txt
deleted file mode 100644
index e421a79fde..0000000000
--- a/test/pool_fwd/CMakeLists.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-add_custom_target(test_pool_fwd)
-
-add_gtest_executable(test_avg_pool3d_fwd test_avg_pool3d_fwd.cpp)
-add_gtest_executable(test_max_pool3d_fwd test_max_pool3d_fwd.cpp)
-
-target_link_libraries(test_avg_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
-target_link_libraries(test_max_pool3d_fwd PRIVATE utility device_pool3d_fwd_instance)
-
-add_dependencies(test_pool_fwd test_avg_pool3d_fwd)
-add_dependencies(test_pool_fwd test_max_pool3d_fwd)

From f5ec04f091fa5c48c67d7bacec36a414d0be06a5 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Thu, 31 Aug 2023 09:22:12 -0500
Subject: [PATCH 107/194] Grouped Gemm with Fixed K and N with SplitK (#818)

* move all arguments into device

* add b2c_tile_map

* add examples

* add SetDeviceKernelArgs

* dedicated fixed_nk solution

* init client api

* add grouped_gemm_bias example

* add a instance

* add instances

* formatting

* fixed cmake

* Update EnableCompilerWarnings.cmake

* Update cmake-ck-dev.sh

* clean; fixed comments

* fixed comment

* add instances for fp32 output

* add instances for fp32 output

* add fp32 out client example

* fixed CI

* init commit for kbatch

* add splitk gridwise

* format

* fixed

* clean deviceop

* clean code

* finish splitk

* fixed instances

* change m_loops to tile_loops

* add setkbatch

* clean code

* add splitK+bias

* add instances

* opt mk_nk instances

* clean examples

* fixed CI

* remove zero

* finished non-zero

* clean

* clean code

* optimized global_barrier

* fixed ci

* fixed CI

* removed AddBias

* format

* fixed CI

* fixed CI

* move 20_grouped_gemm to 21_grouped_gemm

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../21_grouped_gemm_bias/CMakeLists.txt       |    2 +
 .../grouped_gemm_fixed_nk_bias_fp16.cpp       |  244 ++++
 example/15_grouped_gemm/CMakeLists.txt        |    7 +-
 .../grouped_gemm_xdl_fixed_nk_bias_fp16.cpp   |  353 ++++++
 .../grouped_gemm_xdl_fixed_nk_fp16.cpp        |  329 +++++
 .../device/device_grouped_gemm_fixed_nk.hpp   |   63 +
 .../impl/device_grouped_gemm_xdl_fixed_nk.hpp |  836 +++++++++++++
 .../element/binary_element_wise_operation.hpp |    7 +
 .../gpu/grid/block_to_ctile_map.hpp           |    3 +-
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |  172 +++
 ...se_gemm_multiple_d_xdl_splitk_cshuffle.hpp | 1086 +++++++++++++++++
 .../device_operation_instance_factory.hpp     |    1 +
 .../gpu/grouped_gemm_bias.hpp                 |  146 +++
 .../ck/library/utility/device_memory.hpp      |    2 +
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |    1 -
 .../gpu/grouped_gemm_bias/CMakeLists.txt      |    7 +
 ..._nk_bias_f16_f16_f16_mk_kn_mn_instance.cpp |   79 ++
 ..._nk_bias_f16_f16_f16_mk_nk_mn_instance.cpp |   82 ++
 ..._nk_bias_f16_f16_f32_mk_kn_mn_instance.cpp |   83 ++
 ..._nk_bias_f16_f16_f32_mk_nk_mn_instance.cpp |   83 ++
 library/src/utility/device_memory.cpp         |   10 +
 script/cmake-ck-dev.sh                        |    1 -
 22 files changed, 3593 insertions(+), 4 deletions(-)
 create mode 100644 client_example/21_grouped_gemm_bias/CMakeLists.txt
 create mode 100644 client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp
 create mode 100644 example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_bias_fp16.cpp
 create mode 100644 example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_bias.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_bias/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instance.cpp

diff --git a/client_example/21_grouped_gemm_bias/CMakeLists.txt b/client_example/21_grouped_gemm_bias/CMakeLists.txt
new file mode 100644
index 0000000000..a2abd15731
--- /dev/null
+++ b/client_example/21_grouped_gemm_bias/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_executable(client_grouped_gemm_fixed_nk_bias_fp16 grouped_gemm_fixed_nk_bias_fp16.cpp)
+target_link_libraries(client_grouped_gemm_fixed_nk_bias_fp16 PRIVATE composable_kernel::device_operations)
diff --git a/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp b/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp
new file mode 100644
index 0000000000..3b6dd9a2a9
--- /dev/null
+++ b/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp
@@ -0,0 +1,244 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+#include <random>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_bias.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using D0DataType = F32;
+using DsDataType = ck::Tuple<D0DataType>;
+using EDataType  = F32;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using D0Layout = Row;
+using DsLayout = ck::Tuple<D0Layout>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = Add;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideEs;
+
+    int sum_of_m = 0;
+
+    Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    int group_count = Ms.size();
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        Ns.push_back(768);
+        Ks.push_back(4608);
+
+        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
+        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
+        StrideEs.push_back(std::is_same<Row, ELayout>::value ? Ns[i] : Ms[i]);
+
+        sum_of_m += Ms[i];
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    std::vector<SimpleDeviceMem> a_dev_bufs, b_dev_bufs, d0_dev_bufs, e_dev_bufs;
+
+    a_dev_bufs.reserve(group_count);
+    b_dev_bufs.reserve(group_count);
+    d0_dev_bufs.reserve(group_count);
+    e_dev_bufs.reserve(group_count);
+
+    std::vector<void*> p_e;
+
+    p_e.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GroupedGemmKernelArgument<1>>
+        grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        a_dev_bufs.emplace_back(sizeof(ADataType) *
+                                f_matrix_space_size(Ms[i], Ks[i], StrideAs[i], ALayout{}));
+        b_dev_bufs.emplace_back(sizeof(BDataType) *
+                                f_matrix_space_size(Ks[i], Ns[i], StrideBs[i], BLayout{}));
+        d0_dev_bufs.emplace_back(sizeof(D0DataType) *
+                                 f_matrix_space_size(Ms[i], Ns[i], 0, D0Layout{}));
+        e_dev_bufs.emplace_back(sizeof(EDataType) *
+                                f_matrix_space_size(Ms[i], Ns[i], StrideEs[i], ELayout{}));
+
+        gemm_descs.push_back({sum_of_m, Ns[i], Ks[i], 1, StrideBs[i], 1, {0}});
+
+        p_e.push_back(e_dev_bufs[i].GetDeviceBuffer());
+
+        grouped_gemm_kernel_args_.push_back(
+            {a_dev_bufs[i].GetDeviceBuffer(),
+             b_dev_bufs[i].GetDeviceBuffer(),
+             std::array<const void*, 1>{d0_dev_bufs[i].GetDeviceBuffer()},
+             e_dev_bufs[i].GetDeviceBuffer(),
+             Ms[i],
+             Ns[i],
+             Ks[i],
+             StrideAs[i],
+             StrideBs[i],
+             std::array<ck::index_t, 1>{0},
+             StrideEs[i]});
+    }
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                                            BLayout,
+                                                                            DsLayout,
+                                                                            ELayout,
+                                                                            ADataType,
+                                                                            BDataType,
+                                                                            DsDataType,
+                                                                            EDataType,
+                                                                            AElementOp,
+                                                                            BElementOp,
+                                                                            CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    std::vector<const void*> p_a = {}, p_b = {};
+    std::vector<std::array<const void*, 1>> p_ds = {};
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_e, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        SimpleDeviceMem grouped_gemm_kernel_args_dev(
+            op_ptr->GetDeviceKernelArgSize(argument_ptr.get()));
+
+        SimpleDeviceMem grouped_gemm_workspace_dev(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        hipGetErrorString(hipMemcpy(grouped_gemm_kernel_args_dev.GetDeviceBuffer(),
+                                    grouped_gemm_kernel_args_.data(),
+                                    op_ptr->GetDeviceKernelArgSize(argument_ptr.get()),
+                                    hipMemcpyHostToDevice));
+
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(),
+                                    grouped_gemm_workspace_dev.GetDeviceBuffer());
+
+        op_ptr->SetDeviceKernelArgs(argument_ptr.get(),
+                                    grouped_gemm_kernel_args_dev.GetDeviceBuffer());
+
+        op_ptr->SetKBatch(argument_ptr.get(), 2);
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = 0, num_btype = 0;
+            for(std::size_t j = 0; j < gemm_descs.size(); ++j)
+            {
+                flop += std::size_t(2) * Ms[j] * Ns[j] * Ks[j];
+
+                num_btype += sizeof(ADataType) * Ms[j] * Ks[j] + sizeof(BDataType) * Ks[j] * Ns[j] +
+                             sizeof(EDataType) * Ms[j] * Ns[j];
+            }
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index dca60b0e73..2f880af3cf 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -1,4 +1,5 @@
 add_custom_target(example_grouped_gemm_xdl)
+
 if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
   add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
@@ -7,10 +8,14 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
   add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
   add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
   add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
+  add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
+  add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
   add_dependencies(example_grouped_gemm_xdl 
                    example_grouped_gemm_xdl_fp16
                    example_grouped_gemm_multiple_d_dl_fp16
-                   example_grouped_gemm_xdl_splitk_fp16)
+                   example_grouped_gemm_xdl_splitk_fp16
+                   example_grouped_gemm_xdl_fixed_nk_fp16
+                   example_grouped_gemm_xdl_fixed_nk_bias_fp16)
 endif()
 if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
   add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_bias_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_bias_fp16.cpp
new file mode 100644
index 0000000000..a193fc39ba
--- /dev/null
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_bias_fp16.cpp
@@ -0,0 +1,353 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using D0DataType       = F32;
+using DsDataType       = ck::Tuple<D0DataType>;
+using EDataType        = F32;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using D0Layout = Row;
+using DsLayout = ck::Tuple<D0Layout>;
+using ELayout  = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+
+using CDEElementOp = Add;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MPadding;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl_Fixed_NK
+    // clang-format off
+//######| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+//######|        |        |         |        |      Type|      Type|        Type|         DataType|       Type|      Type| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+//######|        |        |         |        |          |          |            |                 |           |          |   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+//######|        |        |         |        |          |          |            |                 |           |          |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,    GemmDefault,        1,   128,    16,   128,    32,   8,   8,   16,   16,    1,    4,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>;
+// clang-format on
+
+struct ProblemSize final
+{
+    std::vector<ck::index_t> Ms;
+    std::vector<ck::index_t> Ns;
+    std::vector<ck::index_t> Ks;
+
+    std::vector<ck::index_t> stride_As;
+    std::vector<ck::index_t> stride_Bs;
+    std::vector<ck::index_t> stride_Cs;
+
+    ck::index_t group_count;
+};
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+    int k_batch          = 1;
+};
+
+bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
+{
+    auto group_count = problem_size.group_count;
+
+    // GEMM shape
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    int sum_of_m = 0;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    std::vector<Tensor<ADataType>> a_tensors;
+    std::vector<Tensor<BDataType>> b_tensors;
+    std::vector<Tensor<D0DataType>> d0_tensors;
+    std::vector<Tensor<EDataType>> c_host_tensors;
+    std::vector<Tensor<EDataType>> c_device_tensors;
+
+    a_tensors.reserve(group_count);
+    b_tensors.reserve(group_count);
+    d0_tensors.reserve(group_count);
+    c_host_tensors.reserve(group_count);
+    c_device_tensors.reserve(group_count);
+
+    using DeviceMemPtr = std::unique_ptr<DeviceMem>;
+
+    std::vector<DeviceMemPtr> a_tensors_device, b_tensors_device, d0_tensors_device,
+        c_tensors_device;
+
+    a_tensors_device.reserve(group_count);
+    b_tensors_device.reserve(group_count);
+    d0_tensors_device.reserve(group_count);
+    c_tensors_device.reserve(group_count);
+
+    std::size_t flop = 0, num_btype = 0;
+
+    for(int i = 0; i < group_count; i++)
+    {
+        sum_of_m += problem_size.Ms[i];
+        a_tensors.push_back(Tensor<ADataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ks[i], problem_size.stride_As[i], ALayout{})));
+        b_tensors.push_back(Tensor<BDataType>(f_host_tensor_descriptor(
+            problem_size.Ks[i], problem_size.Ns[i], problem_size.stride_Bs[i], BLayout{})));
+        d0_tensors.push_back(Tensor<D0DataType>(
+            f_host_tensor_descriptor(problem_size.Ms[i], problem_size.Ns[i], 0, ELayout{})));
+        c_host_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ns[i], problem_size.stride_Cs[i], ELayout{})));
+        c_device_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ns[i], problem_size.stride_Cs[i], ELayout{})));
+        std::cout << "gemm[" << i << "] a_m_k: " << a_tensors[i].mDesc
+                  << " b_k_n: " << b_tensors[i].mDesc << " d_m_n: " << d0_tensors[i].mDesc
+                  << " c_m_n: " << c_device_tensors[i].mDesc << std::endl;
+
+        flop += std::size_t(2) * problem_size.Ms[i] * problem_size.Ks[i] * problem_size.Ns[i];
+        num_btype += sizeof(ADataType) * a_tensors[i].mDesc.GetElementSize() +
+                     sizeof(BDataType) * b_tensors[i].mDesc.GetElementSize() +
+                     sizeof(D0DataType) * d0_tensors[i].mDesc.GetElementSize() +
+                     sizeof(EDataType) * c_device_tensors[i].mDesc.GetElementSize();
+
+        switch(config.init_method)
+        {
+        case 0: break;
+        case 1:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+            break;
+        case 2:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+            break;
+        default:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+        }
+
+        d0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+    }
+
+    using GroupedGemmKernelArgument = ck::tensor_operation::device::GroupedGemmKernelArgument<1>;
+
+    std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; i++)
+    {
+        a_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(ADataType) * sum_of_m * problem_size.Ks[i]));
+
+        b_tensors_device.emplace_back(std::make_unique<DeviceMem>(
+            sizeof(BDataType) * problem_size.Ns[i] * problem_size.Ks[i]));
+
+        d0_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(D0DataType) * problem_size.Ns[i]));
+
+        c_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(EDataType) * sum_of_m * problem_size.Ns[i]));
+
+        a_tensors_device[i]->ToDevice(a_tensors[i].mData.data(),
+                                      a_tensors[i].mDesc.GetElementSpaceSize() * sizeof(ADataType));
+        b_tensors_device[i]->ToDevice(b_tensors[i].mData.data(),
+                                      b_tensors[i].mDesc.GetElementSpaceSize() * sizeof(BDataType));
+        d0_tensors_device[i]->ToDevice(d0_tensors[i].mData.data());
+        c_tensors_device[i]->SetZero();
+
+        gemm_descs.push_back({sum_of_m,
+                              problem_size.Ns[i],
+                              problem_size.Ks[i],
+                              1,
+                              problem_size.stride_Bs[i],
+                              1,
+                              {0}});
+
+        grouped_gemm_kernel_args_.push_back(
+            {a_tensors_device[i]->GetDeviceBuffer(),
+             b_tensors_device[i]->GetDeviceBuffer(),
+             std::array<const void*, 1>{d0_tensors_device[i]->GetDeviceBuffer()},
+             c_tensors_device[i]->GetDeviceBuffer(),
+             problem_size.Ms[i],
+             problem_size.Ns[i],
+             problem_size.Ks[i],
+             problem_size.stride_As[i],
+             problem_size.stride_Bs[i],
+             std::array<ck::index_t, 1>{0},
+             problem_size.stride_Cs[i]});
+    }
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{};
+
+    auto gemm    = DeviceGemmInstance{};
+    auto invoker = gemm.MakeInvoker();
+
+    std::vector<const void*> p_As                = {};
+    std::vector<const void*> p_Bs                = {};
+    std::vector<std::array<const void*, 1>> p_Ds = {};
+    std::vector<void*> p_Cs                      = {};
+
+    // do GEMM
+    auto argument = gemm.MakeArgument(
+        p_As, p_Bs, p_Ds, p_Cs, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    DeviceMem gemm_workspace_dev(gemm.GetWorkSpaceSize(&argument));
+    gemm.SetWorkSpacePointer(&argument, gemm_workspace_dev.GetDeviceBuffer());
+
+    DeviceMem gemm_kernel_args_dev(gemm.GetDeviceKernelArgSize(&argument));
+    hip_check_error(hipMemcpy(gemm_kernel_args_dev.GetDeviceBuffer(),
+                              grouped_gemm_kernel_args_.data(),
+                              gemm.GetDeviceKernelArgSize(&argument),
+                              hipMemcpyHostToDevice));
+
+    gemm.SetDeviceKernelArgs(argument, gemm_kernel_args_dev.GetDeviceBuffer());
+    gemm.SetKBatch(argument, config.k_batch);
+
+    invoker.Run(argument, StreamConfig{nullptr, false});
+
+    if(config.time_kernel)
+    {
+        float ave_time   = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+        float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+        float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                  << " GB/s, " << gemm.GetTypeString() << std::endl;
+    }
+
+    bool pass = true;
+    if(config.do_verification)
+    {
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                EDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+
+        for(std::size_t i = 0; i < gemm_descs.size(); i++)
+        {
+            c_tensors_device[i]->FromDevice(c_device_tensors[i].mData.data(),
+                                            c_device_tensors[i].mDesc.GetElementSize() *
+                                                sizeof(EDataType));
+
+            auto ref_gemm    = ReferenceGemmInstance{};
+            auto ref_invoker = ref_gemm.MakeInvoker();
+
+            auto ref_argument = ref_gemm.MakeArgument(a_tensors[i],
+                                                      b_tensors[i],
+                                                      c_host_tensors[i],
+                                                      a_element_op,
+                                                      b_element_op,
+                                                      PassThrough{});
+
+            ref_invoker.Run(ref_argument);
+
+            for(int m = 0; m < problem_size.Ms[i]; ++m)
+            {
+                for(int n = 0; n < problem_size.Ns[i]; ++n)
+                {
+                    cde_element_op(
+                        c_host_tensors[i](m, n), c_host_tensors[i](m, n), d0_tensors[i](m, n));
+                }
+            }
+
+            pass &= ck::utils::check_err(c_device_tensors[i], c_host_tensors[i]);
+        }
+    }
+
+    return pass;
+}
+
+int main(int argc, char* argv[])
+{
+    ProblemSize problem_size;
+    ExecutionConfig config;
+
+    problem_size.group_count = 16;
+
+    problem_size.Ms = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0};
+
+    for(int i = 0; i < problem_size.group_count; i++)
+    {
+        problem_size.Ns.push_back(768);
+        problem_size.Ks.push_back(4608);
+
+        problem_size.stride_As.push_back(problem_size.Ks[i]);
+        problem_size.stride_Bs.push_back(problem_size.Ns[i]);
+        problem_size.stride_Cs.push_back(problem_size.Ns[i]);
+    }
+
+    if(argc == 5)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+        config.k_batch         = std::stoi(argv[4]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=n0, 1=yes)\n");
+        printf("arg4: k_batch (>0)\n");
+        exit(0);
+    }
+
+    return !run_grouped_gemm(problem_size, config);
+}
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp
new file mode 100644
index 0000000000..89d4789c12
--- /dev/null
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp
@@ -0,0 +1,329 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F32;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl_Fixed_NK
+    // clang-format off
+//######| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+//######|        |        |         |        |      Type|      Type|        Type|         DataType|       Type|      Type| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+//######|        |        |         |        |          |          |            |                 |           |          |   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+//######|        |        |         |        |          |          |            |                 |           |          |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>;
+// clang-format on
+
+struct ProblemSize final
+{
+    std::vector<ck::index_t> Ms;
+    std::vector<ck::index_t> Ns;
+    std::vector<ck::index_t> Ks;
+
+    std::vector<ck::index_t> stride_As;
+    std::vector<ck::index_t> stride_Bs;
+    std::vector<ck::index_t> stride_Cs;
+
+    ck::index_t group_count;
+};
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    int k_batch          = 1;
+    bool time_kernel     = false;
+};
+
+bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
+{
+    auto group_count = problem_size.group_count;
+
+    // GEMM shape
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+    std::vector<void*> p_Cs;
+
+    gemm_descs.reserve(group_count);
+
+    int sum_of_m = 0;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    std::vector<Tensor<ADataType>> a_tensors;
+    std::vector<Tensor<BDataType>> b_tensors;
+    std::vector<Tensor<EDataType>> c_host_tensors;
+    std::vector<Tensor<EDataType>> c_device_tensors;
+
+    a_tensors.reserve(group_count);
+    b_tensors.reserve(group_count);
+    c_host_tensors.reserve(group_count);
+    c_device_tensors.reserve(group_count);
+
+    using DeviceMemPtr = std::unique_ptr<DeviceMem>;
+
+    std::vector<DeviceMemPtr> a_tensors_device, b_tensors_device, c_tensors_device;
+
+    a_tensors_device.reserve(group_count);
+    b_tensors_device.reserve(group_count);
+    c_tensors_device.reserve(group_count);
+
+    std::size_t flop = 0, num_btype = 0;
+
+    for(int i = 0; i < group_count; i++)
+    {
+        sum_of_m += problem_size.Ms[i];
+        a_tensors.push_back(Tensor<ADataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ks[i], problem_size.stride_As[i], ALayout{})));
+        b_tensors.push_back(Tensor<BDataType>(f_host_tensor_descriptor(
+            problem_size.Ks[i], problem_size.Ns[i], problem_size.stride_Bs[i], BLayout{})));
+        c_host_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ns[i], problem_size.stride_Cs[i], ELayout{})));
+        c_device_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ns[i], problem_size.stride_Cs[i], ELayout{})));
+        std::cout << "gemm[" << i << "] a_m_k: " << a_tensors[i].mDesc
+                  << " b_k_n: " << b_tensors[i].mDesc << " c_m_n: " << c_device_tensors[i].mDesc
+                  << std::endl;
+
+        flop += std::size_t(2) * problem_size.Ms[i] * problem_size.Ks[i] * problem_size.Ns[i];
+        num_btype += sizeof(ADataType) * a_tensors[i].mDesc.GetElementSize() +
+                     sizeof(BDataType) * b_tensors[i].mDesc.GetElementSize() +
+                     sizeof(EDataType) * c_device_tensors[i].mDesc.GetElementSize();
+
+        switch(config.init_method)
+        {
+        case 0: break;
+        case 1:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+            break;
+        case 2:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+            break;
+        default:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+        }
+    }
+
+    using GroupedGemmKernelArgument = ck::tensor_operation::device::GroupedGemmKernelArgument<>;
+
+    std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; i++)
+    {
+        a_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(ADataType) * sum_of_m * problem_size.Ks[i]));
+
+        b_tensors_device.emplace_back(std::make_unique<DeviceMem>(
+            sizeof(BDataType) * problem_size.Ns[i] * problem_size.Ks[i]));
+
+        c_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(EDataType) * sum_of_m * problem_size.Ns[i]));
+
+        a_tensors_device[i]->ToDevice(a_tensors[i].mData.data(),
+                                      a_tensors[i].mDesc.GetElementSpaceSize() * sizeof(ADataType));
+        b_tensors_device[i]->ToDevice(b_tensors[i].mData.data(),
+                                      b_tensors[i].mDesc.GetElementSpaceSize() * sizeof(BDataType));
+        c_tensors_device[i]->SetZero();
+
+        p_Cs.push_back(c_tensors_device[i]->GetDeviceBuffer());
+
+        gemm_descs.push_back({sum_of_m,
+                              problem_size.Ns[i],
+                              problem_size.Ks[i],
+                              1,
+                              problem_size.stride_Bs[i],
+                              1,
+                              {}});
+
+        grouped_gemm_kernel_args_.push_back({a_tensors_device[i]->GetDeviceBuffer(),
+                                             b_tensors_device[i]->GetDeviceBuffer(),
+                                             {},
+                                             c_tensors_device[i]->GetDeviceBuffer(),
+                                             problem_size.Ms[i],
+                                             problem_size.Ns[i],
+                                             problem_size.Ks[i],
+                                             problem_size.stride_As[i],
+                                             problem_size.stride_Bs[i],
+                                             {},
+                                             problem_size.stride_Cs[i]});
+    }
+
+    auto a_element_op = AElementOp{};
+    auto b_element_op = BElementOp{};
+    auto c_element_op = CDEElementOp{};
+
+    auto gemm    = DeviceGemmInstance{};
+    auto invoker = gemm.MakeInvoker();
+
+    std::vector<const void*> p_As                = {};
+    std::vector<const void*> p_Bs                = {};
+    std::vector<std::array<const void*, 0>> p_Ds = {};
+
+    // do GEMM
+    auto argument = gemm.MakeArgument(
+        p_As, p_Bs, p_Ds, p_Cs, gemm_descs, a_element_op, b_element_op, c_element_op);
+
+    DeviceMem gemm_arg_dev_mem(gemm.GetDeviceKernelArgSize(&argument));
+    DeviceMem gemm_workspace_dev(gemm.GetWorkSpaceSize(&argument));
+
+    gemm.SetWorkSpacePointer(&argument, gemm_workspace_dev.GetDeviceBuffer());
+
+    hip_check_error(hipMemcpy(gemm_arg_dev_mem.GetDeviceBuffer(),
+                              grouped_gemm_kernel_args_.data(),
+                              gemm.GetDeviceKernelArgSize(&argument),
+                              hipMemcpyHostToDevice));
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    gemm.SetDeviceKernelArgs(argument, gemm_arg_dev_mem.GetDeviceBuffer());
+    gemm.SetKBatch(argument, config.k_batch);
+
+    invoker.Run(argument, StreamConfig{nullptr, false});
+
+    if(config.time_kernel)
+    {
+        float ave_time   = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+        float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+        float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                  << " GB/s, " << gemm.GetTypeString() << std::endl;
+    }
+
+    bool pass = true;
+    if(config.do_verification)
+    {
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                EDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                CDEElementOp>;
+
+        for(std::size_t i = 0; i < gemm_descs.size(); i++)
+        {
+            c_tensors_device[i]->FromDevice(c_device_tensors[i].mData.data(),
+                                            c_device_tensors[i].mDesc.GetElementSize() *
+                                                sizeof(EDataType));
+            auto ref_gemm    = ReferenceGemmInstance{};
+            auto ref_invoker = ref_gemm.MakeInvoker();
+
+            auto ref_argument = ref_gemm.MakeArgument(a_tensors[i],
+                                                      b_tensors[i],
+                                                      c_host_tensors[i],
+                                                      a_element_op,
+                                                      b_element_op,
+                                                      c_element_op);
+
+            ref_invoker.Run(ref_argument);
+
+            pass &= ck::utils::check_err(c_device_tensors[i], c_host_tensors[i]);
+        }
+    }
+
+    return pass;
+}
+
+int main(int argc, char* argv[])
+{
+    ProblemSize problem_size;
+    ExecutionConfig config;
+
+    problem_size.group_count = 16;
+
+    problem_size.Ms = {
+        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    for(int i = 0; i < problem_size.group_count; i++)
+    {
+        problem_size.Ns.push_back(768);
+        problem_size.Ks.push_back(4608);
+
+        problem_size.stride_As.push_back(problem_size.Ks[i]);
+        problem_size.stride_Bs.push_back(problem_size.Ks[i]);
+        problem_size.stride_Cs.push_back(problem_size.Ns[i]);
+    }
+
+    if(argc == 5)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+        config.k_batch         = std::stoi(argv[4]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=n0, 1=yes)\n");
+        printf("arg4: k_batch (> 0)\n");
+        exit(0);
+    }
+
+    return !run_grouped_gemm(problem_size, config);
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp
new file mode 100644
index 0000000000..fcb2ba6a4d
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <array>
+
+#include "device_grouped_gemm.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <index_t NumDTensor = 0>
+struct GroupedGemmKernelArgument
+{
+    const void* p_a_grid;
+    const void* p_b_grid;
+    std::array<const void*, NumDTensor> p_ds_grid;
+    void* p_e_grid;
+
+    index_t M;
+    index_t N;
+    index_t K;
+
+    index_t StrideA;
+    index_t StrideB;
+    std::array<index_t, NumDTensor> StrideDs;
+    index_t StrideE;
+};
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
+struct DeviceGroupedGemmFixedNK : DeviceGroupedGemm<ALayout,
+                                                    BLayout,
+                                                    DsLayout,
+                                                    ELayout,
+                                                    ADataType,
+                                                    BDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    AElementwiseOperation,
+                                                    BElementwiseOperation,
+                                                    CElementwiseOperation>
+{
+    virtual void SetDeviceKernelArgs(BaseArgument* p_arg, const void* kernel_args) const = 0;
+    virtual size_t GetDeviceKernelArgSize(const BaseArgument* p_arg) const               = 0;
+    virtual void SetKBatch(BaseArgument* p_arg, index_t k_batch) const                   = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
new file mode 100644
index 0000000000..8cea09ee54
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
@@ -0,0 +1,836 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename GridwiseGemm,
+          typename GemmDesc,
+          GemmSpecialization GemmSpec,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDataType,
+          typename Block2ETileMap,
+          typename GroupedGemmBlock2ETileMap,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_grouped_gemm_xdl_fixed_nk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
+                                         uint32_t* barrier_count,
+                                         const index_t barrier_size_grp,
+                                         const index_t group_count,
+                                         const index_t grid_size_grp,
+                                         const index_t KBatch,
+                                         const AElementwiseOperation a_element_op,
+                                         const BElementwiseOperation b_element_op,
+                                         const CDEElementwiseOperation c_element_op)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    const index_t block_id = get_block_1d_id();
+
+    const auto gemm_desc_ptr =
+        reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
+
+    const index_t group_id = block_id / grid_size_grp;
+
+    if(group_id >= group_count)
+        return;
+
+    const index_t M = gemm_desc_ptr[group_id].M;
+    const index_t N = gemm_desc_ptr[group_id].N;
+    const index_t K = gemm_desc_ptr[group_id].K;
+
+    if(M * N * K == 0)
+        return;
+
+    const auto StrideA  = gemm_desc_ptr[group_id].StrideA;
+    const auto StrideB  = gemm_desc_ptr[group_id].StrideB;
+    const auto StrideDs = gemm_desc_ptr[group_id].StrideDs;
+    const auto StrideE  = gemm_desc_ptr[group_id].StrideE;
+
+    const auto e_grid_desc_m_n =
+        GridwiseGemm::template MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+    const index_t BlockStart = group_id * grid_size_grp;
+
+    const auto local_b2e_tile_map = Block2ETileMap{e_grid_desc_m_n, KBatch};
+
+    const auto local_grid_size = local_b2e_tile_map.CalculateGridSize(e_grid_desc_m_n);
+
+    constexpr auto NumDTensor = DsDataType::Size();
+
+    using DsGridPointer = decltype(GridwiseGemm::MakeDsGridPointer());
+
+    DsGridPointer p_ds_grid_;
+
+    static_for<0, NumDTensor, 1>{}([&](auto i) {
+        using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+        // D pointer
+        p_ds_grid_(i) = static_cast<const DDataType*>(gemm_desc_ptr[group_id].p_ds_grid[i]);
+    });
+
+    index_t id_off   = 0;
+    index_t id_local = get_block_1d_id() - BlockStart;
+
+    const index_t mn_blocks = local_grid_size / KBatch;
+
+    while(id_local < local_grid_size)
+    {
+        const auto block_2_etile_map =
+            GroupedGemmBlock2ETileMap(local_b2e_tile_map, BlockStart, id_off);
+
+        auto barrier_count_finished =
+            barrier_count + group_id * barrier_size_grp + id_local % mn_blocks;
+
+        GridwiseGemm::template Run<HasMainKBlockLoop,
+                                   EGlobalMemoryDataOperation,
+                                   GemmSpec,
+                                   ALayout,
+                                   BLayout,
+                                   DsLayout,
+                                   ELayout>(gemm_desc_ptr[group_id].p_a_grid,
+                                            gemm_desc_ptr[group_id].p_b_grid,
+                                            p_ds_grid_,
+                                            gemm_desc_ptr[group_id].p_e_grid,
+                                            p_shared,
+                                            barrier_count_finished,
+                                            a_element_op,
+                                            b_element_op,
+                                            c_element_op,
+                                            M,
+                                            N,
+                                            K,
+                                            StrideA,
+                                            StrideB,
+                                            StrideDs,
+                                            StrideE,
+                                            KBatch,
+                                            block_2_etile_map);
+
+        id_off += grid_size_grp;
+        id_local += grid_size_grp;
+    }
+#else
+    ignore = gemm_descs_const;
+    ignore = barrier_count;
+    ignore = barrier_size_grp;
+    ignore = group_count;
+    ignore = grid_size_grp;
+    ignore = KBatch;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = c_element_op;
+#endif
+}
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          ck::index_t NumPrefetch,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t KPerBlock,
+          ck::index_t AK1,
+          ck::index_t BK1,
+          ck::index_t MPerXDL,
+          ck::index_t NPerXDL,
+          ck::index_t MXdlPerWave,
+          ck::index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched = make_default_loop_scheduler()>
+struct DeviceGroupedGemm_Xdl_Fixed_NK : public DeviceGroupedGemmFixedNK<ALayout,
+                                                                        BLayout,
+                                                                        DsLayout,
+                                                                        ELayout,
+                                                                        ADataType,
+                                                                        BDataType,
+                                                                        DsDataType,
+                                                                        EDataType,
+                                                                        AElementwiseOperation,
+                                                                        BElementwiseOperation,
+                                                                        CDEElementwiseOperation>
+{
+    using DeviceOp = DeviceGroupedGemm_Xdl_Fixed_NK;
+
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemmMultipleD_xdl_splitk_cshuffle<
+        ADataType, // TODO: distinguish A/B datatype
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        NumPrefetch, // NumGemmKPrefetchStage
+        BlockSize,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        AK1,
+        BK1,
+        MPerXDL,
+        NPerXDL,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEBlockTransferScalarPerVector_NPerBlock,
+        LoopSched>;
+
+    template <typename UnderlyingBlockToCTileMap>
+    struct OffsettedBlockToCTileMapMLoops
+    {
+        using underlying_type = UnderlyingBlockToCTileMap;
+
+        __host__ __device__ OffsettedBlockToCTileMapMLoops(
+            UnderlyingBlockToCTileMap block_to_ctile_map, index_t block_start, index_t id_off = 0)
+        {
+            block_to_ctile_map_ = block_to_ctile_map;
+            block_start_        = block_start;
+            id_off_             = id_off;
+        }
+
+        template <typename TopIdx>
+        __host__ __device__ constexpr auto CalculateBottomIndex(const TopIdx& idx_top) const
+        {
+            auto idx_bot = block_to_ctile_map_.CalculateBottomIndex(
+                make_multi_index(idx_top[Number<0>{}] - block_start_ + id_off_));
+
+            return make_tuple(idx_bot[Number<0>{}], idx_bot[Number<1>{}], idx_bot[Number<2>{}]);
+        }
+
+        template <typename CTileIdx, typename CTileDim>
+        __host__ __device__ bool ValidCTileIndex(const CTileIdx& c_tile_idx,
+                                                 const CTileDim& c_tile_dim) const
+        {
+            return block_to_ctile_map_.ValidCTileIndex(c_tile_idx, c_tile_dim);
+        }
+
+        template <typename CGridDesc_M_N>
+        __host__ bool CheckValidity(const CGridDesc_M_N& c_grid_desc_m_n) const
+        {
+            return block_to_ctile_map_.CheckValidity(c_grid_desc_m_n);
+        }
+
+        template <typename CGridDesc_M_N>
+        __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+        {
+            return block_to_ctile_map_.CalculateGridSize(c_grid_desc_m_n);
+        }
+
+        UnderlyingBlockToCTileMap block_to_ctile_map_;
+        index_t block_start_;
+        index_t id_off_;
+    };
+
+    template <index_t MPerBlock_, index_t NPerBlock_>
+    struct BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops
+    {
+        static constexpr auto I0 = Number<0>{};
+        static constexpr auto I1 = Number<1>{};
+
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops() = default;
+
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops(
+            const BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops&) = default;
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops(
+            BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops&&) = default;
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops&
+        operator=(const BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops&) = default;
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops&
+        operator=(BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops&&) = default;
+
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops(index_t M,
+                                                                          index_t N,
+                                                                          index_t KBatch,
+                                                                          index_t M01 = 8)
+            : M_(M), N_(N), KBatch_(KBatch), M01_(M01)
+        {
+        }
+
+        template <typename CGridDesc_M_N>
+        __host__ __device__ BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops(
+            const CGridDesc_M_N& c_grid_desc_m_n, index_t KBatch, index_t M01 = 8)
+            : BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops(
+                  c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1), KBatch, M01)
+        {
+        }
+
+        __host__ __device__ constexpr index_t CalculateGridSize(index_t M, index_t N) const
+        {
+            const auto M0 = math::integer_divide_ceil(M, MPerBlock);
+            const auto N0 = math::integer_divide_ceil(N, NPerBlock);
+
+            return M0 * N0 * KBatch_;
+        }
+
+        template <typename CGridDesc_M_N>
+        __host__ __device__ constexpr index_t
+        CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+        {
+            return CalculateGridSize(c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1));
+        }
+
+        template <typename CGridDesc_M_N>
+        __host__ bool CheckValidity(const CGridDesc_M_N& /* c_grid_desc_m_n */) const
+        {
+            return true;
+        }
+
+        template <typename TopIdx>
+        __host__ __device__ constexpr auto CalculateBottomIndex(const TopIdx& idx_top) const
+        {
+            auto block_1d_id = idx_top[I0];
+
+            const auto M0 = math::integer_divide_ceil(M_, MPerBlock_);
+            const auto N0 = math::integer_divide_ceil(N_, NPerBlock_);
+
+            block_1d_id = block_1d_id % (M0 * N0 * KBatch_); // hide groups
+
+            const index_t idx_ksplit = block_1d_id / (M0 * N0);
+            block_1d_id              = block_1d_id % (M0 * N0);
+
+            index_t idx_N0 = block_1d_id % N0;
+            index_t idx_M0 = block_1d_id / N0;
+
+            const auto M01_adapt = (idx_M0 < M0 - M0 % M01_) ? M01_ : M0 % M01_;
+
+            index_t idx_M00          = idx_M0 / M01_;
+            index_t idx_M01          = idx_M0 % M01_;
+            index_t idx_N0_M01_local = idx_N0 + idx_M01 * N0;
+
+            return make_tuple(idx_ksplit,
+                              idx_N0_M01_local % M01_adapt + idx_M00 * M01_,
+                              idx_N0_M01_local / M01_adapt);
+        }
+
+        template <typename CTileIdx, typename CTileDim>
+        __host__ __device__ bool ValidCTileIndex(const CTileIdx& /* c_tile_idx */,
+                                                 const CTileDim& /* c_tile_dim */) const
+        {
+            return true; // always valid provided that user gets grid size from CalculateGridSize()
+        }
+
+        private:
+        index_t M_;
+        index_t N_;
+        index_t KBatch_;
+        index_t M01_;
+    };
+
+    using Block2ETileMap = BlockToCTileMap_KBatch_M00_N0_M01Adapt_MLoops<MPerBlock, NPerBlock>;
+    using GroupedGemmBlock2ETileMap = OffsettedBlockToCTileMapMLoops<Block2ETileMap>;
+
+    struct GemmBiasTransKernelArg
+    {
+        // pointers
+        const void* a_ptr_;
+        const void* b_ptr_;
+        std::array<const void*, NumDTensor> ds_ptr_;
+        void* e_ptr_;
+
+        index_t M_, N_, K_;
+        index_t StrideA_, StrideB_;
+        std::array<index_t, NumDTensor> StrideDs_;
+        index_t StrideE_;
+    };
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+
+        void UpdateKBatch(index_t k_batch)
+        {
+            k_batch_ = k_batch;
+
+            if(k_batch_ < 1)
+            {
+
+                throw std::runtime_error("wrong! k_batch must be > 0");
+            }
+
+            const index_t AverM = math::integer_divide_ceil(sum_of_m, group_count_);
+
+            const index_t StrideE = gemm_desc_kernel_arg_[0].StrideE_;
+            const index_t N       = gemm_desc_kernel_arg_[0].N_;
+
+            const auto e_grid_desc_m_n =
+                GridwiseGemm::template MakeEGridDescriptor_M_N<ELayout, GemmSpec>(
+                    AverM, N, StrideE);
+
+            const auto local_b2c_tile_map = Block2ETileMap{e_grid_desc_m_n, k_batch_};
+
+            grid_size_grp_ = local_b2c_tile_map.CalculateGridSize(e_grid_desc_m_n);
+
+            grid_size_ = grid_size_grp_ * group_count_;
+        }
+
+        Argument(std::vector<const void*>&,
+                 std::vector<const void*>&,
+                 std::vector<std::array<const void*, NumDTensor>>&,
+                 std::vector<void*>&,
+                 std::vector<GemmDesc>& gemm_descs,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation c_element_op)
+            : a_element_op_{a_element_op}, b_element_op_{b_element_op}, c_element_op_{c_element_op}
+        {
+            grid_size_ = 0;
+
+            k_batch_ = 1;
+
+            grouped_gemm_kernel_args_dev = nullptr;
+
+            group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
+
+            gemm_desc_kernel_arg_.reserve(group_count_);
+
+            index_t group_id = 0;
+
+            sum_of_m            = gemm_descs[0].M_;
+            const index_t AverM = math::integer_divide_ceil(sum_of_m, group_count_);
+            const index_t N     = gemm_descs[0].N_;
+            const index_t K     = gemm_descs[0].K_;
+
+            for(std::size_t i = 0; i < gemm_descs.size(); i++)
+            {
+                if(sum_of_m != gemm_descs[i].M_ || N != gemm_descs[i].N_ || K != gemm_descs[i].K_)
+                {
+                    throw std::runtime_error("wrong! M/N/K is not identical");
+                }
+
+                a_mtx_mraw_kraw_.emplace_back(sum_of_m, K);
+                b_mtx_nraw_kraw_.emplace_back(N, K);
+
+                const index_t StrideA = gemm_descs[i].stride_A_;
+                const index_t StrideB = gemm_descs[i].stride_B_;
+                const index_t StrideE = gemm_descs[i].stride_C_;
+
+                // pointer
+                std::array<const void*, NumDTensor> p_ds_grid;
+
+                static_for<0, NumDTensor, 1>{}([&](auto j) { p_ds_grid[j] = nullptr; });
+
+                std::array<index_t, NumDTensor> StrideDs;
+
+                static_for<0, NumDTensor, 1>{}([&](auto j) {
+                    // using DLayout = remove_cvref_t<tuple_element_t<j.value, DsLayout>>;
+
+                    if(gemm_descs[i].stride_Ds_.size() != NumDTensor)
+                    {
+                        throw std::runtime_error(
+                            "wrong! gemm_descs[i].stride_Ds_.size() does not match NumDTensor");
+                    }
+
+                    StrideDs[j] = gemm_descs[i].stride_Ds_[j];
+                });
+
+                const auto e_grid_desc_m_n =
+                    GridwiseGemm::template MakeEGridDescriptor_M_N<ELayout, GemmSpec>(
+                        AverM, N, StrideE);
+
+                // block-to-e-tile map
+                const auto local_b2c_tile_map = Block2ETileMap{e_grid_desc_m_n, k_batch_};
+
+                grid_size_grp_ = local_b2c_tile_map.CalculateGridSize(e_grid_desc_m_n);
+
+                if(group_id * grid_size_grp_ != grid_size_)
+                {
+                    throw std::runtime_error("wrong! grid_size_grp_ is not identical!");
+                }
+
+                grid_size_ += grid_size_grp_;
+
+                // check block-to-E-tile
+                if(!local_b2c_tile_map.CheckValidity(e_grid_desc_m_n))
+                {
+                    throw std::runtime_error("wrong! block_2_etile_map validation failed");
+                }
+
+                if(!GridwiseGemm::
+                       template CheckValidity<ALayout, BLayout, DsLayout, ELayout, GemmSpec>(
+                           AverM, N, K, StrideA, StrideB, StrideDs, StrideE, 1))
+                {
+                    throw std::runtime_error(
+                        "wrong! GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3 has invalid setting");
+                }
+
+                gemm_desc_kernel_arg_.push_back(GemmBiasTransKernelArg{
+                    nullptr,
+                    nullptr,
+                    p_ds_grid,
+                    nullptr,
+                    AverM,
+                    N,
+                    K,
+                    StrideA,
+                    StrideB,
+                    StrideDs,
+                    StrideE,
+                });
+
+                group_id++;
+            }
+
+            const auto e_grid_desc_sum_m_n =
+                GridwiseGemm::template MakeEGridDescriptor_M_N<ELayout, GemmSpec>(
+                    sum_of_m, gemm_desc_kernel_arg_[0].N_, gemm_desc_kernel_arg_[0].StrideE_);
+
+            const auto local_b2c_tile_map = Block2ETileMap{e_grid_desc_sum_m_n, 1};
+
+            barrier_size_grp_ = local_b2c_tile_map.CalculateGridSize(e_grid_desc_sum_m_n);
+        }
+
+        //  private:
+        index_t group_count_;
+
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation c_element_op_;
+
+        std::vector<GemmBiasTransKernelArg> gemm_desc_kernel_arg_;
+        std::vector<Tuple<index_t, index_t>> a_mtx_mraw_kraw_;
+        std::vector<Tuple<index_t, index_t>> b_mtx_nraw_kraw_;
+
+        const void* grouped_gemm_kernel_args_dev;
+
+        index_t grid_size_;
+        index_t grid_size_grp_;
+        index_t barrier_size_grp_;
+        index_t sum_of_m;
+
+        index_t k_batch_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            bool has_main_k_block_loop = true;
+
+            for(std::size_t i = 0; i < arg.gemm_desc_kernel_arg_.size(); i++)
+            {
+                const auto KPad =
+                    GridwiseGemm::CalculateKPadded(arg.gemm_desc_kernel_arg_[i].K_, arg.k_batch_);
+
+                if(GridwiseGemm::CalculateHasMainKBlockLoop(KPad) != has_main_k_block_loop)
+                {
+                    throw std::runtime_error("wrong! not all gemm has_main_k_block_loop");
+                }
+            }
+
+            if(arg.grouped_gemm_kernel_args_dev == nullptr)
+            {
+                throw std::runtime_error("wrong! grouped_gemm_kernel_args_dev is nullpr");
+            }
+
+            float ave_time = 0;
+
+            auto launch_kernel = [&](auto has_main_k_block_loop_, auto e_global_memory_operation_) {
+                const auto kernel =
+                    kernel_grouped_gemm_xdl_fixed_nk<GridwiseGemm,
+                                                     GroupedGemmKernelArgument<NumDTensor>,
+                                                     GemmSpec,
+                                                     ALayout,
+                                                     BLayout,
+                                                     DsLayout,
+                                                     ELayout,
+                                                     DsDataType,
+                                                     Block2ETileMap,
+                                                     GroupedGemmBlock2ETileMap,
+                                                     AElementwiseOperation,
+                                                     BElementwiseOperation,
+                                                     CDEElementwiseOperation,
+                                                     e_global_memory_operation_,
+                                                     has_main_k_block_loop_>;
+
+                return launch_and_time_kernel(
+                    stream_config,
+                    kernel,
+                    dim3(arg.grid_size_),
+                    dim3(BlockSize),
+                    0,
+                    cast_pointer_to_constant_address_space(arg.grouped_gemm_kernel_args_dev),
+                    reinterpret_cast<uint32_t*>(arg.p_workspace_),
+                    arg.barrier_size_grp_,
+                    arg.gemm_desc_kernel_arg_.size(),
+                    arg.grid_size_grp_,
+                    arg.k_batch_,
+                    arg.a_element_op_,
+                    arg.b_element_op_,
+                    arg.c_element_op_);
+            };
+
+            constexpr auto AtomicAdd = InMemoryDataOperationEnum::AtomicAdd;
+            constexpr auto Set       = InMemoryDataOperationEnum::Set;
+
+            if(arg.k_batch_ > 1)
+            {
+                if(has_main_k_block_loop)
+                {
+                    ave_time =
+                        launch_kernel(integral_constant<bool, true>{},
+                                      integral_constant<InMemoryDataOperationEnum, AtomicAdd>{});
+                }
+                else
+                {
+                    ave_time =
+                        launch_kernel(integral_constant<bool, false>{},
+                                      integral_constant<InMemoryDataOperationEnum, AtomicAdd>{});
+                }
+            }
+            else
+            {
+                if(has_main_k_block_loop)
+                {
+                    ave_time = launch_kernel(integral_constant<bool, true>{},
+                                             integral_constant<InMemoryDataOperationEnum, Set>{});
+                }
+                else
+                {
+                    ave_time = launch_kernel(integral_constant<bool, false>{},
+                                             integral_constant<InMemoryDataOperationEnum, Set>{});
+                }
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(ck::type_convert<ck::index_t>(arg.gemm_desc_kernel_arg_.size()) != arg.group_count_)
+        {
+            return false;
+        }
+
+        bool supported = true;
+
+        // If we use padding we do not support vector loads for dimensions not divisible by vector
+        // load size.
+        if constexpr(GemmSpec != GemmSpecialization::Default)
+        {
+            // [A|B]BlockTransferSrcVectorDim value define dimension in the block {K0,M,K1} layout,
+            // thus we have to adapt it to the {M,K} or {N,K} layout.
+            const auto a_raw_vector_dim = ABlockTransferSrcVectorDim != 1 ? 1 : 0;
+            const auto b_raw_vector_dim = BBlockTransferSrcVectorDim != 1 ? 1 : 0;
+
+            for(index_t i = 0; i < arg.group_count_; ++i)
+            {
+                const auto a_vector_dim = arg.a_mtx_mraw_kraw_[i].At(Number<a_raw_vector_dim>{});
+                const auto b_vector_dim = arg.b_mtx_nraw_kraw_[i].At(Number<b_raw_vector_dim>{});
+
+                supported = supported & (a_vector_dim % ABlockTransferSrcScalarPerVector == 0);
+                supported = supported & (b_vector_dim % BBlockTransferSrcScalarPerVector == 0);
+            }
+        }
+
+        return supported;
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::vector<const void*>& p_As,
+                             std::vector<const void*>& p_Bs,
+                             std::vector<std::array<const void*, NumDTensor>>& p_Ds,
+                             std::vector<void*>& p_Es,
+                             std::vector<GemmDesc> gemm_descs,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation c_element_op)
+    {
+        return Argument{
+            p_As, p_Bs, p_Ds, p_Es, gemm_descs, a_element_op, b_element_op, c_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::vector<const void*>& p_As,
+                        std::vector<const void*>& p_Bs,
+                        std::vector<std::array<const void*, NumDTensor>>& p_Ds,
+                        std::vector<void*>& p_Es,
+                        std::vector<GemmDesc>& gemm_descs,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation c_element_op) override
+    {
+        return std::make_unique<Argument>(
+            p_As, p_Bs, p_Ds, p_Es, gemm_descs, a_element_op, b_element_op, c_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedGemm_Xdl_Fixed_NK"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+
+    static void SetDeviceKernelArgs(Argument& arg, const void* kernel_args)
+    {
+        arg.grouped_gemm_kernel_args_dev = kernel_args;
+    }
+
+    // polymorphic
+    void SetDeviceKernelArgs(BaseArgument* p_arg, const void* kernel_args) const override
+    {
+        return SetDeviceKernelArgs(*dynamic_cast<Argument*>(p_arg), kernel_args);
+    }
+
+    size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
+    {
+        auto arg = *dynamic_cast<const Argument*>(p_arg);
+
+        return arg.group_count_ * arg.barrier_size_grp_ * sizeof(uint32_t);
+    }
+
+    size_t GetDeviceKernelArgSize(const BaseArgument* p_arg) const override
+    {
+        auto arg = *dynamic_cast<const Argument*>(p_arg);
+
+        return arg.group_count_ * sizeof(GroupedGemmKernelArgument<NumDTensor>);
+    }
+
+    void SetWorkSpacePointer(BaseArgument* p_arg, void* p_workspace) const override
+    {
+        auto p_arg_          = dynamic_cast<Argument*>(p_arg);
+        p_arg_->p_workspace_ = p_workspace;
+
+        hip_check_error(hipMemset(p_workspace, 0, GetWorkSpaceSize(p_arg)));
+    }
+
+    static void SetKBatch(Argument& arg, index_t k_batch) { arg.UpdateKBatch(k_batch); }
+
+    // polymorphic
+    void SetKBatch(BaseArgument* p_arg, index_t k_batch) const override
+    {
+        return SetKBatch(*dynamic_cast<Argument*>(p_arg), k_batch);
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
index 1dd96809d5..9ade02391f 100644
--- a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
@@ -36,6 +36,13 @@ struct Add
         y = x0 + type_convert<half_t>(x1);
     };
 
+    template <>
+    __host__ __device__ constexpr void
+    operator()<half_t>(half_t& y, const float& x0, const float& x1) const
+    {
+        y = type_convert<half_t>(x0 + x1);
+    };
+
     template <>
     __host__ __device__ constexpr void
     operator()<half_t>(half_t& y, const float& x0, const half_t& x1) const
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index c4f00e83b7..7bb47e9d3c 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -587,7 +587,8 @@ struct OffsettedBlockToCTileMap
 {
     using underlying_type = UnderlyingBlockToCTileMap;
 
-    OffsettedBlockToCTileMap(UnderlyingBlockToCTileMap block_to_ctile_map, index_t block_start)
+    __host__ __device__ OffsettedBlockToCTileMap(UnderlyingBlockToCTileMap block_to_ctile_map,
+                                                 index_t block_start)
     {
         block_to_ctile_map_ = block_to_ctile_map;
         block_start_        = block_start;
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index 8f9e0915d2..6344aa538b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -15,6 +15,9 @@
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+
 namespace ck {
 
 // GEMM:
@@ -74,6 +77,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
 
+    using GemmSpecialization = ck::tensor_operation::device::GemmSpecialization;
+
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
@@ -330,6 +335,94 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
 
     using DsGridPointer = decltype(MakeDsGridPointer());
 
+    template <typename ALayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeAGridDescriptor_M_K(index_t MRaw, index_t KRaw, index_t StrideA)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(I1, StrideA));
+            }
+        }();
+
+        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+    }
+
+    template <typename BLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeBGridDescriptor_N_K(index_t KRaw, index_t NRaw, index_t StrideB)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(StrideB, I1));
+            }
+        }();
+
+        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+    }
+
+    template <typename ELayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+        const auto e_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    template <typename DsLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                             const std::array<index_t, NumDTensor>& NRaws,
+                             const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return MakeEGridDescriptor_M_N<DLayout, GemmSpec>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __device__ __host__ static constexpr auto GetMPerBlock() { return MPerBlock; }
+
     template <bool HasMainKBlockLoop,
               typename AGridDesc_AK0_M_AK1,
               typename BGridDesc_BK0_N_BK1,
@@ -758,6 +851,85 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
             });
         }
     }
+
+    template <bool HasMainKBlockLoop,
+              GemmSpecialization GemmSpec,
+              typename ALayout,
+              typename BLayout,
+              typename DsLayout,
+              typename ELayout,
+              typename Block2ETileMap>
+    __device__ static void Run(const void* __restrict__ p_a_grid_,
+                               const void* __restrict__ p_b_grid_,
+                               DsGridPointer p_ds_grid,
+                               void* __restrict__ p_e_grid_,
+                               void* __restrict__ p_shared,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation& cde_element_op,
+                               const index_t M,
+                               const index_t N,
+                               const index_t K,
+                               const index_t StrideA,
+                               const index_t StrideB,
+                               const std::array<index_t, NumDTensor> StrideDs,
+                               const index_t StrideE,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        const auto p_a_grid = reinterpret_cast<const ADataType*>(p_a_grid_);
+        const auto p_b_grid = reinterpret_cast<const BDataType*>(p_b_grid_);
+        const auto p_e_grid = reinterpret_cast<EDataType*>(p_e_grid_);
+
+        // tensor descriptors for problem definiton
+        const auto a_grid_desc_m_k = MakeAGridDescriptor_M_K<ALayout, GemmSpec>(M, K, StrideA);
+        const auto b_grid_desc_n_k = MakeBGridDescriptor_N_K<BLayout, GemmSpec>(K, N, StrideB);
+
+        using DsGridDesc_M_N =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_M_N<DsLayout, GemmSpec>({}, {}, {}))>;
+
+        DsGridDesc_M_N ds_grid_desc_m_n;
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            using DLayout = remove_cvref_t<tuple_element_t<j.value, DsLayout>>;
+
+            ds_grid_desc_m_n(j) = MakeEGridDescriptor_M_N<DLayout, GemmSpec>(M, N, StrideDs[j]);
+        });
+
+        const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+        // tensor descriptors for block/thread-wise copy
+        const auto a_grid_desc_ak0_m_ak1 = MakeDefaultAGridDescriptor_AK0_M_AK1(a_grid_desc_m_k);
+
+        const auto b_grid_desc_bk0_n_bk1 = MakeDefaultBGridDescriptor_BK0_N_BK1(b_grid_desc_n_k);
+
+        using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                DsGridDesc_M_N{}))>;
+
+        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock ds_grid_desc_mblock_mperblock_nblock_nperblock;
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            ds_grid_desc_mblock_mperblock_nblock_nperblock(j) =
+                MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[j]);
+        });
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(e_grid_desc_m_n);
+
+        Run<HasMainKBlockLoop>(p_a_grid,
+                               p_b_grid,
+                               p_ds_grid,
+                               p_e_grid,
+                               p_shared,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op,
+                               a_grid_desc_ak0_m_ak1,
+                               b_grid_desc_bk0_n_bk1,
+                               ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                               e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               block_2_etile_map);
+    }
 };
 
 } // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
new file mode 100644
index 0000000000..e22391293e
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
@@ -0,0 +1,1086 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+
+namespace ck {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <typename ABDataType, // FIXME: don't assume A/B have same datatype
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t MPerXdl,
+          index_t NPerXdl,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_KBatch_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_KBatch_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched,
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
+{
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    using GemmSpecialization = ck::tensor_operation::device::GemmSpecialization;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto AK1         = Number<AK1Value>{};
+    static constexpr auto BK1         = Number<BK1Value>{};
+    static constexpr auto AK0PerBlock = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0PerBlock = Number<KPerBlock / BK1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+    // denorm test fix, required to work around fp16 mfma issue
+    // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
+    // when mfma if fixed, remove this section and update
+    // ABDataTypeAdjusted -> ABDataType throughout this file
+#if CK_WORKAROUND_DENORM_FIX
+    using ABDataTypeAdjusted =
+        conditional_t<is_same_v<ABDataType, ck::half_t>, ck::bhalf_t, ABDataType>;
+#else
+    using ABDataTypeAdjusted = ABDataType;
+#endif
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_KBatch_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(I1, AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(AK0PerBlock * Number<MPerBlock + ABlockLdsExtraM>{} * AK1,
+                       Number<MPerBlock + ABlockLdsExtraM>{} * AK1,
+                       AK1,
+                       I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_KBatch_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(I1, BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(BK0PerBlock * Number<NPerBlock + BBlockLdsExtraN>{} * BK1,
+                       Number<NPerBlock + BBlockLdsExtraN>{} * BK1,
+                       BK1,
+                       I1));
+    }
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    {
+        constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+        constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl>{},
+                           I1,
+                           Number<CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>{}));
+
+        return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // ck::Tuple<const D0DataType*, const D1DataType*, ...>
+    static constexpr auto MakeDsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                return static_cast<const DDataType*>(nullptr);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        // LDS allocation for C shuffle in LDS
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+        constexpr auto c_block_size =
+            c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
+
+        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
+                             sizeof(ABDataType),
+                         c_block_size * sizeof(CShuffleDataType));
+    }
+
+    __host__ __device__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_least_multiple(M, MPerBlock);
+    }
+
+    __host__ __device__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_least_multiple(N, NPerBlock);
+    }
+
+    __host__ __device__ static auto CalculateKPadded(index_t K, index_t K_Batch)
+    {
+        return math::integer_least_multiple(K, KPerBlock * K_Batch);
+    }
+
+    template <typename ALayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeAGridDescriptor_KBatch_AK0_M_AK1(index_t M, index_t K, index_t StrideA, index_t KBatch)
+    {
+        const auto a_grid_desc_m_k = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto MPad = CalculateMPadded(M);
+        const auto KPad = CalculateKPadded(K, KBatch);
+
+        const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto AK0 = KPad / (KBatch * AK1);
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadM = (MPerBlock - M % MPerBlock) % MPerBlock;
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, AK0, AK1)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, AK0, AK1)),
+                           make_pass_through_transform(M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    template <typename BLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeBGridDescriptor_KBatch_BK0_N_BK1(index_t K, index_t N, index_t StrideB, index_t KBatch)
+    {
+        const auto b_grid_desc_k_n = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(StrideB, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(K, N), make_tuple(I1, StrideB));
+            }
+        }();
+
+        const auto NPad = CalculateNPadded(N);
+        const auto KPad = CalculateKPadded(K, KBatch);
+
+        const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
+            b_grid_desc_k_n,
+            make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        const auto BK0 = KPad / (KBatch * BK1);
+
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            // const auto PadN = (NPerBlock - N % NPerBlock) % NPerBlock;
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, BK0, BK1)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+        else
+        {
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(KBatch, BK0, BK1)),
+                           make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
+        }
+    }
+
+    // E desc for destination in blockwise copy
+    template <typename EGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            e_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return e_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // Ds desc for source in blockwise copy
+    template <typename DsGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const DsGridDesc_M_N& ds_grid_desc_m_n)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // return block_id to E matrix tile idx (m0, n0) mapping
+    template <typename EGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeDefaultBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, EGridDesc_M_N>(
+            e_grid_desc_m_n);
+    }
+
+    template <typename ALayout,
+              typename BLayout,
+              typename DsLayout,
+              typename ELayout,
+              GemmSpecialization GemmSpec>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const index_t M,
+                  const index_t N,
+                  const index_t K,
+                  const index_t StrideA,
+                  const index_t StrideB,
+                  const std::array<index_t, NumDTensor> StrideDs,
+                  const index_t StrideE,
+                  const index_t KBatch)
+    {
+        const auto a_grid_desc_kbatch_ak0_m_ak1 =
+            MakeAGridDescriptor_KBatch_AK0_M_AK1<ALayout, GemmSpec>(M, K, StrideA, KBatch);
+        const auto b_grid_desc_kbatch_bk0_n_bk1 =
+            MakeBGridDescriptor_KBatch_BK0_N_BK1<BLayout, GemmSpec>(K, N, StrideB, KBatch);
+
+        ignore = StrideDs;
+
+        const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+#if 0
+        // check tile size
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && K % KPerBlock == 0))
+        {
+            return false;
+        }
+#endif
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = K / KPerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        // check tensor size: cannot be larger than 2GB each
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        if(!(a_grid_desc_kbatch_ak0_m_ak1.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+             b_grid_desc_kbatch_bk0_n_bk1.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+             e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / KPerBlock;
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    using DsGridPointer = decltype(MakeDsGridPointer());
+
+    template <typename ELayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+        const auto e_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    template <typename DsLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                             const std::array<index_t, NumDTensor>& NRaws,
+                             const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return MakeEGridDescriptor_M_N<DLayout, GemmSpec>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __device__ __host__ static constexpr auto GetMPerBlock() { return MPerBlock; }
+
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+              index_t NumDTensor_,
+              typename DsDataType_,
+              typename AGridDesc_KBatch_AK0_M_AK1,
+              typename BGridDesc_KBatch_BK0_N_BK1,
+              typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename CDEElementwiseOperation_,
+              typename Block2ETileMap>
+    __device__ static void Run(const ABDataType* __restrict__ p_a_grid,
+                               const ABDataType* __restrict__ p_b_grid,
+                               DsGridPointer p_ds_grid,
+                               EDataType* __restrict__ p_e_grid,
+                               void* __restrict__ p_shared,
+                               uint32_t* barrier_count_finished,
+                               const index_t KBatch,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation_& cde_element_op,
+                               const AGridDesc_KBatch_AK0_M_AK1& a_grid_desc_kbatch_ak0_m_ak1,
+                               const BGridDesc_KBatch_BK0_N_BK1& b_grid_desc_kbatch_bk0_n_bk1,
+                               const DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_kbatch_ak0_m_ak1.GetElementSpaceSize());
+
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_kbatch_bk0_n_bk1.GetElementSpaceSize());
+
+        const auto ds_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_ds_grid[i],
+                    ds_grid_desc_mblock_mperblock_nblock_nperblock[i].GetElementSpaceSize());
+            },
+            Number<NumDTensor_>{});
+
+        auto e_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_e_grid, e_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t kbatch_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
+
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * MPerBlock);
+
+        const index_t n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I2] * NPerBlock);
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_kbatch_ak0_m_ak1 =
+            GetABlockDescriptor_KBatch_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_kbatch_bk0_n_bk1 =
+            GetBBlockDescriptor_KBatch_BK0PerBlock_NPerBlock_BK1();
+
+        // A matrix blockwise copy
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                AElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<1, AK0PerBlock, MPerBlock, AK1>,
+                                                ABlockTransferThreadClusterLengths_KBatch_AK0_M_AK1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ABDataType,
+                                                ABDataTypeAdjusted,
+                                                decltype(a_grid_desc_kbatch_ak0_m_ak1),
+                                                decltype(a_block_desc_kbatch_ak0_m_ak1),
+                                                ABlockTransferSrcAccessOrder,
+                                                Sequence<2, 0, 1, 3>,
+                                                ABlockTransferSrcVectorDim,
+                                                3,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_AK1,
+                                                1,
+                                                1,
+                                                AThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                a_grid_desc_kbatch_ak0_m_ak1,
+                make_multi_index(kbatch_id, 0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_kbatch_ak0_m_ak1,
+                make_multi_index(0, 0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // B matrix blockwise copy
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<1, BK0PerBlock, NPerBlock, BK1>,
+                                                BBlockTransferThreadClusterLengths_KBatch_BK0_N_BK1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                ABDataType,
+                                                ABDataTypeAdjusted,
+                                                decltype(b_grid_desc_kbatch_bk0_n_bk1),
+                                                decltype(b_block_desc_kbatch_bk0_n_bk1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<2, 0, 1, 3>,
+                                                BBlockTransferSrcVectorDim,
+                                                3,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_BK1,
+                                                1,
+                                                1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                b_grid_desc_kbatch_bk0_n_bk1,
+                make_multi_index(kbatch_id, 0, n_block_data_idx_on_grid, 0),
+                b_element_op,
+                b_block_desc_kbatch_bk0_n_bk1,
+                make_multi_index(0, 0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // GEMM definition
+        //   c_mtx += transpose(a_mtx) * b_mtx
+        //     a_mtx[K0PerBlock, MPerBlock] is in LDS
+        //     b_mtx[K0PerBlock, NPerBlock] is in LDS
+        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+        //       register
+        // sanity check
+        constexpr index_t KPack =
+            math::max(math::lcm(AK1, BK1),
+                      MfmaSelector<ABDataTypeAdjusted, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
+            BlockSize,
+            ABDataTypeAdjusted,
+            AccDataType,
+            decltype(a_block_desc_ak0_m_ak1),
+            decltype(b_block_desc_bk0_n_bk1),
+            MPerXdl,
+            NPerXdl,
+            MXdlPerWave,
+            NXdlPerWave,
+            KPack,
+            LoopSched>();
+
+#if 1
+        if(block_work_idx[I0] == 0)
+        {
+            const index_t nThreadSize = CDEShuffleBlockTransferScalarPerVector_NPerBlock;
+            const index_t numNThreads = NPerBlock / nThreadSize;
+            const index_t numMThreads = BlockSize / numNThreads;
+            const index_t mThreadSize = MPerBlock / numMThreads;
+
+            const index_t m_tid = get_thread_local_1d_id() / numNThreads;
+            const index_t n_tid = get_thread_local_1d_id() % numNThreads;
+
+            auto c_thread_desc_mblock_mperblock_nblock_nperblock =
+                make_naive_tensor_descriptor_packed(
+                    make_tuple(I1, Number<mThreadSize>{}, I1, Number<nThreadSize>{}));
+
+            StaticBuffer<AddressSpaceEnum::Vgpr,
+                         EDataType,
+                         c_thread_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize(),
+                         true>
+                e_thread_zero_buf;
+
+            auto c_thread_copy = ThreadwiseTensorSliceTransfer_v1r3<
+                EDataType,
+                EDataType,
+                decltype(c_thread_desc_mblock_mperblock_nblock_nperblock),
+                decltype(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                ck::tensor_operation::element_wise::PassThrough,
+                Sequence<1, mThreadSize, 1, nThreadSize>,
+                Sequence<0, 1, 2, 3>,
+                3,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+                InMemoryDataOperationEnum::Set,
+                1,
+                true>{e_grid_desc_mblock_mperblock_nblock_nperblock,
+                      make_multi_index(block_work_idx[I1],
+                                       m_tid * mThreadSize,
+                                       block_work_idx[I2],
+                                       n_tid * nThreadSize),
+                      ck::tensor_operation::element_wise::PassThrough{}};
+
+            c_thread_copy.Run(c_thread_desc_mblock_mperblock_nblock_nperblock,
+                              make_tuple(I0, I0, I0, I0),
+                              e_thread_zero_buf,
+                              e_grid_desc_mblock_mperblock_nblock_nperblock,
+                              e_grid_buf);
+
+            __syncthreads();
+
+            if(threadIdx.x == 0)
+            {
+                atomicAdd(barrier_count_finished, 1);
+            }
+        }
+#endif
+
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataTypeAdjusted*>(p_shared),
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataTypeAdjusted*>(p_shared) + a_block_space_size_aligned,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(0, KPerBlock / AK1, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(0, KPerBlock / BK1, 0, 0);
+
+        // gridwise GEMM pipeline
+        const auto gridwise_gemm_pipeline =
+            GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>();
+
+        const index_t num_k_block_main_loop =
+            __builtin_amdgcn_readfirstlane((a_grid_desc_kbatch_ak0_m_ak1.GetLength(I1) *
+                                            a_grid_desc_kbatch_ak0_m_ak1.GetLength(I3)) /
+                                           KPerBlock);
+
+        gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(a_grid_desc_kbatch_ak0_m_ak1,
+                                                               a_block_desc_kbatch_ak0_m_ak1,
+                                                               a_blockwise_copy,
+                                                               a_grid_buf,
+                                                               a_block_buf,
+                                                               a_block_slice_copy_step,
+                                                               b_grid_desc_kbatch_bk0_n_bk1,
+                                                               b_block_desc_kbatch_bk0_n_bk1,
+                                                               b_blockwise_copy,
+                                                               b_grid_buf,
+                                                               b_block_buf,
+                                                               b_block_slice_copy_step,
+                                                               blockwise_gemm,
+                                                               c_thread_buf,
+                                                               num_k_block_main_loop);
+
+        // shuffle C and write out
+        {
+            if(threadIdx.x == 0)
+            {
+                while(__atomic_load_n(barrier_count_finished, __ATOMIC_RELAXED) == 0) {}
+            }
+
+            __syncthreads();
+
+            static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
+                              NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
+                          "wrong!");
+
+            constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+            constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+            // TODO: hacky, fix it!
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+                blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            // TODO: hacky, fix it!
+            // c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp is only used to get lengths
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp =
+                blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I0);
+            constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I1);
+            constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I2);
+            constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I3);
+            constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I4);
+            constexpr auto M3 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I5);
+            constexpr auto M4 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I6);
+            constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I7);
+
+            constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+                GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<CShuffleDataType*>(p_shared),
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2 = transform_tensor_descriptor(
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMXdlPerWavePerShuffle>{}, // M0 (MXdlPerWave) per shuffle
+                        M1,                                      // M1 = MWave
+                        M2,                                      // M2 * M3 * M4 = MPerXdl
+                        M3,
+                        M4)),
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNXdlPerWavePerShuffle>{}, // N0 (NXdlPerWave) per shuffle
+                        N1,                                      // N1 = NWave
+                        N2))),                                   // N2 = NPerXdl
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(
+                    Sequence<>{}, Sequence<0, 2, 4, 5, 6>{}, Sequence<>{}, Sequence<1, 3, 7>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block =
+                blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
+                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto m_thread_data_on_block_idx =
+                m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
+                    make_multi_index(m_thread_data_on_block));
+
+            const auto n_thread_data_on_block_to_n0_n1_n2_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                    make_tuple(Sequence<0, 1, 2>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_idx =
+                n_thread_data_on_block_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                    make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   CShuffleDataType,
+                                                   decltype(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   decltype(c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMXdlPerWavePerShuffle,
+                                                            CShuffleNXdlPerWavePerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            M2,
+                                                            I1,
+                                                            M4,
+                                                            I1>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                                   7,
+                                                   1,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                    make_multi_index(0,
+                                     0,
+                                     m_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     m_thread_data_on_block_idx[I3],
+                                     m_thread_data_on_block_idx[I4],
+                                     n_thread_data_on_block_idx[I2]),
+                    ck::tensor_operation::element_wise::PassThrough{}};
+
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_desc_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_desc_mblock_mperblock_nblock_nperblock),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_desc_mblock_mperblock_nblock_nperblock[i]; },
+                    Number<NumDTensor_>{}));
+
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_buf_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_buf),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_buf[i]; },
+                    Number<NumDTensor_>{}));
+
+            // tuple of starting index of C/Ds blockwise copy
+            const auto idx_c_ds_block_begin = container_concat(
+                make_tuple(make_multi_index(0, 0, 0, 0)),
+                generate_tuple(
+                    [&](auto) {
+                        return make_multi_index(block_work_idx[I1], 0, block_work_idx[I2], 0);
+                    },
+                    Number<NumDTensor_>{}));
+
+            // space filling curve for threadwise C in VGPR before shuffle
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MXdlPerWave, NXdlPerWave, 1, 1, M2, 1, M4, 1>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                  Sequence<CShuffleMXdlPerWavePerShuffle,
+                                           CShuffleNXdlPerWavePerShuffle,
+                                           1,
+                                           1,
+                                           M2,
+                                           1,
+                                           M4,
+                                           1>>{};
+
+            // space filling curve for shuffled blockwise C/D/E
+            constexpr auto sfc_cde_block =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                                           1,
+                                           CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>>{};
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_cde_block.GetNumOfAccess(), "wrong!");
+
+            // blockwise copy C/D/E between LDS and global
+            auto cde_block_copy_lds_and_global = ThreadGroupTensorSliceTransfer_v7<
+                ThisThreadBlock,
+                decltype(container_concat(make_tuple(CShuffleDataType{}), DsDataType_{})),
+                Tuple<EDataType>,
+                decltype(c_ds_desc_refs),
+                decltype(tie(e_grid_desc_mblock_mperblock_nblock_nperblock)),
+                CDEElementwiseOperation_,
+                Sequence<static_cast<index_t>(EGlobalMemoryDataOperation)>, // FIXME: make
+                                                                            // Sequence support
+                                                                            // arbitray type
+                Sequence<1,
+                         CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                         1,
+                         CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>, // BlockSliceLengths,
+                CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                Sequence<0, 1, 2, 3>, // typename DimAccessOrder,
+                3,                    // index_t VectorDim,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+                sequence_merge_t<
+                    Sequence<true>,
+                    uniform_sequence_gen_t<NumDTensor_,
+                                           false>>, // ThreadTransferSrcResetCoordinateAfterRunFlags
+                Sequence<false>>                    // ThreadTransferDstResetCoordinateAfterRunFlags
+                {c_ds_desc_refs,
+                 idx_c_ds_block_begin,
+                 tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                 make_tuple(make_multi_index(block_work_idx[I1], 0, block_work_idx[I2], 0)),
+                 cde_element_op};
+
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // each thread write its data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each block copy its data from LDS to global
+                cde_block_copy_lds_and_global.Run(
+                    c_ds_desc_refs,
+                    c_ds_buf_refs,
+                    tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                    tie(e_grid_buf));
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto cde_lds_and_global_step =
+                        sfc_cde_block.GetForwardStep(access_id);
+
+                    // move on Ds
+                    static_for<0, NumDTensor_, 1>{}([&](auto i) {
+                        cde_block_copy_lds_and_global.MoveSrcSliceWindow(
+                            c_ds_desc_refs, i + I1, cde_lds_and_global_step);
+                    });
+
+                    // move on E
+                    cde_block_copy_lds_and_global.MoveDstSliceWindow(
+                        tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                        I0,
+                        cde_lds_and_global_step);
+                }
+            });
+
+            if(threadIdx.x == 0)
+            {
+                index_t k_id_finished_t = atomicAdd(barrier_count_finished, 1);
+
+                if(k_id_finished_t == KBatch)
+                {
+                    *barrier_count_finished = 0;
+                }
+            }
+        }
+    }
+
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+              GemmSpecialization GemmSpec,
+              typename ALayout,
+              typename BLayout,
+              typename DsLayout,
+              typename ELayout,
+              typename Block2ETileMap>
+    __device__ static void Run(const void* __restrict__ p_a_grid_,
+                               const void* __restrict__ p_b_grid_,
+                               DsGridPointer p_ds_grid,
+                               void* __restrict__ p_e_grid_,
+                               void* __restrict__ p_shared,
+                               uint32_t* barrier_count_finished,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation& cde_element_op,
+                               const index_t M,
+                               const index_t N,
+                               const index_t K,
+                               const index_t StrideA,
+                               const index_t StrideB,
+                               const std::array<index_t, NumDTensor> StrideDs,
+                               const index_t StrideE,
+                               const index_t KBatch,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        const auto p_a_grid = reinterpret_cast<const ABDataType*>(p_a_grid_);
+        const auto p_b_grid = reinterpret_cast<const ABDataType*>(p_b_grid_);
+        const auto p_e_grid = reinterpret_cast<EDataType*>(p_e_grid_);
+
+        using DsGridDesc_M_N =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_M_N<DsLayout, GemmSpec>({}, {}, {}))>;
+
+        DsGridDesc_M_N ds_grid_desc_m_n;
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            using DLayout = remove_cvref_t<tuple_element_t<j.value, DsLayout>>;
+
+            ds_grid_desc_m_n(j) = MakeEGridDescriptor_M_N<DLayout, GemmSpec>(M, N, StrideDs[j]);
+        });
+
+        const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+        // tensor descriptors for block/thread-wise copy
+        const auto a_grid_desc_kbatch_ak0_m_ak1 =
+            MakeAGridDescriptor_KBatch_AK0_M_AK1<ALayout, GemmSpec>(M, K, StrideA, KBatch);
+
+        const auto b_grid_desc_kbatch_bk0_n_bk1 =
+            MakeBGridDescriptor_KBatch_BK0_N_BK1<BLayout, GemmSpec>(K, N, StrideB, KBatch);
+
+        using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                DsGridDesc_M_N{}))>;
+
+        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock ds_grid_desc_mblock_mperblock_nblock_nperblock;
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            ds_grid_desc_mblock_mperblock_nblock_nperblock(j) =
+                MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[j]);
+        });
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(e_grid_desc_m_n);
+
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        const index_t kbatch_id = __builtin_amdgcn_readfirstlane(block_work_idx[I0]);
+
+        if(kbatch_id == KBatch - 1)
+        {
+            Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, NumDTensor, DsDataType>(
+                p_a_grid,
+                p_b_grid,
+                p_ds_grid,
+                p_e_grid,
+                p_shared,
+                barrier_count_finished,
+                KBatch,
+                a_element_op,
+                b_element_op,
+                cde_element_op,
+                a_grid_desc_kbatch_ak0_m_ak1,
+                b_grid_desc_kbatch_bk0_n_bk1,
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                block_2_etile_map);
+        }
+        else
+        {
+            Run<HasMainKBlockLoop, EGlobalMemoryDataOperation, 0, Tuple<>>(
+                p_a_grid,
+                p_b_grid,
+                p_ds_grid,
+                p_e_grid,
+                p_shared,
+                barrier_count_finished,
+                KBatch,
+                a_element_op,
+                b_element_op,
+                ck::tensor_operation::element_wise::PassThrough{},
+                a_grid_desc_kbatch_ak0_m_ak1,
+                b_grid_desc_kbatch_bk0_n_bk1,
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+                block_2_etile_map);
+        }
+    }
+};
+
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index f874e52ddd..5096f5c30b 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -101,6 +101,7 @@ using MultiplyAdd    = ck::tensor_operation::element_wise::MultiplyAdd;
 using ScaleAdd       = ck::tensor_operation::element_wise::ScaleAdd;
 using Gelu           = ck::tensor_operation::element_wise::Gelu;
 using Swish          = ck::tensor_operation::element_wise::Swish;
+using Add            = ck::tensor_operation::element_wise::Add;
 
 template <typename Activation>
 using Activation_Mul_Clamp = ck::tensor_operation::element_wise::Activation_Mul_Clamp<Activation>;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_bias.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_bias.hpp
new file mode 100644
index 0000000000..9365bf7bc2
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_bias.hpp
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// fp16_output
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+// fp32_output
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F32,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Row_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         F32_Tuple,
+                                                         F32,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                           BLayout,
+                                                           Row_Tuple,
+                                                           ELayout,
+                                                           ADataType,
+                                                           BDataType,
+                                                           F32_Tuple,
+                                                           EDataType,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Add>>
+{
+    using DeviceOp = DeviceGroupedGemmFixedNK<ALayout,
+                                              BLayout,
+                                              Row_Tuple,
+                                              ELayout,
+                                              ADataType,
+                                              BDataType,
+                                              F32_Tuple,
+                                              EDataType,
+                                              PassThrough,
+                                              PassThrough,
+                                              Add>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        // fp16_output
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        // fp32_output
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, float>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/utility/device_memory.hpp b/library/include/ck/library/utility/device_memory.hpp
index b2099acfaa..d2e611a77c 100644
--- a/library/include/ck/library/utility/device_memory.hpp
+++ b/library/include/ck/library/utility/device_memory.hpp
@@ -26,7 +26,9 @@ struct DeviceMem
     void* GetDeviceBuffer() const;
     std::size_t GetBufferSize() const;
     void ToDevice(const void* p) const;
+    void ToDevice(const void* p, const std::size_t cpySize) const;
     void FromDevice(void* p) const;
+    void FromDevice(void* p, const std::size_t cpySize) const;
     void SetZero() const;
     template <typename T>
     void SetValue(T x) const;
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index e2877f2a5c..90223fd9bd 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -83,7 +83,6 @@ using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instanc
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
         DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
     // clang-format on
-    // clang-format on
     >;
 
 void add_device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instances(
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/CMakeLists.txt
new file mode 100644
index 0000000000..ef8a440c1a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/CMakeLists.txt
@@ -0,0 +1,7 @@
+add_instance_library(device_grouped_gemm_bias_instance
+   device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instance.cpp
+
+   device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..28be904e65
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,79 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using D0DataType = F32;
+using DsDataType = ck::Tuple<D0DataType>;
+
+using D0Layout = Row;
+using DsLayout = ck::Tuple<D0Layout>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+        // clang-format on
+        >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..5f5b86a0e4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,82 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using D0DataType = F32;
+using DsDataType = ck::Tuple<D0DataType>;
+
+using D0Layout = Row;
+using DsLayout = ck::Tuple<D0Layout>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,   128,    64,   8,   8,   32,   32,    4,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    64,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    32,   256,    64,   8,   8,   32,   32,    1,    4,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,    64,    64,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,    64,    64,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+        // clang-format on
+        >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..fa8441431a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using D0DataType = F32;
+using DsDataType = ck::Tuple<D0DataType>;
+
+using D0Layout = Row;
+using DsLayout = ck::Tuple<D0Layout>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    32,   128,    32,   8,   8,   32,   32,    1,    1, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    32,   256,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    16,   128,    32,   8,   8,   16,   16,    1,    4, S< 1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>
+        // clang-format on
+        >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         DsDataType,
+                                                         F32,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..386a885642
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_bias/device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using D0DataType = F32;
+using DsDataType = ck::Tuple<D0DataType>;
+
+using D0Layout = Row;
+using DsLayout = ck::Tuple<D0Layout>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Add         = ck::tensor_operation::element_wise::Add;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_irregular_tile_instances =
+    std::tuple<
+        // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   256,    32,   128,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,   128,    64,   8,   8,   32,   32,    4,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    64,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,   128,    32,   256,    64,   8,   8,   32,   32,    1,    4,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,    64,    64,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,    64,    64,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              4>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F32, PassThrough, PassThrough,         Add, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              4>
+        // clang-format on
+        >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         DsDataType,
+                                                         F32,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         Add>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_xdl_fixed_nk_bias_f16_f16_f32_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/utility/device_memory.cpp b/library/src/utility/device_memory.cpp
index e054216dc9..61b6326b57 100644
--- a/library/src/utility/device_memory.cpp
+++ b/library/src/utility/device_memory.cpp
@@ -37,6 +37,11 @@ void DeviceMem::ToDevice(const void* p) const
     }
 }
 
+void DeviceMem::ToDevice(const void* p, const std::size_t cpySize) const
+{
+    hip_check_error(hipMemcpy(mpDeviceBuf, const_cast<void*>(p), cpySize, hipMemcpyHostToDevice));
+}
+
 void DeviceMem::FromDevice(void* p) const
 {
     if(mpDeviceBuf)
@@ -49,6 +54,11 @@ void DeviceMem::FromDevice(void* p) const
     }
 }
 
+void DeviceMem::FromDevice(void* p, const std::size_t cpySize) const
+{
+    hip_check_error(hipMemcpy(p, mpDeviceBuf, cpySize, hipMemcpyDeviceToHost));
+}
+
 void DeviceMem::SetZero() const
 {
     if(mpDeviceBuf)
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index 426f68d443..da9e92b862 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -16,4 +16,3 @@ cmake
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
-

From bd8024b84ab72d2fee145e24e5eb94e2501a792e Mon Sep 17 00:00:00 2001
From: Lauren Wrubleski <Liam.Wrubleski@amd.com>
Date: Mon, 4 Sep 2023 10:49:40 -0600
Subject: [PATCH 108/194] Fix config header installation (#880)

---
 CMakeLists.txt | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index fd03c9b207..4ca54d847d 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -449,7 +449,7 @@ rocm_install(FILES
 )
 
 # Install CK version and configuration files
-install(FILES
+rocm_install(FILES
     ${PROJECT_BINARY_DIR}/include/ck/version.h
     ${PROJECT_BINARY_DIR}/include/ck/config.h
     DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck/

From c981f6d033d5af81aa3809c05beccab219aa8027 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 5 Sep 2023 17:07:41 +0200
Subject: [PATCH 109/194] Fix K padding calculation for grouped conv data
 (#876)

* Fix K padding calculation for grouped conv data

* Restore previous padd for 1x1 specialization
---
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  1 +
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |  2 ++
 .../transform_conv_bwd_data_to_gemm_v1.hpp    | 33 ++++++++++---------
 3 files changed, 21 insertions(+), 15 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 7f6fb908fa..c828269acf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -280,6 +280,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
                                       BK1,
                                       MPerBlock,
                                       NPerBlock,
+                                      KPerBlock,
                                       DoPadGemmM,
                                       DoPadGemmN>{};
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index 6344aa538b..1d920fb44d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -268,6 +268,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
                           (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
                       "Invalid tuning param!");
+        static_assert(KPerBlock % AK1Value == 0 && KPerBlock % BK1Value == 0,
+                      "KPerBlock must be divisible by AK1Value and BK1Value!");
 
         const auto M  = a_grid_desc_m_k.GetLength(I0);
         const auto N  = b_grid_desc_n_k.GetLength(I0);
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
index b1b203b43f..2be0b66812 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp
@@ -164,6 +164,7 @@ template <
     index_t BK1,
     index_t GemmMPerBlock,
     index_t GemmNPerBlock,
+    index_t GemmKPerBlock,
     bool DoPadGemmM,
     bool DoPadGemmN>
 struct TransformConvBwdDataToGemm_v1
@@ -308,9 +309,6 @@ struct TransformConvBwdDataToGemm_v1
             const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
             const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
-            const index_t AK0 =
-                math::integer_divide_ceil(ZDotSlice * YDotSlice * XDotSlice * K, AK1);
-
             if constexpr(NDimSpatial == 2)
             {
                 // A: output tensor
@@ -367,9 +365,11 @@ struct TransformConvBwdDataToGemm_v1
                 const auto out_gemmk_gemmm_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
                         out_gemmk_gemmmraw_grid_desc,
-                        make_tuple(AK1, GemmMPerBlock),
+                        make_tuple(GemmKPerBlock, GemmMPerBlock),
                         Sequence<true, DoPadGemmM>{});
 
+                const index_t AK0 = out_gemmk_gemmm_padded_grid_desc.GetLength(I0) / AK1;
+
                 const auto out_gemmak0_gemmm_gemmak1_grid_desc = transform_tensor_descriptor(
                     out_gemmk_gemmm_padded_grid_desc,
                     make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
@@ -460,9 +460,11 @@ struct TransformConvBwdDataToGemm_v1
                 const auto out_gemmk_gemmm_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
                         out_gemmk_gemmmraw_grid_desc,
-                        make_tuple(AK1, GemmMPerBlock),
+                        make_tuple(GemmKPerBlock, GemmMPerBlock),
                         Sequence<true, DoPadGemmM>{});
 
+                const index_t AK0 = out_gemmk_gemmm_padded_grid_desc.GetLength(I0) / AK1;
+
                 const auto out_gemmak0_gemmm_gemmak1_grid_desc = transform_tensor_descriptor(
                     out_gemmk_gemmm_padded_grid_desc,
                     make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
@@ -568,9 +570,6 @@ struct TransformConvBwdDataToGemm_v1
             const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
             const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
 
-            const index_t BK0 =
-                math::integer_divide_ceil(ZDotSlice * YDotSlice * XDotSlice * K, BK1);
-
             // B weight tensor
             if constexpr(NDimSpatial == 2)
             {
@@ -617,9 +616,11 @@ struct TransformConvBwdDataToGemm_v1
                 const auto wei_gemmk_gemmn_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
                         wei_gemmk_gemmnraw_grid_desc,
-                        make_tuple(BK1, GemmNPerBlock),
+                        make_tuple(GemmKPerBlock, GemmNPerBlock),
                         Sequence<true, DoPadGemmN>{});
 
+                const index_t BK0 = wei_gemmk_gemmn_padded_grid_desc.GetLength(I0) / BK1;
+
                 const auto wei_gemmbk0_gemmn_gemmbk1_grid_desc = transform_tensor_descriptor(
                     wei_gemmk_gemmn_padded_grid_desc,
                     make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
@@ -690,17 +691,19 @@ struct TransformConvBwdDataToGemm_v1
                     make_tuple(Sequence<1, 2, 3, 0>{}, Sequence<4>{}),
                     make_tuple(Sequence<0>{}, Sequence<1>{}));
 
-                const auto wei_gemmk_gemm_padded_grid_desc =
+                const auto wei_gemmk_gemmn_padded_grid_desc =
                     ck::tensor_operation::device::PadTensorDescriptor(
                         wei_gemmk_gemmnraw_grid_desc,
-                        make_tuple(BK1, GemmNPerBlock),
+                        make_tuple(GemmKPerBlock, GemmNPerBlock),
                         Sequence<true, DoPadGemmN>{});
 
+                const index_t BK0 = wei_gemmk_gemmn_padded_grid_desc.GetLength(I0) / BK1;
+
                 const auto wei_gemmbk0_gemm_gemmbk1_grid_desc = transform_tensor_descriptor(
-                    wei_gemmk_gemm_padded_grid_desc,
-                    make_tuple(
-                        make_unmerge_transform(make_tuple(BK0, BK1)),
-                        make_pass_through_transform(wei_gemmk_gemm_padded_grid_desc.GetLength(I1))),
+                    wei_gemmk_gemmn_padded_grid_desc,
+                    make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                               make_pass_through_transform(
+                                   wei_gemmk_gemmn_padded_grid_desc.GetLength(I1))),
                     make_tuple(Sequence<0>{}, Sequence<1>{}),
                     make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
 

From 0c9a1d25b3951736bf1fdc480de5fa21208f8e56 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 5 Sep 2023 17:07:56 +0200
Subject: [PATCH 110/194] Add nhwgc dl generic instances for grouped conv fwd
 (#879)

---
 .../device_grouped_conv_fwd_dl_instance.hpp   |  9 +++
 .../gpu/grouped_convolution_forward.hpp       | 30 ++++++++++
 .../gpu/grouped_conv2d_fwd/CMakeLists.txt     |  5 ++
 ..._fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp | 57 +++++++++++++++++++
 ..._fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp | 57 +++++++++++++++++++
 5 files changed, 158 insertions(+)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp

diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
index ae3bec621c..86ff43e5cf 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
@@ -53,6 +53,11 @@ using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
            // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // generic instances
+        // TODO: Change to ScalarPerVector = 1 when inner_product<half_t, half_t, float> will be supported
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,     4,     2,  2,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 1, 1, 2>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 2>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 2>,
+
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
     >;
@@ -71,6 +76,10 @@ using device_grouped_conv2d_fwd_dl_f32_instances = std::tuple<
            // ########################################|     Spatial|   Type|    Type|        Type|    Type|    Type|         |          |    Layout|          | Elementwise|  Elementwise|   Elementwise|        Forward|    Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
+        // generic instances
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,      4,    2,  1,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 1>,      S<1, 1, 8, 1>,    S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 1, 1, 1>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F32,     F32,  DsDatatype,     F32,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough, PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
     >;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index 627a5ae2aa..2a0da10c65 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -145,6 +145,34 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -394,11 +422,13 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                               is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
index a36e1b47ca..708b37c42d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
@@ -1,4 +1,5 @@
 add_instance_library(device_grouped_conv2d_fwd_instance
+   #xdl
    # GNHWC, GKYXC, GNHWK
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
    device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -8,6 +9,10 @@ add_instance_library(device_grouped_conv2d_fwd_instance
    device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
    device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
    #dl
+   # GNHWC, GKYXC, GNHWK
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+   # NHWGC, GKYXC, NHWGK
+   device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..4a0f7b9b5b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f16_instances<NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..94e57bef46
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_dl_f32_instances<NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 0077eeb3be363ad83c301274b37f994b1163fa1f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 5 Sep 2023 17:11:40 +0200
Subject: [PATCH 111/194] Add image to column kernel (#867)

* Add image to column kernel

* Add instances, tests, profiler, example

* Add client example

* Several fixes of image to column

* Fix variable name in device_image_to_column_impl

* Several fixes of image to column profiler

* Fix num_btype calculation

* Make new mesaurements for correct bytes calculation
---
 .../20_image_to_column/CMakeLists.txt         |   2 +
 .../20_image_to_column/image_to_column.cpp    | 167 +++++++
 example/52_image_to_column/CMakeLists.txt     |  10 +
 example/52_image_to_column/common.hpp         |  95 ++++
 .../image_to_column_f32.cpp                   | 166 +++++++
 .../gpu/device/device_image_to_column.hpp     |  70 +++
 .../impl/device_image_to_column_impl.hpp      | 407 ++++++++++++++++++
 .../gpu/grid/gridwise_image_to_column.hpp     |  97 +++++
 .../cpu/reference_image_to_column.hpp         | 325 ++++++++++++++
 .../gpu/image_to_column.hpp                   | 135 ++++++
 .../device_image_to_column_instance.hpp       | 121 ++++++
 .../gpu/image_to_column/CMakeLists.txt        |   5 +
 ...evice_image_to_column_nhwc_1d_instance.cpp |  39 ++
 ...evice_image_to_column_nhwc_2d_instance.cpp |  39 ++
 ...evice_image_to_column_nhwc_3d_instance.cpp |  39 ++
 profiler/README.md                            |  38 ++
 .../profiler/profile_image_to_column_impl.hpp | 200 +++++++++
 profiler/src/CMakeLists.txt                   |   2 +
 profiler/src/profile_image_to_column.cpp      | 169 ++++++++
 test/CMakeLists.txt                           |   1 +
 test/image_to_column/CMakeLists.txt           |   4 +
 test/image_to_column/test_image_to_column.cpp | 121 ++++++
 .../test_image_to_column_interface.cpp        | 196 +++++++++
 23 files changed, 2448 insertions(+)
 create mode 100644 client_example/20_image_to_column/CMakeLists.txt
 create mode 100644 client_example/20_image_to_column/image_to_column.cpp
 create mode 100644 example/52_image_to_column/CMakeLists.txt
 create mode 100644 example/52_image_to_column/common.hpp
 create mode 100644 example/52_image_to_column/image_to_column_f32.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_image_to_column.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp
 create mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
 create mode 100644 profiler/include/profiler/profile_image_to_column_impl.hpp
 create mode 100644 profiler/src/profile_image_to_column.cpp
 create mode 100644 test/image_to_column/CMakeLists.txt
 create mode 100644 test/image_to_column/test_image_to_column.cpp
 create mode 100644 test/image_to_column/test_image_to_column_interface.cpp

diff --git a/client_example/20_image_to_column/CMakeLists.txt b/client_example/20_image_to_column/CMakeLists.txt
new file mode 100644
index 0000000000..80edcd0416
--- /dev/null
+++ b/client_example/20_image_to_column/CMakeLists.txt
@@ -0,0 +1,2 @@
+add_executable(client_image_to_column image_to_column.cpp)
+target_link_libraries(client_image_to_column PRIVATE composable_kernel::device_operations)
diff --git a/client_example/20_image_to_column/image_to_column.cpp b/client_example/20_image_to_column/image_to_column.cpp
new file mode 100644
index 0000000000..ace4c1a681
--- /dev/null
+++ b/client_example/20_image_to_column/image_to_column.cpp
@@ -0,0 +1,167 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout = ck::tensor_layout::convolution::GNHWC;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t N             = 32; // batch size
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 28; // input H
+static constexpr ck::index_t Wi            = 28; // input W
+static constexpr ck::index_t Ho            = 28; // output H
+static constexpr ck::index_t Wo            = 28; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+
+    std::array<ck::index_t, 2> in_spatial_lengths{Hi, Wi};
+    std::array<ck::index_t, 2> wei_spatial_lengths{Y, X};
+    std::array<ck::index_t, 2> out_spatial_lengths{Ho, Wo};
+
+    // We have NHWGC in memory space (G is dummy)
+    // However, CK's API only accept length and stride with order of GNCHW
+    // Hence, we need to adjust the order of stride
+    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 2> out_strides{Y * X * C, 1};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * Y * X * C);
+
+    using DeviceOp = ck::tensor_operation::device::
+        DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        wei_spatial_lengths,
+                                                        in_strides,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads);
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
+                                    sizeof(OutDataType) * N * Ho * Wo * Y * X * C;
+
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(avg_time < best_avg_time)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_gb_per_sec
+              << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        wei_spatial_lengths,
+                                                        in_strides,
+                                                        out_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+}
diff --git a/example/52_image_to_column/CMakeLists.txt b/example/52_image_to_column/CMakeLists.txt
new file mode 100644
index 0000000000..226e1fc5ae
--- /dev/null
+++ b/example/52_image_to_column/CMakeLists.txt
@@ -0,0 +1,10 @@
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+   add_custom_target(example_image_to_column)
+   add_example_executable(example_image_to_column_f32 image_to_column_f32.cpp)
+   add_dependencies(example_image_to_column example_image_to_column_f32)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/52_image_to_column/common.hpp b/example/52_image_to_column/common.hpp
new file mode 100644
index 0000000000..8510fa1e6d
--- /dev/null
+++ b/example/52_image_to_column/common.hpp
@@ -0,0 +1,95 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <initializer_list>
+#include <iostream>
+#include <numeric>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+static inline constexpr ck::index_t NDimSpatial = 2;
+
+using FP32 = float;
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = true;
+};
+
+#define DefaultConvParams                                                            \
+    ck::utils::conv::ConvParam                                                       \
+    {                                                                                \
+        NDimSpatial, 1, 32, 1, 1, {4, 4}, {64, 64}, {1, 1}, {1, 1}, {0, 0}, { 0, 0 } \
+    }
+
+inline void print_help_msg()
+{
+    std::cerr << "arg1: verification (0=no, 1=yes)\n"
+              << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
+              << "arg3: time kernel (0=no, 1=yes)\n"
+              << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+}
+
+inline bool parse_cmd_args(int argc,
+                           char* argv[],
+                           ExecutionConfig& config,
+                           ck::utils::conv::ConvParam& conv_params)
+{
+    constexpr int num_execution_config_args =
+        3; // arguments for do_verification, init_method, time_kernel
+    constexpr int num_conv_param_leading_args = 5; // arguments for num_dim_spatial_, G_, N_, K_, C_
+
+    constexpr int threshold_to_catch_partial_args = 1 + num_execution_config_args;
+    constexpr int threshold_to_catch_all_args =
+        threshold_to_catch_partial_args + num_conv_param_leading_args;
+
+    if(argc == 1)
+    {
+        // use default
+        config = ExecutionConfig{};
+    }
+    // catch only ExecutionConfig arguments
+    else if(argc == threshold_to_catch_partial_args)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+    }
+    // catch both ExecutionConfig & ConvParam arguments
+    else if(threshold_to_catch_all_args < argc && ((argc - threshold_to_catch_all_args) % 3 == 0))
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t num_dim_spatial = std::stoi(argv[4]);
+        conv_params                       = ck::utils::conv::parse_conv_param(
+            num_dim_spatial, threshold_to_catch_partial_args, argv);
+    }
+    else
+    {
+        print_help_msg();
+        return false;
+    }
+
+    return true;
+}
diff --git a/example/52_image_to_column/image_to_column_f32.cpp b/example/52_image_to_column/image_to_column_f32.cpp
new file mode 100644
index 0000000000..c8a7e5f221
--- /dev/null
+++ b/example/52_image_to_column/image_to_column_f32.cpp
@@ -0,0 +1,166 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+using InDataType  = FP32;
+using OutDataType = FP32;
+
+using InLayout = ck::tensor_layout::convolution::GNHWC;
+
+// clang-format off
+using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumnImpl
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+                              < NDimSpatial, InLayout, InDataType, OutDataType,   256,   128,   128, S<16, 16>,     1>;
+// clang-format on
+
+bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
+{
+
+    const auto N = conv_params.N_;
+    const auto C = conv_params.C_;
+
+    const ck::index_t NDoHoWo =
+        N * ck::accumulate_n<ck::index_t>(
+                conv_params.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+    const ck::index_t CZYX =
+        C * ck::accumulate_n<ck::index_t>(
+                conv_params.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+    const auto in_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_params);
+    const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+
+    std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
+    std::array<ck::index_t, 2> output_m_k_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+
+    copy(conv_params.input_spatial_lengths_, input_spatial_lengths);
+    copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
+    copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
+    copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
+    copy(out_desc.GetStrides(), output_m_k_strides);
+    copy(conv_params.conv_filter_strides_, conv_filter_strides);
+    copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_params.input_left_pads_, input_left_pads);
+    copy(conv_params.input_right_pads_, input_right_pads);
+
+    Tensor<InDataType> in(in_desc);
+    Tensor<OutDataType> out_device(out_desc);
+    Tensor<OutDataType> out_host(out_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "out: " << out_device.mDesc << std::endl;
+
+    switch(config.init_method)
+    {
+    case 0: break;
+    case 1: in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-5, 5}); break;
+    default: in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in.mData.data());
+
+    // reset input to zero
+    out_device_buf.SetZero();
+
+    static_assert(std::is_default_constructible_v<DeviceImgToColInstance>);
+
+    // do conv
+    auto img2col  = DeviceImgToColInstance{};
+    auto invoker  = img2col.MakeInvoker();
+    auto argument = img2col.MakeArgument(in_device_buf.GetDeviceBuffer(),
+                                         out_device_buf.GetDeviceBuffer(),
+                                         N,
+                                         C,
+                                         input_spatial_lengths,
+                                         filter_spatial_lengths,
+                                         output_spatial_lengths,
+                                         input_g_n_c_wis_strides,
+                                         output_m_k_strides,
+                                         conv_filter_strides,
+                                         conv_filter_dilations,
+                                         input_left_pads,
+                                         input_right_pads);
+
+    if(!img2col.IsSupportedArgument(argument))
+    {
+        std::cerr << "wrong! device_img2col with the specified compilation parameters does "
+                     "not support this img2col problem"
+                  << std::endl;
+
+        return false;
+    }
+
+    float ave_time        = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+    std::size_t num_btype = NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
+    float gb_per_sec      = num_btype / 1.E6 / ave_time;
+    std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
+
+    if(config.do_verification)
+    {
+        auto ref_image_to_column = ck::tensor_operation::host::
+            ReferenceImageToColumn<NDimSpatial, InLayout, InDataType, OutDataType>();
+
+        auto ref_invoker = ref_image_to_column.MakeInvoker();
+
+        auto ref_argument = ref_image_to_column.MakeArgument(in,
+                                                             out_host,
+                                                             conv_params.filter_spatial_lengths_,
+                                                             conv_params.conv_filter_strides_,
+                                                             conv_params.conv_filter_dilations_,
+                                                             conv_params.input_left_pads_,
+                                                             conv_params.input_right_pads_);
+
+        if(!ref_image_to_column.IsSupportedArgument(&ref_argument))
+        {
+            std::cerr << "wrong! ref_img2col with the specified compilation parameters does "
+                         "not support this img2col problem"
+                      << std::endl;
+            return false;
+        }
+
+        ref_invoker.Run(ref_argument);
+
+        out_device_buf.FromDevice(out_device.mData.data());
+
+        return ck::utils::check_err(out_device.mData, out_host.mData);
+    }
+
+    return true;
+}
+
+int RunImageToColumnExample(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_params = DefaultConvParams;
+
+    if(!parse_cmd_args(argc, argv, config, conv_params))
+    {
+        return EXIT_FAILURE;
+    }
+
+    if(conv_params.num_dim_spatial_ != NDimSpatial)
+    {
+        std::cerr << "unsupported # of spatial dimensions" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    return !RunImageToColumn(config, conv_params);
+}
+
+int main(int argc, char* argv[]) { return RunImageToColumnExample(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/device/device_image_to_column.hpp b/include/ck/tensor_operation/gpu/device/device_image_to_column.hpp
new file mode 100644
index 0000000000..631d5189dd
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_image_to_column.hpp
@@ -0,0 +1,70 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+/**
+ * \brief Image to column.
+ *
+ * This Device operator converts image ([G, N, Di, Hi, Wi, C]) to the gemm
+ * problem([N * Do * Ho * Wo, Z *  Y * X * C]). G must be equal to 1.
+ *
+ * \tparam NDimSpatial Number of spatial dimensions.
+ * \tparam InputLayout Input Layout.
+ * \tparam InputDataType Input Data Type.
+ * \tparam OutputDataType Output Data Type.
+ */
+template <index_t NDimSpatial,
+          typename InputLayout,
+          typename InputDataType,
+          typename OutputDataType>
+struct DeviceImageToColumn : public BaseOperator
+{
+
+    /**
+     * \brief Make argument pointer for image to column.
+     *
+     * \param p_in A pointer to the device memory of the input image.
+     * \param p_out A pointer to the device memory of the output.
+     * \param N Convolution batch size.
+     * \param C Convolution number of channels.
+     * \param input_spatial_lengths Input spatial lengths.
+     * \param filter_spatial_lengths Filter spatial lengths.
+     * \param output_spatial_lengths Output spatial lengths.
+     * \param input_g_n_c_wis_strides Input strides in order [G, N, C, D, H, W].
+     * \param output_m_k_strides Output strides.
+     * \param conv_filter_strides Convolution filter strides.
+     * \param conv_filter_dilations Convolution filter dilations.
+     * \param input_left_pads Convolution left pads.
+     * \param input_right_pads Convolution right pads.
+     * \return Pointer to the argument.
+     */
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in,
+                        void* p_out,
+                        const ck::index_t N,
+                        const ck::index_t C,
+                        const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
+                        const std::array<index_t, 2>& output_m_k_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<index_t, NDimSpatial>& input_left_pads,
+                        const std::array<index_t, NDimSpatial>& input_right_pads) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
new file mode 100644
index 0000000000..89fcbca1ac
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
@@ -0,0 +1,407 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/host_utility/io.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename InputGridDesc,
+          typename InputDataType,
+          typename OutputGridDesc,
+          typename OutputDataType,
+          typename Block2ETileMap,
+          typename GridwiseImageToColumnKernel>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_image_to_column(const InputGridDesc in_grid_desc,
+                               const InputDataType* __restrict__ p_in_global,
+                               const OutputGridDesc out_grid_desc,
+                               OutputDataType* __restrict__ p_out_global,
+                               const Block2ETileMap block_2_tile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
+    GridwiseImageToColumnKernel::Run(
+        in_grid_desc, p_in_global, out_grid_desc, p_out_global, block_2_tile_map);
+#else
+    ignore = in_grid_desc;
+    ignore = p_in_global;
+    ignore = out_grid_desc;
+    ignore = p_out_global;
+    ignore = block_2_tile_map;
+#endif
+}
+
+// Image to column for input layout NDHWC:
+//   input : input image [N, Di, Hi, Wi, C],
+//   output : output image [N * Do * Ho * Wo, Z * Y * X * C]
+template <index_t NDimSpatial,
+          typename InputLayout,
+          typename InputDataType,
+          typename OutputDataType,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t KPerBlock,
+          typename ThreadClusterLengths,
+          index_t ScalarPerVector>
+struct DeviceImageToColumnImpl
+    : public DeviceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>
+{
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    static constexpr auto conv_to_gemm_transformer =
+        TransformConvFwdToGemm<NDimSpatial, ConvolutionForwardSpecialization::Default>{};
+
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpecialization::MKPadding, index_t, index_t, index_t>{
+            MPerBlock, 0 /* NPerBlock*/, KPerBlock};
+
+    // Use MakeADescriptor_M_K from grouped convolution forward
+    static auto
+    MakeInputDescriptor_M_K(const ck::index_t N,
+                            const ck::index_t C,
+                            const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                            const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                            const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                            const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
+                            const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                            const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                            const std::array<index_t, NDimSpatial>& input_left_pads,
+                            const std::array<index_t, NDimSpatial>& input_right_pads)
+    {
+        std::array<index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{1};
+        std::array<index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{1};
+        std::array<index_t, NDimSpatial + 3> c_g_n_k_wos_lengths{1};
+
+        auto copy = [](const auto& x, auto& y, index_t dst_offset) {
+            std::copy(x.begin(), x.end(), y.begin() + dst_offset);
+        };
+
+        constexpr index_t spatial_offset = 3;
+
+        copy(input_spatial_lengths, a_g_n_c_wis_lengths, spatial_offset);
+        copy(filter_spatial_lengths, b_g_k_c_xs_lengths, spatial_offset);
+        copy(output_spatial_lengths, c_g_n_k_wos_lengths, spatial_offset);
+
+        // fill only significant values (C and N)
+        a_g_n_c_wis_lengths[I1] = N;
+        a_g_n_c_wis_lengths[I2] = C;
+        b_g_k_c_xs_lengths[I2]  = C;
+        c_g_n_k_wos_lengths[I1] = N;
+
+        const auto in_gemmmraw_gemmkraw_desc =
+            conv_to_gemm_transformer.template MakeADescriptor_M_K<InputLayout>(
+                a_g_n_c_wis_lengths,
+                input_g_n_c_wis_strides,
+                b_g_k_c_xs_lengths,
+                {}, // not needed for A Descriptor
+                c_g_n_k_wos_lengths,
+                {}, // not needed for A Descriptor
+                conv_filter_strides,
+                conv_filter_dilations,
+                input_left_pads,
+                input_right_pads);
+
+        const auto in_gemmm_gemmk_desc =
+            matrix_padder.PadADescriptor_M_K(in_gemmmraw_gemmkraw_desc);
+        return in_gemmm_gemmk_desc;
+    }
+
+    static auto
+    MakeOutDescriptor_M_K(const ck::index_t N,
+                          const ck::index_t C,
+                          const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                          const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                          const std::array<index_t, 2>& output_m_k_strides)
+    {
+        const index_t NDoHoWo =
+            N * ck::accumulate_n<index_t>(
+                    output_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
+        const index_t CZYX =
+            C * ck::accumulate_n<index_t>(
+                    filter_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
+        const auto desc_mraw_kraw = make_naive_tensor_descriptor(
+            make_tuple(NDoHoWo, CZYX), make_tuple(output_m_k_strides[I0], output_m_k_strides[I1]));
+
+        const auto desc_m_k = matrix_padder.PadADescriptor_M_K(desc_mraw_kraw);
+        return desc_m_k;
+    }
+
+    using InputGridDesc =
+        remove_cvref_t<decltype(MakeInputDescriptor_M_K(1, 1, {}, {}, {}, {}, {}, {}, {}, {}))>;
+    using OutputGridDesc = remove_cvref_t<decltype(MakeOutDescriptor_M_K(1, 1, {}, {}, {}))>;
+
+    using Block2ETileMap = remove_cvref_t<decltype(
+        BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(OutputGridDesc{}))>;
+
+    using GridwiseImageToColumnKernel = GridwiseImageToColumn<InputGridDesc,
+                                                              InputDataType,
+                                                              OutputGridDesc,
+                                                              OutputDataType,
+                                                              BlockSize,
+                                                              MPerBlock,
+                                                              KPerBlock,
+                                                              ThreadClusterLengths,
+                                                              ScalarPerVector,
+                                                              Block2ETileMap>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_in, // input image
+                 void* p_out,      // output image
+                 const ck::index_t N,
+                 const ck::index_t C,
+                 const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                 const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                 const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
+                 const std::array<index_t, 2>& output_m_k_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads)
+            : C_(C),
+              X_(filter_spatial_lengths[NDimSpatial - I1]),
+              p_in_{static_cast<const InputDataType*>(p_in)},
+              p_out_{static_cast<OutputDataType*>(p_out)},
+              input_g_n_c_wis_strides_{input_g_n_c_wis_strides},
+              conv_filter_strides_{conv_filter_strides},
+              conv_filter_dilations_{conv_filter_dilations},
+              input_left_pads_{input_left_pads},
+              input_right_pads_{input_right_pads}
+        {
+
+            in_grid_desc_m_k_ = MakeInputDescriptor_M_K(N,
+                                                        C,
+                                                        input_spatial_lengths,
+                                                        filter_spatial_lengths,
+                                                        output_spatial_lengths,
+                                                        input_g_n_c_wis_strides,
+
+                                                        conv_filter_strides,
+                                                        conv_filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads);
+
+            out_grid_desc_m_k_ = MakeOutDescriptor_M_K(
+                N, C, filter_spatial_lengths, output_spatial_lengths, output_m_k_strides);
+        }
+
+        void Print() const
+        {
+            std::cout << in_grid_desc_m_k_ << std::endl;
+            std::cout << out_grid_desc_m_k_ << std::endl;
+        }
+
+        const ck::index_t C_;
+        const ck::index_t X_;
+
+        const InputDataType* p_in_;
+        OutputDataType* p_out_;
+
+        const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides_;
+        const std::array<index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<index_t, NDimSpatial>& conv_filter_dilations_;
+        const std::array<index_t, NDimSpatial>& input_left_pads_;
+        const std::array<index_t, NDimSpatial>& input_right_pads_;
+
+        InputGridDesc in_grid_desc_m_k_;
+        OutputGridDesc out_grid_desc_m_k_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                arg.Print();
+            }
+
+            const auto block_2_tile_map =
+                BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
+                    arg.out_grid_desc_m_k_);
+            const index_t grid_size = block_2_tile_map.CalculateGridSize(arg.out_grid_desc_m_k_);
+            const auto kernel       = kernel_image_to_column<InputGridDesc,
+                                                       InputDataType,
+                                                       OutputGridDesc,
+                                                       OutputDataType,
+                                                       Block2ETileMap,
+                                                       GridwiseImageToColumnKernel>;
+
+            float elapsed_time = launch_and_time_kernel(stream_config,
+                                                        kernel,
+                                                        dim3(grid_size),
+                                                        dim3(BlockSize),
+                                                        0,
+                                                        arg.in_grid_desc_m_k_,
+                                                        arg.p_in_,
+                                                        arg.out_grid_desc_m_k_,
+                                                        arg.p_out_,
+                                                        block_2_tile_map);
+            return elapsed_time;
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    bool IsSupportedArgument(const Argument& arg)
+    {
+        using namespace tensor_layout::convolution;
+        if(!(std::is_same_v<InputLayout, GNWC> || std::is_same_v<InputLayout, GNHWC> ||
+             std::is_same_v<InputLayout, GNDHWC>))
+        {
+            return false;
+        }
+        if(!(NDimSpatial >= 1 && NDimSpatial <= 3))
+        {
+            return false;
+        }
+
+        const auto w_pad_left  = arg.input_left_pads_[NDimSpatial - I1];
+        const auto w_pad_right = arg.input_right_pads_[NDimSpatial - I1];
+        const auto dilation_x  = arg.conv_filter_dilations_[NDimSpatial - I1];
+        const auto stride_x    = arg.conv_filter_strides_[NDimSpatial - I1];
+        bool is_w_packed       = arg.input_g_n_c_wis_strides_[NDimSpatial + I2] == arg.C_;
+        bool is_c_packed       = arg.input_g_n_c_wis_strides_[I2] == 1;
+
+        // check vector acces with c not packed
+        if(!is_c_packed && ScalarPerVector != 1)
+            return false;
+        // check vector access of filter window row (only C if C is not packed)
+        if(!is_w_packed && arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of filter window row (X * C)
+        if(arg.X_ * arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of pads (w_pad_left/w_pad_right * C)
+        if(w_pad_left * arg.C_ % ScalarPerVector != 0 ||
+           w_pad_right * arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of with stride and pad
+        if((w_pad_left != 0 || w_pad_right != 0) && stride_x > 1 && arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of with dilation
+        if(dilation_x > 1 && arg.C_ % ScalarPerVector != 0)
+            return false;
+
+        return GridwiseImageToColumnKernel::CheckValidity(arg.in_grid_desc_m_k_,
+                                                          arg.out_grid_desc_m_k_);
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_in, // input image
+                             void* p_out,      // output image
+                             const ck::index_t N,
+                             const ck::index_t C,
+                             const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                             const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                             const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                             const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
+                             const std::array<index_t, 2>& output_m_k_strides,
+                             const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                             const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                             const std::array<index_t, NDimSpatial>& input_left_pads,
+                             const std::array<index_t, NDimSpatial>& input_right_pads)
+    {
+        return Argument{static_cast<const InputDataType*>(p_in),
+                        static_cast<OutputDataType*>(p_out),
+                        N,
+                        C,
+                        input_spatial_lengths,
+                        filter_spatial_lengths,
+                        output_spatial_lengths,
+                        input_g_n_c_wis_strides,
+                        output_m_k_strides,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in, // input image
+                        void* p_out,      // output image
+                        const ck::index_t N,
+                        const ck::index_t C,
+                        const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
+                        const std::array<index_t, 2>& output_m_k_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<index_t, NDimSpatial>& input_left_pads,
+                        const std::array<index_t, NDimSpatial>& input_right_pads) override
+    {
+        return std::make_unique<Argument>(static_cast<const InputDataType*>(p_in),
+                                          static_cast<OutputDataType*>(p_out),
+                                          N,
+                                          C,
+                                          input_spatial_lengths,
+                                          filter_spatial_lengths,
+                                          output_spatial_lengths,
+                                          input_g_n_c_wis_strides,
+                                          output_m_k_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceImageToColumn"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << KPerBlock << ", "
+            << ScalarPerVector
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp
new file mode 100644
index 0000000000..93625a324e
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp
@@ -0,0 +1,97 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename InputGridDesc,
+          typename InputDataType,
+          typename OutputGridDesc,
+          typename OutputDataType,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t KPerBlock,
+          typename ThreadClusterLengths,
+          index_t ScalarPerVector,
+          typename Block2ETileMap>
+struct GridwiseImageToColumn
+{
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    __device__ static void Run(const InputGridDesc& in_grid_desc,
+                               const InputDataType* __restrict__ p_in_global,
+                               const OutputGridDesc& out_grid_desc,
+                               OutputDataType* __restrict__ p_out_global,
+                               const Block2ETileMap& block_2_tile_map)
+    {
+        const auto block_work_idx =
+            block_2_tile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+
+        const index_t k_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * KPerBlock);
+
+        // Global Memory
+        const auto in_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_global, in_grid_desc.GetElementSpaceSize());
+        auto out_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_out_global, out_grid_desc.GetElementSpaceSize());
+
+        auto copy_global_to_global = ThreadGroupTensorSliceTransfer_v7<
+            ThisThreadBlock,
+            Tuple<InputDataType>,
+            Tuple<OutputDataType>,
+            decltype(tie(in_grid_desc)),
+            decltype(tie(out_grid_desc)),
+            tensor_operation::element_wise::PassThrough,
+            Sequence<static_cast<index_t>(InMemoryDataOperationEnum::Set)>,
+            Sequence<MPerBlock, KPerBlock>,
+            ThreadClusterLengths,
+            Sequence<0, 1>,
+            Sequence<0, 1>,
+            I1,
+            ScalarPerVector,
+            Sequence<true>,
+            Sequence<true>>{
+            in_grid_desc,
+            make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
+            out_grid_desc,
+            make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
+            tensor_operation::element_wise::PassThrough{}};
+
+        copy_global_to_global.Run(
+            tie(in_grid_desc), tie(in_global_buf), tie(out_grid_desc), tie(out_global_buf));
+    }
+
+    __host__ static constexpr bool CheckValidity(const InputGridDesc& in_grid_desc,
+                                                 const OutputGridDesc& out_grid_desc)
+    {
+        if(in_grid_desc.GetLength(I0) % MPerBlock != 0 ||
+           in_grid_desc.GetLength(I1) % KPerBlock != 0)
+            return false;
+        if(out_grid_desc.GetLength(I0) % MPerBlock != 0 ||
+           out_grid_desc.GetLength(I1) % KPerBlock != 0)
+            return false;
+        return true;
+    }
+};
+
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
new file mode 100644
index 0000000000..3f50ab88b3
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
@@ -0,0 +1,325 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <type_traits>
+#include <sstream>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+/**
+ * \brief Reference implementation for image to column.
+ *
+ * Tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
+ * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
+ *
+ * \tparam NDimSpatial Number of spatial dimensions.
+ * \tparam InputLayout Input Layout.
+ * \tparam InDataType Input Data Type.
+ * \tparam OutDataType Output Data Type.
+ */
+template <ck::index_t NDimSpatial,
+          typename InputLayout,
+          typename InDataType,
+          typename OutDataType,
+          typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
+struct ReferenceImageToColumn : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        public:
+        Argument(const Tensor<InDataType>& input,
+                 Tensor<OutDataType>& output,
+                 std::vector<ck::index_t> filter_spatial_lengths,
+                 std::vector<ck::index_t> conv_filter_strides,
+                 std::vector<ck::index_t> conv_filter_dilations,
+                 std::vector<ck::index_t> input_left_pads,
+                 std::vector<ck::index_t> input_right_pads)
+            : input_{input},
+              output_{output},
+              conv_strides_{conv_filter_strides},
+              conv_dilations_{conv_filter_dilations},
+              in_left_pads_{input_left_pads},
+              in_right_pads_{input_right_pads},
+              filter_spatial_lengths_{filter_spatial_lengths}
+        {
+            initOutputSpatialLengths();
+        }
+
+        const Tensor<InDataType>& input_;
+        Tensor<OutDataType>& output_;
+
+        std::vector<index_t> conv_strides_;
+        std::vector<index_t> conv_dilations_;
+        std::vector<index_t> in_left_pads_;
+        std::vector<index_t> in_right_pads_;
+
+        std::vector<index_t> filter_spatial_lengths_;
+        std::vector<index_t> output_spatial_lengths_;
+
+        private:
+        void initOutputSpatialLengths()
+        {
+            constexpr auto input_offset_to_spatial = 3;
+
+            for(ck::index_t i = 0; i < NDimSpatial; ++i)
+            {
+                // XEff = (X - 1) * conv_dilation_w + 1;
+                // Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
+                const ck::index_t x_eff = (filter_spatial_lengths_[i] - 1) * conv_dilations_[i] + 1;
+
+                output_spatial_lengths_.push_back(
+                    (input_.GetLengths()[i + input_offset_to_spatial] + in_left_pads_[i] +
+                     in_right_pads_[i] - x_eff) /
+                        conv_strides_[i] +
+                    1);
+            }
+        }
+    };
+
+    struct Invoker : public device::BaseInvoker
+    {
+        using Argument = ReferenceImageToColumn::Argument;
+
+        float Run(const Argument& arg)
+        {
+            if(!(arg.input_.GetNumOfDimension() == NDimSpatial + 3 &&
+                 arg.output_.GetNumOfDimension() == 2))
+            {
+                throw std::runtime_error("wrong! inconsistent dimension");
+            }
+
+            const index_t N = arg.input_.GetLengths()[1];
+            const index_t C = arg.input_.GetLengths()[2];
+
+            if constexpr(NDimSpatial == 1)
+            {
+                const index_t Wo = arg.output_spatial_lengths_[0];
+                auto func        = [&](auto n, auto wo) {
+                    index_t row    = n * Wo + wo;
+                    index_t column = 0;
+
+                    for(index_t x = 0; x < arg.filter_spatial_lengths_[0]; ++x)
+                    {
+                        auto wi = static_cast<ck::long_index_t>(wo * arg.conv_strides_[0]) +
+                                  static_cast<ck::long_index_t>(x * arg.conv_dilations_[0]) -
+                                  static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+
+                        for(index_t c = 0; c < C; ++c)
+                        {
+                            if(wi >= 0 &&
+                               ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[3])
+                            {
+                                InDataType v_in          = arg.input_(0, n, c, wi);
+                                arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                            }
+                            column++;
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N, Wo)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+            else if constexpr(NDimSpatial == 2)
+            {
+                const index_t Ho = arg.output_spatial_lengths_[0];
+                const index_t Wo = arg.output_spatial_lengths_[1];
+
+                auto func = [&](auto n, auto ho, auto wo) {
+                    index_t row    = n * Ho * Wo + ho * Wo + wo;
+                    index_t column = 0;
+
+                    for(index_t y = 0; y < arg.filter_spatial_lengths_[0]; ++y)
+                    {
+                        auto hi = static_cast<ck::long_index_t>(ho * arg.conv_strides_[0]) +
+                                  static_cast<ck::long_index_t>(y * arg.conv_dilations_[0]) -
+                                  static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+
+                        for(index_t x = 0; x < arg.filter_spatial_lengths_[1]; ++x)
+                        {
+                            auto wi = static_cast<ck::long_index_t>(wo * arg.conv_strides_[1]) +
+                                      static_cast<ck::long_index_t>(x * arg.conv_dilations_[1]) -
+                                      static_cast<ck::long_index_t>(arg.in_left_pads_[1]);
+
+                            for(index_t c = 0; c < C; ++c)
+                            {
+
+                                if(hi >= 0 &&
+                                   ck::type_convert<std::size_t>(hi) < arg.input_.GetLengths()[3] &&
+                                   wi >= 0 &&
+                                   ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[4])
+                                {
+                                    InDataType v_in          = arg.input_(0, n, c, hi, wi);
+                                    arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                                }
+                                column++;
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N, Ho, Wo)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+            else if constexpr(NDimSpatial == 3)
+            {
+                const index_t Do = arg.output_spatial_lengths_[0];
+                const index_t Ho = arg.output_spatial_lengths_[1];
+                const index_t Wo = arg.output_spatial_lengths_[2];
+
+                auto func = [&](auto n, auto d_o, auto ho, auto wo) {
+                    index_t row    = n * Do * Ho * Wo + d_o * Ho * Wo + ho * Wo + wo;
+                    index_t column = 0;
+
+                    for(index_t z = 0; z < arg.filter_spatial_lengths_[0]; ++z)
+                    {
+                        auto di = static_cast<ck::long_index_t>(d_o * arg.conv_strides_[0]) +
+                                  static_cast<ck::long_index_t>(z * arg.conv_dilations_[0]) -
+                                  static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+                        for(index_t y = 0; y < arg.filter_spatial_lengths_[1]; ++y)
+                        {
+                            auto hi = static_cast<ck::long_index_t>(ho * arg.conv_strides_[1]) +
+                                      static_cast<ck::long_index_t>(y * arg.conv_dilations_[1]) -
+                                      static_cast<ck::long_index_t>(arg.in_left_pads_[1]);
+                            for(index_t x = 0; x < arg.filter_spatial_lengths_[2]; ++x)
+                            {
+                                auto wi =
+                                    static_cast<ck::long_index_t>(wo * arg.conv_strides_[2]) +
+                                    static_cast<ck::long_index_t>(x * arg.conv_dilations_[2]) -
+                                    static_cast<ck::long_index_t>(arg.in_left_pads_[2]);
+                                for(index_t c = 0; c < C; ++c)
+                                {
+                                    if(di >= 0 &&
+                                       ck::type_convert<std::size_t>(di) <
+                                           arg.input_.GetLengths()[3] &&
+                                       hi >= 0 &&
+                                       ck::type_convert<std::size_t>(hi) <
+                                           arg.input_.GetLengths()[4] &&
+                                       wi >= 0 &&
+                                       ck::type_convert<std::size_t>(wi) <
+                                           arg.input_.GetLengths()[5])
+                                    {
+                                        InDataType v_in = arg.input_(0, n, c, di, hi, wi);
+                                        arg.output_(row, column) =
+                                            ck::type_convert<OutDataType>(v_in);
+                                    }
+                                    column++;
+                                }
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N, Do, Ho, Wo)(
+                    std::thread::hardware_concurrency());
+
+                return 0;
+            }
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /*stream_config*/ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        using namespace tensor_layout::convolution;
+
+        if constexpr(!(std::is_same_v<InputLayout, GNWC> || std::is_same_v<InputLayout, GNHWC> ||
+                       std::is_same_v<InputLayout, GNDHWC>))
+        {
+            return false;
+        }
+        if constexpr(!(NDimSpatial >= 1 && NDimSpatial <= 3))
+        {
+            return false;
+        }
+        return true;
+    }
+
+    bool IsSupportedArgument(const Argument& arg)
+    {
+        const ck::index_t G = arg.input_.GetLengths()[0];
+        const ck::index_t N = arg.input_.GetLengths()[1];
+        const ck::index_t C = arg.input_.GetLengths()[2];
+
+        const index_t NDoHoWo =
+            N * ck::accumulate_n<index_t>(
+                    arg.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+        const index_t CZYX =
+            C * ck::accumulate_n<index_t>(
+                    arg.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+        if(!(arg.output_.GetLengths()[0] == static_cast<std::size_t>(NDoHoWo) &&
+             arg.output_.GetLengths()[1] == static_cast<std::size_t>(CZYX)))
+        {
+            return false;
+        }
+
+        if(G != 1)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    bool IsSupportedArgument(const device::BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const Tensor<InDataType>& input,
+                             Tensor<OutDataType>& output,
+                             std::vector<ck::index_t> filter_spatial_lengths,
+                             std::vector<ck::index_t> conv_filter_strides,
+                             std::vector<ck::index_t> conv_filter_dilations,
+                             std::vector<ck::index_t> input_left_pads,
+                             std::vector<ck::index_t> input_right_pads)
+    {
+        return Argument{input,
+                        output,
+                        filter_spatial_lengths,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceImageToColumn"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp b/library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp
new file mode 100644
index 0000000000..6c4526ba4e
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp
@@ -0,0 +1,135 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// nhwc, 1d
+void add_device_image_to_column_nhwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances);
+
+void add_device_image_to_column_nhwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances);
+
+void add_device_image_to_column_nhwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances);
+
+void add_device_image_to_column_nhwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances);
+// nhwc, 2d
+void add_device_image_to_column_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances);
+
+void add_device_image_to_column_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances);
+
+void add_device_image_to_column_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances);
+
+void add_device_image_to_column_nhwc_2d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances);
+// nhwc, 3d
+void add_device_image_to_column_nhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances);
+
+void add_device_image_to_column_nhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances);
+
+void add_device_image_to_column_nhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances);
+
+void add_device_image_to_column_nhwc_3d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances);
+
+template <ck::index_t NumDimSpatial, typename InLayout, typename InDataType, typename OutDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::
+        DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>>
+{
+    using DeviceOp = DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+            {
+                add_device_image_to_column_nhwc_1d_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+            {
+                add_device_image_to_column_nhwc_1d_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_image_to_column_nhwc_1d_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+            {
+                add_device_image_to_column_nhwc_1d_i8_instances(op_ptrs);
+            }
+        }
+        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+            {
+                add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+            {
+                add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+            {
+                add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
+            }
+        }
+        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC>)
+        {
+            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+            {
+                add_device_image_to_column_nhwc_3d_f32_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+            {
+                add_device_image_to_column_nhwc_3d_f16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                              is_same_v<OutDataType, ck::bhalf_t>)
+            {
+                add_device_image_to_column_nhwc_3d_bf16_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+            {
+                add_device_image_to_column_nhwc_3d_i8_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp
new file mode 100644
index 0000000000..a2603218b2
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp
@@ -0,0 +1,121 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_bf16_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_f16_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_f32_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     4>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_image_to_column_i8_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,     8,     8,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    16,    16,   S<8, 8>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    16,    16,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,    64,  S<8, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    16,    16, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     1>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     4>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     8>,
+        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   256,   256, S<16, 16>,     16>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt
new file mode 100644
index 0000000000..de10369374
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_instance_library(device_image_to_column_instance
+   device_image_to_column_nhwc_1d_instance.cpp
+   device_image_to_column_nhwc_2d_instance.cpp
+   device_image_to_column_nhwc_3d_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
new file mode 100644
index 0000000000..c8463623c3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_image_to_column_nhwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_bf16_instances<1, GNWC>{});
+}
+
+void add_device_image_to_column_nhwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_f16_instances<1, GNWC>{});
+}
+
+void add_device_image_to_column_nhwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_f32_instances<1, GNWC>{});
+}
+
+void add_device_image_to_column_nhwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_i8_instances<1, GNWC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
new file mode 100644
index 0000000000..652c7fac2a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_image_to_column_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_bf16_instances<2, GNHWC>{});
+}
+
+void add_device_image_to_column_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_f16_instances<2, GNHWC>{});
+}
+
+void add_device_image_to_column_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_f32_instances<2, GNHWC>{});
+}
+
+void add_device_image_to_column_nhwc_2d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_i8_instances<2, GNHWC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
new file mode 100644
index 0000000000..07774504d7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_image_to_column_nhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_bf16_instances<3, GNDHWC>{});
+}
+
+void add_device_image_to_column_nhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_f16_instances<3, GNDHWC>{});
+}
+
+void add_device_image_to_column_nhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_f32_instances<3, GNDHWC>{});
+}
+
+void add_device_image_to_column_nhwc_3d_i8_instances(
+    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances)
+{
+    add_device_operation_instances(instances, device_image_to_column_i8_instances<3, GNDHWC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/README.md b/profiler/README.md
index e8ac3a4d2b..d03bfa7fc4 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -184,3 +184,41 @@ tflops: 95.337
 GB/s: 69.2301
 ```
 Note: This kernel use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
+
+## Profile image to column kernels
+```bash
+# arg1: tensor operation (" OP_NAME ": " OP_DESC ")
+# arg2: data type (0: Input fp32, Weight fp32, Output fp32
+#                  1: Input fp16, Weight fp16, Output fp16
+#                  2: Input bf16, Weight bf16, Output bf16
+#                  3: Input int8, Weight int8, Output int8)
+# arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])
+# arg4: verification (0: no, 1: yes)
+# arg5: initialization (0: no init, 1: integer value, 2: decimal value)
+# arg6: print tensor value (0: no; 1: yes)
+# arg7: time kernel (0: no, 1: yes)
+# Following arguments (depending on number of spatial dims):
+#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  G, N, K, C, 
+#  <filter spatial dimensions>, (ie Y, X for 2D)
+#  <input image spatial dimensions>, (ie Hi, Wi for 2D)
+#  <strides>, (ie Sy, Sx for 2D)
+#  <dilations>, (ie Dy, Dx for 2D)
+#  <left padding>, (ie LeftPy, LeftPx for 2D)
+#  <right padding>, (ie RightPy, RightPx for 2D)
+
+ ################             op   datatype  layout  verify  init  log  time  Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx
+./bin/ckProfiler image_to_column          0       0       1     1    0     1      2  1 256   1 512  3  3   28  28   1   1   1   1        0       0       0        0
+
+ ```
+
+Result (MI210, FP32, NHWC)
+```
+input: dim 5, lengths {1, 256, 512, 28, 28}, strides {102760448, 401408, 1, 14336, 512}
+output: dim 2, lengths {173056, 4608}, strides {4608, 1}
+....
+Best configuration parameters:
+name: DeviceImageToColumn<128, 32, 64, 4>
+avg_time: 3.12326
+GB/s: 2042.59
+```
diff --git a/profiler/include/profiler/profile_image_to_column_impl.hpp b/profiler/include/profiler/profile_image_to_column_impl.hpp
new file mode 100644
index 0000000000..cc929e9220
--- /dev/null
+++ b/profiler/include/profiler/profile_image_to_column_impl.hpp
@@ -0,0 +1,200 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iomanip>
+#include <iostream>
+#include <typeinfo>
+#include <limits>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
+#include "ck/library/tensor_operation_instance/gpu/image_to_column.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp"
+
+namespace ck {
+namespace profiler {
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+template <index_t NDimSpatial,
+          typename InputLayout,
+          typename InputDataType,
+          typename OutputDataType>
+bool profile_image_to_column_impl(int do_verification,
+                                  int init_method,
+                                  bool do_log,
+                                  bool time_kernel,
+                                  const ck::utils::conv::ConvParam& conv_param)
+{
+    const ck::index_t NDoHoWo =
+        conv_param.N_ *
+        ck::accumulate_n<ck::index_t>(
+            conv_param.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+    const ck::index_t CZYX =
+        conv_param.C_ *
+        ck::accumulate_n<ck::index_t>(
+            conv_param.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+    const auto in_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InputLayout>(
+            conv_param);
+    const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+
+    std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
+    std::array<ck::index_t, 2> output_m_k_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+
+    copy(conv_param.input_spatial_lengths_, input_spatial_lengths);
+    copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
+    copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
+    copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
+    copy(out_desc.GetStrides(), output_m_k_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    Tensor<InputDataType> input(in_desc);
+    Tensor<OutputDataType> host_output(out_desc);
+    Tensor<OutputDataType> device_output(out_desc);
+
+    std::cout << "input: " << input.mDesc << std::endl;
+    std::cout << "output: " << host_output.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1: input.GenerateTensorValue(GeneratorTensor_2<InputDataType>{-5, 5}); break;
+    default: input.GenerateTensorValue(GeneratorTensor_3<InputDataType>{0.0, 1.0});
+    }
+
+    DeviceMem in_device_buf(sizeof(InputDataType) * input.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutputDataType) * device_output.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(input.mData.data());
+
+    // run reference op
+    if(do_verification)
+    {
+        auto ref_image_to_column = ck::tensor_operation::host::
+            ReferenceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>{};
+
+        auto ref_invoker  = ref_image_to_column.MakeInvoker();
+        auto ref_argument = ref_image_to_column.MakeArgument(input,
+                                                             host_output,
+                                                             conv_param.filter_spatial_lengths_,
+                                                             conv_param.conv_filter_strides_,
+                                                             conv_param.conv_filter_dilations_,
+                                                             conv_param.input_left_pads_,
+                                                             conv_param.input_right_pads_);
+
+        // init host output to zero
+        host_output.SetZero();
+
+        ref_invoker.Run(ref_argument);
+    }
+
+    using DeviceOp = ck::tensor_operation::device::
+        DeviceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device op instances
+    bool pass                   = true;
+    bool is_supporting_instance = false;
+
+    for(auto& op_ptr : op_ptrs)
+    {
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            static_cast<InputDataType*>(in_device_buf.GetDeviceBuffer()),
+            static_cast<OutputDataType*>(out_device_buf.GetDeviceBuffer()),
+            conv_param.N_,
+            conv_param.C_,
+            input_spatial_lengths,
+            filter_spatial_lengths,
+            output_spatial_lengths,
+            input_g_n_c_wis_strides,
+            output_m_k_strides,
+            conv_filter_strides,
+            conv_filter_dilations,
+            input_left_pads,
+            input_right_pads);
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            is_supporting_instance = true;
+            // re-init output to zero before profiling next kernel
+            out_device_buf.SetZero();
+            std::string op_name = op_ptr->GetTypeString();
+            auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+            float avg_time =
+                invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
+            std::size_t num_btype =
+                NDoHoWo * CZYX * (sizeof(OutputDataType) + sizeof(InputDataType));
+            float gb_per_sec = num_btype / 1.E6 / avg_time;
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(avg_time < best_avg_time)
+            {
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+
+            if(do_verification)
+            {
+                out_device_buf.FromDevice(device_output.mData.data());
+                pass = pass & ck::utils::check_err(device_output, host_output);
+
+                if(do_log)
+                {
+                    LogRangeAsType<float>(std::cout << "input : ", input.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "host_output  : ", host_output.mData, ",")
+                        << std::endl;
+                    LogRangeAsType<float>(std::cout << "device_output: ", device_output.mData, ",")
+                        << std::endl;
+                }
+            }
+        }
+        else
+        {
+            std::cout << op_ptr->GetTypeString() << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best configuration parameters:"
+              << "\nname: " << best_op_name << "\navg_time: " << best_avg_time
+              << "\nGB/s: " << best_gb_per_sec << std::endl;
+
+    return is_supporting_instance && pass;
+}
+
+} // namespace profiler
+} // namespace ck
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 66755c47c5..7da7613f26 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -28,6 +28,7 @@ set(PROFILER_SOURCES
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
     profile_grouped_conv_bwd_data.cpp
+    profile_image_to_column.cpp
 )
 if(DL_KERNELS)
   list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
@@ -82,6 +83,7 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_avg_pool3d_bwd_insta
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_image_to_column_instance)
 if(DL_KERNELS)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 endif()
diff --git a/profiler/src/profile_image_to_column.cpp b/profiler/src/profile_image_to_column.cpp
new file mode 100644
index 0000000000..bf4312a6cf
--- /dev/null
+++ b/profiler/src/profile_image_to_column.cpp
@@ -0,0 +1,169 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_image_to_column_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+namespace {
+
+enum struct ConvLayout
+{
+    NHWC, // 0
+};
+
+enum struct DataType
+{
+    F32_F32,   // 0
+    F16_F16,   // 1
+    BF16_BF16, // 2
+    INT8_INT8, // 3
+};
+
+#define OP_NAME "image_to_column"
+#define OP_DESC "Image To Column"
+
+static void print_helper_msg()
+{
+    std::cout
+        // clang-format off
+        << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+        << "arg2: data type (0: Input fp32, Weight fp32, Output fp32\n"
+        << "                 1: Input fp16, Weight fp16, Output fp16\n"
+        << "                 2: Input bf16, Weight bf16, Output bf16\n"
+        << "                 3: Input int8, Weight int8, Output int8)\n"
+        << "arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])\n"
+        << "arg4: verification (0: no, 1: yes)\n"
+        << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
+        << "arg6: print tensor value (0: no; 1: yes)\n"
+        << "arg7: time kernel (0: no, 1: yes)\n"
+        << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+    // clang-format on
+}
+
+} // namespace
+
+int profile_image_to_column(int argc, char* argv[])
+{
+    // 8 for control, 1 for num_dim_spatial
+    if(argc < 9)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto data_type       = static_cast<DataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<ConvLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+    const int num_dim_spatial  = std::stoi(argv[8]);
+
+    // 8 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
+    if(argc != 8 + 1 + 4 + 6 * num_dim_spatial)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto params = ck::utils::conv::parse_conv_param(num_dim_spatial, 9, argv);
+
+    using F32  = float;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+    using INT8 = int8_t;
+
+    using namespace ck::tensor_layout::convolution;
+
+    constexpr auto I1 = ck::Number<1>{};
+    constexpr auto I2 = ck::Number<2>{};
+    constexpr auto I3 = ck::Number<3>{};
+
+    auto profile = [&](auto num_dim_spatial_tmp, auto in_layout, auto in_type, auto out_type) {
+        constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
+
+        using InLayout = decltype(in_layout);
+
+        using InDataType  = decltype(in_type);
+        using OutDataType = decltype(out_type);
+
+        bool pass = ck::profiler::
+            profile_image_to_column_impl<NDimSpatial, InLayout, InDataType, OutDataType>(
+                do_verification, init_method, do_log, time_kernel, params);
+
+        return pass ? 0 : 1;
+    };
+
+    // NHWC
+    if(layout == ConvLayout::NHWC)
+    {
+        if(num_dim_spatial == 1)
+        {
+            if(data_type == DataType::F32_F32)
+            {
+                return profile(I1, GNWC{}, F32{}, F32{});
+            }
+            else if(data_type == DataType::F16_F16)
+            {
+                return profile(I1, GNWC{}, F16{}, F16{});
+            }
+            else if(data_type == DataType::BF16_BF16)
+            {
+                return profile(I1, GNWC{}, BF16{}, BF16{});
+            }
+            else if(data_type == DataType::INT8_INT8)
+            {
+                return profile(I1, GNWC{}, INT8{}, INT8{});
+            }
+        }
+        else if(num_dim_spatial == 2)
+        {
+            if(data_type == DataType::F32_F32)
+            {
+                return profile(I2, GNHWC{}, F32{}, F32{});
+            }
+            else if(data_type == DataType::F16_F16)
+            {
+                return profile(I2, GNHWC{}, F16{}, F16{});
+            }
+            else if(data_type == DataType::BF16_BF16)
+            {
+                return profile(I2, GNHWC{}, BF16{}, BF16{});
+            }
+            else if(data_type == DataType::INT8_INT8)
+            {
+                return profile(I2, GNHWC{}, INT8{}, INT8{});
+            }
+        }
+        else if(num_dim_spatial == 3)
+        {
+            if(data_type == DataType::F32_F32)
+            {
+                return profile(I3, GNDHWC{}, F32{}, F32{});
+            }
+            else if(data_type == DataType::F16_F16)
+            {
+                return profile(I3, GNDHWC{}, F16{}, F16{});
+            }
+            else if(data_type == DataType::BF16_BF16)
+            {
+                return profile(I3, GNDHWC{}, BF16{}, BF16{});
+            }
+            else if(data_type == DataType::INT8_INT8)
+            {
+                return profile(I3, GNDHWC{}, INT8{}, INT8{});
+            }
+        }
+    }
+
+    std::cout << "this data_type & layout is not implemented" << std::endl;
+
+    return 1;
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_image_to_column);
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index a359c529a4..8fddd60858 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -60,6 +60,7 @@ add_subdirectory(contraction)
 add_subdirectory(pool)
 add_subdirectory(batched_gemm_multi_d)
 add_subdirectory(grouped_convnd_bwd_data)
+add_subdirectory(image_to_column)
 if(GPU_TARGETS MATCHES "gfx11")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/image_to_column/CMakeLists.txt b/test/image_to_column/CMakeLists.txt
new file mode 100644
index 0000000000..0feb827b55
--- /dev/null
+++ b/test/image_to_column/CMakeLists.txt
@@ -0,0 +1,4 @@
+add_gtest_executable(test_image_to_column test_image_to_column.cpp)
+target_link_libraries(test_image_to_column PRIVATE utility device_image_to_column_instance)
+add_gtest_executable(test_image_to_column_interface test_image_to_column_interface.cpp)
+target_link_libraries(test_image_to_column_interface PRIVATE utility)
diff --git a/test/image_to_column/test_image_to_column.cpp b/test/image_to_column/test_image_to_column.cpp
new file mode 100644
index 0000000000..0b17cac2d0
--- /dev/null
+++ b/test/image_to_column/test_image_to_column.cpp
@@ -0,0 +1,121 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "profiler/profile_image_to_column_impl.hpp"
+
+template <typename Tuple>
+class TestImageToColumn : public ::testing::Test
+{
+    protected:
+    using InDataType  = std::tuple_element_t<0, Tuple>;
+    using OutDataType = std::tuple_element_t<1, Tuple>;
+    using InLayout    = std::tuple_element_t<2, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+
+    template <ck::index_t NDimSpatial>
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_image_to_column_impl<NDimSpatial,
+                                                                      InLayout,
+                                                                      InDataType,
+                                                                      OutDataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+using namespace ck::tensor_layout::convolution;
+
+using KernelTypes1d = ::testing::Types<std::tuple<float, float, GNWC>,
+                                       std::tuple<ck::bhalf_t, ck::bhalf_t, GNWC>,
+                                       std::tuple<ck::half_t, ck::half_t, GNWC>,
+                                       std::tuple<int8_t, int8_t, GNWC>>;
+
+using KernelTypes2d = ::testing::Types<std::tuple<float, float, GNHWC>,
+                                       std::tuple<ck::bhalf_t, ck::bhalf_t, GNHWC>,
+                                       std::tuple<ck::half_t, ck::half_t, GNHWC>,
+                                       std::tuple<int8_t, int8_t, GNHWC>>;
+
+using KernelTypes3d = ::testing::Types<std::tuple<float, float, GNDHWC>,
+                                       std::tuple<ck::bhalf_t, ck::bhalf_t, GNDHWC>,
+                                       std::tuple<ck::half_t, ck::half_t, GNDHWC>,
+                                       std::tuple<int8_t, int8_t, GNDHWC>>;
+
+template <typename Tuple>
+class TestImageToColumn1d : public TestImageToColumn<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestImageToColumn2d : public TestImageToColumn<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestImageToColumn3d : public TestImageToColumn<Tuple>
+{
+};
+
+TYPED_TEST_SUITE(TestImageToColumn1d, KernelTypes1d);
+TYPED_TEST_SUITE(TestImageToColumn2d, KernelTypes2d);
+TYPED_TEST_SUITE(TestImageToColumn3d, KernelTypes3d);
+
+TYPED_TEST(TestImageToColumn1d, Test1D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back({1, 1, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {7}, {2}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
+    // ScalarPerVector should be 1
+    this->conv_params.push_back({1, 1, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
+    // stride != 1
+    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
+    // dilation != 1
+    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
+    this->template Run<1>();
+}
+
+TYPED_TEST(TestImageToColumn2d, Test2D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back(
+        {2, 1, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 1, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->template Run<2>();
+}
+
+TYPED_TEST(TestImageToColumn3d, Test3D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {3, 1, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 1, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->template Run<3>();
+}
diff --git a/test/image_to_column/test_image_to_column_interface.cpp b/test/image_to_column/test_image_to_column_interface.cpp
new file mode 100644
index 0000000000..ea8b9632e1
--- /dev/null
+++ b/test/image_to_column/test_image_to_column_interface.cpp
@@ -0,0 +1,196 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using DataType = float;
+using InLayout = ck::tensor_layout::convolution::GNWC;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+template <ck::index_t ScalarPerVector, bool IsCPacked>
+class TestImageToColumnInterface : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 1;
+
+    // clang-format off
+    using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumnImpl
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread|         Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|            Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths|         Vector|
+        //#####################|           |         |           |            |      |      |      |          |               |
+                              < NDimSpatial, InLayout,   DataType,    DataType,   256,   128,   128, S<16, 16>,ScalarPerVector>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    bool Run()
+    {
+
+        const auto N = conv_param.N_;
+        const auto C = conv_param.C_;
+        const auto FakeC =
+            conv_param.C_ / 2; // Fake C to simulate the behavior that C is not packed
+
+        const ck::index_t NDoHoWo =
+            N *
+            ck::accumulate_n<ck::index_t>(
+                conv_param.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+        const ck::index_t CZYX =
+            C *
+            ck::accumulate_n<ck::index_t>(
+                conv_param.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+        const auto in_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+        const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+
+        std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
+        std::array<ck::index_t, 2> output_m_k_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+
+        copy(conv_param.input_spatial_lengths_, input_spatial_lengths);
+        copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
+        copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
+        copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
+        copy(out_desc.GetStrides(), output_m_k_strides);
+        copy(conv_param.conv_filter_strides_, conv_filter_strides);
+        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+        copy(conv_param.input_left_pads_, input_left_pads);
+        copy(conv_param.input_right_pads_, input_right_pads);
+
+        auto img2col  = DeviceImgToColInstance{};
+        auto argument = img2col.MakeArgument(nullptr,
+                                             nullptr,
+                                             N,
+                                             IsCPacked ? C : FakeC,
+                                             input_spatial_lengths,
+                                             filter_spatial_lengths,
+                                             output_spatial_lengths,
+                                             input_g_n_c_wis_strides,
+                                             output_m_k_strides,
+                                             conv_filter_strides,
+                                             conv_filter_dilations,
+                                             input_left_pads,
+                                             input_right_pads);
+
+        return img2col.IsSupportedArgument(argument);
+    }
+};
+
+class TestImageToColumnInterface1ScalarPerVector : public TestImageToColumnInterface<1, true>
+{
+};
+
+class TestImageToColumnInterface4ScalarPerVector : public TestImageToColumnInterface<4, true>
+{
+};
+
+class TestImageToColumnInterface4ScalarPerVectorFakeC : public TestImageToColumnInterface<4, false>
+{
+};
+
+TEST_F(TestImageToColumnInterface1ScalarPerVector, X1ScalarPerVector)
+{
+    // vector load C * X % ScalarPerVector
+    this->conv_param  = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
+    bool is_supported = this->Run();
+    EXPECT_TRUE(is_supported);
+    // vector load C * left_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+    // vector load C * right_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+
+    // vector load C % ScalarPerVector, right_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+    // vector load C % ScalarPerVector, left_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+    // vector load C % ScalarPerVector, dilation
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+
+    // C = 4
+    this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+}
+
+TEST_F(TestImageToColumnInterface4ScalarPerVector, X4ScalarPerVector)
+{
+    // vector load C * X % ScalarPerVector
+    this->conv_param  = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
+    bool is_supported = this->Run();
+    EXPECT_FALSE(is_supported);
+    // vector load C * left_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
+    is_supported     = this->Run();
+    EXPECT_FALSE(is_supported);
+    // vector load C * right_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
+    is_supported     = this->Run();
+    EXPECT_FALSE(is_supported);
+
+    // vector load C % ScalarPerVector, right_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
+    is_supported     = this->Run();
+    EXPECT_FALSE(is_supported);
+    // vector load C % ScalarPerVector, left_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
+    is_supported     = this->Run();
+    EXPECT_FALSE(is_supported);
+    // vector load C % ScalarPerVector, dilation
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
+    is_supported     = this->Run();
+    EXPECT_FALSE(is_supported);
+
+    // C = 4
+    this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+}
+
+TEST_F(TestImageToColumnInterface4ScalarPerVectorFakeC, X4ScalarPerVectorFakeC)
+{
+    // C = 3
+    this->conv_param  = {1, 1, 1, 1, 3, {4}, {3}, {1}, {1}, {0}, {0}};
+    bool is_supported = this->Run();
+    EXPECT_FALSE(is_supported);
+    // C = 4
+    this->conv_param = {1, 1, 1, 1, 8, {4}, {3}, {1}, {1}, {0}, {0}};
+    is_supported     = this->Run();
+    EXPECT_TRUE(is_supported);
+}

From 7dcb14d9d495e6329477f5ee27a27cfbb4ce49a6 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 5 Sep 2023 11:29:44 -0700
Subject: [PATCH 112/194] fix syntax (#890)

---
 .../gpu/device/impl/device_image_to_column_impl.hpp          | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
index 89fcbca1ac..19f126e66f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
@@ -151,8 +151,9 @@ struct DeviceImageToColumnImpl
         remove_cvref_t<decltype(MakeInputDescriptor_M_K(1, 1, {}, {}, {}, {}, {}, {}, {}, {}))>;
     using OutputGridDesc = remove_cvref_t<decltype(MakeOutDescriptor_M_K(1, 1, {}, {}, {}))>;
 
-    using Block2ETileMap = remove_cvref_t<decltype(
-        BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(OutputGridDesc{}))>;
+    using Block2ETileMap = remove_cvref_t<
+        decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
+            OutputGridDesc{}))>;
 
     using GridwiseImageToColumnKernel = GridwiseImageToColumn<InputGridDesc,
                                                               InputDataType,

From 1e1f82d9b0c3282d721ed09603091b72c75fd48d Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Tue, 5 Sep 2023 21:25:28 +0200
Subject: [PATCH 113/194] Add contribution guidelines to the documentation
 (#843)

Add contribution guidelines to the documentation
---
 docs/Contributors_Guide.rst | 100 ++++++++++++++++++++++++++++++++++--
 1 file changed, 97 insertions(+), 3 deletions(-)

diff --git a/docs/Contributors_Guide.rst b/docs/Contributors_Guide.rst
index b2ddff398c..41cb8f1915 100644
--- a/docs/Contributors_Guide.rst
+++ b/docs/Contributors_Guide.rst
@@ -2,7 +2,101 @@
 Contributor's Guide
 ===================
 
-Pull-request guidelines
-=======================
+This chapter explains how to get started contributing to the Composable Kernel project and what are
+the contributing rules.
 
-[TODO]
+Getting started
+===============
+
+#. **Documentation:** Before contributing to the library, familiarize yourself with the
+   `Composable Kernel User Guide <https://rocm.docs.amd.com/projects/composable_kernel/en/latest/>`_.
+   It provides insight into the core concepts, environment configuration, and steps to obtain or
+   build the library. You can also find some of this information in the
+   `README file <https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/README.md>`_
+   on the project's GitHub page.
+#. **Additional reading:** We also recommend reading a `blog post
+   <https://community.amd.com/t5/instinct-accelerators/amd-composable-kernel-library-efficient-fused-kernels-for-ai/ba-p/553224>`_
+   from the AMD Community portal. It offers a deeper understanding of the library's objectives and
+   showcases its performance capabilities.
+#. **General information:** For broader information about AMD products, consider exploring the
+   `AMD Developer Central portal <https://www.amd.com/en/developer.html>`_.
+
+How do I contribute
+===================
+
+We deeply value contributions from our users. You can make an impact by reporting issues or
+proposing code enhancements through pull requests.
+
+Reporting issues
+----------------
+
+We use `Github issues <https://github.com/ROCmSoftwarePlatform/composable_kernel/issues>`_
+to track public bugs and enhancement requests.
+
+If you encounter an issue with the library, please check if the problem has already been
+reported by searching existing issues on GitHub. If your issue seems unique, please submit a new
+issue. All reported issues must include:
+
+* A comprehensive description of the problem, including:
+
+  * What did you observe?
+  * Why do you think it is a bug (if it seems like one)?
+  * What did you expect to happen? What would indicate the resolution of the problem?
+  * Are there any known workarounds?
+
+* Your configuration details, including:
+
+  * Which GPU are you using?
+  * Which OS version are you on?
+  * Which ROCm version are you using?
+  * Are you using a Docker image? If so, which one?
+
+* Steps to reproduce the issue, including:
+
+  * What actions trigger the issue? What are the reproduction steps?
+
+    * If you build the library from scratch, what CMake command did you use?
+
+  * How frequently does this issue happen? Does it reproduce every time? Or is it a sporadic issue?
+
+Before sumbitting any issue, ensure you have addressed all relevant questions from the checklist.
+
+Creating Pull Requests
+----------------------
+
+You can submit `Pull Requests (PR) on GitHub
+<https://github.com/ROCmSoftwarePlatform/composable_kernel/pulls>`_.
+
+All contributors are required to develop their changes on a separate branch and then create a
+pull requrest to merge their changes into the `develop` branch, which is the default
+development branch in the Composable Kernel project. All external contributors must use their own
+forks of the project to develop their changes.
+
+When submitting a Pull Request you should:
+
+* Describe the change providing information about the motivation for the change and a general
+  description of all code modifications.
+
+* Verify and test the change:
+
+  * Run any relevant existing tests.
+  * Write new tests if added functionality is not covered by current tests.
+
+* Ensure your changes align with the coding style defined in the ``.clang-format`` file located in
+  the project's root directory. We leverage `pre-commit` to run `clang-format` automatically. We
+  highly recommend contributors utilize this method to maintain consistent code formatting.
+  Instructions on setting up `pre-commit` can be found in the project's
+  `README file <https://github.com/ROCmSoftwarePlatform/composable_kernel/blob/develop/README.md>`_
+
+* Link your PR to any related issues:
+
+  * If there is an issue that is resolved by your change, please provide a link to the issue in
+    the description of your pull request.
+
+* For larger contributions, structure your change into a sequence of smaller, focused commits, each
+  addressing a particular aspect or fix.
+
+Following the above guidelines ensures a seamless review process and faster assistance from our
+end.
+
+Thank you for your commitment to enhancing the Composable Kernel project! We look forward to collaborating with you.

From aae4df559615e611bafd5850be5168287d97edd9 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 5 Sep 2023 14:29:37 -0700
Subject: [PATCH 114/194] set warnings as errors in doxygen (#864)

---
 cmake/DoxygenDoc.cmake | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/cmake/DoxygenDoc.cmake b/cmake/DoxygenDoc.cmake
index 2e3669fcdf..c91308b5bb 100644
--- a/cmake/DoxygenDoc.cmake
+++ b/cmake/DoxygenDoc.cmake
@@ -309,6 +309,8 @@ XML_OUTPUT
 XML_PROGRAMLISTING
 )
 
+set(WARN_AS_ERROR YES)
+
 set(DOXYGEN_CONFIG_FILE "${CMAKE_CURRENT_BINARY_DIR}/doxygen/doxygen.conf" CACHE PATH "Path to generated doxygen configuration file")
 
 function(add_doxygen_doc)

From a61b8b785eae5f4432fbf62e85b3b119cf019ea0 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 6 Sep 2023 09:59:20 -0500
Subject: [PATCH 115/194] Fixed fp8 gemm (#882)

* add generic instances; fixed initi with fp8

* fixed comment

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../library/utility/host_tensor_generator.hpp | 35 ++++++++++++++++
 ...f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp | 16 ++++++++
 ...f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp | 18 ++++++++-
 ...dl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp | 12 ++++++
 ...dl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp | 40 ++++++++++++-------
 5 files changed, 106 insertions(+), 15 deletions(-)

diff --git a/library/include/ck/library/utility/host_tensor_generator.hpp b/library/include/ck/library/utility/host_tensor_generator.hpp
index 31ff13aec2..2fdb0b141d 100644
--- a/library/include/ck/library/utility/host_tensor_generator.hpp
+++ b/library/include/ck/library/utility/host_tensor_generator.hpp
@@ -95,6 +95,22 @@ struct GeneratorTensor_2<int8_t>
     }
 };
 
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+template <>
+struct GeneratorTensor_2<ck::f8_t>
+{
+    int min_value = 0;
+    int max_value = 1;
+
+    template <typename... Is>
+    ck::f8_t operator()(Is...)
+    {
+        float tmp = (std::rand() % (max_value - min_value)) + min_value;
+        return ck::type_convert<ck::f8_t>(tmp);
+    }
+};
+#endif
+
 template <typename T>
 struct GeneratorTensor_3
 {
@@ -127,6 +143,25 @@ struct GeneratorTensor_3<ck::bhalf_t>
     }
 };
 
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+template <>
+struct GeneratorTensor_3<ck::f8_t>
+{
+    float min_value = 0;
+    float max_value = 1;
+
+    template <typename... Is>
+    ck::f8_t operator()(Is...)
+    {
+        float tmp = float(std::rand()) / float(RAND_MAX);
+
+        float fp32_tmp = min_value + tmp * (max_value - min_value);
+
+        return ck::type_convert<ck::f8_t>(fp32_tmp);
+    }
+};
+#endif
+
 template <typename T>
 struct GeneratorTensor_4
 {
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
index 7946bafc28..10c4453463 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
@@ -33,6 +33,18 @@ using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
 
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
+using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_generic_instances =
+    std::tuple<
+        // clang-format off
+        // M/N/K padding
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Row, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               1>
+        // clang-format on
+        >;
+
 using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -73,6 +85,10 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_m
                                                     PassThrough,
                                                     MultiplyAdd>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_generic_instances{});
+
     add_device_operation_instances(
         instances,
         device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
index 99db9f4270..71a3e42496 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
@@ -33,6 +33,19 @@ using MultiplyAdd = ck::tensor_operation::element_wise::MultiplyAdd;
 
 static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
+using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_generic_instances =
+    std::tuple<
+        // clang-format off
+        // M/N/K padding
+        // N % 8 == 0 && K % 1 == 0
+        //##############################|      A|      B|        Ds|      E| AData| BData| AccData| CShuffle|    DsData| EData|           A|           B|            CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //##############################| Layout| Layout|    Layout| Layout|  Type|  Type|    Type| DataType|      Type|  Type| Elementwise| Elementwise|    Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //##############################|       |       |          |       |      |      |        |         |          |      |   Operation|   Operation|      Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //##############################|       |       |          |       |      |      |        |         |          |      |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>
+        // clang-format on
+        >;
+
 using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances =
     std::tuple<
         // clang-format off
@@ -55,7 +68,6 @@ using device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_i
         DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 64>,               1>,
         DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>,
         DeviceGemmMultipleD_Xdl_CShuffle<    Row,    Col, Row_Tuple,    Row,   F16,    F8,     F32,      F32, F32_Tuple,   F16, PassThrough, PassThrough,    MultiplyAdd, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,              S<1,  2, 1, 32>,               1>
-
         // clang-format on
         >;
 
@@ -72,6 +84,10 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_m
                                                     PassThrough,
                                                     MultiplyAdd>>>& instances)
 {
+    add_device_operation_instances(
+        instances,
+        device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_generic_instances{});
+
     add_device_operation_instances(
         instances,
         device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instances{});
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
index c0b6ec67bb..6168695b7f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
@@ -30,6 +30,16 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 // static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
 
+using device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_generic_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               2,    F16>
+    // clang-format on
+    >;
+
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
@@ -61,6 +71,8 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances(
         DeviceGemmSplitK<Row, Row, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instances{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
index fa38640ae2..a2dd9532a1 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
@@ -27,7 +27,17 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_generic_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               2,    F16>
+    // clang-format on
+    >;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances = std::tuple<
@@ -36,19 +46,19 @@ using device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
-        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,   GemmDefault,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>,
+        DeviceGemmXdlSplitKCShuffle<  F16,    F8,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F16>
     // clang-format on
     >;
 
@@ -57,6 +67,8 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
         DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances{});
 }

From 3786bfe1cce4b895494261a19a31e41f6b43fc7a Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 6 Sep 2023 10:15:52 -0500
Subject: [PATCH 116/194] added padding of K into gemm_v2r3 (#887)

* added kpad support into v2r3

* add generic instances

* fixed comments

* fixed mnk padding

* Update device_batched_gemm_xdl.hpp

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../device/impl/device_batched_gemm_xdl.hpp   | 14 +++---
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    | 45 +++++++++++++++++--
 ...m_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp | 12 +++++
 ...m_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp | 12 +++++
 4 files changed, 72 insertions(+), 11 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
index 408c56d490..303eba156e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_xdl.hpp
@@ -185,7 +185,7 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
         AElementwiseOperation,
         BElementwiseOperation,
         CElementwiseOperation,
-        GemmSpecialization::MNPadding,
+        GemmSpecialization::MNKPadding,
         MPerBlock,
         NPerBlock,
         K0PerBlock,
@@ -315,11 +315,6 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
             return false;
         }
 
-        if(problem.K % K1 != 0)
-        {
-            return false;
-        }
-
         return GridwiseGemm::CheckValidity(problem);
     }
 
@@ -416,7 +411,12 @@ struct DeviceBatchedGemmXdl : public DeviceBatchedGemm<ALayout,
             << BlockSize << ", "
             << MPerBlock << ", "
             << NPerBlock << ", "
-            << K0PerBlock
+            << K0PerBlock << ", "
+            << K1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
             << ">"
             << " NumGemmKPrefetchStage: "
             << NumGemmKPrefetchStage << ", "
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index 31aee70db2..b6c146ae61 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -194,7 +194,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
               StrideC{StrideC_},
               MPadded{CalculateMPadded(M_)},
               NPadded{CalculateNPadded(N_)},
-              K0{CalculateK0(K)}
+              K0{CalculateK0(K_)}
         {
         }
 
@@ -383,7 +383,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
 
     __host__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
     {
-        const index_t num_loop = K / (K0PerBlock * K1);
+        const index_t num_loop = math::integer_divide_ceil(K, K0PerBlock * K1);
 
         return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
     }
@@ -840,7 +840,25 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext
             }
         }();
 
-        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            const auto K0Pad = math::integer_divide_ceil(K0, K0PerBlock) * K0PerBlock;
+            const auto KPad  = K0Pad * K1Value;
+
+            const auto a_grid_desc_m_kpad = transform_tensor_descriptor(
+                a_grid_desc_m_k,
+                make_tuple(make_pass_through_transform(M), make_right_pad_transform(K, KPad - K)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return transform_tensor_descriptor(
+                a_grid_desc_m_kpad,
+                make_tuple(make_unmerge_transform(make_tuple(K0Pad, K1Value)),
+                           make_right_pad_transform(M, MPad - M)),
+                make_tuple(Sequence<1>{}, Sequence<0>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+        else if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
         {
             return transform_tensor_descriptor(
                 a_grid_desc_m_k,
@@ -874,7 +892,26 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext
             }
         }();
 
-        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
+        {
+            const auto K0Pad = math::integer_divide_ceil(K0, K0PerBlock) * K0PerBlock;
+            const auto KPad  = K0Pad * K1Value;
+
+            const auto b_grid_desc_kpad_n = transform_tensor_descriptor(
+                b_grid_desc_k_n,
+                make_tuple(make_right_pad_transform(K, KPad - K), make_pass_through_transform(N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return transform_tensor_descriptor(
+                b_grid_desc_kpad_n,
+                make_tuple(make_unmerge_transform(make_tuple(K0Pad, K1Value)),
+                           make_right_pad_transform(N, NPad - N)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+        }
+
+        else if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
         {
             return transform_tensor_descriptor(
                 b_grid_desc_k_n,
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
index 2f42f62b0b..dc7de8c689 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
@@ -25,6 +25,16 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
+using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_generic_instances = std::tuple<
+    // clang-format off
+        //#################| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumGemmK|          LoopScheduler|                    Pipeline|
+        //#################|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar| Prefetch|                       |                            |
+        //#################|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector| Stage   |                       |                            |
+        //#################|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |         |                       |                            |
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough,    64,    16,    16,     4,  8,   16,   16,    1,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances = std::tuple<
     // clang-format off
@@ -100,6 +110,8 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances(
         DeviceBatchedGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(
+        instances, device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instances{});
 }
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
index 10b4cea7d7..cccad7ca19 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
@@ -25,6 +25,16 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
+using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_generic_instances = std::tuple<
+    // clang-format off
+        //#################| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumGemmK|          LoopScheduler|                    Pipeline|
+        //#################|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar| Prefetch|                       |                            |
+        //#################|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector| Stage   |                       |                            |
+        //#################|      |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |         |                       |                            |
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    64,    32,    64,     4,  8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v1>
+    // clang-format on
+    >;
+
 // Compilation parameters for a[m, k] * b[n, k] = c[m, n]
 using device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances = std::tuple<
     // clang-format off
@@ -88,6 +98,8 @@ void add_device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances(
         DeviceBatchedGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
+    add_device_operation_instances(
+        instances, device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_generic_instances{});
     add_device_operation_instances(instances,
                                    device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instances{});
 }

From 37a8c1f756f94e95b996fc3c26ff39fcf4e435c5 Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Wed, 6 Sep 2023 18:44:09 +0200
Subject: [PATCH 117/194] Redesign the DPP8 GEMM kernel to use warp-wise
 component (#863)

* Redesign the DPP8 GEMM kernel to use warp-wise component

* Review: Improve error messages

* Review: Remove unnecessary empty lines

* Review: Fix M, N per thread names

* Review: Rename mfma_input_type to dpp_input_type

* Review: Fix tensor adaptor; remove unnecessary element

* Review: Remove calls to dpp_gemm's MakeCDescriptor

* Review: Add blockwise doc, change function names to include dimension names

* Review: Remove duplicated code; Move Block2CtileMap alias to the top of the file

* Review: Add __restrict__ keywords

* Review: Use MatrixPadder for padding A, B, C matrices

* Review: Remove hardcoded datatypes

* Review: Change names from FloatX to XDataType

* Review: Introduce AK0 and BK0 instead of a single K0

* Review: Remove construction of dpp_datatypes object

* Review: Rename DppInstrRunner to DppLanegroupGemm
---
 example/01_gemm/CMakeLists.txt                |   3 +-
 example/01_gemm/gemm_dl_dpp8_fp16.cpp         |  37 -
 example/01_gemm/gemm_dpp_fp16.cpp             |  39 +
 .../gpu/block/blockwise_gemm_dl_dpp8.hpp      | 370 ---------
 .../gpu/block/blockwise_gemm_dpp.hpp          | 348 +++++++++
 .../gpu/block/blockwise_gemm_xdlops.hpp       |  16 +-
 .../gpu/device/gemm_dl_algorithm.hpp          |  18 -
 .../gpu/device/impl/device_gemm_dl.hpp        |  27 +-
 .../gpu/device/impl/device_gemm_dl_dpp8.hpp   | 133 ----
 .../gpu/device/impl/device_gemm_dpp.hpp       | 271 +++++++
 .../gpu/grid/gridwise_gemm_dl_v1r3.hpp        |  71 +-
 .../gpu/grid/gridwise_gemm_dpp.hpp            | 701 ++++++++++++++++++
 .../gpu/grid/gridwise_gemm_pipeline_v1.hpp    |   3 +-
 .../thread/threadwise_contraction_dl_dpp8.hpp | 136 ----
 .../ck/tensor_operation/gpu/warp/dpp_gemm.hpp | 322 ++++++++
 include/ck/utility/amd_gemm_dpp.hpp           |  56 +-
 include/ck/utility/inner_product_dpp8.hpp     |   4 +
 include/ck/utility/loop_scheduler.hpp         |  26 +
 .../tensor_operation_instance/gpu/gemm.hpp    |  16 +-
 .../gpu/gemm/CMakeLists.txt                   |   8 +-
 ..._dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp |  61 --
 ..._dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp |  61 --
 ..._dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp |  60 --
 ..._dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp |  61 --
 ...gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp |  58 ++
 ...gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp |  58 ++
 ...gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp |  57 ++
 ...gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp |  58 ++
 28 files changed, 2030 insertions(+), 1049 deletions(-)
 delete mode 100644 example/01_gemm/gemm_dl_dpp8_fp16.cpp
 create mode 100644 example/01_gemm/gemm_dpp_fp16.cpp
 delete mode 100644 include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp
 create mode 100644 include/ck/tensor_operation/gpu/block/blockwise_gemm_dpp.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp
 create mode 100644 include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp
 create mode 100644 include/ck/utility/loop_scheduler.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index a671877f68..3dc2a0966e 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -6,8 +6,7 @@ if(DL_KERNELS)
   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
     add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
     add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-    add_example_executable(example_gemm_dl_dpp8_fp16 gemm_dl_dpp8_fp16.cpp)
-    add_dependencies(example_gemm_dl example_gemm_dl_dpp8_fp16)
+    add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
   endif()
   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
     add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
diff --git a/example/01_gemm/gemm_dl_dpp8_fp16.cpp b/example/01_gemm/gemm_dl_dpp8_fp16.cpp
deleted file mode 100644
index ea0ba39076..0000000000
--- a/example/01_gemm/gemm_dl_dpp8_fp16.cpp
+++ /dev/null
@@ -1,37 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "common.hpp"
-
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
-
-using ADataType   = ck::half_t;
-using BDataType   = ck::half_t;
-using CDataType   = ck::half_t;
-using AccDataType = float;
-
-using ALayout = Col;
-using BLayout = Row;
-using CLayout = Row;
-
-using AElementOp = PassThrough;
-using BElementOp = PassThrough;
-using CElementOp = PassThrough;
-
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
-
-// clang-format off
-using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDlDpp8
-// ######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|      ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|      BBlockTransfer|     CThreadTransfer| CThreadTransfer|    CThreadTransfer|
-// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise| Spacialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|     DstVectorTensor|        SrcDstAccess| SrcDstVectorDim| DstScalarPerVector|
-// ######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
-// ######|          |          |          |            |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
-         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,   256,   128,   128,    16,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,               5,                  4>;
-// clang-format on
-
-using ReferenceGemmInstance = ck::tensor_operation::host::
-    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
-
-#include "run_gemm_example.inc"
-
-int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/example/01_gemm/gemm_dpp_fp16.cpp b/example/01_gemm/gemm_dpp_fp16.cpp
new file mode 100644
index 0000000000..7a9e3f6186
--- /dev/null
+++ b/example/01_gemm/gemm_dpp_fp16.cpp
@@ -0,0 +1,39 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+
+using ADataType   = ck::half_t;
+using BDataType   = ck::half_t;
+using AccDataType = float;
+using CDataType   = ck::half_t;
+
+using F16 = ck::half_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemmDpp
+// ######|     AData|     BData|     CData|     AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+// ######|      Type|      Type|      Type|        Type|        |        |        | Elementwise| Elementwise| Elementwise|Spacialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+// ######|          |          |          |            |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+// ######|          |          |          |            |        |        |        |            |            |            |              |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+         < ADataType, BDataType, CDataType, AccDataType, ALayout, BLayout, CLayout,  AElementOp,  BElementOp,  CElementOp,   GemmDefault,   128,    64,    64,    64,   8,   2,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,      true,               5,               1>;
+// // clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::
+    ReferenceGemm<ADataType, BDataType, CDataType, AccDataType, AElementOp, BElementOp, CElementOp>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp
deleted file mode 100644
index e527509f57..0000000000
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp
+++ /dev/null
@@ -1,370 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/utility/amd_gemm_dpp.hpp"
-#include "ck/utility/common_header.hpp"
-#include "ck/tensor_description/tensor_adaptor.hpp"
-#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v4r1.hpp"
-#include "ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp"
-
-namespace ck {
-
-/**
- * DPP8 version of blockwise GEMM algorithm. It uses DPP8 instruction modifier to limit
- * the data loaded from LDS to registers.
- *
- * The algorithm groups threads into groups of size `dpp8::lane_group_size` and splits the matrix C
- * between them in such a way that threads from the same group need the same chunk of either
- * matrix A (or B, respectively). Without the usage of DPP8, each thread would need to load the
- * whole chunk from LDS to its own register space.
- * Usage of DPP8 modifiers allow each thread to load less data, exactly `1 / dpp8::lane_group_size`
- * of the chunk, and then share that data with other threads from the same lane group.
- *
- * Assumptions coming from the usage of DPP8:
- *   1. `BM10BN10ThreadClusterBM10Xs[1] == dpp8::lane_group_size` or
- *      `BM10BN10ThreadClusterBN10Xs[1] == dpp8::lane_group_size` -
- *        - it makes consecutive `dpp8::lane_group_size` threads use the same chunk of either
- *          matrix A or B;
- *        - based on these values we determine which matrix to share.
- *   2. `BM1PerThreadBM11 % dpp8::lane_group_size == 0` (if sharing A) or
- *      `BN1PerThreadBN11 % dpp8::lane_group_size == 0` (if sharing B) -
- *        - we have to make sure that the data to split is divisible by the number of
- *          threads in the group.
- *
- * General algorithm:
- * C[BM0, BM1, BN0, BN1] += transpose(A[K, BM0, BM1]) * B[K, BN0, BN1]
- * A and B are visible to the whole block, C is distributed among each thread
- * Assume:
- *   1. A:
- *     1. ABlockDesc_BK0_BM_BK1 is known at compile-time
- *     2. ABlockBuffer is DynamicBuffer
- *   2. B:
- *     1. BBlockDesc_BK0_BN_BK1 is known at compile-time
- *     2. BBlockBuffer is DynamicBuffer
- *   3. C:
- *     1. CThreadDesc_BM0_BM11_BN0_BN11 is known at compile-time
- *     2. CThreadBuffer is StaticBuffer
- *   4. BM10BN10ThreadClusterBM10Xs::Size() = BM10BN10ThreadClusterBN10Xs::Size() == 2
- */
-template <index_t BlockSize,
-          typename FloatA,
-          typename FloatB,
-          typename FloatC,
-          typename ABlockDesc_BK0_BM_BK1,
-          typename BBlockDesc_BK0_BN_BK1,
-          index_t BM1PerThreadBM11,
-          index_t BN1PerThreadBN11,
-          index_t BK0PerThread,
-          typename BM10BN10ThreadClusterBM10Xs, // Sequence<BM10BN10ThreadClusterBM100,
-                                                //          BM10BN10ThreadClusterBM101, ...>
-          typename BM10BN10ThreadClusterBN10Xs, // Sequence<BM10BN10ThreadClusterBN100,
-                                                //          BM10BN10ThreadClusterBN101, ...>
-          index_t AThreadCopyScalarPerVector_BM11,
-          index_t BThreadCopyScalarPerVector_BN11,
-          typename enable_if<ABlockDesc_BK0_BM_BK1::IsKnownAtCompileTime() &&
-                                 BBlockDesc_BK0_BN_BK1::IsKnownAtCompileTime(),
-                             bool>::type = false>
-struct BlockwiseGemmDlDpp8_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_loop_BM0_BN0
-{
-    using AIndex = MultiIndex<4>;
-    using BIndex = MultiIndex<4>;
-    using CIndex = MultiIndex<4>;
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    static constexpr index_t BK0 = ABlockDesc_BK0_BM_BK1{}.GetLength(I0);
-    static constexpr index_t BK1 = ABlockDesc_BK0_BM_BK1{}.GetLength(I2);
-    static constexpr index_t BM  = ABlockDesc_BK0_BM_BK1{}.GetLength(I1);
-    static constexpr index_t BN  = BBlockDesc_BK0_BN_BK1{}.GetLength(I1);
-
-    static constexpr index_t BM100 = BM10BN10ThreadClusterBM10Xs{}[I0];
-    static constexpr index_t BN100 = BM10BN10ThreadClusterBN10Xs{}[I0];
-
-    static constexpr index_t BM101 = BM10BN10ThreadClusterBM10Xs{}[I1];
-    static constexpr index_t BN101 = BM10BN10ThreadClusterBN10Xs{}[I1];
-
-    static constexpr index_t BM11 = BM1PerThreadBM11;
-    static constexpr index_t BN11 = BN1PerThreadBN11;
-
-    static constexpr index_t BM1 = BM100 * BM101 * BM11;
-    static constexpr index_t BN1 = BN100 * BN101 * BN11;
-
-    static constexpr index_t BM0 = BM / BM1;
-    static constexpr index_t BN0 = BN / BN1;
-
-    // We assume that either `BM101` or `BN101` is equal to `dpp8::lane_group_size`. It makes all
-    // threads in a lane group need the same chunk of B or A matrices and we can share them using
-    // DPP.
-    static_assert(BM101 == dpp8::lane_group_size || BN101 == dpp8::lane_group_size);
-    static constexpr bool ShareB = BM101 == dpp8::lane_group_size ? true : false;
-    static constexpr bool ShareA = !ShareB;
-
-    // If DPP shares A (B, respectively), lane group gets `BM1PerThreadBM11` (`BN1PerThreadBN11`,
-    // respectively) elements, so we split them between threads in lane group so each thread loads
-    // less data from LDS.
-    static constexpr index_t BM1PerThread =
-        ShareA ? BM1PerThreadBM11 / dpp8::lane_group_size : BM1PerThreadBM11;
-    static constexpr index_t BN1PerThread =
-        ShareB ? BN1PerThreadBN11 / dpp8::lane_group_size : BN1PerThreadBN11;
-
-    __host__ __device__ static constexpr auto
-    MakeABlockDescriptor_BK0_BM0_BM1_BK1(const ABlockDesc_BK0_BM_BK1& a_block_desc_bk0_bm_bk1)
-    {
-        const auto a_block_bk0_bm0_bm1_bk1 = transform_tensor_descriptor(
-            a_block_desc_bk0_bm_bk1,
-            make_tuple(make_pass_through_transform(Number<BK0>{}),
-                       make_unmerge_transform(make_tuple(Number<BM0>{}, Number<BM1>{})),
-                       make_pass_through_transform(Number<BK1>{})),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}));
-
-        return a_block_bk0_bm0_bm1_bk1;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeBBlockDescriptor_BK0_BN0_BN1_BK1(const BBlockDesc_BK0_BN_BK1& b_block_desc_bk0_bn_bk1)
-    {
-        const auto b_block_desc_bk0_bn0_bn1_bk1 = transform_tensor_descriptor(
-            b_block_desc_bk0_bn_bk1,
-            make_tuple(make_pass_through_transform(Number<BK0>{}),
-                       make_unmerge_transform(make_tuple(Number<BN0>{}, Number<BN1>{})),
-                       make_pass_through_transform(Number<BK1>{})),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-            make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}));
-
-        return b_block_desc_bk0_bn0_bn1_bk1;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCBlockAdaptor_BM0_BM100_BM101_BM11_BN0_BN100_BN101_BN11_To_BM_BN()
-    {
-        // upper: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
-        // lower: [BM, BN]
-        constexpr auto c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m_n =
-            make_single_stage_tensor_adaptor(
-                make_tuple(make_unmerge_transform(make_tuple(
-                               Number<BM0>{}, Number<BM100>{}, Number<BM101>{}, Number<BM11>{})),
-                           make_unmerge_transform(make_tuple(
-                               Number<BN0>{}, Number<BN100>{}, Number<BN101>{}, Number<BN11>{}))),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0, 1, 2, 3>{}, Sequence<4, 5, 6, 7>{}));
-
-        return c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m_n;
-    }
-
-    __host__ __device__ static constexpr auto
-    MakeCBlockAdaptor_BM0_BM100_BM101_BM11_BN0_BN100_BN101_BN11_To_BM0_BM1_BN0_BN1()
-    {
-        // upper: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
-        // lower: [BM0, BM1, BN0, BN1]
-        constexpr auto c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m0_m1_n0_n1 =
-            make_single_stage_tensor_adaptor(
-                make_tuple(make_pass_through_transform(Number<BM0>{}),
-                           make_unmerge_transform(
-                               make_tuple(Number<BM100>{}, Number<BM101>{}, Number<BM11>{})),
-                           make_pass_through_transform(Number<BN0>{}),
-                           make_unmerge_transform(
-                               make_tuple(Number<BN100>{}, Number<BN101>{}, Number<BN11>{}))),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2, 3>{}, Sequence<4>{}, Sequence<5, 6, 7>{}));
-
-        return c_block_adaptor_m0_m100_m101_m11_n0_n100_n101_n11_to_m0_m1_n0_n1;
-    }
-
-    __host__ __device__ static constexpr auto GetCThreadTensorLengths_BM0_BM1_BN0_BN1()
-    {
-        return Sequence<BM0, BM11, BN0, BN11>{};
-    }
-
-    static constexpr auto a_block_desc_bk0_bm0_bm1_bk1_ =
-        MakeABlockDescriptor_BK0_BM0_BM1_BK1(ABlockDesc_BK0_BM_BK1{});
-
-    static constexpr auto b_block_desc_bk0_bn0_bn1_bk1_ =
-        MakeBBlockDescriptor_BK0_BN0_BN1_BK1(BBlockDesc_BK0_BN_BK1{});
-
-    public:
-    __device__ BlockwiseGemmDlDpp8_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_loop_BM0_BN0()
-        : c_thread_origin_data_idx_{CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(
-              get_thread_local_1d_id())},
-          a_thread_copy_{CalculateAThreadOriginOnBlock_BK0_BM0_BM1_BK1()},
-          b_thread_copy_{CalculateBThreadOriginOnBlock_BK0_BN0_BN1_BK1()}
-    {
-        static_assert(ABlockDesc_BK0_BM_BK1::IsKnownAtCompileTime() &&
-                          BBlockDesc_BK0_BN_BK1::IsKnownAtCompileTime(),
-                      "wrong! Desc should be known at compile-time");
-
-        static_assert(BM % BM1 == 0 && BN % BN1 == 0, "wrong!");
-
-        static_assert(ABlockDesc_BK0_BM_BK1{}.GetLength(I0) ==
-                          BBlockDesc_BK0_BN_BK1{}.GetLength(I0),
-                      "wrong! K dimension not consistent");
-
-        static_assert(BM10BN10ThreadClusterBM10Xs::Size() == 2 &&
-                          BM10BN10ThreadClusterBN10Xs::Size() == 2,
-                      "wrong!");
-    }
-
-    __device__ static CIndex CalculateCThreadOriginOnBlock_BM0_BM1_BN0_BN1(index_t thread_id)
-    {
-        // lower: [BM0, BM1, BN0, BN1]
-        // upper: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
-        constexpr auto adaptor0 =
-            MakeCBlockAdaptor_BM0_BM100_BM101_BM11_BN0_BN100_BN101_BN11_To_BM0_BM1_BN0_BN1();
-
-        // lower: [BM0, BM100, BM101, BM11, BN0, BN100, BN101, BN11]
-        // upper: [Tid, BM0, BM11, BN0, BN11]
-        constexpr auto adaptor1 = make_single_stage_tensor_adaptor(
-            make_tuple(make_merge_transform(make_tuple(BM100, BN100, BM101, BN101)),
-                       make_pass_through_transform(BM0),
-                       make_pass_through_transform(BM11),
-                       make_pass_through_transform(BN0),
-                       make_pass_through_transform(BN11)),
-            make_tuple(
-                Sequence<1, 5, 2, 6>{}, Sequence<0>{}, Sequence<3>{}, Sequence<4>{}, Sequence<7>{}),
-            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
-
-        constexpr auto adaptor = chain_tensor_adaptors(adaptor0, adaptor1);
-
-        return adaptor.CalculateBottomIndex(make_multi_index(thread_id, 0, 0, 0, 0));
-    }
-
-    __device__ AIndex CalculateAThreadOriginOnBlock_BK0_BM0_BM1_BK1()
-    {
-        const auto offsetBM0 = c_thread_origin_data_idx_[I0];
-        // If sharing matrix A, we need a separate BM1 offset for each thread in lane group.
-        const auto offsetBM1 = ShareA ? c_thread_origin_data_idx_[I1] +
-                                            dpp8::get_thread_idx_in_lane_group() * BM1PerThread
-                                      : c_thread_origin_data_idx_[I1];
-        return make_tuple(0, offsetBM0, offsetBM1, 0);
-    }
-
-    __device__ BIndex CalculateBThreadOriginOnBlock_BK0_BN0_BN1_BK1()
-    {
-        const auto offsetBN0 = c_thread_origin_data_idx_[I2];
-        // If sharing matrix B, we need a separate BN1 offset for each thread in lane group.
-        const auto offsetBN1 = ShareB ? c_thread_origin_data_idx_[I3] +
-                                            dpp8::get_thread_idx_in_lane_group() * BN1PerThread
-                                      : c_thread_origin_data_idx_[I3];
-        return make_tuple(0, offsetBN0, offsetBN1, 0);
-    }
-
-    template <typename CThreadDesc_BM0_BM11_BN0_BN11,
-              typename ABlockBuffer,
-              typename BBlockBuffer,
-              typename CThreadBuffer>
-    __device__ void Run(const CThreadDesc_BM0_BM11_BN0_BN11&,
-                        const ABlockBuffer& a_block_buf,
-                        const BBlockBuffer& b_block_buf,
-                        CThreadBuffer& c_thread_buf) const
-    {
-        static_assert(CThreadDesc_BM0_BM11_BN0_BN11::IsKnownAtCompileTime(),
-                      "wrong! Desc should be known at compile-time");
-
-        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatA>(
-            a_thread_desc_bk0_bm0_bm1_bk1_.GetElementSpaceSize());
-        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
-            b_thread_desc_bk0_bn0_bn1_bk1_.GetElementSpaceSize());
-
-        constexpr auto threadwise_contraction =
-            ThreadwiseContractionDlDpp8_A_TK0_TM0_TM1_TK1_B_TK0_TN0_TN1_TK1_C_TM0_TM1_TN0_TN1<
-                FloatA,
-                FloatB,
-                FloatC,
-                decltype(a_thread_desc_bk0_bm0_bm1_bk1_),
-                decltype(b_thread_desc_bk0_bn0_bn1_bk1_),
-                CThreadDesc_BM0_BM11_BN0_BN11,
-                Sequence<BK0PerThread, BK1>,
-                Sequence<1, BM1PerThreadBM11>,
-                Sequence<1, BN1PerThreadBN11>,
-                ShareA>{};
-
-        static_for<0, BN0, 1>{}([&](auto bn0) {
-            static_for<0, BM0, 1>{}([&](auto bm0) {
-                a_thread_copy_.Run(a_block_desc_bk0_bm0_bm1_bk1_,
-                                   make_tuple(I0, bm0, I0, I0),
-                                   a_block_buf,
-                                   a_thread_desc_bk0_bm0_bm1_bk1_,
-                                   make_tuple(I0, I0, I0, I0),
-                                   a_thread_buf);
-
-                b_thread_copy_.Run(b_block_desc_bk0_bn0_bn1_bk1_,
-                                   make_tuple(I0, bn0, I0, I0),
-                                   b_block_buf,
-                                   b_thread_desc_bk0_bn0_bn1_bk1_,
-                                   make_tuple(I0, I0, I0, I0),
-                                   b_thread_buf);
-
-                threadwise_contraction.Run(a_thread_buf,
-                                           make_tuple(I0, I0, I0, I0),
-                                           b_thread_buf,
-                                           make_tuple(I0, I0, I0, I0),
-                                           c_thread_buf,
-                                           make_tuple(bm0, I0, bn0, I0));
-
-                static_for<BK0PerThread, BK0, BK0PerThread>{}([&](auto bk0) {
-                    a_thread_copy_.Run(a_block_desc_bk0_bm0_bm1_bk1_,
-                                       make_tuple(bk0, bm0, I0, I0),
-                                       a_block_buf,
-                                       a_thread_desc_bk0_bm0_bm1_bk1_,
-                                       make_tuple(I0, I0, I0, I0),
-                                       a_thread_buf);
-
-                    b_thread_copy_.Run(b_block_desc_bk0_bn0_bn1_bk1_,
-                                       make_tuple(bk0, bn0, I0, I0),
-                                       b_block_buf,
-                                       b_thread_desc_bk0_bn0_bn1_bk1_,
-                                       make_tuple(I0, I0, I0, I0),
-                                       b_thread_buf);
-
-                    threadwise_contraction.Run(a_thread_buf,
-                                               make_tuple(I0, I0, I0, I0),
-                                               b_thread_buf,
-                                               make_tuple(I0, I0, I0, I0),
-                                               c_thread_buf,
-                                               make_tuple(bm0, I0, bn0, I0));
-                });
-            });
-        });
-    }
-
-    private:
-    // A[BK0, BM0, BM1, BK1]
-    static constexpr auto a_thread_desc_bk0_bm0_bm1_bk1_ = make_naive_tensor_descriptor_packed(
-        make_tuple(Number<BK0PerThread>{}, Number<BM0>{}, Number<BM1PerThread>{}, Number<BK1>{}));
-
-    // B[BK0, BN0, BN1, BK1]
-    static constexpr auto b_thread_desc_bk0_bn0_bn1_bk1_ = make_naive_tensor_descriptor_packed(
-        make_tuple(Number<BK0PerThread>{}, Number<BN0>{}, Number<BN1PerThread>{}, Number<BK1>{}));
-
-    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4r1<
-        FloatA,
-        FloatA,
-        decltype(a_block_desc_bk0_bm0_bm1_bk1_),
-        decltype(a_thread_desc_bk0_bm0_bm1_bk1_),
-        Sequence<BK0PerThread, 1, BM1PerThread, BK1>, // SliceLengths
-        Sequence<0, 1, 2, 3>,                         // DimAccessOrder
-        Sequence<1, 1, BM1PerThread, BK1>,            // SrcVectorTensorLengths
-        Sequence<0, 1, 2, 3>>;                        // SrcVectorTensorContiguousDimOrder
-
-    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4r1<
-        FloatB,
-        FloatB,
-        decltype(b_block_desc_bk0_bn0_bn1_bk1_),
-        decltype(b_thread_desc_bk0_bn0_bn1_bk1_),
-        Sequence<BK0PerThread, 1, BN1PerThread, BK1>, // SliceLengths
-        Sequence<0, 1, 2, 3>,                         // DimAccessOrder
-        Sequence<1, 1, BN1PerThread, BK1>,            // SrcVectorTensorLengths
-        Sequence<0, 1, 2, 3>>;                        // SrcVectorTensorContiguousDimOrder
-
-    CIndex c_thread_origin_data_idx_;
-
-    AThreadCopy a_thread_copy_;
-    BThreadCopy b_thread_copy_;
-};
-
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_dpp.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dpp.hpp
new file mode 100644
index 0000000000..d62ed4b15d
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_dpp.hpp
@@ -0,0 +1,348 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_adaptor.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/warp/dpp_gemm.hpp"
+
+namespace ck {
+
+/**
+ * Blockwise GEMM that uses DPP instruction modifier to limit the amount of data loaded for each
+ * thread by sharing the data between threads in a lanegroup.
+ *
+ * In every iteration, each wave calculates a C tile of size `MPerDpp` * `NPerDpp`, there are
+ * `MRepeat` iterations for `M` dimension and `NRepeat` for `N` one.
+ * In total, the algorithm runs using
+ * `MPerBlock / (MRepeat * MPerDpp) * NPerBlock / (NRepeat * NPerDpp)` waves.
+ */
+template <index_t BlockSize,
+          typename ABDataType,
+          typename AccDataType,
+          typename AK0MK1BlockDesc,
+          typename BK0NK1BlockDesc,
+          index_t MPerDpp,
+          index_t NPerDpp,
+          index_t MRepeat,
+          index_t NRepeat,
+          index_t KPack>
+struct BlockwiseGemmDpp_ak0mak1_bk0nbk1_m0n0m1n1m2n2
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    static constexpr index_t WaveSize = get_warp_size();
+
+    static constexpr index_t MPerBlock = AK0MK1BlockDesc{}.GetLength(I1);
+    static constexpr index_t NPerBlock = BK0NK1BlockDesc{}.GetLength(I1);
+    static constexpr index_t KPerBlock =
+        BK0NK1BlockDesc{}.GetLength(I0) * BK0NK1BlockDesc{}.GetLength(I2);
+
+    static constexpr index_t A_K0 = AK0MK1BlockDesc{}.GetLength(I0);
+    static constexpr index_t B_K0 = BK0NK1BlockDesc{}.GetLength(I0);
+    static constexpr index_t A_K1 = AK0MK1BlockDesc{}.GetLength(I2);
+    static constexpr index_t B_K1 = BK0NK1BlockDesc{}.GetLength(I2);
+
+    static constexpr auto dpp_gemm = DppGemm<ABDataType, MPerDpp, NPerDpp, KPack>{};
+
+    static constexpr index_t KPerThread = KPerBlock / dpp_gemm.K0PerDpp;
+
+    static constexpr index_t MWaves = MPerBlock / (MRepeat * MPerDpp);
+    static constexpr index_t NWaves = NPerBlock / (NRepeat * NPerDpp);
+
+    StaticBufferTupleOfVector<AddressSpaceEnum::Vgpr,
+                              AccDataType,
+                              MRepeat * NRepeat,
+                              dpp_gemm.GetRegSizePerDpp(),
+                              true>
+        c_thread_buf_;
+
+    __host__ __device__ constexpr auto& GetCThreadBuffer() { return c_thread_buf_; }
+
+    __device__ static auto GetWaveIdx()
+    {
+        const index_t thread_id = ThisThreadBlock::GetThreadId();
+
+        constexpr auto threadid_to_wave_idx_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_merge_transform(make_tuple(MWaves, NWaves, WaveSize))),
+            make_tuple(Sequence<0, 1, 2>{}),
+            make_tuple(Sequence<0>{}));
+
+        return threadid_to_wave_idx_adaptor.CalculateBottomIndex(make_multi_index(thread_id));
+    }
+
+    __device__ static auto CalculateAThreadOriginDataIndex_M0_M1_M2_K()
+    {
+        const auto wave_idx    = GetWaveIdx();
+        const auto waveId_m    = wave_idx[I0];
+        const auto dpp_a_idx   = dpp_gemm.CalculateAThreadOriginDataIndex_K_M();
+        const auto dpp_a_idx_k = dpp_a_idx[I0];
+        const auto dpp_a_idx_m = dpp_a_idx[I1];
+        return make_tuple(0, waveId_m, dpp_a_idx_m, KPerThread * dpp_a_idx_k);
+    }
+
+    __device__ static auto CalculateBThreadOriginDataIndex_N0_N1_N2_K()
+    {
+        const auto wave_idx    = GetWaveIdx();
+        const auto waveId_n    = wave_idx[I1];
+        const auto dpp_b_idx   = dpp_gemm.CalculateBThreadOriginDataIndex_K_N();
+        const auto dpp_b_idx_k = dpp_b_idx[I0];
+        const auto dpp_b_idx_n = dpp_b_idx[I1];
+        return make_tuple(0, waveId_n, dpp_b_idx_n, KPerThread * dpp_b_idx_k);
+    }
+
+    template <index_t m0, index_t n0>
+    __device__ static auto CalculateCThreadOriginDataIndex(Number<m0>, Number<n0>)
+    {
+        const auto wave_idx = GetWaveIdx();
+        const auto waveId_m = wave_idx[I0];
+        const auto waveId_n = wave_idx[I1];
+
+        const auto blk_idx      = dpp_gemm.GetBeginOfThreadBlk();
+        const auto blk_m_offset = blk_idx[I0];
+        const auto blk_n_offset = blk_idx[I1];
+
+        constexpr auto mrepeat_mwave_MPerDpp_to_m_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(MRepeat, MWaves, MPerDpp))),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0, 1, 2>{}));
+
+        constexpr auto nrepeat_nwave_NPerDpp_to_n_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(make_unmerge_transform(make_tuple(NRepeat, NWaves, NPerDpp))),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0, 1, 2>{}));
+
+        const index_t c_thread_m = mrepeat_mwave_MPerDpp_to_m_adaptor.CalculateBottomIndex(
+            make_tuple(m0, waveId_m, blk_m_offset))[I0];
+        const index_t c_thread_n = nrepeat_nwave_NPerDpp_to_n_adaptor.CalculateBottomIndex(
+            make_tuple(n0, waveId_n, blk_n_offset))[I0];
+
+        return make_tuple(c_thread_m, c_thread_n);
+    }
+
+    __host__ __device__ BlockwiseGemmDpp_ak0mak1_bk0nbk1_m0n0m1n1m2n2()
+    {
+        static_assert(AK0MK1BlockDesc::IsKnownAtCompileTime() &&
+                          BK0NK1BlockDesc::IsKnownAtCompileTime(),
+                      "Wrong! Block descriptors should be known at the time of compilation.");
+
+#if defined(__HIP_DEVICE_COMPILE__)
+        // Host wave size can be different than the device one and this assert could fail for host,
+        // but it does matter only for device.
+        static_assert(ThisThreadBlock::GetNumOfThread() == MWaves * NWaves * WaveSize,
+                      "ThisThreadBlock::GetNumOfThread() != MWaves * NWaves * WaveSize\n");
+#endif
+
+        static_assert(MPerBlock % (MPerDpp * MRepeat) == 0,
+                      "Invalid parameters. MPerBlock must be divisible by MPerDpp * MRepeat.");
+        static_assert(NPerBlock % (NPerDpp * NRepeat) == 0,
+                      "Invalid parameters. NPerBlock must be divisible by NPerDpp * NRepeat.");
+    }
+
+    __host__ __device__ static constexpr auto GetCThreadDescriptor_M0_N0_M1_N1_M2_N2()
+    {
+        constexpr auto c_m_n_tblk_lens = dpp_gemm.GetCMNThreadBlkLengths();
+        constexpr auto M               = c_m_n_tblk_lens[I0];
+        constexpr auto N               = c_m_n_tblk_lens[I1];
+
+        return make_naive_tensor_descriptor_packed(
+            make_tuple(Number<MRepeat>{}, Number<NRepeat>{}, I1, I1, M, N));
+    }
+
+    __host__ __device__ static constexpr auto GetCThreadDescriptor_G_M0_N0_M1_N1_M2_N2()
+    {
+        constexpr auto c_m_n_tblk_lens = dpp_gemm.GetCMNThreadBlkLengths();
+        constexpr auto M               = c_m_n_tblk_lens[I0];
+        constexpr auto N               = c_m_n_tblk_lens[I1];
+
+        return make_naive_tensor_descriptor_packed(
+            make_tuple(I1, Number<MRepeat>{}, Number<NRepeat>{}, I1, I1, M, N));
+    }
+
+    __host__ __device__ static constexpr auto GetCBlockDescriptor_M0_N0_M1_N1_M2_N2()
+    {
+        constexpr auto c_block_desc_m0_n0_m1_n1_m2_n2 =
+            make_naive_tensor_descriptor_packed(make_tuple(Number<MRepeat>{},
+                                                           Number<NRepeat>{},
+                                                           Number<MWaves>{},
+                                                           Number<NWaves>{},
+                                                           Number<MPerDpp>{},
+                                                           Number<NPerDpp>{}));
+
+        return c_block_desc_m0_n0_m1_n1_m2_n2;
+    }
+
+    __host__ __device__ static constexpr auto GetCBlockDescriptor_G_M0_N0_M1_N1_M2_N2()
+    {
+        constexpr auto c_block_desc_g_m0_n0_m1_n1_m2_n2 =
+            make_naive_tensor_descriptor_packed(make_tuple(I1,
+                                                           Number<MRepeat>{},
+                                                           Number<NRepeat>{},
+                                                           Number<MWaves>{},
+                                                           Number<NWaves>{},
+                                                           Number<MPerDpp>{},
+                                                           Number<NPerDpp>{}));
+        return c_block_desc_g_m0_n0_m1_n1_m2_n2;
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeCGridDescriptor_M0_N0_M1_N1_M2_N2(const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        const auto M = c_grid_desc_m_n.GetLength(I0);
+        const auto N = c_grid_desc_m_n.GetLength(I1);
+
+        const auto c_grid_desc_m0_n0_m1_n1_m2_n2 = transform_tensor_descriptor(
+            c_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(M / (MWaves * MPerDpp), MWaves, MPerDpp)),
+                       make_unmerge_transform(make_tuple(N / (NWaves * NPerDpp), NWaves, NPerDpp))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 2, 4>{}, Sequence<1, 3, 5>{}));
+
+        return c_grid_desc_m0_n0_m1_n1_m2_n2;
+    }
+
+    template <typename CGridDesc_G_M_N>
+    __host__ __device__ static constexpr auto
+    MakeCGridDescriptor_G_M0_N0_M1_N1_M2_N2(const CGridDesc_G_M_N& c_grid_desc_g_m_n)
+    {
+        const auto G = c_grid_desc_g_m_n.GetLength(I0);
+        const auto M = c_grid_desc_g_m_n.GetLength(I1);
+        const auto N = c_grid_desc_g_m_n.GetLength(I2);
+
+        const auto c_grid_desc_g_m0_n0_m1_n1_m2_n2 = transform_tensor_descriptor(
+            c_grid_desc_g_m_n,
+            make_tuple(make_pass_through_transform(G),
+                       make_unmerge_transform(make_tuple(M / (MWaves * MPerDpp), MWaves, MPerDpp)),
+                       make_unmerge_transform(make_tuple(N / (NWaves * NPerDpp), NWaves, NPerDpp))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+            make_tuple(Sequence<0>{}, Sequence<1, 3, 5>{}, Sequence<2, 4, 6>{}));
+
+        return c_grid_desc_g_m0_n0_m1_n1_m2_n2;
+    }
+
+    __host__ __device__ static constexpr auto MakeABlockDescriptor_M0_M1_M2_K()
+    {
+        return transform_tensor_descriptor(
+            AK0MK1BlockDesc{},
+            make_tuple(
+                make_merge_transform_v3_division_mod(make_tuple(Number<A_K0>{}, Number<A_K1>{})),
+                make_unmerge_transform(
+                    make_tuple(Number<MRepeat>{}, Number<MWaves>{}, Number<MPerDpp>{}))),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}),
+            make_tuple(Sequence<3>{}, Sequence<0, 1, 2>{}));
+    }
+
+    __host__ __device__ static constexpr auto MakeBBlockDescriptor_N0_N1_N2_K()
+    {
+        return transform_tensor_descriptor(
+            BK0NK1BlockDesc{},
+            make_tuple(
+                make_merge_transform_v3_division_mod(make_tuple(Number<B_K0>{}, Number<B_K1>{})),
+                make_unmerge_transform(
+                    make_tuple(Number<NRepeat>{}, Number<NWaves>{}, Number<NPerDpp>{}))),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}),
+            make_tuple(Sequence<3>{}, Sequence<0, 1, 2>{}));
+    }
+
+    static constexpr auto a_block_desc_m0_m1_m2_k = MakeABlockDescriptor_M0_M1_M2_K();
+    static constexpr auto b_block_desc_n0_n1_n2_k = MakeBBlockDescriptor_N0_N1_N2_K();
+
+    template <typename ABlockBuffer, typename BBlockBuffer, typename CThreadBuffer>
+    __device__ void Run(const ABlockBuffer& a_block_buf,
+                        const BBlockBuffer& b_block_buf,
+                        CThreadBuffer& c_thread_buf) const
+    {
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ABDataType>(
+            a_thread_desc_.GetElementSpaceSize());
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, ABDataType>(
+            b_thread_desc_.GetElementSpaceSize());
+
+        static_for<0, MRepeat, 1>{}([&](auto m0) {
+            // read A
+            a_thread_copy_.Run(a_block_desc_m0_m1_m2_k,
+                               make_tuple(m0, I0, I0, I0),
+                               a_block_buf,
+                               a_thread_desc_,
+                               make_tuple(I0, I0, I0, I0),
+                               a_thread_buf);
+
+            static_for<0, NRepeat, 1>{}([&](auto n0) {
+                // read B
+                b_thread_copy_.Run(b_block_desc_n0_n1_n2_k,
+                                   make_tuple(n0, I0, I0, I0),
+                                   b_block_buf,
+                                   b_thread_desc_,
+                                   make_tuple(I0, I0, I0, I0),
+                                   b_thread_buf);
+
+                static_for<0, KPerThread, KPack>{}([&](auto k) {
+                    vector_type<ABDataType, KPack> a_thread_vec;
+                    vector_type<ABDataType, KPack> b_thread_vec;
+
+                    static_for<0, KPack, 1>{}([&](auto i) {
+                        a_thread_vec.template AsType<ABDataType>()(i) = a_thread_buf
+                            [Number<a_thread_desc_.CalculateOffset(make_tuple(0, 0, 0, k + i))>{}];
+                        b_thread_vec.template AsType<ABDataType>()(i) = b_thread_buf
+                            [Number<b_thread_desc_.CalculateOffset(make_tuple(0, 0, 0, k + i))>{}];
+                    });
+
+                    using dpp_input_type =
+                        typename vector_type<ABDataType, dpp_gemm.K1PerDpp>::type;
+
+                    constexpr index_t c_offset =
+                        c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
+
+                    dpp_gemm.template Run(a_thread_vec.template AsType<dpp_input_type>(),
+                                          b_thread_vec.template AsType<dpp_input_type>(),
+                                          c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                });
+            });
+        });
+    }
+
+    protected:
+    // A[M0, M1, M2, KPerThread]
+    static constexpr auto a_thread_desc_ =
+        make_naive_tensor_descriptor_packed(make_tuple(I1, I1, I1, Number<KPerThread>{}));
+
+    // B[N0, N1, N2, KPerThread]
+    static constexpr auto b_thread_desc_ =
+        make_naive_tensor_descriptor_packed(make_tuple(I1, I1, I1, Number<KPerThread>{}));
+
+    // C[M, N, NumRegDpp]
+    static constexpr auto c_thread_desc_ = make_naive_tensor_descriptor_packed(
+        make_tuple(Number<MRepeat>{}, Number<NRepeat>{}, dpp_gemm.GetRegSizePerDpp()));
+
+    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<ABDataType,
+                                                         ABDataType,
+                                                         decltype(a_block_desc_m0_m1_m2_k),
+                                                         decltype(a_thread_desc_),
+                                                         Sequence<1, 1, 1, KPerThread>,
+                                                         Sequence<0, 1, 2, 3>,
+                                                         3,
+                                                         A_K1,
+                                                         A_K1>;
+
+    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<ABDataType,
+                                                         ABDataType,
+                                                         decltype(b_block_desc_n0_n1_n2_k),
+                                                         decltype(b_thread_desc_),
+                                                         Sequence<1, 1, 1, KPerThread>,
+                                                         Sequence<0, 1, 2, 3>,
+                                                         3,
+                                                         B_K1,
+                                                         B_K1>;
+
+    AThreadCopy a_thread_copy_{CalculateAThreadOriginDataIndex_M0_M1_M2_K()};
+    BThreadCopy b_thread_copy_{CalculateBThreadOriginDataIndex_N0_N1_N2_K()};
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
index d5a64d7aa6..1fee9c3225 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
@@ -4,27 +4,13 @@
 #pragma once
 
 #include "ck/utility/common_header.hpp"
+#include "ck/utility/loop_scheduler.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/warp/xdlops_gemm.hpp"
 #include "ck/tensor_description/tensor_adaptor.hpp"
 
 namespace ck {
 
-enum struct LoopScheduler
-{
-    Default,
-    Interwave,
-};
-
-constexpr LoopScheduler make_default_loop_scheduler()
-{
-#if CK_EXPERIMENTAL_DEFAULT_TO_INTER_WAVE_SCHEDULING
-    return LoopScheduler::Interwave;
-#else
-    return LoopScheduler::Default;
-#endif // if CK_EXPERIMENTAL_DEFAULT_TO_INTER_WAVE_SCHEDULING
-}
-
 template <index_t MNXdlPerWave, index_t MNWaves, index_t MNPerXdl, typename TileDesc_K0_MN_K1>
 __host__ __device__ static constexpr auto
 MakeGemmMmaTileDescriptor_MN0_MN1_MN2_K(const TileDesc_K0_MN_K1&)
diff --git a/include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp b/include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp
deleted file mode 100644
index 16ca582b89..0000000000
--- a/include/ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp
+++ /dev/null
@@ -1,18 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-enum struct GemmDlAlgorithm
-{
-    Default, // Uses DOT vector instructions
-    Dpp8,    // Uses DOT vector instructions with DPP8 SEL modifier to reduce data loads from LDS
-};
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
index eedf384cd9..d9f8324bd2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
@@ -11,7 +11,6 @@
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
 #include "ck/host_utility/device_prop.hpp"
@@ -60,7 +59,6 @@ template <
     typename CThreadTransferSrcDstAccessOrder,
     index_t CThreadTransferSrcDstVectorDim,
     index_t CThreadTransferDstScalarPerVector,
-    GemmDlAlgorithm GemmDlAlg = GemmDlAlgorithm::Default,
     enable_if_t<
         is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
             is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
@@ -238,8 +236,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                      BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
                                      CThreadTransferSrcDstAccessOrder,
                                      CThreadTransferSrcDstVectorDim,
-                                     CThreadTransferDstScalarPerVector,
-                                     GemmDlAlg>;
+                                     CThreadTransferDstScalarPerVector>;
 
     using AGridDesc_K0_M0_M1_K1 =
         decltype(GridwiseGemm::MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
@@ -375,8 +372,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         true,
-                                        true,
-                                        GemmDlAlg>;
+                                        true>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -402,8 +398,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         true,
-                                        false,
-                                        GemmDlAlg>;
+                                        false>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -429,8 +424,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         false,
-                                        true,
-                                        GemmDlAlg>;
+                                        true>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -456,8 +450,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
                                         remove_reference_t<CGridDesc_M0_M10_M11_N0_N10_N11>,
                                         remove_reference_t<DefaultBlock2CTileMap>,
                                         false,
-                                        false,
-                                        GemmDlAlg>;
+                                        false>;
 
                 ave_time = launch_and_time_kernel(stream_config,
                                                   kernel,
@@ -492,16 +485,6 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if constexpr(GemmDlAlg == GemmDlAlgorithm::Dpp8)
-        {
-            if(ck::get_device_name() == "gfx1030")
-            {
-                return GridwiseGemm::CheckValidity(
-                    arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);
-            }
-            return false;
-        }
-
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
            ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
            ck::get_device_name() == "gfx1102")
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp
deleted file mode 100644
index 336ce31e82..0000000000
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp
+++ /dev/null
@@ -1,133 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/utility/common_header.hpp"
-#include "ck/tensor_description/tensor_descriptor.hpp"
-#include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
-#include "ck/host_utility/device_prop.hpp"
-#include "ck/host_utility/kernel_launch.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-template <
-    typename ADataType,
-    typename BDataType,
-    typename CDataType,
-    typename AccDataType,
-    typename ALayout,
-    typename BLayout,
-    typename CLayout,
-    typename AElementwiseOperation,
-    typename BElementwiseOperation,
-    typename CElementwiseOperation,
-    GemmSpecialization GemmSpec,
-    index_t BlockSize,
-    index_t MPerBlock,
-    index_t NPerBlock,
-    index_t K0PerBlock,
-    index_t K1,
-    index_t M1PerThread,
-    index_t N1PerThread,
-    index_t KPerThread,
-    typename M1N1ThreadClusterM1Xs,
-    typename M1N1ThreadClusterN1Xs,
-    typename ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
-    typename ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
-    typename ABlockTransferThreadClusterArrangeOrder,
-    typename ABlockTransferSrcAccessOrder,
-    typename ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
-    typename ABlockTransferSrcVectorTensorContiguousDimOrder,
-    typename ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
-    typename BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
-    typename BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
-    typename BBlockTransferThreadClusterArrangeOrder,
-    typename BBlockTransferSrcAccessOrder,
-    typename BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
-    typename BBlockTransferSrcVectorTensorContiguousDimOrder,
-    typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
-    typename CThreadTransferSrcDstAccessOrder,
-    index_t CThreadTransferSrcDstVectorDim,
-    index_t CThreadTransferDstScalarPerVector,
-    enable_if_t<
-        is_same_v<AElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
-            is_same_v<BElementwiseOperation, ck::tensor_operation::element_wise::PassThrough> &&
-            is_same_v<CElementwiseOperation, ck::tensor_operation::element_wise::PassThrough>,
-        bool> = false>
-struct DeviceGemmDlDpp8 : public DeviceGemmDl<ADataType,
-                                              BDataType,
-                                              CDataType,
-                                              AccDataType,
-                                              ALayout,
-                                              BLayout,
-                                              CLayout,
-                                              AElementwiseOperation,
-                                              BElementwiseOperation,
-                                              CElementwiseOperation,
-                                              GemmSpec,
-                                              BlockSize,
-                                              MPerBlock,
-                                              NPerBlock,
-                                              K0PerBlock,
-                                              K1,
-                                              M1PerThread,
-                                              N1PerThread,
-                                              KPerThread,
-                                              M1N1ThreadClusterM1Xs,
-                                              M1N1ThreadClusterN1Xs,
-                                              ABlockTransferThreadSliceLengths_K0_M0_M1_K1,
-                                              ABlockTransferThreadClusterLengths_K0_M0_M1_K1,
-                                              ABlockTransferThreadClusterArrangeOrder,
-                                              ABlockTransferSrcAccessOrder,
-                                              ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1,
-                                              ABlockTransferSrcVectorTensorContiguousDimOrder,
-                                              ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1,
-                                              BBlockTransferThreadSliceLengths_K0_N0_N1_K1,
-                                              BBlockTransferThreadClusterLengths_K0_N0_N1_K1,
-                                              BBlockTransferThreadClusterArrangeOrder,
-                                              BBlockTransferSrcAccessOrder,
-                                              BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1,
-                                              BBlockTransferSrcVectorTensorContiguousDimOrder,
-                                              BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
-                                              CThreadTransferSrcDstAccessOrder,
-                                              CThreadTransferSrcDstVectorDim,
-                                              CThreadTransferDstScalarPerVector,
-                                              GemmDlAlgorithm::Dpp8>
-
-{
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        // clang-format off
-        str << "DeviceGemmDlDpp8"
-            << "<"
-            << BlockSize << ", "
-            << MPerBlock << ", "
-            << NPerBlock << ", "
-            << K0PerBlock << ", "
-            << K1 << ", "
-            << M1PerThread << ", "
-            << N1PerThread << ", "
-            << KPerThread
-            << ">";
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
new file mode 100644
index 0000000000..162f0690b8
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
@@ -0,0 +1,271 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <typename ADataType,
+          typename BDataType,
+          typename CDataType,
+          typename AccDataType,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t KPerBlock,
+          ck::index_t AK1,
+          ck::index_t BK1,
+          ck::index_t MPerDpp,
+          ck::index_t NPerDpp,
+          ck::index_t MDppPerWave,
+          ck::index_t NDppPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          ck::index_t ABlockTransferSrcVectorDim,
+          ck::index_t ABlockTransferSrcScalarPerVector,
+          ck::index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          ck::index_t BBlockTransferSrcVectorDim,
+          ck::index_t BBlockTransferSrcScalarPerVector,
+          ck::index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsAddExtraN,
+          ck::index_t CThreadTransferSrcDstVectorDim,
+          ck::index_t CThreadTransferDstScalarPerVector,
+          ck::index_t NumPrefetch         = 1,
+          ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
+struct DeviceGemmDpp : public DeviceGemm<ALayout,
+                                         BLayout,
+                                         CLayout,
+                                         ADataType,
+                                         BDataType,
+                                         CDataType,
+                                         AElementwiseOperation,
+                                         BElementwiseOperation,
+                                         CElementwiseOperation>
+{
+    using GridwiseGemm = GridwiseGemm_ak0mak1_bk0nbk1_mn_dpp<
+        BlockSize,
+        ADataType,
+        AccDataType,
+        CDataType,
+        InMemoryDataOperationEnum::Set,
+        ALayout,
+        BLayout,
+        CLayout,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        GemmSpec,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        MPerDpp,
+        NPerDpp,
+        AK1,
+        BK1,
+        MDppPerWave,
+        NDppPerWave,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false, // AThreadTransferSrcResetCoordinateAfterRun,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false, // BThreadTransferSrcResetCoordinateAfterRun,
+        BBlockLdsAddExtraN,
+        Sequence<0, 2, 4, 1, 3, 5>, // CThreadTransferSrcDstAccessOrder,
+        CThreadTransferSrcDstVectorDim,
+        CThreadTransferDstScalarPerVector,
+        NumPrefetch,
+        PipelineVer>;
+
+    using Argument = typename GridwiseGemm::Argument;
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& karg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                karg.Print();
+            }
+
+            if(!GridwiseGemm::CheckValidity(karg))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemm_k0mk1_k0nk1_mn_dpp has invalid setting");
+            }
+
+            const auto [gdx, gdy, gdz] = GridwiseGemm::CalculateGridSize(karg.M, karg.N);
+
+            float ave_time = 0;
+
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(karg.K))
+            {
+                const auto kernel = kernel_gemm_dpp<GridwiseGemm, true>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
+            }
+            else
+            {
+                const auto kernel = kernel_gemm_dpp<GridwiseGemm, false>;
+
+                ave_time = launch_and_time_kernel(
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
+            }
+
+            return ave_time;
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& karg)
+    {
+        if(ck::get_device_name() == "gfx1030")
+        {
+            return GridwiseGemm::CheckValidity(karg);
+        }
+        return false;
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const ADataType* p_a,
+                             const BDataType* p_b,
+                             CDataType* p_c,
+                             index_t M,
+                             index_t N,
+                             index_t K,
+                             index_t StrideA,
+                             index_t StrideB,
+                             index_t StrideC,
+                             AElementwiseOperation,
+                             BElementwiseOperation,
+                             CElementwiseOperation)
+    {
+        return Argument{p_a, p_b, p_c, M, N, K, StrideA, StrideB, StrideC};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
+                                                      const void* p_b,
+                                                      void* p_c,
+                                                      index_t M,
+                                                      index_t N,
+                                                      index_t K,
+                                                      index_t StrideA,
+                                                      index_t StrideB,
+                                                      index_t StrideC,
+                                                      AElementwiseOperation,
+                                                      BElementwiseOperation,
+                                                      CElementwiseOperation) override
+    {
+        return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
+                                          static_cast<const BDataType*>(p_b),
+                                          static_cast<CDataType*>(p_c),
+                                          M,
+                                          N,
+                                          K,
+                                          StrideA,
+                                          StrideB,
+                                          StrideC);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceGemmDpp"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerDpp << ", "
+            << NPerDpp << ", "
+            << MDppPerWave << ", "
+            << MDppPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1
+            << ">"
+            << " NumPrefetch: "
+            << NumPrefetch << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
index 1d1bb6ed2d..1da7236978 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp
@@ -7,11 +7,9 @@
 #include "ck/tensor_description/multi_index_transform_helper.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_dl_algorithm.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_gemm_dl_v2r3.hpp"
-#include "ck/tensor_operation/gpu/block/blockwise_gemm_dl_dpp8.hpp"
 #include "ck/tensor_operation/gpu/block/blockwise_tensor_slice_transfer_v5r1.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_set.hpp"
@@ -19,8 +17,6 @@
 
 namespace ck {
 
-using GemmDlAlgorithm = tensor_operation::device::GemmDlAlgorithm;
-
 template <typename GridwiseGemm,
           typename FloatAB,
           typename FloatC,
@@ -29,8 +25,7 @@ template <typename GridwiseGemm,
           typename CGridDesc_M0_M10_M11_N0_N10_N11,
           typename Block2CTileMap,
           bool HasMainKBlockLoop,
-          bool HasDoubleTailKBlockLoop,
-          GemmDlAlgorithm GemmDlAlg = GemmDlAlgorithm::Default>
+          bool HasDoubleTailKBlockLoop>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
@@ -43,13 +38,6 @@ __global__ void
                             const CGridDesc_M0_M10_M11_N0_N10_N11 c_grid_desc_m0_m10_m11_n0_n10_n11,
                             const Block2CTileMap block_2_ctile_map)
 {
-// DPP8 is currently only supported on gfx1030
-#if !defined(__gfx1030__)
-    if(GemmDlAlg == GemmDlAlgorithm::Dpp8)
-    {
-        return;
-    }
-#endif
     constexpr index_t shared_block_size =
         GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB);
 
@@ -100,8 +88,7 @@ template <index_t BlockSize,
           typename BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1,
           typename CThreadTransferSrcDstAccessOrder,
           index_t CThreadTransferSrcDstVectorDim,
-          index_t CThreadTransferDstScalarPerVector,
-          GemmDlAlgorithm GemmDlAlg = GemmDlAlgorithm::Default>
+          index_t CThreadTransferDstScalarPerVector>
 struct GridwiseGemmDl_km_kn_mn_v1r3
 {
     static constexpr auto I0 = Number<0>{};
@@ -257,45 +244,6 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
             c_grid_desc_m_n);
     }
 
-    template <typename ABlockDesc_BK0_BM_BK1, typename BBlockDesc_BK0_BN_BK1>
-    __host__ __device__ static constexpr auto GetBlockwiseGemm()
-    {
-        if constexpr(GemmDlAlg == GemmDlAlgorithm::Dpp8)
-        {
-            return BlockwiseGemmDlDpp8_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_loop_BM0_BN0<
-                BlockSize,
-                FloatAB,
-                FloatAB,
-                FloatAcc,
-                ABlockDesc_BK0_BM_BK1,
-                BBlockDesc_BK0_BN_BK1,
-                M1PerThreadM111,
-                N1PerThreadN111,
-                KPerThread,
-                M11N11ThreadClusterM110Xs,
-                M11N11ThreadClusterN110Xs,
-                M1PerThreadM111,
-                N1PerThreadN111>{};
-        }
-        else
-        {
-            return BlockwiseGemmDl_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
-                BlockSize,
-                FloatAB,
-                FloatAB,
-                FloatAcc,
-                ABlockDesc_BK0_BM_BK1,
-                BBlockDesc_BK0_BN_BK1,
-                M1PerThreadM111,
-                N1PerThreadN111,
-                KPerThread,
-                M11N11ThreadClusterM110Xs,
-                M11N11ThreadClusterN110Xs,
-                M1PerThreadM111,
-                N1PerThreadN111>{};
-        }
-    }
-
     using AGridDesc_K0_M0_M1_K1 = decltype(MakeAGridDescriptor_K0_M0_M1_K1(AGridDesc_K0_M_K1{}));
     using BGridDesc_K0_N0_N1_K1 = decltype(MakeBGridDescriptor_K0_N0_N1_K1(BGridDesc_K0_N_K1{}));
     using CGridDesc_M0_M10_M11_N0_N10_N11 =
@@ -424,7 +372,20 @@ struct GridwiseGemmDl_km_kn_mn_v1r3
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         const auto blockwise_gemm =
-            GetBlockwiseGemm<decltype(a_k0_m_k1_block_desc), decltype(b_k0_n_k1_block_desc)>();
+            BlockwiseGemmDl_A_BK0_BM_BK1_B_BK0_BN_BK1_C_BM0_BM1_BN0_BN1_pipeline_BM0_2_BN0_2<
+                BlockSize,
+                FloatAB,
+                FloatAB,
+                FloatAcc,
+                decltype(a_k0_m_k1_block_desc),
+                decltype(b_k0_n_k1_block_desc),
+                M1PerThreadM111,
+                N1PerThreadN111,
+                KPerThread,
+                M11N11ThreadClusterM110Xs,
+                M11N11ThreadClusterN110Xs,
+                M1PerThreadM111,
+                N1PerThreadN111>{};
 
         constexpr auto c_m10_m11_n10_n11_thread_tensor_lengths =
             decltype(blockwise_gemm)::GetCThreadTensorLengths_BM0_BM1_BN0_BN1();
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
new file mode 100644
index 0000000000..d39f0ba30b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
@@ -0,0 +1,701 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_dpp.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v4r1.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm, bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+#if CK_USE_WAVES_PER_EU
+        __attribute__((amdgpu_waves_per_eu(CK_MIN_WAVES_PER_EU, CK_MAX_WAVES_PER_EU)))
+#endif
+        kernel_gemm_dpp(const typename GridwiseGemm::Argument karg)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1030__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    const auto a_grid_desc_ak0_m_ak1 = amd_wave_read_first_lane(
+        GridwiseGemm::MakeAGridDescriptor_AK0_M_AK1(karg.M, karg.K, karg.AK0, karg.StrideA));
+    const auto b_grid_desc_bk0_n_bk1 = amd_wave_read_first_lane(
+        GridwiseGemm::MakeBGridDescriptor_BK0_N_BK1(karg.K, karg.N, karg.BK0, karg.StrideB));
+    const auto c_grid_desc_m_n = amd_wave_read_first_lane(
+        GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC));
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(karg.p_a_grid,
+                                                  karg.p_b_grid,
+                                                  karg.p_c_grid,
+                                                  p_shared,
+                                                  a_grid_desc_ak0_m_ak1,
+                                                  b_grid_desc_bk0_n_bk1,
+                                                  c_grid_desc_m_n);
+#else
+    ignore = karg;
+#endif
+}
+
+template <index_t BlockSize,
+          typename ABDataType,
+          typename AccDataType,
+          typename CDataType,
+          InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation,
+          tensor_operation::device::GemmSpecialization GemmSpec,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t MPerDpp,
+          index_t NPerDpp,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t MDppPerWave,
+          index_t NDppPerWave,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_K1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_K1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          bool BBlockLdsExtraN,
+          typename CThreadTransferSrcDstAccessOrder,
+          index_t CThreadTransferSrcDstVectorDim,
+          index_t CThreadTransferDstScalarPerVector,
+          index_t NumGemmKPrefetchStage = 1,
+          PipelineVersion PipelineVer   = PipelineVersion::v1>
+struct GridwiseGemm_ak0mak1_bk0nbk1_mn_dpp
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto AK1         = Number<AK1Value>{};
+    static constexpr auto BK1         = Number<BK1Value>{};
+    static constexpr auto AK0PerBlock = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0PerBlock = Number<KPerBlock / BK1Value>{};
+
+    static constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    using Block2CTileMap = BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock>;
+
+    __host__ static auto CalculateGridSize(index_t M, index_t N)
+    {
+        return std::make_tuple(Block2CTileMap::CalculateGridSize(M, N), 1, 1);
+    }
+
+    __host__ static auto CalculateMPadded(index_t M)
+    {
+        return math::integer_divide_ceil(M, MPerBlock) * MPerBlock;
+    }
+
+    __host__ static auto CalculateNPadded(index_t N)
+    {
+        return math::integer_divide_ceil(N, NPerBlock) * NPerBlock;
+    }
+
+    __host__ static auto CalculateAK0(index_t K) { return math::integer_divide_floor(K, AK1Value); }
+    __host__ static auto CalculateBK0(index_t K) { return math::integer_divide_floor(K, BK1Value); }
+
+    // Argument
+    struct Problem
+    {
+        __host__ Problem(index_t M_,
+                         index_t N_,
+                         index_t K_,
+                         index_t StrideA_,
+                         index_t StrideB_,
+                         index_t StrideC_)
+            : M{M_},
+              N{N_},
+              K{K_},
+              StrideA{StrideA_},
+              StrideB{StrideB_},
+              StrideC{StrideC_},
+              MPadded{CalculateMPadded(M_)},
+              NPadded{CalculateNPadded(N_)},
+              AK0{CalculateAK0(K)},
+              BK0{CalculateBK0(K)}
+        {
+        }
+
+        __host__ void Print() const
+        {
+            std::cout << "problem {"
+                      << "M:" << M << ", "
+                      << "N:" << N << ", "
+                      << "K:" << K << ", "
+                      << "SA:" << StrideA << ", "
+                      << "SB:" << StrideB << ", "
+                      << "SC:" << StrideC << ", "
+                      << "MP:" << MPadded << ", "
+                      << "NP:" << NPadded << ", "
+                      << "AK0:" << AK0 << ", "
+                      << "BK0:" << BK0 << "}" << std::endl;
+        }
+
+        index_t M;
+        index_t N;
+        index_t K;
+        index_t StrideA;
+        index_t StrideB;
+        index_t StrideC;
+        index_t MPadded;
+        index_t NPadded;
+        index_t AK0;
+        index_t BK0;
+    };
+
+    // Argument
+    struct Argument : public Problem, public tensor_operation::device::BaseArgument
+    {
+        __host__ Argument(const ABDataType* p_a_grid_,
+                          const ABDataType* p_b_grid_,
+                          CDataType* p_c_grid_,
+                          index_t M_,
+                          index_t N_,
+                          index_t K_,
+                          index_t StrideA_,
+                          index_t StrideB_,
+                          index_t StrideC_)
+            : Problem{M_, N_, K_, StrideA_, StrideB_, StrideC_},
+              p_a_grid{p_a_grid_},
+              p_b_grid{p_b_grid_},
+              p_c_grid{p_c_grid_}
+        {
+        }
+
+        const ABDataType* p_a_grid;
+        const ABDataType* p_b_grid;
+        CDataType* p_c_grid;
+    };
+
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage>())>;
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = [&]() {
+            if constexpr(ABlockLdsExtraM)
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(Number<AK0PerBlock>{}, Number<MPerBlock>{}, AK1),
+                    make_tuple(Number<MPerBlock + 1>{} * AK1, AK1, I1));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_aligned(
+                    make_tuple(Number<AK0PerBlock>{}, Number<MPerBlock>{}, AK1), max_lds_align);
+            }
+        }();
+
+        return a_block_desc_ak0_m_ak1;
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = [&]() {
+            if constexpr(BBlockLdsExtraN)
+            {
+                return make_naive_tensor_descriptor(
+                    make_tuple(Number<BK0PerBlock>{}, Number<NPerBlock>{}, BK1),
+                    make_tuple(Number<NPerBlock + 1>{} * BK1, BK1, I1));
+            }
+            else
+            {
+                return make_naive_tensor_descriptor_aligned(
+                    make_tuple(Number<BK0PerBlock>{}, Number<NPerBlock>{}, BK1), max_lds_align);
+            }
+        }();
+
+        return b_block_desc_bk0_n_bk1;
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        return (a_block_space_size_aligned + b_block_space_size_aligned) * sizeof(ABDataType);
+    }
+
+    __host__ static constexpr bool CheckValidity(const Problem& problem)
+    {
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(AK1)>>::value,
+                      "Wrong! AK1 must be known at the time of compilation.");
+        static_assert(is_known_at_compile_time<remove_cv_t<decltype(BK1)>>::value,
+                      "Wrong! BK1 must be known at the time of compilation.");
+
+        static_assert(
+            MPerBlock % (MPerDpp * MDppPerWave) == 0,
+            "Invalid tuning parameters! MPerBlock must be divisible by MPerDpp * MDppPerWave.");
+        static_assert(
+            NPerBlock % (NPerDpp * NDppPerWave) == 0,
+            "Invalid tuning parameters! NPerBlock must be divisible by NPerDpp * NDppPerWave.");
+
+        static_assert(
+            KPerBlock % AK1Value == 0 && KPerBlock % BK1Value == 0,
+            "Invalid tuning parameters! KPerBlock must be divisible by both AK1 and BK1.");
+
+        static_assert(AK1Value % ABlockTransferDstScalarPerVector_K1 == 0,
+                      "Invalid tuning parameters! AK1Value must be divisible by "
+                      "ABlockTransferDstScalarPerVector_K1");
+
+        static_assert(BK1Value % BBlockTransferDstScalarPerVector_K1 == 0,
+                      "Invalid tuning parameters! BK1Value must be divisible by "
+                      "BBlockTransferDstScalarPerVector_K1");
+
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::MPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.M % MPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(!(GemmSpec == tensor_operation::device::GemmSpecialization::NPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::NKPadding ||
+                       GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding))
+        {
+            if(!(problem.N % NPerBlock == 0))
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            if(problem.K % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.M % ABlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            if(problem.N % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            if(problem.K % BBlockTransferSrcScalarPerVector != 0)
+            {
+                return false;
+            }
+        }
+
+        if(problem.K % KPerBlock != 0)
+        {
+            return false;
+        }
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = problem.K / KPerBlock;
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const auto num_loop = K / KPerBlock;
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    template <typename CGridDesc>
+    __host__ __device__ static constexpr auto
+    MakeCGridDescriptor_M0_N0_M1_N1_M2_N2(const CGridDesc& c_grid_desc_m_n)
+    {
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        constexpr index_t KPack = math::max(
+            math::lcm(AK1, BK1), DppSelector<ABDataType, MPerDpp, NPerDpp>::selected_dpp.k_per_dpp);
+
+        using BlockwiseGemm =
+            BlockwiseGemmDpp_ak0mak1_bk0nbk1_m0n0m1n1m2n2<BlockSize,
+                                                          ABDataType,
+                                                          AccDataType,
+                                                          decltype(a_block_desc_ak0_m_ak1),
+                                                          decltype(b_block_desc_bk0_n_bk1),
+                                                          MPerDpp,
+                                                          NPerDpp,
+                                                          MDppPerWave,
+                                                          NDppPerWave,
+                                                          KPack>;
+
+        return BlockwiseGemm::MakeCGridDescriptor_M0_N0_M1_N1_M2_N2(c_grid_desc_m_n);
+    }
+
+    static constexpr auto matrix_padder =
+        ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+            MPerBlock, NPerBlock, KPerBlock};
+
+    __device__ static auto
+    MakeAGridDescriptor_AK0_M_AK1(index_t M, index_t K, index_t AK0, index_t StrideA)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ALayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I1, StrideA));
+            }
+        }();
+
+        const auto a_grid_desc_m_k = matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+        return transform_tensor_descriptor(
+            a_grid_desc_m_k,
+            make_tuple(make_unmerge_transform(make_tuple(AK0, AK1Value)),
+                       make_pass_through_transform(M)),
+            make_tuple(Sequence<1>{}, Sequence<0>{}),
+            make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    __device__ static auto
+    MakeBGridDescriptor_BK0_N_BK1(index_t K, index_t N, index_t BK0, index_t StrideB)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N, K), make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(N, K), make_tuple(StrideB, I1));
+            }
+        }();
+
+        const auto b_grid_desc_n_k = matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+        return transform_tensor_descriptor(
+            b_grid_desc_n_k,
+            make_tuple(make_pass_through_transform(N),
+                       make_unmerge_transform(make_tuple(BK0, BK1Value))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<1>{}, Sequence<0, 2>{}));
+    }
+
+    __device__ static auto MakeCGridDescriptor_M_N(index_t M, index_t N, index_t StrideC)
+    {
+        const auto c_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(StrideC, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, CLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(M, N), make_tuple(I1, StrideC));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(c_grid_desc_mraw_nraw);
+    }
+
+    template <bool HasMainKBlockLoop,
+              typename AGridDesc_AK0_M_AK1,
+              typename BGridDesc_BK0_N_BK1,
+              typename CGridDesc_M_N>
+    __device__ static void Run(const ABDataType* __restrict__ p_a_grid,
+                               const ABDataType* __restrict__ p_b_grid,
+                               CDataType* __restrict__ p_c_grid,
+                               void* __restrict__ p_shared,
+                               const AGridDesc_AK0_M_AK1& a_grid_desc_ak0_m_ak1,
+                               const BGridDesc_BK0_N_BK1& b_grid_desc_bk0_n_bk1,
+                               const CGridDesc_M_N& c_grid_desc_m_n)
+    {
+        const auto c_grid_desc_m0_n0_m1_n1_m2_n2 =
+            MakeCGridDescriptor_M0_N0_M1_N1_M2_N2(c_grid_desc_m_n);
+
+        const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_a_grid, a_grid_desc_ak0_m_ak1.GetElementSpaceSize());
+        const auto b_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_b_grid, b_grid_desc_bk0_n_bk1.GetElementSpaceSize());
+        auto c_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_c_grid, c_grid_desc_m0_n0_m1_n1_m2_n2.GetElementSpaceSize());
+
+        const AElementwiseOperation a_element_op{};
+        const BElementwiseOperation b_element_op{};
+        const CElementwiseOperation c_element_op{};
+
+        const auto block_2_ctile_map =
+            Block2CTileMap{c_grid_desc_m_n.GetLength(I0), c_grid_desc_m_n.GetLength(I1)};
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        if(!block_2_ctile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(c_grid_desc_m0_n0_m1_n1_m2_n2.GetLength(I0),
+                          c_grid_desc_m0_n0_m1_n1_m2_n2.GetLength(I1))))
+        {
+            return;
+        }
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+        const index_t n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        auto a_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                AElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<AK0PerBlock, MPerBlock, AK1>,
+                                                ABlockTransferThreadClusterLengths_K0_M_K1,
+                                                ABlockTransferThreadClusterArrangeOrder,
+                                                ABDataType,
+                                                ABDataType,
+                                                decltype(a_grid_desc_ak0_m_ak1),
+                                                decltype(a_block_desc_ak0_m_ak1),
+                                                ABlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                ABlockTransferSrcVectorDim,
+                                                2,
+                                                ABlockTransferSrcScalarPerVector,
+                                                ABlockTransferDstScalarPerVector_K1,
+                                                1,
+                                                1,
+                                                AThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                a_grid_desc_ak0_m_ak1,
+                make_multi_index(0, m_block_data_idx_on_grid, 0),
+                a_element_op,
+                a_block_desc_ak0_m_ak1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        auto b_blockwise_copy =
+            ThreadGroupTensorSliceTransfer_v4r1<ThisThreadBlock,
+                                                BElementwiseOperation,
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                InMemoryDataOperationEnum::Set,
+                                                Sequence<BK0PerBlock, NPerBlock, BK1>,
+                                                BBlockTransferThreadClusterLengths_K0_N_K1,
+                                                BBlockTransferThreadClusterArrangeOrder,
+                                                ABDataType,
+                                                ABDataType,
+                                                decltype(b_grid_desc_bk0_n_bk1),
+                                                decltype(b_block_desc_bk0_n_bk1),
+                                                BBlockTransferSrcAccessOrder,
+                                                Sequence<1, 0, 2>,
+                                                BBlockTransferSrcVectorDim,
+                                                2,
+                                                BBlockTransferSrcScalarPerVector,
+                                                BBlockTransferDstScalarPerVector_K1,
+                                                1,
+                                                1,
+                                                BThreadTransferSrcResetCoordinateAfterRun,
+                                                true,
+                                                NumGemmKPrefetchStage>(
+                b_grid_desc_bk0_n_bk1,
+                make_multi_index(0, n_block_data_idx_on_grid, 0),
+                b_element_op,
+                b_block_desc_bk0_n_bk1,
+                make_multi_index(0, 0, 0),
+                ck::tensor_operation::element_wise::PassThrough{});
+
+        // GEMM definition
+        //   c_mtx += transpose(a_mtx) * b_mtx
+        //     a_mtx[AK0PerBlock, MPerBlock] is in LDS
+        //     b_mtx[BK0PerBlock, NPerBlock] is in LDS
+        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+        //       register
+        constexpr index_t KPack = math::max(
+            math::lcm(AK1, BK1), DppSelector<ABDataType, MPerDpp, NPerDpp>::selected_dpp.k_per_dpp);
+        auto blockwise_gemm =
+            BlockwiseGemmDpp_ak0mak1_bk0nbk1_m0n0m1n1m2n2<BlockSize,
+                                                          ABDataType,
+                                                          AccDataType,
+                                                          decltype(a_block_desc_ak0_m_ak1),
+                                                          decltype(b_block_desc_bk0_n_bk1),
+                                                          MPerDpp,
+                                                          NPerDpp,
+                                                          MDppPerWave,
+                                                          NDppPerWave,
+                                                          KPack>();
+
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ABDataType*>(p_shared) + a_block_space_size_aligned,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(AK0PerBlock, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(BK0PerBlock, 0, 0);
+
+        // gridwise GEMM pipeline
+        const auto AK0 = a_grid_desc_ak0_m_ak1.GetLength(I0);
+        // (AK0 / AK0PerBlock) is always equal to (BK0 / BK0PerBlock)
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(AK0 / AK0PerBlock);
+
+        GridwiseGemmPipe::template Run<HasMainKBlockLoop>(a_grid_desc_ak0_m_ak1,
+                                                          a_block_desc_ak0_m_ak1,
+                                                          a_blockwise_copy,
+                                                          a_grid_buf,
+                                                          a_block_buf,
+                                                          a_block_slice_copy_step,
+                                                          b_grid_desc_bk0_n_bk1,
+                                                          b_block_desc_bk0_n_bk1,
+                                                          b_blockwise_copy,
+                                                          b_grid_buf,
+                                                          b_block_buf,
+                                                          b_block_slice_copy_step,
+                                                          blockwise_gemm,
+                                                          c_thread_buf,
+                                                          num_k_block_main_loop);
+
+        // output: register to global memory
+        {
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_n2 =
+                blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_N2();
+
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_n2 =
+                blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_N2();
+
+            constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_n2.GetLength(I0);
+            constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_n2.GetLength(I1);
+            constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_n2.GetLength(I2);
+            constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_n2.GetLength(I3);
+            constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_n2.GetLength(I4);
+            constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_n2.GetLength(I5);
+
+            constexpr auto MPerThread = c_thread_desc_m0_n0_m1_n1_m2_n2.GetLength(I4);
+            constexpr auto NPerThread = c_thread_desc_m0_n0_m1_n1_m2_n2.GetLength(I5);
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block =
+                blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0);
+
+            const index_t m_thread_data_on_grid =
+                m_block_data_idx_on_grid + c_thread_mtx_on_block[I0];
+
+            const index_t n_thread_data_on_grid =
+                n_block_data_idx_on_grid + c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_grid_to_m0_m1_m2_adaptor = make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(M0, M1, M2))),
+                make_tuple(Sequence<0, 1, 2>{}),
+                make_tuple(Sequence<0>{}));
+
+            const auto m_thread_data_on_grid_idx =
+                m_thread_data_on_grid_to_m0_m1_m2_adaptor.CalculateBottomIndex(
+                    make_multi_index(m_thread_data_on_grid));
+
+            const auto n_thread_data_on_grid_to_n0_n1_n2_adaptor = make_single_stage_tensor_adaptor(
+                make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                make_tuple(Sequence<0, 1, 2>{}),
+                make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_grid_idx =
+                n_thread_data_on_grid_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                    make_multi_index(n_thread_data_on_grid));
+
+            auto c_thread_copy =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   CDataType,
+                                                   decltype(c_thread_desc_m0_n0_m1_n1_m2_n2),
+                                                   decltype(c_grid_desc_m0_n0_m1_n1_m2_n2),
+                                                   CElementwiseOperation,
+                                                   Sequence<M0, N0, I1, I1, MPerThread, NPerThread>,
+                                                   CThreadTransferSrcDstAccessOrder,
+                                                   CThreadTransferSrcDstVectorDim,
+                                                   CThreadTransferDstScalarPerVector,
+                                                   CGlobalMemoryDataOperation,
+                                                   1,
+                                                   true>{
+                    c_grid_desc_m0_n0_m1_n1_m2_n2,
+                    make_multi_index(m_thread_data_on_grid_idx[I0],
+                                     n_thread_data_on_grid_idx[I0],
+                                     m_thread_data_on_grid_idx[I1],
+                                     n_thread_data_on_grid_idx[I1],
+                                     m_thread_data_on_grid_idx[I2],
+                                     n_thread_data_on_grid_idx[I2]),
+                    c_element_op};
+
+            c_thread_copy.Run(c_thread_desc_m0_n0_m1_n1_m2_n2,
+                              make_tuple(I0, I0, I0, I0, I0, I0),
+                              c_thread_buf,
+                              c_grid_desc_m0_n0_m1_n1_m2_n2,
+                              c_grid_buf);
+        }
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp
index d1209636de..754a3e89c9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp
@@ -4,7 +4,8 @@
 #pragma once
 
 #include "ck/utility/common_header.hpp"
-#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/utility/loop_scheduler.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 
 namespace ck {
 
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp
deleted file mode 100644
index d0d214381d..0000000000
--- a/include/ck/tensor_operation/gpu/thread/threadwise_contraction_dl_dpp8.hpp
+++ /dev/null
@@ -1,136 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/utility/amd_gemm_dpp.hpp"
-#include "ck/utility/common_header.hpp"
-#include "ck/utility/inner_product_dpp8.hpp"
-#include "ck/utility/math.hpp"
-
-namespace ck {
-
-/**
- * Threadwise contraction using dot instructions with DPP8 modifier.
- *
- * Assumptions:
- *   1. `AThreadDesc_TK0_TM0_TM1_TK1`, `BThreadDesc_TK0_TN0_TN1_TK1`, `CThreadDesc_TM0_TM1_TN0_TN1`
- *      are known at compile-time;
- *   2. `AOriginIdx`, `BOriginIdx`, `COriginIdx` are known at compile-time;
- *   3. `TM0` is equal to 1 and `TN0` is equal to 1;
- *   4. When `ShareA` is set (unset, respectively), `TM1` (`TN1`, respectively) is divisible by
- *      the size of the lane group (`dpp8::lane_group_size`).
- */
-template <typename FloatA,
-          typename FloatB,
-          typename FloatC,
-          typename AThreadDesc_TK0_TM0_TM1_TK1,
-          typename BThreadDesc_TK0_TN0_TN1_TK1,
-          typename CThreadDesc_TM0_TM1_TN0_TN1,
-          typename TKLengths,
-          typename TMLengths,
-          typename TNLengths,
-          bool ShareA,
-          typename enable_if<AThreadDesc_TK0_TM0_TM1_TK1::IsKnownAtCompileTime() &&
-                                 BThreadDesc_TK0_TN0_TN1_TK1::IsKnownAtCompileTime() &&
-                                 CThreadDesc_TM0_TM1_TN0_TN1::IsKnownAtCompileTime(),
-                             bool>::type = false>
-struct ThreadwiseContractionDlDpp8_A_TK0_TM0_TM1_TK1_B_TK0_TN0_TN1_TK1_C_TM0_TM1_TN0_TN1
-{
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-
-    static constexpr index_t TK0 = TKLengths{}[I0];
-    static constexpr index_t TK1 = TKLengths{}[I1];
-    static constexpr index_t TM0 = TMLengths{}[I0];
-    static constexpr index_t TM1 = TMLengths{}[I1];
-    static constexpr index_t TN0 = TNLengths{}[I0];
-    static constexpr index_t TN1 = TNLengths{}[I1];
-
-    static_assert(TM0 == 1 && TN0 == 1);
-
-    static_assert((ShareA && TM1 % dpp8::lane_group_size == 0) ||
-                  (!ShareA && TN1 % dpp8::lane_group_size == 0));
-    static constexpr index_t shared_elems_per_lane =
-        ShareA ? TM1 / dpp8::lane_group_size : TN1 / dpp8::lane_group_size;
-
-    __device__ constexpr ThreadwiseContractionDlDpp8_A_TK0_TM0_TM1_TK1_B_TK0_TN0_TN1_TK1_C_TM0_TM1_TN0_TN1()
-    {
-        static_assert(AThreadDesc_TK0_TM0_TM1_TK1::IsKnownAtCompileTime() &&
-                          BThreadDesc_TK0_TN0_TN1_TK1::IsKnownAtCompileTime() &&
-                          CThreadDesc_TM0_TM1_TN0_TN1::IsKnownAtCompileTime(),
-                      "wrong! Desc should be known at compile-time");
-
-        static_assert(TKLengths::Size() == 2 && TMLengths::Size() == 2 && TNLengths::Size() == 2,
-                      "wrong!");
-    }
-
-    template <typename ABuffer,
-              typename AOriginIdx,
-              typename BBuffer,
-              typename BOriginIdx,
-              typename CBuffer,
-              typename COriginIdx>
-    __device__ static void Run(const ABuffer& a_buf,
-                               AOriginIdx,
-                               const BBuffer& b_buf,
-                               BOriginIdx,
-                               CBuffer& c_buf,
-                               COriginIdx)
-    {
-        static_assert(is_known_at_compile_time<remove_cvref_t<AOriginIdx>>::value &&
-                          is_known_at_compile_time<remove_cvref_t<BOriginIdx>>::value &&
-                          is_known_at_compile_time<remove_cvref_t<COriginIdx>>::value,
-                      "wrong! AOriginIdx, BOriginIdx, COringinIdx should be known at compile-time");
-
-        static_assert(
-            is_same<remove_cvref_t<typename ABuffer::type>, remove_cvref_t<FloatA>>::value &&
-            is_same<remove_cvref_t<typename BBuffer::type>, remove_cvref_t<FloatB>>::value &&
-            is_same<remove_cvref_t<typename CBuffer::type>, remove_cvref_t<FloatC>>::value &&
-            "wrong! inconsistent type");
-
-        constexpr auto a_origin_idx = to_multi_index(AOriginIdx{});
-        constexpr auto b_origin_idx = to_multi_index(BOriginIdx{});
-        constexpr auto c_origin_idx = to_multi_index(COriginIdx{});
-
-        static_for<0, TK0, 1>{}([&](auto tk0) {
-            static_for<0, TM1, 1>{}([&](auto tm1) {
-                static_for<0, TN1, 1>{}([&](auto tn1) {
-                    vector_type<FloatA, TK1> a_vec;
-                    vector_type<FloatB, TK1> b_vec;
-
-                    static_for<0, TK1, 1>{}([&](auto tk1) {
-                        constexpr index_t local_tm1 = ShareA ? tm1 % shared_elems_per_lane : tm1;
-                        constexpr index_t a_offset  = AThreadDesc_TK0_TM0_TM1_TK1{}.CalculateOffset(
-                            a_origin_idx + make_multi_index(tk0, 0, local_tm1, tk1));
-
-                        constexpr index_t local_tn1 = ShareA ? tn1 : tn1 % shared_elems_per_lane;
-                        constexpr index_t b_offset  = BThreadDesc_TK0_TN0_TN1_TK1{}.CalculateOffset(
-                            b_origin_idx + make_multi_index(tk0, 0, local_tn1, tk1));
-
-                        a_vec.template AsType<FloatA>()(tk1) = a_buf[Number<a_offset>{}];
-                        b_vec.template AsType<FloatB>()(tk1) = b_buf[Number<b_offset>{}];
-                    });
-
-                    using a_vector_t = typename vector_type<FloatA, TK1>::type;
-                    using b_vector_t = typename vector_type<FloatB, TK1>::type;
-
-                    constexpr index_t c_offset = CThreadDesc_TM0_TM1_TN0_TN1{}.CalculateOffset(
-                        c_origin_idx + make_multi_index(0, tm1, 0, tn1));
-
-                    constexpr int src_lane =
-                        ShareA ? (tm1 / shared_elems_per_lane) % dpp8::lane_group_size
-                               : (tn1 / shared_elems_per_lane) % dpp8::lane_group_size;
-
-                    dpp8::inner_product_dpp<a_vector_t, b_vector_t, FloatC, src_lane, ShareA>(
-                        a_vec.template AsType<a_vector_t>()[I0],
-                        b_vec.template AsType<b_vector_t>()[I0],
-                        c_buf(Number<c_offset>{}));
-                });
-            });
-        });
-    }
-};
-
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp b/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp
new file mode 100644
index 0000000000..03a4d17c9b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp
@@ -0,0 +1,322 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/amd_gemm_dpp.hpp"
+#include "ck/utility/common_header.hpp"
+#include "ck/utility/math.hpp"
+
+namespace ck {
+
+enum struct DppInstr
+{
+    dpp8_f16_16x16x2 = 0,
+    dpp8_f16_8x32x2,
+    dpp8_f16_32x8x2
+};
+
+/**
+ * Structure representing DPP GEMM executed by a single wavefront.
+ *
+ * Each structure instantiation must contain the following fields:
+ * - wave_size - number of threads that execute a single DPP GEMM operation, usually equal to the
+ *               number of threads in a wavefront;
+ * - lanegroup_size - number of threads (lanes) that share data using DPP instruction modifier,
+ *                    it's 8 in case of DPP8;
+ * - m_per_wave - size along M dimension of matrix C that is processed in a single DPP GEMM
+ *                operation;
+ * - n_per_wave - size along N dimension of matrix C that is processed in a single DPP GEMM
+ *                operation;
+ * - m_per_lanegroup - size along M dimension that is processed by a single lanegroup;
+ * - n_per_lanegroup - size along N dimension that is processed by a single lanegroup;
+ * - m_per_thread - size along M dimension of the tile calculated by a single thread;
+ * - n_per_thread - size along N dimension of the tile calculated by a single thread;
+ * - k_per_dpp - size along K dimension that is reduced in a single DPP GEMM operation;
+ * - share_a - indicates whether we share matrix A or matrix B between lanes using DPP modifiers.
+ *
+ * Not all the combinarions are supported now, for current restrictions see the static asserts
+ * in the DppSelector's contructor.
+ */
+template <DppInstr instr>
+struct dpp_type;
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_32x8x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 32;
+    static constexpr index_t n_per_wave      = 8;
+    static constexpr index_t m_per_lanegroup = 8;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 8;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_8x32x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 8;
+    static constexpr index_t n_per_wave      = 32;
+    static constexpr index_t m_per_lanegroup = 8;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 8;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_16x16x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 16;
+    static constexpr index_t n_per_wave      = 16;
+    static constexpr index_t m_per_lanegroup = 8;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 8;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <typename BaseType, index_t MPerDpp, index_t NPerDpp>
+struct DppSelector
+{
+    template <typename BaseType_, index_t MPerDpp_, index_t NPerDpp_>
+    static constexpr auto GetDpp();
+
+    template <>
+    static constexpr auto GetDpp<half_t, 8, 32>()
+    {
+        return DppInstr::dpp8_f16_8x32x2;
+    }
+
+    template <>
+    static constexpr auto GetDpp<half_t, 16, 16>()
+    {
+        return DppInstr::dpp8_f16_16x16x2;
+    }
+
+    template <>
+    static constexpr auto GetDpp<half_t, 32, 8>()
+    {
+        return DppInstr::dpp8_f16_32x8x2;
+    }
+
+    static constexpr auto selected_dpp = dpp_type<GetDpp<BaseType, MPerDpp, NPerDpp>()>{};
+
+    __host__ __device__ constexpr DppSelector()
+    {
+        static_assert(selected_dpp.m_per_wave % selected_dpp.m_per_lanegroup == 0);
+        static_assert(selected_dpp.n_per_wave % selected_dpp.n_per_lanegroup == 0);
+
+        static_assert(selected_dpp.k_per_dpp % 2 == 0);
+
+        static_assert(selected_dpp.wave_size % selected_dpp.lanegroup_size == 0);
+        constexpr index_t num_dpp_per_wave = selected_dpp.wave_size / selected_dpp.lanegroup_size;
+        constexpr index_t num_wave_c_elems = selected_dpp.m_per_wave * selected_dpp.n_per_wave;
+        constexpr index_t num_dpp_c_elems =
+            selected_dpp.m_per_lanegroup * selected_dpp.n_per_lanegroup;
+        static_assert(num_wave_c_elems % num_dpp_c_elems == 0);
+        static_assert(num_dpp_per_wave == num_wave_c_elems / num_dpp_c_elems);
+
+        if constexpr(selected_dpp.share_a)
+        {
+            static_assert(selected_dpp.m_per_lanegroup == selected_dpp.m_per_thread);
+            static_assert(selected_dpp.n_per_lanegroup % selected_dpp.n_per_thread == 0);
+            static_assert(selected_dpp.n_per_lanegroup / selected_dpp.n_per_thread ==
+                          selected_dpp.lanegroup_size);
+        }
+        else
+        {
+            static_assert(selected_dpp.m_per_lanegroup % selected_dpp.n_per_thread == 0);
+            static_assert(selected_dpp.m_per_lanegroup / selected_dpp.n_per_thread ==
+                          selected_dpp.lanegroup_size);
+            static_assert(selected_dpp.n_per_lanegroup == selected_dpp.n_per_thread);
+        }
+
+        // Below checks come from the restrictions of the current implementation, could be removed
+        // in the future when the implementation is more generalized.
+        static_assert(selected_dpp.share_a);
+        static_assert(selected_dpp.n_per_thread == 1);
+        static_assert(selected_dpp.m_per_thread == selected_dpp.lanegroup_size);
+        static_assert(selected_dpp.m_per_lanegroup == selected_dpp.m_per_thread);
+        static_assert(selected_dpp.n_per_lanegroup ==
+                      selected_dpp.n_per_thread * selected_dpp.lanegroup_size);
+    }
+
+    static constexpr index_t GetK1PerDpp() { return selected_dpp.k_per_dpp; }
+};
+
+template <typename BaseType, index_t MPerDpp, index_t NPerDpp, index_t KPack>
+struct DppGemm
+{
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    using CIndex   = MultiIndex<2>;
+    using CIndex4D = MultiIndex<4>;
+
+    __host__ __device__ constexpr DppGemm()
+    {
+        static_assert(MPerDpp == 8 || MPerDpp == 16 || MPerDpp == 32,
+                      "MPerDpp must be either 8, 16 or 32.");
+        static_assert(NPerDpp == 8 || NPerDpp == 16 || NPerDpp == 32,
+                      "NPerDpp must be either 8, 16 or 32.");
+
+        static_assert(KPack % dpp_instr.k_per_dpp == 0, "KPack must be divisible by k_per_dpp.");
+    }
+
+    __device__ static constexpr index_t GetRegSizePerDpp()
+    {
+        return MPerDpp * NPerDpp / dpp_instr.wave_size;
+    }
+
+    template <class ADataType, class BDataType, class CDataType>
+    __device__ void
+    Run(const ADataType& p_a_wave, const BDataType& p_b_wave, CDataType& p_c_thread) const
+    {
+        static_assert(is_same<BaseType, double>::value || is_same<BaseType, float>::value ||
+                          is_same<BaseType, half_t>::value || is_same<BaseType, bhalf_t>::value ||
+                          is_same<BaseType, int8_t>::value || is_same<BaseType, f8_t>::value,
+                      "base BaseType must be double, float, half, bfloat16, and int8_t!");
+
+        static_for<0, KPack / dpp_instr.k_per_dpp, 1>{}([&](auto k) {
+            dpp_instr.template run<MPerDpp, NPerDpp>(p_a_wave[k], p_b_wave[k], p_c_thread);
+        });
+    }
+
+    __device__ static auto GetLaneIdInWave()
+    {
+        return get_thread_local_1d_id() % dpp_instr.wave_size;
+    }
+
+    __device__ static auto GetWaveId() { return get_thread_local_1d_id() / dpp_instr.wave_size; }
+
+    __device__ static auto GetLaneIdInLaneGroup()
+    {
+        return get_thread_local_1d_id() % dpp_instr.lanegroup_size;
+    }
+
+    __device__ static auto GetLaneGroupIdInWave()
+    {
+        return GetLaneIdInWave() / dpp_instr.lanegroup_size;
+    }
+
+    __device__ static auto GetDppOpIdx()
+    {
+        const auto lanegroupId = GetLaneGroupIdInWave();
+
+        constexpr auto lanegroup_idx_1d_to_dpp_idx_2d_adaptor = make_single_stage_tensor_adaptor(
+            make_tuple(
+                make_merge_transform(make_tuple(dpp_instr.m_per_wave / dpp_instr.m_per_lanegroup,
+                                                dpp_instr.n_per_wave / dpp_instr.n_per_lanegroup))),
+            make_tuple(Sequence<0, 1>{}),
+            make_tuple(Sequence<0>{}));
+
+        const auto dpp_idx = lanegroup_idx_1d_to_dpp_idx_2d_adaptor.CalculateBottomIndex(
+            make_multi_index(lanegroupId));
+
+        const auto m_dpp_idx = dpp_idx[I0];
+        const auto n_dpp_idx = dpp_idx[I1];
+
+        return make_tuple(m_dpp_idx, n_dpp_idx);
+    }
+
+    __host__ __device__ static auto CalculateAThreadOriginDataIndex_K_M()
+    {
+        const auto laneId   = get_thread_local_1d_id();
+        const auto wave_row = laneId / dpp_instr.n_per_wave;
+        auto m_idx          = dpp_instr.m_per_thread * wave_row + GetLaneIdInLaneGroup();
+        return make_tuple(0, m_idx % dpp_instr.m_per_wave);
+    }
+
+    __host__ __device__ static auto CalculateBThreadOriginDataIndex_K_N()
+    {
+        const auto laneId = get_thread_local_1d_id();
+        return make_tuple(0, laneId % dpp_instr.n_per_wave);
+    }
+
+    __device__ static CIndex GetBeginOfThreadBlk()
+    {
+        const auto dpp_op_idx = GetDppOpIdx();
+
+        const auto m_dpp_op_idx = dpp_op_idx[I0];
+        const auto n_dpp_op_idx = dpp_op_idx[I1];
+
+        index_t n_offset = n_dpp_op_idx * dpp_instr.n_per_lanegroup + GetLaneIdInLaneGroup();
+        index_t m_offset = m_dpp_op_idx * dpp_instr.m_per_lanegroup;
+
+        return CIndex{m_offset, n_offset};
+    }
+
+    static constexpr auto dpp = DppSelector<BaseType, MPerDpp, NPerDpp>{};
+
+    static constexpr auto dpp_instr = dpp.selected_dpp;
+
+    static constexpr auto K0PerDpp = 1;
+    static constexpr auto K1PerDpp = dpp.GetK1PerDpp();
+
+    __host__ __device__ static constexpr auto GetCMNThreadBlkLengths()
+    {
+        return make_tuple(Number<dpp_instr.m_per_thread>{}, Number<dpp_instr.n_per_thread>{});
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/utility/amd_gemm_dpp.hpp b/include/ck/utility/amd_gemm_dpp.hpp
index 8d6c7eede9..a28292dade 100644
--- a/include/ck/utility/amd_gemm_dpp.hpp
+++ b/include/ck/utility/amd_gemm_dpp.hpp
@@ -5,17 +5,63 @@
 
 #include "ck/utility/common_header.hpp"
 #include "ck/utility/math.hpp"
-#include "ck/utility/amd_gemm_dpp.hpp"
+#include "ck/utility/inner_product_dpp8.hpp"
 
 namespace ck {
 
 namespace dpp8 {
 
-/// Number of lanes that can share data using DPP8 modifiers.
-constexpr index_t lane_group_size = 8;
+template <class ABDataType>
+struct dpp_datatypes;
 
-__device__ index_t get_lane_group_local_idx() { return threadIdx.x / lane_group_size; }
-__device__ index_t get_thread_idx_in_lane_group() { return threadIdx.x % lane_group_size; }
+template <>
+struct dpp_datatypes<half_t>
+{
+    // Dot product of `half2_t` and `half2_t` to get `float`. Reducing 2 elements from K in a
+    // single instruction.
+    using a_dtype                        = half_t;
+    using b_dtype                        = half_t;
+    using c_dtype                        = float;
+    static constexpr index_t k_per_instr = 2;
+};
+
+template <index_t MPerThread,
+          index_t NPerThread,
+          index_t KPerThread,
+          class BaseInputType,
+          class AVecDataType,
+          class BVecDataType,
+          class CVecDataType,
+          bool ShareA>
+struct DppLanegroupGemm
+{
+    using datatypes_conf = dpp_datatypes<BaseInputType>;
+    using ADataType      = typename datatypes_conf::a_dtype;
+    using BDataType      = typename datatypes_conf::b_dtype;
+    using CDataType      = typename datatypes_conf::c_dtype;
+
+    __device__ void Run(const AVecDataType& a_vec, const BVecDataType& b_vec, CVecDataType& c_vec)
+    {
+        constexpr index_t num_c_elems_per_thread = ShareA ? MPerThread : NPerThread;
+
+        const vector_type<ADataType, KPerThread> a_vector{a_vec};
+        const vector_type<BDataType, KPerThread> b_vector{b_vec};
+
+        static_for<0, num_c_elems_per_thread, 1>{}([&](auto c_idx) {
+            float c = c_vec.template AsType<CDataType>()(c_idx);
+            // Next `c_idx` implies that we need to pull data from the next lane.
+            constexpr index_t source_lane = c_idx;
+            static_for<0, KPerThread / datatypes_conf::k_per_instr, 1>{}([&](auto k_chunk) {
+                const auto a_k_vec = a_vector.template AsType<AVecDataType>()[k_chunk];
+                const auto b_k_vec = b_vector.template AsType<BVecDataType>()[k_chunk];
+                ck::dpp8::
+                    inner_product_dpp<AVecDataType, BVecDataType, CDataType, source_lane, ShareA>(
+                        a_k_vec, b_k_vec, c);
+            });
+            c_vec.template AsType<CDataType>()(c_idx) = c;
+        });
+    }
+};
 
 } // namespace dpp8
 
diff --git a/include/ck/utility/inner_product_dpp8.hpp b/include/ck/utility/inner_product_dpp8.hpp
index ccd7a4e628..f079e2ca64 100644
--- a/include/ck/utility/inner_product_dpp8.hpp
+++ b/include/ck/utility/inner_product_dpp8.hpp
@@ -2,6 +2,7 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
+
 #include "amd_gemm_dpp.hpp"
 #include "data_type.hpp"
 #include "type_convert.hpp"
@@ -10,6 +11,9 @@ namespace ck {
 
 namespace dpp8 {
 
+/// Number of lanes that can share data using DPP8 modifiers.
+constexpr index_t lane_group_size = 8;
+
 template <int SrcLaneIdx>
 __device__ void inline_v_dot2c_dpp8_instr(const half2_t& a, const half2_t& b, float& c);
 
diff --git a/include/ck/utility/loop_scheduler.hpp b/include/ck/utility/loop_scheduler.hpp
new file mode 100644
index 0000000000..b2eb0ddb93
--- /dev/null
+++ b/include/ck/utility/loop_scheduler.hpp
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_adaptor.hpp"
+
+namespace ck {
+
+enum struct LoopScheduler
+{
+    Default,
+    Interwave,
+};
+
+constexpr LoopScheduler make_default_loop_scheduler()
+{
+#if CK_EXPERIMENTAL_DEFAULT_TO_INTER_WAVE_SCHEDULING
+    return LoopScheduler::Interwave;
+#else
+    return LoopScheduler::Default;
+#endif // if CK_EXPERIMENTAL_DEFAULT_TO_INTER_WAVE_SCHEDULING
+}
+
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 9310fd433b..b4fa746485 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -23,7 +23,7 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
+void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -38,7 +38,7 @@ void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
+void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -53,7 +53,7 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
+void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -68,7 +68,7 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
+void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -374,7 +374,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
-                add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
@@ -385,7 +385,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
-                add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
@@ -397,7 +397,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
-                add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
@@ -408,7 +408,7 @@ struct DeviceOperationInstanceFactory<
 #ifdef DL_KERNELS
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
-                add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 5162617b18..48dd292f05 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -31,10 +31,10 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp)
    endif()
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp
deleted file mode 100644
index 53fc74a392..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instance.cpp
+++ /dev/null
@@ -1,61 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
-
-// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances = std::tuple<
-    // clang-format off
-        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
-        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
-        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
-        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<1, 1, 4, 2>,       S<4, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,      S<16, 1, 2, 2>,       S<1, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,        S<1, 1, 2, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 8, 2>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<2, 1, 4, 2>,      S<4, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<2, 1, 4, 2>,      S<8, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,      S<4, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<1, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
-    // clang-format on
-    >;
-
-void add_device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_km_kn_mn_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp
deleted file mode 100644
index 35d46b32f5..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instance.cpp
+++ /dev/null
@@ -1,61 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
-
-// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances = std::tuple<
-    // clang-format off
-        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
-        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
-        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
-        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<1, 1, 4, 2>,       S<4, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 1, 2>,      S<1, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,      S<16, 1, 2, 2>,      S<1, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<16, 1,  2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 8, 2>,      S<4, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<2, 1, 4, 2>,      S<4, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<2, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<1, 1, 2, 2>,      S<16, 1,  4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,       S<1, 1, 2, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<2, 1, 4, 2>,      S<8, 1,  16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,      S<4, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<1, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<2, 1, 4, 2>,      S<8, 1,  32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,       S<1, 1, 4, 2>,       S<4, 1, 2, 2>,      S<4, 1,  64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
-    // clang-format on
-    >;
-
-void add_device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_km_nk_mn_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp
deleted file mode 100644
index 22e2e5f398..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instance.cpp
+++ /dev/null
@@ -1,60 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
-
-// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances = std::tuple<
-    // clang-format off
-        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
-        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
-        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
-        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<4, 1, 1, 2>,       S<1, 1,  8, 1>,  S<1, 2, 0 ,3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<16, 1, 2, 2>,       S<1, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 2, 1>,      S<0, 3, 1, 2>,        S<1, 1, 2, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 8, 2>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<4, 1, 2, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<1, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<4, 1, 2, 2>,       S<4, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
-    // clang-format on
-    >;
-
-void add_device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_mk_kn_mn_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp
deleted file mode 100644
index 49e9416c81..0000000000
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instance.cpp
+++ /dev/null
@@ -1,61 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl_dpp8.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
-
-// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
-using device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances = std::tuple<
-    // clang-format off
-        // ##########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer| K0Per| K1|      M1Per|      N1Per|   KPer|  M11N11Thread|  M11N11Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|      ABlockTransfer|     ABlockTransfer|       ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|      BBlockTransfer|     BBlockTransfer|       BBlockTransfer|      CThreadTransfer|  CThreadTransfer|    CThreadTransfer|
-        // ##########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|   | ThreadM111| ThreadN111| Thread| ClusterM110Xs| ClusterN110Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster|      SrcAccess|     SrcVectorTensor|    SrcVectorTensor|      DstVectorTensor|         SrcDstAccess|  SrcDstVectorDim| DstScalarPerVector|
-        // ##########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|                Order|                 |                   |
-        // ##########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                     |                 |                   |
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,     8,     8,     8,     4,  2,          1,          8,      1,       S<1, 8>,       S<1, 1>,      S<4, 1, 1, 2>,       S<1, 1,  8, 1>,  S<1, 2, 0 ,3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 1, 2>,      S<1, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<16, 1, 2, 2>,      S<1, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<4, 1>,      S<4, 1, 1, 2>,        S<4, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 8, 2>,      S<4, 1,   8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          1,          8,      1,       S<1, 8>,       S<8, 1>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,     8,  2,          4,          8,      1,       S<2, 8>,       S<4, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<2, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     8,    64,    16,  2,          8,          1,      1,       S<1, 1>,       S<8, 8>,      S<2, 1, 1, 2>,        S<8, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<2, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 4, 2>,      S<4, 1,  16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    16,  2,          4,          8,      1,       S<2, 8>,       S<8, 1>,      S<4, 1, 2, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,   128,    64,    16,  2,          1,          8,      8,       S<4, 8>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<4, 1,  32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,     8,  2,          1,          8,      8,       S<8, 8>,       S<4, 1>,      S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>,
-        DeviceGemmDlDpp8< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    16,  2,          4,          8,      8,       S<2, 8>,      S<16, 1>,      S<4, 1, 2, 2>,       S<4, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,       S<4, 1, 2, 2>,      S<4, 1,  64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,  S<0, 1, 2, 3, 4, 5>,                5,                  4>
-    // clang-format on
-    >;
-
-void add_device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances(
-    std::vector<std::unique_ptr<
-        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
-        instances)
-{
-    add_device_operation_instances(instances, device_gemm_dl_dpp8_f16_f16_f16_mk_nk_mn_instances{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000..80da6d4c3b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
@@ -0,0 +1,58 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_km_kn_mn_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   4,   16,   16,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   4,   32,    8,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   4,   4,   32,    8,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   4,   4,   32,    8,       1,       1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   4,   4,   32,    8,       2,       4,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   4,   4,   32,    8,       1,       4,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   4,   4,   16,   16,       1,       1,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_km_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000..97b85fd1a0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
@@ -0,0 +1,58 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_km_nk_mn_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   8,   16,   16,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   8,   32,    8,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   4,   8,   32,    8,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   4,   8,   32,    8,       1,       1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   4,   8,   32,    8,       2,       4,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   4,   8,   32,    8,       1,       4,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   4,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_km_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..370ebfccc4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   4,   16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   4,   32,    8,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   8,   4,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   8,   4,   32,    8,       1,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   8,   4,   32,    8,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   8,   4,   32,    8,       1,       4,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   8,   4,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..6053275492
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,58 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   8,   16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   8,   32,    8,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   8,   8,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   8,   8,   32,    8,       1,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   8,   8,   32,    8,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   8,   8,   32,    8,       1,       4,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   8,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances, device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 562b4cec485ff735a46f46dd7da894b0c3ebdf59 Mon Sep 17 00:00:00 2001
From: Haocong WANG <haocwang@amd.com>
Date: Fri, 8 Sep 2023 10:59:26 +0800
Subject: [PATCH 118/194] [Navi3x] Add fp16/int8 wmma conv forward instances
 (#746)

* fix wmma gemm int8; add grouped conv int8 example

* Add int8 gemm-bilinear instances

* compile sanity check unknown

* Sanity pass + clang-format

* add int8 conv profiler instances

* solve merge conflict

---------

Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: Chao Liu <chao.liu2@amd.com>
---
 example/02_gemm_bilinear/CMakeLists.txt       |   3 +
 .../gemm_bilinear_wmma_int8.cpp               | 304 ++++++++++++++++++
 .../gpu/block/blockwise_gemm_wmma.hpp         | 131 +++++---
 ...uped_conv_fwd_multiple_d_wmma_cshuffle.hpp |   4 +-
 .../element/binary_element_wise_operation.hpp |   7 +
 .../device_operation_instance_factory.hpp     |   1 +
 .../gpu/gemm_bilinear.hpp                     |  76 +++++
 ...evice_grouped_conv2d_fwd_wmma_instance.hpp | 134 ++++++++
 .../gpu/grouped_convolution_forward.hpp       |  32 ++
 .../gpu/gemm_bilinear/CMakeLists.txt          |   4 +
 ...uffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp |  89 +++++
 ...uffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp |  89 +++++
 ...uffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp |  89 +++++
 ...uffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp | 115 +++++++
 .../gpu/grouped_conv2d_fwd/CMakeLists.txt     |   3 +
 ...wd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  66 ++++
 ...fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp |  66 ++++
 .../profile_grouped_conv_fwd_impl.hpp         |   2 +-
 profiler/src/profile_gemm_bilinear.cpp        |  19 ++
 19 files changed, 1192 insertions(+), 42 deletions(-)
 create mode 100644 example/02_gemm_bilinear/gemm_bilinear_wmma_int8.cpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp

diff --git a/example/02_gemm_bilinear/CMakeLists.txt b/example/02_gemm_bilinear/CMakeLists.txt
index 9989e45e0c..52e6330523 100644
--- a/example/02_gemm_bilinear/CMakeLists.txt
+++ b/example/02_gemm_bilinear/CMakeLists.txt
@@ -5,6 +5,9 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
     add_example_executable(example_gemm_bilinear_wmma_fp16 gemm_bilinear_wmma_fp16.cpp)
+    add_example_executable(example_gemm_bilinear_wmma_int8 gemm_bilinear_wmma_int8.cpp)
+endif()
+if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
    set(target 1)
  endif()
 endforeach()
diff --git a/example/02_gemm_bilinear/gemm_bilinear_wmma_int8.cpp b/example/02_gemm_bilinear/gemm_bilinear_wmma_int8.cpp
new file mode 100644
index 0000000000..9f23ad2652
--- /dev/null
+++ b/example/02_gemm_bilinear/gemm_bilinear_wmma_int8.cpp
@@ -0,0 +1,304 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(int alpha, int beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<std::int8_t, std::int32_t, std::int8_t>(
+        std::int8_t& e, const std::int32_t& c, const std::int8_t& d) const
+    {
+        e = ck::type_convert<std::int8_t>(alpha_ * c + beta_ * ck::type_convert<std::int32_t>(d));
+    };
+
+    int alpha_;
+    int beta_;
+};
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using I8  = std::int8_t;
+using I32 = std::int32_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = I8;
+using BDataType        = I8;
+using AccDataType      = I32;
+using CShuffleDataType = I32;
+using DDataType        = I8;
+using EDataType        = I8;
+
+using ALayout = Row;
+using BLayout = Row;
+using DLayout = Row;
+using ELayout = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default;
+
+using DeviceOpInstance =
+    ck::tensor_operation::device::DeviceGemmMultipleD_Wmma_CShuffle<ALayout,
+                                                                    BLayout,
+                                                                    ck::Tuple<DLayout>,
+                                                                    ELayout,
+                                                                    ADataType,
+                                                                    BDataType,
+                                                                    ck::Tuple<DDataType>,
+                                                                    EDataType,
+                                                                    AccDataType,
+                                                                    CShuffleDataType,
+                                                                    AElementOp,
+                                                                    BElementOp,
+                                                                    CDEElementOp,
+                                                                    GemmSpec,
+                                                                    32,
+                                                                    16,
+                                                                    16,
+                                                                    4,
+                                                                    16,
+                                                                    16,
+                                                                    16,
+                                                                    1,
+                                                                    1,
+                                                                    S<2, 16, 1>,
+                                                                    S<1, 0, 2>,
+                                                                    S<1, 0, 2>,
+                                                                    2,
+                                                                    16,
+                                                                    16,
+                                                                    1,
+                                                                    S<4, 1, 8>,
+                                                                    S<0, 2, 1>,
+                                                                    S<0, 2, 1>,
+                                                                    1,
+                                                                    16,
+                                                                    2,
+                                                                    1,
+                                                                    1,
+                                                                    1,
+                                                                    S<1, 16, 1, 2>,
+                                                                    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = true;
+
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideD = 4096;
+    ck::index_t StrideE = 4096;
+
+    int alpha = 1;
+    int beta  = 1;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 6)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        alpha = std::stof(argv[4]);
+        beta  = std::stof(argv[5]);
+    }
+    else if(argc == 13)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        StrideA = std::stoi(argv[7]);
+        StrideB = std::stoi(argv[8]);
+        StrideD = std::stoi(argv[9]);
+        StrideE = std::stoi(argv[10]);
+
+        alpha = std::stof(argv[11]);
+        beta  = std::stof(argv[12]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
+               "beta\n");
+        exit(0);
+    }
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
+        break;
+    default:
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d_device_buf.ToDevice(d_m_n.mData.data());
+    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                               b_device_buf.GetDeviceBuffer(),
+                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                               e_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               StrideA,
+                               StrideB,
+                               std::array<ck::index_t, 1>{StrideD},
+                               StrideE,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_m_n({M, N});
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, a_element_op, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
index 5ec964bd3f..b3d45f3d0c 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
@@ -221,49 +221,102 @@ struct BlockwiseGemmWMMA_k0mk1_k0nk1_m0m1m2n0n1n2m3_CShuffle
         auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
             b_thread_desc_.GetElementSpaceSize());
 
-        static_for<0, KPerBlock / WmmaK, 1>{}([&](auto k) { // k=0,1,2 instead of k=0,kpack*1, ...
-            static_for<0, MRepeat, 1>{}([&](auto m0) {
-                // read A
-                a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
-                                   make_tuple(Number<k * WmmaK / A_K1>{}, m0, I0, I0, I0),
-                                   a_block_buf,
-                                   a_thread_desc_,
-                                   make_tuple(I0, m0, I0, I0, I0),
-                                   a_thread_buf);
+        // basic intrinsic to determine loopover direction
+        if constexpr(MRepeat < NRepeat)
+        {
+            static_for<0, KPerBlock / WmmaK, 1>{}(
+                [&](auto k) { // k=0,1,2 instead of k=0,kpack*1, ...
+                    static_for<0, MRepeat, 1>{}([&](auto m0) {
+                        // read A
+                        a_thread_copy_.Run(a_block_desc_k0_m0_m1_m2_k1,
+                                           make_tuple(Number<k * WmmaK / A_K1>{}, m0, I0, I0, I0),
+                                           a_block_buf,
+                                           a_thread_desc_,
+                                           make_tuple(I0, m0, I0, I0, I0),
+                                           a_thread_buf);
 
-                static_for<0, NRepeat, 1>{}([&](auto n0) {
-                    // read B
-                    b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
-                                       make_tuple(Number<k * WmmaK / B_K1>{}, n0, I0, I0, I0),
-                                       b_block_buf,
-                                       b_thread_desc_,
-                                       make_tuple(I0, n0, I0, I0, I0),
-                                       b_thread_buf);
-                    vector_type<FloatA, WmmaK> a_thread_vec;
-                    vector_type<FloatB, WmmaK> b_thread_vec;
+                        static_for<0, NRepeat, 1>{}([&](auto n0) {
+                            // read B
+                            b_thread_copy_.Run(
+                                b_block_desc_k0_n0_n1_n2_k1,
+                                make_tuple(Number<k * WmmaK / B_K1>{}, n0, I0, I0, I0),
+                                b_block_buf,
+                                b_thread_desc_,
+                                make_tuple(I0, n0, I0, I0, I0),
+                                b_thread_buf);
+                            vector_type<FloatA, WmmaK> a_thread_vec;
+                            vector_type<FloatB, WmmaK> b_thread_vec;
 
-                    static_for<0, WmmaK, 1>{}([&](auto i) {
-                        a_thread_vec.template AsType<FloatA>()(i) =
-                            a_thread_buf[Number<a_thread_desc_.CalculateOffset(
-                                make_tuple(i / A_K1, m0, 0, 0, i % A_K1))>{}];
-                        b_thread_vec.template AsType<FloatB>()(i) =
-                            b_thread_buf[Number<b_thread_desc_.CalculateOffset(
-                                make_tuple(i / B_K1, n0, 0, 0, i % B_K1))>{}];
+                            static_for<0, WmmaK, 1>{}([&](auto i) {
+                                a_thread_vec.template AsType<FloatA>()(i) =
+                                    a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                        make_tuple(i / A_K1, m0, 0, 0, i % A_K1))>{}];
+                                b_thread_vec.template AsType<FloatB>()(i) =
+                                    b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                        make_tuple(i / B_K1, n0, 0, 0, i % B_K1))>{}];
+                            });
+
+                            using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                            using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                            constexpr index_t c_offset =
+                                c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
+
+                            wmma_gemm.template Run(
+                                a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                                b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                                c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                        });
                     });
-
-                    using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
-                    using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
-
-                    constexpr index_t c_offset =
-                        c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
-
-                    wmma_gemm.template Run(
-                        a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
-                        b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
-                        c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
                 });
-            });
-        });
+        }
+        else
+        {
+            static_for<0, KPerBlock / WmmaK, 1>{}(
+                [&](auto k) { // k=0,1,2 instead of k=0,kpack*1, ...
+                    static_for<0, NRepeat, 1>{}([&](auto n0) {
+                        // read B
+                        b_thread_copy_.Run(b_block_desc_k0_n0_n1_n2_k1,
+                                           make_tuple(Number<k * WmmaK / B_K1>{}, n0, I0, I0, I0),
+                                           b_block_buf,
+                                           b_thread_desc_,
+                                           make_tuple(I0, n0, I0, I0, I0),
+                                           b_thread_buf);
+                        static_for<0, MRepeat, 1>{}([&](auto m0) {
+                            // read A
+                            a_thread_copy_.Run(
+                                a_block_desc_k0_m0_m1_m2_k1,
+                                make_tuple(Number<k * WmmaK / A_K1>{}, m0, I0, I0, I0),
+                                a_block_buf,
+                                a_thread_desc_,
+                                make_tuple(I0, m0, I0, I0, I0),
+                                a_thread_buf);
+                            vector_type<FloatA, WmmaK> a_thread_vec;
+                            vector_type<FloatB, WmmaK> b_thread_vec;
+
+                            static_for<0, WmmaK, 1>{}([&](auto i) {
+                                a_thread_vec.template AsType<FloatA>()(i) =
+                                    a_thread_buf[Number<a_thread_desc_.CalculateOffset(
+                                        make_tuple(i / A_K1, m0, 0, 0, i % A_K1))>{}];
+                                b_thread_vec.template AsType<FloatB>()(i) =
+                                    b_thread_buf[Number<b_thread_desc_.CalculateOffset(
+                                        make_tuple(i / B_K1, n0, 0, 0, i % B_K1))>{}];
+                            });
+
+                            using wmma_input_type_a = typename vector_type<FloatA, WmmaK>::type;
+                            using wmma_input_type_b = typename vector_type<FloatB, WmmaK>::type;
+
+                            constexpr index_t c_offset =
+                                c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
+
+                            wmma_gemm.template Run(
+                                a_thread_vec.template AsType<wmma_input_type_a>()(Number<0>{}),
+                                b_thread_vec.template AsType<wmma_input_type_b>()(Number<0>{}),
+                                c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
+                        });
+                    });
+                });
+        }
     }
 
     protected:
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
index 7c726cd851..1d73a72387 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -599,7 +599,7 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
             // check if it's 1x1, stride=1 conv
             for(index_t i = 0; i < NDimSpatial; ++i)
             {
-                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t X          = arg.b_g_k_c_xs_lengths_[i + 3];
                 const index_t ConvStride = arg.conv_filter_strides_[i];
                 const index_t LeftPad    = arg.input_left_pads_[i];
                 const index_t RightPad   = arg.input_right_pads_[i];
@@ -616,7 +616,7 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
             // check if it's 1x1 conv
             for(index_t i = 0; i < NDimSpatial; ++i)
             {
-                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 2];
+                const index_t X        = arg.b_g_k_c_xs_lengths_[i + 3];
                 const index_t LeftPad  = arg.input_left_pads_[i];
                 const index_t RightPad = arg.input_right_pads_[i];
 
diff --git a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
index 9ade02391f..9fe0931cba 100644
--- a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
@@ -186,6 +186,13 @@ struct Bilinear
         y = type_convert<half_t>(alpha_ * x0 + beta_ * ck::type_convert<float>(x1));
     };
 
+    template <>
+    __host__ __device__ constexpr void operator()<std::int8_t, std::int32_t, std::int8_t>(
+        std::int8_t& y, const std::int32_t& x0, const std::int8_t& x1) const
+    {
+        y = type_convert<std::int8_t>(x0 + ck::type_convert<std::int32_t>(x1));
+    };
+
     float alpha_;
     float beta_;
 };
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index 5096f5c30b..84d31ce267 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -31,6 +31,7 @@ using F64_Tuple     = ck::Tuple<F64>;
 using F32_Tuple     = ck::Tuple<F32>;
 using I32_Tuple     = ck::Tuple<I32>;
 using I32_F32_Tuple = ck::Tuple<I32, F32>;
+using I8_Tuple      = ck::Tuple<I8>;
 
 using F32_F32_Tuple = ck::Tuple<F32, F32>;
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index 9d8eaada22..387499e584 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -69,6 +69,58 @@ void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance
                                                     PassThrough,
                                                     Bilinear>>>& instances);
 
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances);
+
 // GEMM + Bilinear
 template <typename ALayout,
           typename BLayout,
@@ -135,6 +187,30 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
+        else if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
+                          is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
+            {
+                add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(op_ptrs);
+            }
+        }
 
         return op_ptrs;
     }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp
new file mode 100644
index 0000000000..c9cf0f8e1d
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp
@@ -0,0 +1,134 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+using I8   = int8_t;
+using I32  = int32_t;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_wmma_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // blocksize=256
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  8,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  8,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  8,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  8,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  8,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  8,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  8,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  8,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  8,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  8,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  8,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  8,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  8,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+template <typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv2d_fwd_wmma_i8_instances = std::tuple<
+    // clang-format off
+        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|        |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|        |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // blocksize=256
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  16,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  16,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  16,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  16,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  16,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  16,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  16,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  16,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  16,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  16,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index 2a0da10c65..576b9d8983 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -145,6 +145,19 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -159,6 +172,19 @@ void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -407,6 +433,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
             }
             else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                               is_same_v<WeiDataType, ck::bhalf_t> &&
@@ -414,6 +441,11 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             {
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                              is_same_v<OutDataType, int8_t>)
+            {
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
+            }
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
index b9f1210b2e..aef8fe86dd 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
@@ -4,5 +4,9 @@ add_instance_library(device_gemm_bilinear_instance
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
+   device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp
+   device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp
+   device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
+   device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
 )
 endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp
new file mode 100644
index 0000000000..73ea9cac07
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instance.cpp
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I8       = std::int8_t;
+using I32      = std::int32_t;
+using I8_Tuple = ck::Tuple<std::int8_t>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e[m, n] = bilinear(a[m, k] * b[k, n], d[m, n])
+using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        
+        // M/N/K padding
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   8,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           2,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   8,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           2,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           2,               S<1, 32, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   8,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   4,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           2,               S<1, 32, 1, 8>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   4,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           2,               S<1, 32, 1, 4>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   4,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           2,               S<1, 32, 1, 2>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   4,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,               S<1, 16, 1, 2>,               4>
+
+    // clang-format on
+    >;
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_kn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp
new file mode 100644
index 0000000000..1f36113e62
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instance.cpp
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I8       = std::int8_t;
+using I32      = std::int32_t;
+using I8_Tuple = ck::Tuple<std::int8_t>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e[m, n] = bilinear(a[m, k] * b[k, n], d[m, n])
+using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        
+        // M/N/K padding
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   8,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,           1,           2,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   8,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,           1,           2,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,           1,           2,               S<1, 32, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   8,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   4,   16,   16,       4,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,           1,           2,               S<1, 32, 1, 8>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   4,   16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,           1,           2,               S<1, 32, 1, 4>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   4,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,           1,           2,               S<1, 32, 1, 2>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Col,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   4,   16,   16,       1,       1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,           1,           1,               S<1, 16, 1, 2>,               4>
+
+    // clang-format on
+    >;
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
new file mode 100644
index 0000000000..688c463369
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I8       = std::int8_t;
+using I32      = std::int32_t;
+using I8_Tuple = ck::Tuple<std::int8_t>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e[m, n] = bilinear(a[m, k] * b[k, n], d[m, n])
+using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances = std::tuple<
+    // clang-format off
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        
+        // M/N/K padding
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   8,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           2,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   8,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           2,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           2,               S<1, 32, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   4,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           2,               S<1, 32, 1, 8>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   4,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           2,               S<1, 32, 1, 4>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   4,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           2,               S<1, 32, 1, 2>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Row, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   4,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4,  8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,               S<1, 16, 1, 2>,               4>
+
+    // clang-format on
+    >;
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Row,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
new file mode 100644
index 0000000000..5319bd8605
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
@@ -0,0 +1,115 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I8       = std::int8_t;
+using I32      = std::int32_t;
+using I8_Tuple = ck::Tuple<std::int8_t>;
+
+using Row       = ck::tensor_layout::gemm::RowMajor;
+using Col       = ck::tensor_layout::gemm::ColumnMajor;
+using Row_Tuple = ck::Tuple<Row>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmDefault    = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// e[m, n] = bilinear(a[m, k] * b[n, k], d[m, n])
+using device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances = std::tuple<
+    // clang-format off
+        // no padding
+        // N % 16 == 0 && K % 16 == 0
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear,    GemmDefault,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        // M/N/K padding
+        // N % 16 == 0 && K % 16 == 0
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     4,  16,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           2,               S<1, 32, 1, 8>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     4,  16,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           2,               S<1, 32, 1, 4>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     4,  16,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           2,               S<1, 32, 1, 2>,              16>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     4,  16,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        // M/N/K padding
+        // N % 8 == 0 && K % 8 == 0
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   8,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           2,               S<1, 32, 1, 8>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   8,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           2,               S<1, 32, 1, 4>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           2,               S<1, 32, 1, 2>,               8>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        
+        // M/N/K padding
+        // N % 8 == 0 && K % 8 == 0
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   4,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           2,               S<1, 32, 1, 8>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   4,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           2,               S<1, 32, 1, 4>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   4,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           2,               S<1, 32, 1, 2>,               4>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   4,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 2>,               4>,
+        
+        // M/N/K padding
+        // N % 1 == 0 && K % 8 == 0
+        //################################|      A|      B|        Ds|      E| AData| BData|    DsData| EData| AccData| CShuffle|           A|           B|         CDE|           GEMM| Block|  MPer|  NPer| K0Per|  K1| MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|    Layout| Layout|  Type|  Type|      Type|  Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    | WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |          |       |      |      |          |      |        |         |   Operation|   Operation|   Operation|               |      |      |      |      |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |          |       |      |      |          |      |        |         |            |            |            |               |      |      |      |      |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   256,   128,   128,     8,   8,   16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           2,               S<1, 32, 1, 8>,               1>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,   128,    64,    64,     8,   8,   16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           2,               S<1, 32, 1, 4>,               1>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    64,    32,    32,     8,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           2,               S<1, 32, 1, 2>,               1>,
+        DeviceGemmMultipleD_Wmma_CShuffle<     Row,    Col, Row_Tuple,    Row,    I8,    I8,  I8_Tuple,    I8,     I32,      I32, PassThrough, PassThrough,    Bilinear, GemmMNKPadding,    32,    16,    16,     8,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               1>
+
+    // clang-format on
+    >;
+
+void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
+                                                    Col,
+                                                    Row_Tuple,
+                                                    Row,
+                                                    I8,
+                                                    I8,
+                                                    I8_Tuple,
+                                                    I8,
+                                                    PassThrough,
+                                                    PassThrough,
+                                                    Bilinear>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
index 708b37c42d..6b0ad99d69 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
@@ -12,6 +12,9 @@ add_instance_library(device_grouped_conv2d_fwd_instance
    # GNHWC, GKYXC, GNHWK
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
    device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+   # WMMA
+   device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
    # NHWGC, GKYXC, NHWGK
    device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
    device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
new file mode 100644
index 0000000000..3904592fca
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
+                                                                                GKYXC,
+                                                                                Empty_Tuple,
+                                                                                GNHWK,
+                                                                                Empty_Tuple,
+                                                                                PassThrough,
+                                                                                ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
new file mode 100644
index 0000000000..0cded93d67
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdDefault>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwd1x1S1P0>{});
+
+    add_device_operation_instances(instances,
+                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
+                                                                               GKYXC,
+                                                                               Empty_Tuple,
+                                                                               GNHWK,
+                                                                               Empty_Tuple,
+                                                                               PassThrough,
+                                                                               ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp b/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
index 9fadfe9699..8d37c28881 100644
--- a/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_fwd_impl.hpp
@@ -215,7 +215,7 @@ bool profile_grouped_conv_fwd_impl(int do_verification,
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
 
-    std::cout << "xdl found " << op_ptrs.size() << " instances" << std::endl;
+    std::cout << "ckProfiler found " << op_ptrs.size() << " instances" << std::endl;
 
     for(auto& op_ptr : op_ptrs)
     {
diff --git a/profiler/src/profile_gemm_bilinear.cpp b/profiler/src/profile_gemm_bilinear.cpp
index a1a48616b4..4527a2fa00 100644
--- a/profiler/src/profile_gemm_bilinear.cpp
+++ b/profiler/src/profile_gemm_bilinear.cpp
@@ -71,6 +71,9 @@ int profile_gemm_bilinear(int argc, char* argv[])
     using F16 = ck::half_t;
     using F32 = float;
 
+    using I8  = std::int8_t;
+    using I32 = std::int32_t;
+
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
 
@@ -141,6 +144,22 @@ int profile_gemm_bilinear(int argc, char* argv[])
     {
         return profile(F16{}, F16{}, F32{}, F16{}, F16{}, Col{}, Col{}, Row{}, Row{});
     }
+    else if(data_type == MatrixDataType::INT8_INT8_INT8_INT8 && layout == MatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(I8{}, I8{}, I32{}, I8{}, I8{}, Row{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::INT8_INT8_INT8_INT8 && layout == MatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(I8{}, I8{}, I32{}, I8{}, I8{}, Row{}, Col{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::INT8_INT8_INT8_INT8 && layout == MatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(I8{}, I8{}, I32{}, I8{}, I8{}, Col{}, Row{}, Row{}, Row{});
+    }
+    else if(data_type == MatrixDataType::INT8_INT8_INT8_INT8 && layout == MatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(I8{}, I8{}, I32{}, I8{}, I8{}, Col{}, Col{}, Row{}, Row{});
+    }
     else
     {
         std::cout << "this data_type & layout is not implemented" << std::endl;

From 8f84a01237e0ea12d63ff59e08ee90b8afd5eecc Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Fri, 8 Sep 2023 18:14:57 +0200
Subject: [PATCH 119/194] Enable DPP8 GEMM on Navi3 (#892)

---
 .../ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp    | 3 ++-
 include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp     | 3 ++-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
index 162f0690b8..24393511c5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp
@@ -168,7 +168,8 @@ struct DeviceGemmDpp : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& karg)
     {
-        if(ck::get_device_name() == "gfx1030")
+        if(ck::get_device_name() == "gfx1030" || ck::get_device_name() == "gfx1100" ||
+           ck::get_device_name() == "gfx1101" || ck::get_device_name() == "gfx1102")
         {
             return GridwiseGemm::CheckValidity(karg);
         }
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
index d39f0ba30b..6eca77c89c 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_dpp.hpp
@@ -28,7 +28,8 @@ __global__ void
 #endif
         kernel_gemm_dpp(const typename GridwiseGemm::Argument karg)
 {
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1030__))
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx1030__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__))
     __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
 
     const auto a_grid_desc_ak0_m_ak1 = amd_wave_read_first_lane(

From 85e2e1e2e2071f0a0488ab03cbcdc5e26bbd1de6 Mon Sep 17 00:00:00 2001
From: Sam Wu <sam.wu2@amd.com>
Date: Mon, 11 Sep 2023 11:01:36 -0600
Subject: [PATCH 120/194] Add codeowners for documentation (#902)

Co-authored-by: samjwu <samjwu@users.noreply.github.com>
---
 .github/CODEOWNERS | 6 ++++++
 1 file changed, 6 insertions(+)
 create mode 100644 .github/CODEOWNERS

diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
new file mode 100644
index 0000000000..30f0dedd8d
--- /dev/null
+++ b/.github/CODEOWNERS
@@ -0,0 +1,6 @@
+# Documentation files
+docs/* @saadrahim @LisaDelaney
+*.md  @saadrahim @LisaDelaney
+*.rst  @saadrahim @LisaDelaney
+# Header directory
+library/include/*  @saadrahim @LisaDelaney

From 547dbcfbc2514768e7c3b8b71bff053238cd878c Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Tue, 12 Sep 2023 17:05:23 +0200
Subject: [PATCH 121/194] Add new instances and support for small cases in DPP8
 GEMM (#896)

---
 .../ck/tensor_operation/gpu/warp/dpp_gemm.hpp | 230 +++++++++++++++++-
 .../tensor_operation_instance/gpu/gemm.hpp    |  32 ++-
 .../gpu/gemm/CMakeLists.txt                   |   4 +
 ...gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp |  16 +-
 ...16_f16_f16_km_kn_mn_irregular_instance.cpp |  64 +++++
 ...gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp |  16 +-
 ...16_f16_f16_km_nk_mn_irregular_instance.cpp |  64 +++++
 ...gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp |  16 +-
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |  63 +++++
 ...gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp |  16 +-
 ...16_f16_f16_mk_nk_mn_irregular_instance.cpp |  64 +++++
 11 files changed, 562 insertions(+), 23 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp

diff --git a/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp b/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp
index 03a4d17c9b..a184431648 100644
--- a/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/dpp_gemm.hpp
@@ -11,8 +11,14 @@ namespace ck {
 
 enum struct DppInstr
 {
-    dpp8_f16_16x16x2 = 0,
+    dpp8_f16_1x32x2 = 0,
+    dpp8_f16_2x16x2,
+    dpp8_f16_2x32x2,
+    dpp8_f16_4x16x2,
+    dpp8_f16_4x32x2,
+    dpp8_f16_8x16x2,
     dpp8_f16_8x32x2,
+    dpp8_f16_16x16x2,
     dpp8_f16_32x8x2
 };
 
@@ -101,6 +107,36 @@ struct dpp_type<DppInstr::dpp8_f16_8x32x2>
     }
 };
 
+template <>
+struct dpp_type<DppInstr::dpp8_f16_8x16x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 8;
+    static constexpr index_t n_per_wave      = 16;
+    static constexpr index_t m_per_lanegroup = 4;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 4;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
 template <>
 struct dpp_type<DppInstr::dpp8_f16_16x16x2>
 {
@@ -131,6 +167,156 @@ struct dpp_type<DppInstr::dpp8_f16_16x16x2>
     }
 };
 
+template <>
+struct dpp_type<DppInstr::dpp8_f16_4x32x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 4;
+    static constexpr index_t n_per_wave      = 32;
+    static constexpr index_t m_per_lanegroup = 4;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 4;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_4x16x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 4;
+    static constexpr index_t n_per_wave      = 16;
+    static constexpr index_t m_per_lanegroup = 2;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 2;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_1x32x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 1;
+    static constexpr index_t n_per_wave      = 32;
+    static constexpr index_t m_per_lanegroup = 1;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 1;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_2x32x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 2;
+    static constexpr index_t n_per_wave      = 32;
+    static constexpr index_t m_per_lanegroup = 2;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 2;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct dpp_type<DppInstr::dpp8_f16_2x16x2>
+{
+    static constexpr index_t wave_size       = 32;
+    static constexpr index_t lanegroup_size  = 8;
+    static constexpr index_t m_per_wave      = 2;
+    static constexpr index_t n_per_wave      = 16;
+    static constexpr index_t m_per_lanegroup = 1;
+    static constexpr index_t n_per_lanegroup = 8;
+    static constexpr index_t m_per_thread    = 1;
+    static constexpr index_t n_per_thread    = 1;
+    static constexpr index_t k_per_dpp       = 2;
+    static constexpr bool share_a            = true;
+    using BaseType                           = half_t;
+
+    template <index_t MPerDpp, index_t NPerDpp, class ADataType, class BDataType, class CDataType>
+    __device__ void run(const ADataType& a, const BDataType& b, CDataType& reg_c) const
+    {
+        dpp8::DppLanegroupGemm<m_per_thread,
+                               n_per_thread,
+                               k_per_dpp,
+                               BaseType,
+                               ADataType,
+                               BDataType,
+                               CDataType,
+                               share_a>{}
+            .Run(a, b, reg_c);
+    }
+};
+
 template <typename BaseType, index_t MPerDpp, index_t NPerDpp>
 struct DppSelector
 {
@@ -143,6 +329,12 @@ struct DppSelector
         return DppInstr::dpp8_f16_8x32x2;
     }
 
+    template <>
+    static constexpr auto GetDpp<half_t, 8, 16>()
+    {
+        return DppInstr::dpp8_f16_8x16x2;
+    }
+
     template <>
     static constexpr auto GetDpp<half_t, 16, 16>()
     {
@@ -155,6 +347,36 @@ struct DppSelector
         return DppInstr::dpp8_f16_32x8x2;
     }
 
+    template <>
+    static constexpr auto GetDpp<half_t, 1, 32>()
+    {
+        return DppInstr::dpp8_f16_1x32x2;
+    }
+
+    template <>
+    static constexpr auto GetDpp<half_t, 2, 32>()
+    {
+        return DppInstr::dpp8_f16_2x32x2;
+    }
+
+    template <>
+    static constexpr auto GetDpp<half_t, 2, 16>()
+    {
+        return DppInstr::dpp8_f16_2x16x2;
+    }
+
+    template <>
+    static constexpr auto GetDpp<half_t, 4, 16>()
+    {
+        return DppInstr::dpp8_f16_4x16x2;
+    }
+
+    template <>
+    static constexpr auto GetDpp<half_t, 4, 32>()
+    {
+        return DppInstr::dpp8_f16_4x32x2;
+    }
+
     static constexpr auto selected_dpp = dpp_type<GetDpp<BaseType, MPerDpp, NPerDpp>()>{};
 
     __host__ __device__ constexpr DppSelector()
@@ -191,7 +413,6 @@ struct DppSelector
         // in the future when the implementation is more generalized.
         static_assert(selected_dpp.share_a);
         static_assert(selected_dpp.n_per_thread == 1);
-        static_assert(selected_dpp.m_per_thread == selected_dpp.lanegroup_size);
         static_assert(selected_dpp.m_per_lanegroup == selected_dpp.m_per_thread);
         static_assert(selected_dpp.n_per_lanegroup ==
                       selected_dpp.n_per_thread * selected_dpp.lanegroup_size);
@@ -215,11 +436,6 @@ struct DppGemm
 
     __host__ __device__ constexpr DppGemm()
     {
-        static_assert(MPerDpp == 8 || MPerDpp == 16 || MPerDpp == 32,
-                      "MPerDpp must be either 8, 16 or 32.");
-        static_assert(NPerDpp == 8 || NPerDpp == 16 || NPerDpp == 32,
-                      "NPerDpp must be either 8, 16 or 32.");
-
         static_assert(KPack % dpp_instr.k_per_dpp == 0, "KPack must be divisible by k_per_dpp.");
     }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index b4fa746485..3ceea58c1a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -23,12 +23,17 @@ void add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -38,12 +43,17 @@ void add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -53,12 +63,17 @@ void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -68,12 +83,17 @@ void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances);
+
 void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
-void add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(
+void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
@@ -375,6 +395,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
             }
@@ -386,6 +407,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
                 add_device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
@@ -398,6 +420,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
                 add_device_gemm_dpp_f16_f16_f16_km_kn_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instances(op_ptrs);
             }
@@ -409,6 +432,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_instances(op_ptrs);
                 add_device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
                 add_device_gemm_dpp_f16_f16_f16_km_nk_mn_instances(op_ptrs);
+                add_device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances(op_ptrs);
 #endif
                 add_device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 48dd292f05..6c76ee843a 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -32,9 +32,13 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp)
      list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp)
+     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp)
    endif()
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
    list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
index 80da6d4c3b..19e4d875c2 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
@@ -36,11 +36,21 @@ using device_gemm_dpp_f16_f16_f16_km_kn_mn_instances = std::tuple<
     // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   4,   16,   16,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   4,   32,    8,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,    64,    64,   4,   4,   16,   16,       1,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   4,   4,   32,    8,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    64,    64,   4,   4,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   4,   4,   32,    8,       1,       1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   4,   4,   32,    8,       2,       4,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   4,   4,   32,    8,       1,       4,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   4,   4,   16,   16,       1,       1,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,     1,   128,    64,   4,   4,    1,   32,       1,       1,      S<4, 1, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    32,    64,   4,   4,    8,   32,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    16,    64,   4,   4,    8,   16,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     4,    32,    64,   4,   4,    2,   32,       1,       1,      S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    64,    64,   4,   4,    2,   32,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    32,    64,   4,   4,    1,   32,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     1,    64,    64,   4,   4,    1,   32,       1,       1,      S<4, 1, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    64,   4,   4,    4,   16,       2,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    32,   4,   4,    8,   16,       1,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     4,    32,    32,   4,   4,    4,   32,       1,       1,      S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     2,    16,    32,   4,   4,    2,   16,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>
     >;
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..23e3152a13
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp
@@ -0,0 +1,64 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto MNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,   128,   128,    64,   4,   4,   16,   16,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,    64,    64,    64,   4,   4,   16,   16,       1,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    64,    64,    64,   4,   4,   32,    8,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    32,    32,    64,   4,   4,   32,    8,       1,       1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,     4,    32,    64,   4,   4,    1,   32,       1,       1,      S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    64,    64,    32,   4,   4,   32,    8,       2,       4,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    32,    64,   4,   4,    8,   32,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    16,    64,   4,   4,    8,   16,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,     2,    32,    64,   4,   4,    1,   32,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,    32,    16,    64,   4,   4,   16,   16,       2,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     8,    16,    64,   4,   4,    4,   16,       2,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     1,    32,    64,   4,   4,    1,   32,       1,       1,      S<4, 1, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
index 97b85fd1a0..b1533b01ad 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
@@ -36,11 +36,21 @@ using device_gemm_dpp_f16_f16_f16_km_nk_mn_instances = std::tuple<
     // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   8,   16,   16,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   4,   8,   32,    8,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,    64,    64,   4,   8,   16,   16,       1,       2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   4,   8,   32,    8,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    64,    64,   4,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   4,   8,   32,    8,       1,       1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   4,   8,   32,    8,       2,       4,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   4,   8,   32,    8,       1,       4,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   4,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,     1,   128,    64,   4,   8,    1,   32,       1,       1,      S<4, 1, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    32,    64,   4,   8,    8,   32,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    16,    64,   4,   8,    8,   16,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     4,    32,    64,   4,   8,    2,   32,       1,       1,      S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    64,    64,   4,   8,    2,   32,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    32,    64,   4,   8,    1,   32,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     1,    64,    64,   4,   8,    1,   32,       1,       1,      S<4, 1, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    64,   4,   8,    4,   16,       2,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    32,   4,   8,    8,   16,       1,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     4,    32,    32,   4,   8,    4,   32,       1,       1,      S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     2,    16,    32,   4,   8,    2,   16,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
     >;
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..a2fc5926a6
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp
@@ -0,0 +1,64 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto MNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,   128,   128,    64,   4,   8,   16,   16,       2,       4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    64,    64,    64,   4,   8,   32,    8,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    32,    64,    64,   4,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    32,    32,    64,   4,   8,   32,    8,       1,       1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,     4,    32,    64,   4,   8,    1,   32,       1,       1,      S<4, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    64,    64,    32,   4,   8,   32,    8,       2,       4,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    32,    64,   4,   8,    8,   32,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    16,    64,   4,   8,    8,   16,       1,       1,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,     2,    32,    64,   4,   8,    1,   32,       1,       1,      S<4, 2, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,    32,    16,    64,   4,   8,   16,   16,       2,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     8,    16,    64,   4,   8,    4,   16,       2,       1,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     1,    32,    64,   4,   8,    1,   32,       1,       1,      S<4, 1, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
index 370ebfccc4..2afa28a457 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -35,11 +35,21 @@ using device_gemm_dpp_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   4,   16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   4,   32,    8,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,    64,    64,   8,   4,   16,   16,       1,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   8,   4,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    64,    64,   8,   4,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   8,   4,   32,    8,       1,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   8,   4,   32,    8,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   8,   4,   32,    8,       1,       4,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   8,   4,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,     1,   128,    64,   8,   4,    1,   32,       1,       1,      S<4, 1, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    32,    64,   8,   4,    8,   32,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    16,    64,   8,   4,    8,   16,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     4,    32,    64,   8,   4,    2,   32,       1,       1,      S<4, 4, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    64,    64,   8,   4,    2,   32,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    32,    64,   8,   4,    1,   32,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     1,    64,    64,   8,   4,    1,   32,       1,       1,      S<4, 1, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    64,   8,   4,    4,   16,       2,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    32,   8,   4,    8,   16,       1,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     4,    32,    32,   8,   4,    4,   32,       1,       1,      S<4, 4, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     2,    16,    32,   8,   4,    2,   16,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>
     >;
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..508b2e8df5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -0,0 +1,63 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto MNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,   128,   128,    64,   8,   4,   16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,    64,    64,    64,   8,   4,   16,   16,       1,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    64,    64,    64,   8,   4,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    32,    32,    64,   8,   4,   32,    8,       1,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,     4,    32,    64,   8,   4,    1,   32,       1,       1,      S<4, 4, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    64,    64,    32,   8,   4,   32,    8,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    32,    64,   8,   4,    8,   32,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    16,    64,   8,   4,    8,   16,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,     2,    32,    64,   8,   4,    1,   32,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,    32,    16,    64,   8,   4,   16,   16,       2,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     8,    16,    64,   8,   4,    4,   16,       2,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     1,    32,    64,   8,   4,    1,   32,       1,       1,      S<4, 1, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
index 6053275492..242ff07cb6 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -36,11 +36,21 @@ using device_gemm_dpp_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   8,   16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,   128,   128,    64,   8,   8,   32,    8,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   256,    64,    64,    64,   8,   8,   16,   16,       1,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    64,    64,    64,   8,   8,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    64,    64,   8,   8,   16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
     DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,    32,    32,    32,   8,   8,   32,    8,       1,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    64,    64,    64,   8,   8,   32,    8,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    32,    32,    32,   8,   8,   32,    8,       1,       4,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
-    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,    16,    16,    16,   8,   8,   16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,   128,     1,   128,    64,   8,   8,    1,   32,       1,       1,      S<4, 1, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    32,    64,   8,   8,    8,   32,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,    16,    16,    64,   8,   8,    8,   16,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     4,    32,    64,   8,   8,    2,   32,       1,       1,      S<4, 4, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    64,    64,   8,   8,    2,   32,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     2,    32,    64,   8,   8,    1,   32,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    64,     1,    64,    64,   8,   8,    1,   32,       1,       1,      S<4, 1, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    64,   8,   8,    4,   16,       2,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     8,    16,    32,   8,   8,    8,   16,       1,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     4,    32,    32,   8,   8,    4,   32,       1,       1,      S<4, 4, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,    32,     2,    16,    32,   8,   8,    2,   16,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
     >;
 // clang-format on
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
new file mode 100644
index 0000000000..241fd40b7a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -0,0 +1,64 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dpp.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto MNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+// clang-format off
+using device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances = std::tuple<
+    // ########| AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MDpp|    NDpp|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer|
+    // ########|  Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise| Specialization|  Size| Block| Block| Block|    |    |  Dpp|  Dpp| PerWave| PerWave|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar|
+    // ########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |    |    |     |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector|
+    // ########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |    |    |     |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,   128,   128,    64,   8,   8,   16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   256,    64,    64,    64,   8,   8,   16,   16,       1,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    64,    64,    64,   8,   8,   32,    8,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,    32,    32,    64,   8,   8,   32,    8,       1,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,   128,     4,    32,    64,   8,   8,    1,   32,       1,       1,      S<4, 4, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    64,    64,    32,   8,   8,   32,    8,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    32,    64,   8,   8,    8,   32,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,    16,    16,    64,   8,   8,    8,   16,       1,       1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    64,     2,    32,    64,   8,   8,    1,   32,       1,       1,      S<4, 2, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,    32,    16,    64,   8,   8,   16,   16,       2,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     8,    16,    32,   8,   8,    8,   16,       1,       1,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>,
+    DeviceGemmDpp< F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,      MNPadding,    32,     1,    32,    64,   8,   8,    1,   32,       1,       1,      S<4, 1, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,      true,      S<4, 8, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,      true,               5,               1>
+    >;
+// clang-format on
+
+void add_device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 56c0279bbd822b9736b4e97d73331e1e5ccb27bb Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 12 Sep 2023 08:57:12 -0700
Subject: [PATCH 122/194] clean up the workspace after every stage (#909)

---
 Jenkinsfile | 21 +++++++++++----------
 1 file changed, 11 insertions(+), 10 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 668d9a613b..87d757e5c4 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -367,8 +367,6 @@ def runCKProfiler(Map conf=[:]){
             withDockerContainer(image: image, args: dockerOpts + ' -v=/var/jenkins/:/var/jenkins') {
                 timeout(time: 24, unit: 'HOURS')
                 {
-                    //cmake_build(conf)
-                    //instead of building, just unstash the ckProfiler and install it
                     sh """
                         rm -rf build
                         mkdir build
@@ -663,15 +661,12 @@ pipeline {
     }
     stages{
         stage("Build Docker"){
-            //when {
-            //    beforeAgent true
-            //    expression { params.BUILD_DOCKER.toBoolean() }
-            //}
             parallel{
                 stage('Docker /opt/rocm'){
                     agent{ label rocmnode("nogpu") }
                     steps{
                         buildDocker('/opt/rocm')
+                        cleanWs()
                     }
                 }
             }
@@ -693,6 +688,7 @@ pipeline {
                     }
                     steps{
                         buildHipClangJobAndReboot(setup_cmd: "", build_cmd: "", execute_cmd: execute_cmd, no_reboot:true)
+                        cleanWs()
                     }
                 }
             }
@@ -715,6 +711,7 @@ pipeline {
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                        cleanWs()
                     }
                 }
                 stage("Build CK and run Tests on MI100/MI200")
@@ -730,6 +727,7 @@ pipeline {
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                        cleanWs()
                     }
                 }
                 stage("Build CK and run Tests on Navi21")
@@ -742,10 +740,10 @@ pipeline {
                     environment{
                         setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1030" -DDL_KERNELS=ON """ 
                         execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1030" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
-
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                        cleanWs()
                     }
                 }
                 stage("Build CK and run Tests on Navi32")
@@ -756,12 +754,12 @@ pipeline {
                     }
                     agent{ label rocmnode("navi32") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DDTYPES="fp16;fp32;bf16" -DGPU_TARGETS="gfx1101" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -DDTYPES="fp16;fp32;bf16" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
-
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1101" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
+                        cleanWs()
                     }
                 }
             }
@@ -784,6 +782,7 @@ pipeline {
                    }
                     steps{
                         runPerfTest(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release')
+                        cleanWs()
                     }
                 }
                 stage("Run ckProfiler: gfx90a")
@@ -799,6 +798,7 @@ pipeline {
                     }
                     steps{
                         runPerfTest(setup_args:setup_args, config_targets: "ckProfiler", no_reboot:true, build_type: 'Release')
+                        cleanWs()
                     }
                 }
             }
@@ -811,6 +811,7 @@ pipeline {
                     agent { label 'mici' }
                     steps{
                         process_results()
+                        cleanWs()
                     }
                 }
             }

From 62d4af74491c153c196237575087843792553714 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Tue, 12 Sep 2023 17:04:27 -0500
Subject: [PATCH 123/194] Refactor f8_t, add bf8_t (#792)

* Refactor f8_t to add bf8_t

* Add check_err impl for f8_t

* Update fp8 test

* Format

* Revert the fix

* Update vector_type implementation

* Add bf8 test

* Add bf8, use BitInt types

* Add bf8 conversion methods

* Update type_convert for fp8/bf8

* Add check_err fp8/bf8 support

* Add subnorm fp8 tests

* Add subnorm bf8 tests

* Fix conversion

* Add bf8 cmake bindings

* Add macros to enable build with disabled fp8/bf8

* Remove is_native method

* Update flag combination for mixed precision instances

* Add more flag checks

* Add another flag to a client example

* Add type traits, decouple f8/bf8 casting

* Clean up

* Decouple fp8 and bf8 flags

* Remove more redundant flags

* Remove leftover comments
---
 CMakeLists.txt                                |  13 +-
 client_example/20_splitk_gemm/CMakeLists.txt  |   6 +-
 example/01_gemm/CMakeLists.txt                |   6 +-
 include/ck/config.h.in                        |   7 +
 .../element/unary_element_wise_operation.hpp  |   4 +
 .../tensor_operation/gpu/warp/xdlops_gemm.hpp |  10 +-
 include/ck/utility/amd_buffer_addressing.hpp  |  18 +-
 include/ck/utility/amd_xdlops.hpp             |   2 +
 include/ck/utility/data_type.hpp              | 131 ++++++++++-
 include/ck/utility/f8_utils.hpp               | 215 +++++++++---------
 include/ck/utility/type_convert.hpp           | 104 ++++++++-
 .../device_operation_instance_factory.hpp     |   7 +-
 .../gpu/gemm_multiply_add.hpp                 |   4 +
 .../gpu/gemm_splitk.hpp                       |   4 +
 .../include/ck/library/utility/check_err.hpp  |  94 ++++++++
 .../gpu/gemm_multiply_add/CMakeLists.txt      |  18 +-
 .../gpu/gemm_splitk/CMakeLists.txt            |   2 +-
 .../profiler/profile_gemm_splitk_impl.hpp     |   4 +
 profiler/src/profile_gemm_multiply_add.cpp    |   6 +-
 profiler/src/profile_gemm_splitk.cpp          |   6 +-
 test/data_type/CMakeLists.txt                 |  11 +-
 test/data_type/bf8.cpp                        | 158 +++++++++++++
 test/data_type/{fp8.cpp => f8.cpp}            |  81 +++++--
 23 files changed, 739 insertions(+), 172 deletions(-)
 create mode 100644 test/data_type/bf8.cpp
 rename test/data_type/{fp8.cpp => f8.cpp} (58%)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4ca54d847d..a79af9cc32 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -15,6 +15,12 @@ if (DTYPES)
     if (DTYPES MATCHES "fp8")
         add_definitions(-DCK_ENABLE_FP8)
         set(CK_ENABLE_FP8 "ON")
+        add_compile_options(-Wno-bit-int-extension)
+    endif()
+    if (DTYPES MATCHES "bf8")
+        add_definitions(-DCK_ENABLE_BF8)
+        set(CK_ENABLE_BF8 "ON")
+        add_compile_options(-Wno-bit-int-extension)
     endif()
     if (DTYPES MATCHES "fp16")
         add_definitions(-DCK_ENABLE_FP16)
@@ -34,8 +40,9 @@ if (DTYPES)
     endif()
     message("DTYPES macro set to ${DTYPES}")
 else()
-    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
+    add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_BF8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
     set(CK_ENABLE_ALL_DTYPES "ON")
+    add_compile_options(-Wno-bit-int-extension) # enable fp8 and bf8
 endif()
 
 if(DL_KERNELS)
@@ -365,6 +372,10 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
         #message("fp8 instance found!")
         set(add_inst 1)
     endif()
+    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf8\" " AND DTYPES MATCHES "bf8")
+        #message("bf8 instance found!")
+        set(add_inst 1)
+    endif()
     if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
         #message("fp16 instance found!")
         set(add_inst 1)
diff --git a/client_example/20_splitk_gemm/CMakeLists.txt b/client_example/20_splitk_gemm/CMakeLists.txt
index a60bada473..5571ed1d70 100644
--- a/client_example/20_splitk_gemm/CMakeLists.txt
+++ b/client_example/20_splitk_gemm/CMakeLists.txt
@@ -1,2 +1,4 @@
-add_executable(client_splitK_gemm splitK_gemm_fp16_f8.cpp)
-target_link_libraries(client_splitK_gemm PRIVATE composable_kernel::device_operations)
+if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+  add_executable(client_splitK_gemm splitK_gemm_fp16_f8.cpp)
+  target_link_libraries(client_splitK_gemm PRIVATE composable_kernel::device_operations)
+endif()
diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 3dc2a0966e..5574d09001 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -69,5 +69,7 @@ if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
   endif()
 endif()
 
-add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
-add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
+if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+  add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
+endif()
diff --git a/include/ck/config.h.in b/include/ck/config.h.in
index 13dc5da5d1..1748344756 100644
--- a/include/ck/config.h.in
+++ b/include/ck/config.h.in
@@ -43,6 +43,9 @@
 #ifndef CK_ENABLE_FP8
 #define CK_ENABLE_FP8 "ON"
 #endif
+#ifndef CK_ENABLE_BF8
+#define CK_ENABLE_BF8 "ON"
+#endif
 #ifndef CK_ENABLE_FP16
 #define CK_ENABLE_FP16 "ON"
 #endif
@@ -66,6 +69,10 @@
 #cmakedefine CK_ENABLE_FP8 @CK_ENABLE_FP8@
 #endif
 
+#ifndef CK_ENABLE_BF8
+#cmakedefine CK_ENABLE_BF8 @CK_ENABLE_BF8@
+#endif
+
 #ifndef CK_ENABLE_FP16
 #cmakedefine CK_ENABLE_FP16 @CK_ENABLE_FP16@
 #endif
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 905908a1c3..34ac08b665 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -89,6 +89,7 @@ struct PassThrough
     }
 #endif
 
+#if defined CK_ENABLE_FP8
     template <>
     __host__ __device__ void operator()<f8_t, f8_t>(f8_t& y, const f8_t& x) const
     {
@@ -118,6 +119,7 @@ struct PassThrough
     {
         y = type_convert<f8_t>(x);
     }
+#endif
 };
 
 struct UnaryConvert
@@ -146,6 +148,7 @@ struct ConvertBF16RTN
     }
 };
 
+#if defined CK_ENABLE_FP8
 struct ConvertF8SR
 {
     // convert to fp8 using stochastic rounding (SR)
@@ -162,6 +165,7 @@ struct ConvertF8SR
         y = f8_convert_sr<Y>(x);
     }
 };
+#endif
 
 struct Scale
 {
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index 814969ef42..9ee07b84ae 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -456,6 +456,7 @@ struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
     }
 };
 
+#if defined CK_ENABLE_FP8
 template <>
 struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8f8>
 {
@@ -499,6 +500,7 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8f8>
         intrin_mfma_f32_16x16x32f8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
     }
 };
+#endif
 
 template <typename base_type, index_t MPerXdlops, index_t NPerXdlops>
 struct MfmaSelector
@@ -640,6 +642,7 @@ struct MfmaSelector
     }
 #endif
 
+#if defined CK_ENABLE_FP8
     template <>
     static constexpr auto GetMfma<f8_t, 32, 32>()
     {
@@ -651,6 +654,7 @@ struct MfmaSelector
     {
         return MfmaInstr::mfma_f32_16x16x32f8f8;
     }
+#endif
 
     static constexpr auto selected_mfma = mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops>()>{};
 
@@ -852,7 +856,11 @@ struct XdlopsGemm
     {
         static_assert(is_same<base_type, double>::value || is_same<base_type, float>::value ||
                           is_same<base_type, half_t>::value || is_same<base_type, bhalf_t>::value ||
-                          is_same<base_type, int8_t>::value || is_same<base_type, f8_t>::value,
+                          is_same<base_type, int8_t>::value
+#if defined CK_ENABLE_FP8
+                          || is_same<base_type, f8_t>::value
+#endif
+                      ,
                       "base base_type must be double, float, half, bfloat16, and int8_t!");
 
         static_for<0, KPack / mfma_instr.k_per_blk, 1>{}([&](auto k) {
diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index 897cb4f249..694027100f 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -1127,7 +1127,7 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 
 #if CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK
     uint32_t src_addr_shift = src_thread_element_valid ? 0 : 0x80000000;
-
+#if defined CK_ENABLE_FP8
     if constexpr(is_same<scalar_t, f8_t>::value)
     {
         auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
@@ -1136,10 +1136,14 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
     }
     else
     {
+#endif
         return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
             src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+#if defined CK_ENABLE_FP8
     }
+#endif
 #else
+#if defined CK_ENABLE_FP8
     if constexpr(is_same<scalar_t, f8_t>::value)
     {
         auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
@@ -1148,11 +1152,14 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
     }
     else
     {
+#endif
         vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
             src_wave_buffer_resource, src_thread_addr_offset, 0);
         return src_thread_element_valid ? tmp : vector_t(0);
+#if defined CK_ENABLE_FP8
     }
 #endif
+#endif
 }
 
 // buffer_load requires:
@@ -1209,7 +1216,7 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
 
 #if CK_EXPERIMENTAL_USE_BUFFER_STORE_OOB_CHECK_OFFSET_TRICK
     uint32_t dst_addr_shift = dst_thread_element_valid ? 0 : 0x80000000;
-
+#if defined CK_ENABLE_FP8
     if constexpr(is_same<scalar_t, f8_t>::value)
     {
         auto tmp =
@@ -1219,12 +1226,16 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
     }
     else
     {
+#endif
         amd_buffer_store_impl<scalar_t, vector_size, coherence>(
             src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+#if defined CK_ENABLE_FP8
     }
+#endif
 #else
     if(dst_thread_element_valid)
     {
+#if defined CK_ENABLE_FP8
         if constexpr(is_same<scalar_t, f8_t>::value)
         {
             auto tmp = bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
@@ -1234,9 +1245,12 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
         }
         else
         {
+#endif
             amd_buffer_store_impl<scalar_t, vector_size, coherence>(
                 src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+#if defined CK_ENABLE_FP8
         }
+#endif
     }
 #endif
 }
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index ea7755036f..a80540515a 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -355,6 +355,7 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
     }
 };
 
+#if defined CK_ENABLE_FP8
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f32_32x32x16f8f8;
 
@@ -417,5 +418,6 @@ struct intrin_mfma_f32_16x16x32f8f8<16, 16>
 #endif
     }
 };
+#endif
 } // namespace ck
 #endif
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index c240afa2b8..89100577aa 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -12,7 +12,12 @@ using half_t  = _Float16;
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 using int4_t = _BitInt(4);
 #endif
-using f8_t = uint8_t;
+#if defined CK_ENABLE_FP8
+using f8_t = _BitInt(8);
+#endif
+#if defined CK_ENABLE_BF8
+using bf8_t = unsigned _BitInt(8);
+#endif
 
 // vector_type
 template <typename T, index_t N>
@@ -143,14 +148,24 @@ struct scalar_type<int4_t>
 };
 #endif
 
+#if defined CK_ENABLE_FP8
 template <>
 struct scalar_type<f8_t>
 {
     using type                           = f8_t;
     static constexpr index_t vector_size = 1;
 };
+#endif
+
+#if defined CK_ENABLE_BF8
+template <>
+struct scalar_type<bf8_t>
+{
+    using type                           = bf8_t;
+    static constexpr index_t vector_size = 1;
+};
+#endif
 
-//
 template <typename T>
 struct vector_type<T, 1>
 {
@@ -953,12 +968,24 @@ using int8x32_t = typename vector_type<int8_t, 32>::type;
 using int8x64_t = typename vector_type<int8_t, 64>::type;
 
 // f8
+#if defined CK_ENABLE_FP8
 using f8x2_t  = typename vector_type<f8_t, 2>::type;
 using f8x4_t  = typename vector_type<f8_t, 4>::type;
 using f8x8_t  = typename vector_type<f8_t, 8>::type;
 using f8x16_t = typename vector_type<f8_t, 16>::type;
 using f8x32_t = typename vector_type<f8_t, 32>::type;
 using f8x64_t = typename vector_type<f8_t, 64>::type;
+#endif
+
+// bf8
+#if defined CK_ENABLE_BF8
+using bf8x2_t  = typename vector_type<bf8_t, 2>::type;
+using bf8x4_t  = typename vector_type<bf8_t, 4>::type;
+using bf8x8_t  = typename vector_type<bf8_t, 8>::type;
+using bf8x16_t = typename vector_type<bf8_t, 16>::type;
+using bf8x32_t = typename vector_type<bf8_t, 32>::type;
+using bf8x64_t = typename vector_type<bf8_t, 64>::type;
+#endif
 
 template <typename T>
 struct NumericLimits
@@ -1006,21 +1033,109 @@ struct NumericLimits<int4_t>
 };
 #endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 
+#if defined CK_ENABLE_FP8
 template <>
 struct NumericLimits<f8_t>
 {
+    // negative zero nan mode with exp bias = 8
     static constexpr uint8_t binary_min    = 0x08; // 0b00001000
-    static constexpr uint8_t binary_max    = 0x77; // 0b01110111
-    static constexpr uint8_t binary_lowest = 0xF7; // 0b11110111
+    static constexpr uint8_t binary_max    = 0x7F; // 0b01111111
+    static constexpr uint8_t binary_lowest = 0xFF; // 0b11111111
     static constexpr uint8_t binary_qnan   = 0x80; // 0b10000000
+    // ieee mode with exp bias = 7
+    // static constexpr uint8_t binary_min    = 0x08; // 0b00001000
+    // static constexpr uint8_t binary_max    = 0x77; // 0b01110111
+    // static constexpr uint8_t binary_lowest = 0xF7; // 0b11110111
+    // static constexpr uint8_t binary_qnan   = 0x79; // any sign, exp=1111, mant!=0
 
-    __host__ __device__ static constexpr f8_t Min() { return bit_cast<f8_t>(binary_min); }
+    __host__ __device__ static constexpr f8_t Min() { return f8_t(binary_min); }
 
-    __host__ __device__ static constexpr f8_t Max() { return bit_cast<f8_t>(binary_max); }
+    __host__ __device__ static constexpr f8_t Max() { return f8_t(binary_max); }
 
-    __host__ __device__ static constexpr f8_t Lowest() { return bit_cast<f8_t>(binary_lowest); }
+    __host__ __device__ static constexpr f8_t Lowest() { return f8_t(binary_lowest); }
 
-    __host__ __device__ static constexpr f8_t QuietNaN() { return bit_cast<f8_t>(binary_qnan); }
+    __host__ __device__ static constexpr f8_t QuietNaN() { return f8_t(binary_qnan); }
+};
+#endif
+
+#if defined CK_ENABLE_BF8
+template <>
+struct NumericLimits<bf8_t>
+{
+    // negative zero nan mode with exp bias = 16
+    static constexpr uint8_t binary_min    = 0x04; // 0b00000100
+    static constexpr uint8_t binary_max    = 0x7F; // 0b01111111
+    static constexpr uint8_t binary_lowest = 0xFF; // 0b11111111
+    static constexpr uint8_t binary_qnan   = 0x80; // 0b10000000
+    // ieee mode with exp bias = 15
+    // static constexpr uint8_t binary_min    = 0x04; // 0b00000100
+    // static constexpr uint8_t binary_max    = 0x7B; // 0b01111011
+    // static constexpr uint8_t binary_lowest = 0xFB; // 0b11111011
+    // static constexpr uint8_t binary_qnan   = 0x79; // any sign, exp=1111, mant!=
+
+    __host__ __device__ static constexpr bf8_t Min() { return bf8_t(binary_min); }
+
+    __host__ __device__ static constexpr bf8_t Max() { return bf8_t(binary_max); }
+
+    __host__ __device__ static constexpr bf8_t Lowest() { return bf8_t(binary_lowest); }
+
+    __host__ __device__ static constexpr bf8_t QuietNaN() { return bf8_t(binary_qnan); }
+};
+#endif
+
+template <typename T>
+struct NumericUtils
+{
 };
 
+template <>
+struct NumericUtils<float>
+{
+    static constexpr int exp            = 8;
+    static constexpr int mant           = 23;
+    static constexpr uint32_t nan_mask  = 0x7F800000;
+    static constexpr uint32_t head_mask = 0xFF800000;
+    static constexpr uint32_t mant_mask = 0x7FFFFF;
+    static constexpr uint32_t exp_mask  = 0xFF;
+    static constexpr uint32_t Inf       = 0x7F800000;
+    static constexpr uint32_t NegInf    = 0xFF800000;
+    static constexpr uint32_t NaN       = 0x7F800001;
+    static constexpr uint32_t Neg0      = 0x80000000;
+    using bitwise_type                  = uint32_t;
+};
+
+template <>
+struct NumericUtils<half_t>
+{
+    static constexpr int exp            = 5;
+    static constexpr int mant           = 10;
+    static constexpr uint16_t nan_mask  = 0x7C00;
+    static constexpr uint16_t head_mask = 0xFC00;
+    static constexpr uint16_t mant_mask = 0x3FF;
+    static constexpr uint16_t exp_mask  = 0x1F;
+    static constexpr uint32_t Inf       = 0x7C00;
+    static constexpr uint32_t NegInf    = 0xFC00;
+    static constexpr uint32_t NaN       = 0x7C01;
+    static constexpr uint32_t Neg0      = 0x8000;
+    using bitwise_type                  = uint16_t;
+};
+
+#if defined CK_ENABLE_FP8
+template <>
+struct NumericUtils<f8_t>
+{
+    static constexpr int exp  = 4;
+    static constexpr int mant = 3;
+};
+#endif
+
+#if defined CK_ENABLE_BF8
+template <>
+struct NumericUtils<bf8_t>
+{
+    static constexpr int exp  = 5;
+    static constexpr int mant = 2;
+};
+#endif
+
 } // namespace ck
diff --git a/include/ck/utility/f8_utils.hpp b/include/ck/utility/f8_utils.hpp
index bb13f98154..5fbebb708d 100644
--- a/include/ck/utility/f8_utils.hpp
+++ b/include/ck/utility/f8_utils.hpp
@@ -5,6 +5,7 @@
 
 #include "ck/utility/data_type.hpp"
 
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
 namespace ck {
 
 // fp8 rounding modes
@@ -22,53 +23,38 @@ namespace ck::utils {
 
 namespace {
 
-template <typename T, bool negative_zero_nan, bool clip, bool stoch>
-__host__ __device__ f8_t run_cast_to_f8(T x, uint32_t rng)
+template <typename X, typename Y, bool negative_zero_nan, bool clip, bool stoch>
+__host__ __device__ Y run_cast_to_f8(X x, uint32_t rng)
 {
-    // check data type
-    constexpr bool is_half  = std::is_same<T, half_t>::value;
-    constexpr bool is_float = std::is_same<T, float>::value;
+    // fp8/bf8 exponent/mantissa layout
+    constexpr int out_exp  = NumericUtils<Y>::exp;
+    constexpr int out_mant = NumericUtils<Y>::mant;
 
-    // fp8 exponent/mantissa layout
-    constexpr int f8_exp  = 4;
-    constexpr int f8_mant = 3;
-
-    // resulting type exponent/mantissa layout
-    constexpr int type_exp  = is_half ? 5 : 8;
-    constexpr int type_mant = is_half ? 10 : 23;
+    // original type exponent/mantissa layout
+    constexpr int in_exp  = NumericUtils<X>::exp;
+    constexpr int in_mant = NumericUtils<X>::mant;
 
     int exponent;
     uint32_t head, mantissa, sign;
     // nan code is same for float and half
-    constexpr uint8_t nan_code  = 0x80;
-    constexpr uint32_t nan_mask = is_half ? 0x7C00 : 0x7F800000;
+    constexpr Y nan_code        = 0x80;
+    constexpr uint32_t nan_mask = NumericUtils<X>::nan_mask;
 
     // convert to bitwise
-    typedef typename std::conditional<std::is_same<T, half_t>::value, uint16_t, uint32_t>::type
-        T_bitwise;
+    using T_bitwise     = typename NumericUtils<X>::bitwise_type;
     T_bitwise x_bitwise = *(reinterpret_cast<T_bitwise*>(&x));
 
     // unpack the input, depends on datatype
-    if constexpr(is_float)
-    {
-        head     = x_bitwise & 0xFF800000;
-        mantissa = x_bitwise & 0x7FFFFF;
-        exponent = (head >> type_mant) & 0xFF;
-        sign     = head >> (type_exp + type_mant);
-    }
-    else if constexpr(is_half)
-    {
-        head     = x_bitwise & 0xFC00;
-        mantissa = x_bitwise & 0x3FF;
-        exponent = (head >> type_mant) & 0x1F;
-        sign     = head >> (type_exp + type_mant);
-    }
+    head     = x_bitwise & NumericUtils<X>::head_mask;
+    mantissa = x_bitwise & NumericUtils<X>::mant_mask;
+    exponent = (head >> in_mant) & NumericUtils<X>::exp_mask;
+    sign     = head >> (in_exp + in_mant);
 
-    uint32_t signed_inf   = (sign << (type_exp + type_mant)) + (((1 << type_exp) - 1) << type_mant);
-    uint32_t drop_mask    = (1 << (type_mant - f8_mant)) - 1;
-    constexpr int max_exp = (1 << f8_exp) - (negative_zero_nan ? 1 : 2);
+    uint32_t signed_inf   = (sign << (in_exp + in_mant)) + (((1 << in_exp) - 1) << in_mant);
+    uint32_t drop_mask    = (1 << (in_mant - out_mant)) - 1;
+    constexpr int max_exp = (1 << out_exp) - (negative_zero_nan ? 1 : 2);
     constexpr int exp_low_cutoff =
-        (1 << (type_exp - 1)) - (1 << (f8_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
+        (1 << (in_exp - 1)) - (1 << (out_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
 
     if constexpr(negative_zero_nan)
     {
@@ -81,22 +67,35 @@ __host__ __device__ f8_t run_cast_to_f8(T x, uint32_t rng)
             return signed_inf + (mantissa != 0 ? 1 : 0);
     }
 
+    // if input is half and output is bf8
+    if((NumericUtils<X>::mant == 10) && (NumericUtils<Y>::mant == 2) && negative_zero_nan &&
+       exponent == 0)
+    {
+        exponent += 1;
+        while(mantissa < (1 << in_mant))
+        {
+            mantissa <<= 1;
+            exponent -= 1;
+        }
+        mantissa &= ~(1 << in_mant);
+    }
+
     // check if x is 0.0
     if(x_bitwise == 0)
         return 0;
 
     exponent -= exp_low_cutoff - 1;
     if(exponent <= 0)
-        drop_mask = (1 << (type_mant - f8_mant + 1 - exponent)) - 1;
-    mantissa += 1 << type_mant;
+        drop_mask = (1 << (in_mant - out_mant + 1 - exponent)) - 1;
+    mantissa += 1 << in_mant;
     // apply random number if needed
     mantissa += (stoch ? rng : mantissa) & drop_mask;
-    if(mantissa >= (2 << type_mant))
+    if(mantissa >= (2 << in_mant))
     {
         mantissa >>= 1;
         exponent++;
     }
-    mantissa >>= (type_mant - f8_mant);
+    mantissa >>= (in_mant - out_mant);
 
     // check negative exponent
     if(exponent <= 0)
@@ -116,7 +115,7 @@ __host__ __device__ f8_t run_cast_to_f8(T x, uint32_t rng)
     {
         if(clip)
         {
-            mantissa = (1 << f8_mant) - 1;
+            mantissa = (1 << out_mant) - 1;
             exponent = max_exp;
         }
         else
@@ -127,124 +126,120 @@ __host__ __device__ f8_t run_cast_to_f8(T x, uint32_t rng)
 
     // check if x is 0.0 or -0.0
     if(exponent == 0 && mantissa == 0)
-        return negative_zero_nan ? 0 : (sign << (f8_exp + f8_mant));
-    mantissa &= (1 << f8_mant) - 1;
-    return (sign << (f8_exp + f8_mant)) | (exponent << f8_mant) | mantissa;
+        return negative_zero_nan ? 0 : (sign << (out_exp + out_mant));
+    mantissa &= (1 << out_mant) - 1;
+    return (sign << (out_exp + out_mant)) | (exponent << out_mant) | mantissa;
 }
 
-template <typename T, bool negative_zero_nan>
-__host__ __device__ T run_cast_from_f8(f8_t x)
+template <typename X, typename Y, bool negative_zero_nan>
+__host__ __device__ Y run_cast_from_f8(X x)
 {
-    // check data type
-    constexpr bool is_half  = std::is_same<T, half_t>::value;
-    constexpr bool is_float = std::is_same<T, float>::value;
-
-    // fp8 exponent/mantissa layout
-    constexpr int f8_exp  = 4;
-    constexpr int f8_mant = 3;
+    // fp8/bf8 exponent/mantissa layout
+    constexpr int in_exp  = NumericUtils<X>::exp;
+    constexpr int in_mant = NumericUtils<X>::mant;
 
     // resulting type exponent/mantissa layout
-    constexpr int type_exp  = is_half ? 5 : 8;
-    constexpr int type_mant = is_half ? 10 : 23;
+    constexpr int out_exp  = NumericUtils<Y>::exp;
+    constexpr int out_mant = NumericUtils<Y>::mant;
 
     // prepare the codes
-    constexpr uint8_t nan_code = 0x80;
-    T fInf, fNegInf, fNaN, fNeg0;
-    if constexpr(is_half)
-    {
-        constexpr uint16_t ihInf    = 0x7C00;
-        constexpr uint16_t ihNegInf = 0xFC00;
-        constexpr uint16_t ihNaN    = 0x7C01;
-        constexpr uint16_t ihNeg0   = 0x8000;
-        fInf                        = *(reinterpret_cast<const half_t*>(&ihInf));
-        fNegInf                     = *(reinterpret_cast<const half_t*>(&ihNegInf));
-        fNaN                        = *(reinterpret_cast<const half_t*>(&ihNaN));
-        fNeg0                       = *(reinterpret_cast<const half_t*>(&ihNeg0));
-    }
-    else if constexpr(is_float)
-    {
-        constexpr uint32_t ifInf    = 0x7F800000;
-        constexpr uint32_t ifNegInf = 0xFF800000;
-        constexpr uint32_t ifNaN    = 0x7F800001;
-        constexpr uint32_t ifNeg0   = 0x80000000;
-        fInf                        = *(reinterpret_cast<const float*>(&ifInf));
-        fNegInf                     = *(reinterpret_cast<const float*>(&ifNegInf));
-        fNaN                        = *(reinterpret_cast<const float*>(&ifNaN));
-        fNeg0                       = *(reinterpret_cast<const float*>(&ifNeg0));
-    }
+    constexpr X nan_code = 0x80;
+    Y Inf, NegInf, NaN, Neg0;
+    using T_bitwise = typename NumericUtils<Y>::bitwise_type;
+
+    constexpr T_bitwise Inf_bitwise    = NumericUtils<Y>::Inf;
+    constexpr T_bitwise NegInf_bitwise = NumericUtils<Y>::NegInf;
+    constexpr T_bitwise NaN_bitwise    = NumericUtils<Y>::NaN;
+    constexpr T_bitwise Neg0_bitwise   = NumericUtils<Y>::Neg0;
+
+    Inf    = *(reinterpret_cast<const Y*>(&Inf_bitwise));
+    NegInf = *(reinterpret_cast<const Y*>(&NegInf_bitwise));
+    NaN    = *(reinterpret_cast<const Y*>(&NaN_bitwise));
+    Neg0   = *(reinterpret_cast<const Y*>(&Neg0_bitwise));
+
+    // check if x is 0.0
+    if(x == 0)
+        return static_cast<Y>(0);
 
     // unpack the input
-    uint32_t sign     = x >> (f8_exp + f8_mant);
-    uint32_t mantissa = x & ((1 << f8_mant) - 1);
-    int exponent      = (x & 0x7F) >> f8_mant;
+    uint32_t sign     = x >> (in_exp + in_mant);
+    uint32_t mantissa = x & ((1 << in_mant) - 1);
+    int exponent      = (x & 0x7F) >> in_mant;
 
     constexpr int exp_low_cutoff =
-        (1 << (type_exp - 1)) - (1 << (f8_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
-    typename std::conditional<std::is_same<T, half_t>::value, uint16_t, uint32_t>::type retval;
+        (1 << (out_exp - 1)) - (1 << (in_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
+    T_bitwise retval;
 
     if constexpr(negative_zero_nan)
     {
         if(x == nan_code)
-            return fNaN;
+            return NaN;
     }
     else
     {
         if(x == nan_code)
-            return fNeg0;
-        if(exponent == ((1 << f8_exp) - 1))
-            return (mantissa == 0) ? (sign ? fNegInf : fInf) : fNaN;
+            return Neg0;
+        if(exponent == ((1 << in_exp) - 1))
+            return (mantissa == 0) ? (sign ? NegInf : Inf) : NaN;
+    }
+
+    if((NumericUtils<Y>::mant == 10) && (NumericUtils<X>::mant == 2) && !negative_zero_nan)
+    {
+        retval = x;
+        retval <<= 8;
+        return *(reinterpret_cast<const Y*>(&retval));
     }
 
     // subnormal input
     if(exponent == 0)
     {
         // guaranteed mantissa!=0 since cases 0x0 and 0x80 are handled above
-        int sh = 1 + __builtin_clz(mantissa) - ((1 + type_exp + type_mant) - f8_mant);
-        mantissa <<= sh;
-        mantissa &= ((1 << f8_mant) - 1);
-        exponent += 1 - sh;
+        exponent++;
+        while(mantissa < (1 << in_mant))
+        {
+            mantissa <<= 1;
+            exponent--;
+        }
+        mantissa &= ((1 << in_mant) - 1);
     }
     exponent += exp_low_cutoff - 1;
-    mantissa <<= type_mant - f8_mant;
+    mantissa <<= out_mant - in_mant;
 
     // subnormal output (occurs when T=half, we=5, negative_zero_nan=true)
     if(exponent <= 0)
     {
-        mantissa |= 1 << type_mant;
+        mantissa |= 1 << out_mant;
         mantissa >>= 1 - exponent;
         exponent = 0;
     }
 
-    retval = (sign << (type_exp + type_mant)) | (exponent << type_mant) | mantissa;
-    return *(reinterpret_cast<const T*>(&retval));
+    retval = (sign << (out_exp + out_mant)) | (exponent << out_mant) | mantissa;
+    return *(reinterpret_cast<const Y*>(&retval));
 }
 
 } // namespace
 
-template <typename T, bool negative_zero_nan, bool clip, bool stoch>
-__host__ __device__ f8_t cast_to_f8(T x, uint32_t rng)
+template <typename X, typename Y, bool negative_zero_nan, bool clip, bool stoch>
+__host__ __device__ Y cast_to_f8(X x, uint32_t rng)
 {
-    // check datatype
-    constexpr bool is_half  = std::is_same<T, half_t>::value;
-    constexpr bool is_float = std::is_same<T, float>::value;
-    static_assert(is_half || is_float, "Only half and float can be casted to f8.");
+    // check datatypes
+    constexpr bool is_half  = std::is_same<X, half_t>::value;
+    constexpr bool is_float = std::is_same<X, float>::value;
+    static_assert(is_half || is_float, "Only half and float can be casted.");
 
-    return run_cast_to_f8<T, negative_zero_nan, clip, stoch>(x, rng);
+    return run_cast_to_f8<X, Y, negative_zero_nan, clip, stoch>(x, rng);
 }
 
-template <typename T, bool negative_zero_nan>
-__host__ __device__ T cast_from_f8(f8_t x)
+template <typename X, typename Y, bool negative_zero_nan>
+__host__ __device__ Y cast_from_f8(X x)
 {
     // check datatype
-    constexpr bool is_half  = std::is_same<T, half_t>::value;
-    constexpr bool is_float = std::is_same<T, float>::value;
+    constexpr bool is_half  = std::is_same<Y, half_t>::value;
+    constexpr bool is_float = std::is_same<Y, float>::value;
     static_assert(is_half || is_float, "only half and float are supported.");
 
-    // check if x is 0.0
-    if(x == 0)
-        return static_cast<T>(0);
-
-    return run_cast_from_f8<T, negative_zero_nan>(x);
+    return run_cast_from_f8<X, Y, negative_zero_nan>(x);
 }
 
 } // namespace ck::utils
+#endif
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 65d8940377..5c5447f94e 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -80,6 +80,7 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_
     return type_convert<bhalf_t>(x_fp32);
 }
 
+#if defined CK_ENABLE_FP8
 // convert fp32 to fp8
 template <>
 inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
@@ -88,8 +89,9 @@ inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
     constexpr uint32_t rng           = 0;
-    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
-        x, rng);
+    return utils::
+        cast_to_f8<float, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(x,
+                                                                                               rng);
 }
 
 // convert fp8 to fp32
@@ -97,7 +99,7 @@ template <>
 inline __host__ __device__ float type_convert<float, f8_t>(f8_t x)
 {
     constexpr bool negative_zero_nan = true;
-    return utils::cast_from_f8<float, negative_zero_nan>(x);
+    return utils::cast_from_f8<f8_t, float, negative_zero_nan>(x);
 }
 
 // convert fp16 to fp8
@@ -108,8 +110,9 @@ inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
     constexpr uint32_t rng           = 0;
-    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
-        x, rng);
+    return utils::
+        cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
 }
 
 // convert fp8 to fp16
@@ -117,8 +120,53 @@ template <>
 inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
 {
     constexpr bool negative_zero_nan = true;
-    return utils::cast_from_f8<half_t, negative_zero_nan>(x);
+    return utils::cast_from_f8<f8_t, half_t, negative_zero_nan>(x);
 }
+#endif
+
+#if defined CK_ENABLE_BF8
+// convert fp32 to bf8
+template <>
+inline __host__ __device__ bf8_t type_convert<bf8_t, float>(float x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::
+        cast_to_f8<float, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+}
+
+// convert bf8 to fp32
+template <>
+inline __host__ __device__ float type_convert<float, bf8_t>(bf8_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<bf8_t, float, negative_zero_nan>(x);
+}
+
+// convert fp16 to bf8
+template <>
+inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
+    constexpr uint32_t rng           = 0;
+    return utils::
+        cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+}
+
+// convert bf8 to fp16
+template <>
+inline __host__ __device__ half_t type_convert<half_t, bf8_t>(bf8_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    return utils::cast_from_f8<bf8_t, half_t, negative_zero_nan>(x);
+}
+#endif
 
 // Declare a template function for bf16 conversion using RTN
 template <typename Y, typename X>
@@ -181,6 +229,7 @@ inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(h
 template <typename Y, typename X>
 __host__ __device__ constexpr Y f8_convert_sr(X x);
 
+#if defined CK_ENABLE_FP8
 // convert fp32 to fp8 with stochastic rounding
 template <>
 inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
@@ -191,8 +240,9 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
     constexpr int seed               = 42;
     // as thread id is not available on host, use 0 for prn generation
     uint32_t rng = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
-    return utils::cast_to_f8<float, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
-        x, rng);
+    return utils::
+        cast_to_f8<float, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(x,
+                                                                                               rng);
 }
 
 // convert fp16 to fp8 with stochastic rounding
@@ -205,8 +255,42 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
     constexpr int seed               = 42;
     // as thread id is not available on host, use 0 for prn generation
     uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
-    return utils::cast_to_f8<half_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
-        x, rng);
+    return utils::
+        cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+}
+#endif
+
+#if defined CK_ENABLE_BF8
+// convert fp32 to bf8 with stochastic rounding
+template <>
+inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, float>(float x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    constexpr int seed               = 42;
+    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::
+        cast_to_f8<float, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
 }
 
+// convert fp16 to bf8 with stochastic rounding
+template <>
+inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
+{
+    constexpr bool negative_zero_nan = true;
+    constexpr bool clip              = true;
+    constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
+    constexpr int seed               = 42;
+    // as thread id is not available on host, use 0 for prn generation
+    uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
+    return utils::
+        cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
+            x, rng);
+}
+#endif
+
 } // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index 84d31ce267..ea11fd2e1a 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -17,10 +17,15 @@ namespace instance {
 using F64  = double;
 using F32  = float;
 using F16  = ck::half_t;
-using F8   = ck::f8_t;
 using BF16 = ck::bhalf_t;
 using I8   = int8_t;
 using I32  = int32_t;
+#if defined CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+#if defined CK_ENABLE_BF8
+using BF8 = ck::bf8_t;
+#endif
 
 using Empty_Tuple = ck::Tuple<>;
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
index badc06dd6f..64c74d4795 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_multiply_add.hpp
@@ -45,6 +45,7 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_
                                                     PassThrough,
                                                     MultiplyAdd>>>&);
 
+#if defined CK_ENABLE_FP8
 void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
                                                     Row,
@@ -70,6 +71,7 @@ void add_device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_m
                                                     PassThrough,
                                                     PassThrough,
                                                     MultiplyAdd>>>&);
+#endif
 
 // GEMM + Multiply + Add
 template <typename ALayout,
@@ -131,6 +133,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
             }
         }
 
+#if defined CK_ENABLE_FP8
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
                      is_same_v<D0DataType, float> && is_same_v<D1DataType, float> &&
                      is_same_v<EDataType, half_t>)
@@ -150,6 +153,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
+#endif
 
         return op_ptrs;
     }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index dbd2b8f656..d239cd7706 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -57,6 +57,7 @@ void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(
         DeviceGemmSplitK<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
 
+#if defined CK_ENABLE_FP8
 void add_device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -96,6 +97,7 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
+#endif
 
 template <typename ADataType,
           typename BDataType,
@@ -176,6 +178,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+#if defined CK_ENABLE_FP8
         else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -224,6 +227,7 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+#endif
 
         return op_ptrs;
     }
diff --git a/library/include/ck/library/utility/check_err.hpp b/library/include/ck/library/utility/check_err.hpp
index 8a72631376..c0f9ba2edc 100644
--- a/library/include/ck/library/utility/check_err.hpp
+++ b/library/include/ck/library/utility/check_err.hpp
@@ -230,5 +230,99 @@ check_err(const Range& out,
     return res;
 }
 
+#if defined CK_ENABLE_FP8
+template <typename Range, typename RefRange>
+std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
+                  std::is_same_v<ranges::range_value_t<Range>, f8_t>),
+                 bool>
+check_err(const Range& out,
+          const RefRange& ref,
+          const std::string& msg = "Error: Incorrect results!",
+          double rtol            = 1e-3,
+          double atol            = 1e-3)
+{
+    if(out.size() != ref.size())
+    {
+        std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
+                  << std::endl;
+        return false;
+    }
+
+    bool res{true};
+    int err_count  = 0;
+    double err     = 0;
+    double max_err = std::numeric_limits<float>::min();
+    for(std::size_t i = 0; i < ref.size(); ++i)
+    {
+        const double o = type_convert<float>(*std::next(std::begin(out), i));
+        const double r = type_convert<float>(*std::next(std::begin(ref), i));
+        err            = std::abs(o - r);
+        if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
+        {
+            max_err = err > max_err ? err : max_err;
+            err_count++;
+            if(err_count < 5)
+            {
+                std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
+                          << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
+            }
+            res = false;
+        }
+    }
+    if(!res)
+    {
+        std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
+    }
+    return res;
+}
+#endif
+
+#if defined CK_ENABLE_BF8
+template <typename Range, typename RefRange>
+std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
+                  std::is_same_v<ranges::range_value_t<Range>, bf8_t>),
+                 bool>
+check_err(const Range& out,
+          const RefRange& ref,
+          const std::string& msg = "Error: Incorrect results!",
+          double rtol            = 1e-3,
+          double atol            = 1e-3)
+{
+    if(out.size() != ref.size())
+    {
+        std::cerr << msg << " out.size() != ref.size(), :" << out.size() << " != " << ref.size()
+                  << std::endl;
+        return false;
+    }
+
+    bool res{true};
+    int err_count  = 0;
+    double err     = 0;
+    double max_err = std::numeric_limits<float>::min();
+    for(std::size_t i = 0; i < ref.size(); ++i)
+    {
+        const double o = type_convert<float>(*std::next(std::begin(out), i));
+        const double r = type_convert<float>(*std::next(std::begin(ref), i));
+        err            = std::abs(o - r);
+        if(err > atol + rtol * std::abs(r) || !std::isfinite(o) || !std::isfinite(r))
+        {
+            max_err = err > max_err ? err : max_err;
+            err_count++;
+            if(err_count < 5)
+            {
+                std::cerr << msg << std::setw(12) << std::setprecision(7) << " out[" << i
+                          << "] != ref[" << i << "]: " << o << " != " << r << std::endl;
+            }
+            res = false;
+        }
+    }
+    if(!res)
+    {
+        std::cerr << std::setw(12) << std::setprecision(7) << "max err: " << max_err << std::endl;
+    }
+    return res;
+}
+#endif
+
 } // namespace utils
 } // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
index a4f7443455..36bd6a4aa3 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
@@ -1,7 +1,13 @@
-add_instance_library(device_gemm_multiply_add_instance
-   device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
-   device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+set(GEMM_MULTIPLY_ADD_INSTANCES)
 
-   device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
-   device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp
-)
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp)
+  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp)
+endif()
+
+if((DTYPES MATCHES "fp16" AND DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
+  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp)
+  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp)
+endif()
+
+add_instance_library(device_gemm_multiply_add_instance ${GEMM_MULTIPLY_ADD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
index 89dfa8f2ed..043b28a1be 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
@@ -14,7 +14,7 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
   list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp)
 endif()
 
-if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
+if((DTYPES MATCHES "fp16" AND DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
   list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp)
   list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp)
   list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp)
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index fb68bb8811..495513f665 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -214,6 +214,7 @@ bool profile_gemm_splitk_impl(int do_verification,
                           << " TFlops, " << gb_per_sec << " GB/s, " << op_name << ", KBatch "
                           << kbatch_curr << std::endl;
 
+#if defined CK_ENABLE_FP8
                 // set softer tolerances for fp8
                 if constexpr(is_same_v<ADataType, f8_t> || is_same_v<BDataType, f8_t> ||
                              is_same_v<CDataType, f8_t>)
@@ -226,8 +227,11 @@ bool profile_gemm_splitk_impl(int do_verification,
                 }
                 else
                 {
+#endif
                     pass = pass & ck::utils::check_err(c_m_n_device_result, c_m_n_host_result);
+#if defined CK_ENABLE_FP8
                 }
+#endif
 
                 if(tflops > best_tflops)
                 {
diff --git a/profiler/src/profile_gemm_multiply_add.cpp b/profiler/src/profile_gemm_multiply_add.cpp
index fd1f5c65c1..98973b2f01 100644
--- a/profiler/src/profile_gemm_multiply_add.cpp
+++ b/profiler/src/profile_gemm_multiply_add.cpp
@@ -59,9 +59,11 @@ int profile_gemm_multiply_add(int argc, char* argv[])
     const int StrideD1 = std::stoi(argv[14]);
     const int StrideE  = std::stoi(argv[15]);
 
-    using F8  = ck::f8_t;
     using F16 = ck::half_t;
     using F32 = float;
+#if defined CK_ENABLE_FP8
+    using F8 = ck::f8_t;
+#endif
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -132,6 +134,7 @@ int profile_gemm_multiply_add(int argc, char* argv[])
     {
         return profile(F16{}, F16{}, F32{}, F16{}, F16{}, F16{}, Row{}, Col{}, Row{}, Row{}, Row{});
     }
+#if defined CK_ENABLE_FP8
     else if(data_type == MatrixDataType::F16_F8_F32_F32_F16 &&
             layout == MatrixLayout::MK_KN_MN_MN_MN)
     {
@@ -142,6 +145,7 @@ int profile_gemm_multiply_add(int argc, char* argv[])
     {
         return profile(F16{}, F8{}, F32{}, F32{}, F32{}, F16{}, Row{}, Col{}, Row{}, Row{}, Row{});
     }
+#endif
     else
     {
         std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/profiler/src/profile_gemm_splitk.cpp b/profiler/src/profile_gemm_splitk.cpp
index 617e0b9cd4..9c805fc1d1 100644
--- a/profiler/src/profile_gemm_splitk.cpp
+++ b/profiler/src/profile_gemm_splitk.cpp
@@ -67,7 +67,9 @@ int profile_gemm_splitk(int argc, char* argv[])
 
     using F32 = float;
     using F16 = ck::half_t;
-    using F8  = ck::f8_t;
+#if defined CK_ENABLE_FP8
+    using F8 = ck::f8_t;
+#endif
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -146,6 +148,7 @@ int profile_gemm_splitk(int argc, char* argv[])
     {
         return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{});
     }
+#if defined CK_ENABLE_FP8
     else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
         return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{});
@@ -178,6 +181,7 @@ int profile_gemm_splitk(int argc, char* argv[])
     {
         return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Col{}, Row{});
     }
+#endif
     else
     {
         std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/test/data_type/CMakeLists.txt b/test/data_type/CMakeLists.txt
index 2b63727f19..baf0174556 100644
--- a/test/data_type/CMakeLists.txt
+++ b/test/data_type/CMakeLists.txt
@@ -3,5 +3,12 @@ if (USE_BITINT_EXTENSION_INT4)
   target_link_libraries(test_int4 PRIVATE utility)
 endif()
 
-add_gtest_executable(test_fp8 fp8.cpp)
-target_link_libraries(test_fp8 PRIVATE utility)
+if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
+  add_gtest_executable(test_f8 f8.cpp)
+  target_link_libraries(test_f8 PRIVATE utility)
+endif()
+
+if(DTYPES MATCHES "bf8" OR NOT DEFINED DTYPES)
+  add_gtest_executable(test_bf8 bf8.cpp)
+  target_link_libraries(test_bf8 PRIVATE utility)
+endif()
diff --git a/test/data_type/bf8.cpp b/test/data_type/bf8.cpp
new file mode 100644
index 0000000000..6a5fa281e8
--- /dev/null
+++ b/test/data_type/bf8.cpp
@@ -0,0 +1,158 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/type_convert.hpp"
+
+using ck::bf8_t;
+using ck::f8_convert_sr;
+using ck::half_t;
+using ck::type_convert;
+
+TEST(BF8, NumericLimits)
+{
+    // constants given for negative zero nan mode
+    EXPECT_EQ(ck::NumericLimits<bf8_t>::Min(), type_convert<bf8_t>(0x04));
+    EXPECT_EQ(ck::NumericLimits<bf8_t>::Max(), type_convert<bf8_t>(0x7F));
+    EXPECT_EQ(ck::NumericLimits<bf8_t>::Lowest(), type_convert<bf8_t>(0xFF));
+    EXPECT_EQ(ck::NumericLimits<bf8_t>::QuietNaN(), type_convert<bf8_t>(0x80));
+}
+
+TEST(BF8, ConvertFP32Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to bf8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(type_convert<bf8_t>(0.0f)), abs_tol);
+    // convert minimal float to bf8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(type_convert<bf8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal bf8_t to float and check if equal to 57344.0
+    ASSERT_NEAR(57344.0f, type_convert<float>(type_convert<bf8_t>(57344.0f)), abs_tol);
+    // convert maximal float to bf8 and back, check if clipped to 57344.0
+    ASSERT_NEAR(57344.0f,
+                type_convert<float>(type_convert<bf8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to bf8_t and check if it is qNan
+    ASSERT_NEAR(type_convert<bf8_t>(0x80),
+                type_convert<bf8_t>(std::numeric_limits<float>::infinity()),
+                abs_tol);
+    // positive norm float value to bf8 and back, check if holds
+    float pos_float = 0.0000762939f;
+    ASSERT_NEAR(pos_float, type_convert<float>(type_convert<bf8_t>(pos_float)), abs_tol);
+    // negative norm float value to bf8 and back, check if holds
+    float neg_float = -0.0000610351f;
+    ASSERT_NEAR(neg_float, type_convert<float>(type_convert<bf8_t>(neg_float)), abs_tol);
+    // positive subnorm float value to bf8 and back, check if holds
+    pos_float = 0.0000305175f;
+    ASSERT_NEAR(pos_float, type_convert<float>(type_convert<bf8_t>(pos_float)), abs_tol);
+    // negative subnorm float value to bf8 and back, check if holds
+    neg_float = -0.0000152587f;
+    ASSERT_NEAR(neg_float, type_convert<float>(type_convert<bf8_t>(neg_float)), abs_tol);
+}
+
+TEST(BF8, ConvertFP32Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-6;
+    // convert 0 float to bf8 and back, check if holds
+    ASSERT_NEAR(0.0f, type_convert<float>(f8_convert_sr<bf8_t>(0.0f)), abs_tol);
+    // convert minimal float to bf8 and back, check if holds
+    ASSERT_NEAR(std::numeric_limits<float>::min(),
+                type_convert<float>(f8_convert_sr<bf8_t>(std::numeric_limits<float>::min())),
+                abs_tol);
+    // convert maximal bf8_t to float and check if equal to 57344.0
+    ASSERT_NEAR(57344.0f, type_convert<float>(f8_convert_sr<bf8_t>(57344.0f)), abs_tol);
+    // convert maximal float to bf8 and back, check if clipped to 57344.0
+    ASSERT_NEAR(57344.0f,
+                type_convert<float>(f8_convert_sr<bf8_t>(std::numeric_limits<float>::max())),
+                abs_tol);
+    // convert inf float to bf8_t and check if it is qNan
+    ASSERT_NEAR(type_convert<bf8_t>(0x80),
+                f8_convert_sr<bf8_t>(std::numeric_limits<float>::infinity()),
+                abs_tol);
+    // positive norm float value to bf8 and back, check if holds
+    float pos_float = 0.0000762939f;
+    ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<bf8_t>(pos_float)), abs_tol);
+    // negative norm float value to bf8 and back, check if holds
+    float neg_float = -0.0000610351f;
+    ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<bf8_t>(neg_float)), abs_tol);
+    // positive subnorm float value to bf8 and back, check if holds
+    pos_float = 0.0000305175f;
+    ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<bf8_t>(pos_float)), abs_tol);
+    // negative subnorm float value to bf8 and back, check if holds
+    neg_float = -0.0000152587f;
+    ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<bf8_t>(neg_float)), abs_tol);
+}
+
+TEST(BF8, ConvertFP16Nearest)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to bf8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(type_convert<bf8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to bf8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(type_convert<bf8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal bf8_t to fp16 and check if equal to 57344.0
+    ASSERT_NEAR(
+        half_t{57344.0}, type_convert<half_t>(type_convert<bf8_t>(half_t{57344.0})), abs_tol);
+    // convert maximal fp16 to bf8 and back, check if clipped to 57344.0
+    ASSERT_NEAR(half_t{57344.0},
+                type_convert<half_t>(type_convert<bf8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to bf8_t and check if it is QuietNaN
+    ASSERT_NEAR(type_convert<bf8_t>(0x80),
+                type_convert<bf8_t>(ck::NumericLimits<half_t>::QuietNaN()),
+                abs_tol);
+    // positive norm fp16 value to bf8 and back, check if holds
+    half_t pos_half = half_t{0.0000762939};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<bf8_t>(pos_half)), abs_tol);
+    // negative norm fp16 value to bf8 and back, check if holds
+    half_t neg_half = half_t{-0.0000610351};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<bf8_t>(neg_half)), abs_tol);
+    // positive subnorm fp16 value to bf8 and back, check if holds
+    pos_half = half_t{0.0000305175};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<bf8_t>(pos_half)), abs_tol);
+    // negative subnorm fp16 value to bf8 and back, check if holds
+    neg_half = half_t{-0.0000152587};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<bf8_t>(neg_half)), abs_tol);
+}
+
+TEST(BF8, ConvertFP16Stochastic)
+{
+    // fix the tolerance value
+    float abs_tol = 1e-3;
+    // convert 0 fp16 to bf8 and back, check if holds
+    ASSERT_NEAR(half_t{0.0}, type_convert<half_t>(f8_convert_sr<bf8_t>(half_t{0.0})), abs_tol);
+    // convert minimal fp16 to bf8 and back, check if holds
+    ASSERT_NEAR(ck::NumericLimits<half_t>::Min(),
+                type_convert<half_t>(f8_convert_sr<bf8_t>(ck::NumericLimits<half_t>::Min())),
+                abs_tol);
+    // convert maximal bf8_t to fp16 and check if equal to 57344.0
+    ASSERT_NEAR(
+        half_t{57344.0}, type_convert<half_t>(f8_convert_sr<bf8_t>(half_t{57344.0})), abs_tol);
+    // convert maximal fp16 to bf8 and back, check if clipped to 57344.0
+    ASSERT_NEAR(half_t{57344.0},
+                type_convert<half_t>(f8_convert_sr<bf8_t>(ck::NumericLimits<half_t>::Max())),
+                abs_tol);
+    // convert QuietNaN fp16 to bf8_t and check if it is QuietNaN
+    ASSERT_NEAR(type_convert<bf8_t>(0x80),
+                f8_convert_sr<bf8_t>(ck::NumericLimits<half_t>::QuietNaN()),
+                abs_tol);
+    // positive norm fp16 value to bf8 and back, check if holds
+    half_t pos_half = half_t{0.0000762939};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<bf8_t>(pos_half)), abs_tol);
+    // negative norm fp16 value to bf8 and back, check if holds
+    half_t neg_half = half_t{-0.0000610351};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<bf8_t>(neg_half)), abs_tol);
+    // positive subnorm fp16 value to bf8 and back, check if holds
+    pos_half = half_t{0.0000305175};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<bf8_t>(pos_half)), abs_tol);
+    // negative subnorm fp16 value to bf8 and back, check if holds
+    neg_half = half_t{-0.0000152587};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<bf8_t>(neg_half)), abs_tol);
+}
diff --git a/test/data_type/fp8.cpp b/test/data_type/f8.cpp
similarity index 58%
rename from test/data_type/fp8.cpp
rename to test/data_type/f8.cpp
index 5004fe9527..0612a1cf44 100644
--- a/test/data_type/fp8.cpp
+++ b/test/data_type/f8.cpp
@@ -12,10 +12,11 @@ using ck::type_convert;
 
 TEST(FP8, NumericLimits)
 {
-    EXPECT_EQ(ck::NumericLimits<f8_t>::Min(), 0x08);
-    EXPECT_EQ(ck::NumericLimits<f8_t>::Max(), 0x77);
-    EXPECT_EQ(ck::NumericLimits<f8_t>::Lowest(), 0xF7);
-    EXPECT_EQ(ck::NumericLimits<f8_t>::QuietNaN(), 0x80);
+    // constants given for negative zero nan mode
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Min(), type_convert<f8_t>(0x08));
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Max(), type_convert<f8_t>(0x7F));
+    EXPECT_EQ(ck::NumericLimits<f8_t>::Lowest(), type_convert<f8_t>(0xFF));
+    EXPECT_EQ(ck::NumericLimits<f8_t>::QuietNaN(), type_convert<f8_t>(0x80));
 }
 
 TEST(FP8, ConvertFP32Nearest)
@@ -35,12 +36,20 @@ TEST(FP8, ConvertFP32Nearest)
                 type_convert<float>(type_convert<f8_t>(std::numeric_limits<float>::max())),
                 abs_tol);
     // convert inf float to f8_t and check if it is qNan
-    ASSERT_NEAR(0x80, type_convert<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
-    // positive float value to fp8 and back, check if holds
-    float pos_float = 0.0078125f;
+    ASSERT_NEAR(type_convert<f8_t>(0x80),
+                type_convert<f8_t>(std::numeric_limits<float>::infinity()),
+                abs_tol);
+    // positive norm float value to fp8 and back, check if holds
+    float pos_float = 0.017578125f;
     ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
-    // negative float value to fp8 and back, check if holds
-    float neg_float = -0.0156250f;
+    // negative norm float value to fp8 and back, check if holds
+    float neg_float = -0.015625f;
+    ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
+    // positive subnorm float value to fp8 and back, check if holds
+    pos_float = 0.00390625f;
+    ASSERT_NEAR(pos_float, type_convert<float>(type_convert<f8_t>(pos_float)), abs_tol);
+    // negative subnorm float value to fp8 and back, check if holds
+    neg_float = -0.001953125f;
     ASSERT_NEAR(neg_float, type_convert<float>(type_convert<f8_t>(neg_float)), abs_tol);
 }
 
@@ -61,12 +70,20 @@ TEST(FP8, ConvertFP32Stochastic)
                 type_convert<float>(f8_convert_sr<f8_t>(std::numeric_limits<float>::max())),
                 abs_tol);
     // convert inf float to f8_t and check if it is qNan
-    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(std::numeric_limits<float>::infinity()), abs_tol);
-    // positive float value to fp8 and back, check if holds
-    float pos_float = 0.0078125f;
+    ASSERT_NEAR(type_convert<f8_t>(0x80),
+                f8_convert_sr<f8_t>(std::numeric_limits<float>::infinity()),
+                abs_tol);
+    // positive norm float value to fp8 and back, check if holds
+    float pos_float = 0.017578125f;
     ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<f8_t>(pos_float)), abs_tol);
-    // negative float value to fp8 and back, check if holds
-    float neg_float = -0.0156250f;
+    // negative norm float value to fp8 and back, check if holds
+    float neg_float = -0.015625f;
+    ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<f8_t>(neg_float)), abs_tol);
+    // positive subnorm float value to fp8 and back, check if holds
+    pos_float = 0.00390625f;
+    ASSERT_NEAR(pos_float, type_convert<float>(f8_convert_sr<f8_t>(pos_float)), abs_tol);
+    // negative subnorm float value to fp8 and back, check if holds
+    neg_float = -0.001953125f;
     ASSERT_NEAR(neg_float, type_convert<float>(f8_convert_sr<f8_t>(neg_float)), abs_tol);
 }
 
@@ -87,12 +104,20 @@ TEST(FP8, ConvertFP16Nearest)
                 type_convert<half_t>(type_convert<f8_t>(ck::NumericLimits<half_t>::Max())),
                 abs_tol);
     // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
-    ASSERT_NEAR(0x80, type_convert<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
-    // positive fp16 value to fp8 and back, check if holds
-    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(type_convert<f8_t>(0x80),
+                type_convert<f8_t>(ck::NumericLimits<half_t>::QuietNaN()),
+                abs_tol);
+    // positive norm fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.017578125};
     ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
-    // negative fp16 value to fp8 and back, check if holds
-    half_t neg_half = half_t{-0.0156250};
+    // negative norm fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.015625};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
+    // positive subnorm fp16 value to fp8 and back, check if holds
+    pos_half = half_t{0.00390625};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(type_convert<f8_t>(pos_half)), abs_tol);
+    // negative subnorm fp16 value to fp8 and back, check if holds
+    neg_half = half_t{-0.001953125};
     ASSERT_NEAR(neg_half, type_convert<half_t>(type_convert<f8_t>(neg_half)), abs_tol);
 }
 
@@ -113,11 +138,19 @@ TEST(FP8, ConvertFP16Stochastic)
                 type_convert<half_t>(f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::Max())),
                 abs_tol);
     // convert QuietNaN fp16 to f8_t and check if it is QuietNaN
-    ASSERT_NEAR(0x80, f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::QuietNaN()), abs_tol);
-    // positive fp16 value to fp8 and back, check if holds
-    half_t pos_half = half_t{0.0078125};
+    ASSERT_NEAR(type_convert<f8_t>(0x80),
+                f8_convert_sr<f8_t>(ck::NumericLimits<half_t>::QuietNaN()),
+                abs_tol);
+    // positive norm fp16 value to fp8 and back, check if holds
+    half_t pos_half = half_t{0.017578125};
     ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<f8_t>(pos_half)), abs_tol);
-    // negative fp16 value to fp8 and back, check if holds
-    half_t neg_half = half_t{-0.0156250};
+    // negative norm fp16 value to fp8 and back, check if holds
+    half_t neg_half = half_t{-0.015625};
+    ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<f8_t>(neg_half)), abs_tol);
+    // positive subnorm fp16 value to fp8 and back, check if holds
+    pos_half = half_t{0.00390625};
+    ASSERT_NEAR(pos_half, type_convert<half_t>(f8_convert_sr<f8_t>(pos_half)), abs_tol);
+    // negative subnorm fp16 value to fp8 and back, check if holds
+    neg_half = half_t{-0.001953125};
     ASSERT_NEAR(neg_half, type_convert<half_t>(f8_convert_sr<f8_t>(neg_half)), abs_tol);
 }

From 74d32f0719eeaa2ebdc465bc6c1ed525f785c30c Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 12 Sep 2023 18:14:33 -0700
Subject: [PATCH 124/194] Add a switch to build DL kernels and build them with
 staging compiler. (#907)

* enable building DL kernels with the daily staging compiler

* move the DL_KERNELS flag to another function
---
 Jenkinsfile | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 87d757e5c4..c1a28a74ed 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -210,6 +210,9 @@ def cmake_build(Map conf=[:]){
     } else{
         setup_args = ' -DBUILD_DEV=On' + setup_args
     }
+    if (params.DL_KERNELS){
+        setup_args = setup_args + " -DDL_KERNELS=ON "
+    }
 
     if(build_type_debug){
         setup_args = " -DCMAKE_BUILD_TYPE=debug -DCMAKE_CXX_FLAGS_DEBUG='${debug_flags}'" + setup_args
@@ -612,7 +615,7 @@ def process_results(Map conf=[:]){
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
 CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=rc1
                                               0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
-                                              0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
+                                              0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {
     agent none
@@ -647,6 +650,10 @@ pipeline {
             name: "RUN_FULL_QA",
             defaultValue: false,
             description: "Select whether to run small set of performance tests (default) or full QA")
+        booleanParam(
+            name: "DL_KERNELS",
+            defaultValue: false,
+            description: "Select whether to build DL kernels (default: OFF)")
     }
     environment{
         dbuser = "${dbuser}"

From a66d14edf25d34cf2c4b021f8c27f12e5267d566 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 12 Sep 2023 22:17:56 -0500
Subject: [PATCH 125/194] fixed fp8 issues (#894)

* fixed fp8 init; and reference gemm

* Update host_tensor_generator.hpp

* fixed convert

* fixed reference gemm

* fixed comments

* fixed comments

* fixed ci

* fixed computeType

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../groupnorm_sigmoid_mul_fp16.cpp            | 16 ++++++-----
 .../device_gemm_multiple_d_xdl_cshuffle.hpp   |  7 +++--
 .../element/unary_element_wise_operation.hpp  | 27 +++++++++++++++----
 .../threadwise_tensor_slice_transfer.hpp      |  9 +++----
 .../cpu/reference_gemm.hpp                    |  7 ++---
 .../profile_gemm_multiply_add_impl.hpp        |  4 +--
 6 files changed, 45 insertions(+), 25 deletions(-)

diff --git a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
index cc107b63dc..b36bd761b3 100644
--- a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
@@ -14,18 +14,22 @@ using ComputeDataType = float;
 
 struct YElementOp
 {
-    template <typename T>
-    __host__ __device__ void operator()(T& y, const T& x) const
+    template <typename Y, typename X>
+    __host__ __device__ void operator()(Y& y, const X& x) const
     {
-        static_assert(ck::is_same<T, float>::value || ck::is_same<T, double>::value ||
-                          ck::is_same<T, ck::half_t>::value,
+        static_assert(ck::is_same<X, float>::value || ck::is_same<X, double>::value ||
+                          ck::is_same<X, ck::half_t>::value,
                       "Data type is not supported by this operation!");
 
-        T a;
+        static_assert(ck::is_same<Y, float>::value || ck::is_same<Y, double>::value ||
+                          ck::is_same<Y, ck::half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        X a;
 
         ck::tensor_operation::element_wise::Sigmoid{}(a, x);
 
-        y = x * a;
+        y = ck::type_convert<Y>(x * a);
     };
 };
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
index c90c28f5a8..d98725cf9d 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp
@@ -144,7 +144,8 @@ template <typename ALayout,
           typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CDEBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched     = make_default_loop_scheduler(),
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          PipelineVersion PipelineVer = PipelineVersion::v1,
+          typename ComputeDataType    = EDataType>
 struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
                                                                      BLayout,
                                                                      DsLayout,
@@ -243,11 +244,9 @@ struct DeviceGemmMultipleD_Xdl_CShuffle : public DeviceGemmMultipleD<ALayout,
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}, {}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N<ELayout>(1, 1, 1));
 
-    using ComputeDataType = EDataType;
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
         BDataType,
         ComputeDataType,
         AccDataType,
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 34ac08b665..69a6540a0c 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -27,6 +27,12 @@ struct PassThrough
         y = x;
     }
 
+    template <>
+    __host__ __device__ void operator()<float, double>(float& y, const double& x) const
+    {
+        y = type_convert<float>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<float, float>(float& y, const float& x) const
     {
@@ -81,6 +87,12 @@ struct PassThrough
         y = type_convert<int8_t>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<int8_t, float>(int8_t& y, const float& x) const
+    {
+        y = type_convert<int8_t>(x);
+    }
+
 #ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
     template <>
     __host__ __device__ void operator()<int4_t, int4_t>(int4_t& y, const int4_t& x) const
@@ -416,14 +428,19 @@ struct Swish
 {
     Swish(float beta = 1.0f) : beta_(beta) {}
 
-    template <typename T>
-    __host__ __device__ void operator()(T& y, const T& x) const
+    template <typename Y, typename X>
+    __host__ __device__ void operator()(Y& y, const X& x) const
     {
-        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
-                          is_same<T, ck::half_t>::value,
+        static_assert(is_same<X, float>::value || is_same<X, double>::value ||
+                          is_same<X, ck::half_t>::value,
                       "Data type is not supported by this operation!");
 
-        y = x / (ck::type_convert<T>(1) + ck::math::exp(-beta_ * x));
+        static_assert(is_same<Y, float>::value || is_same<Y, double>::value ||
+                          is_same<Y, ck::half_t>::value,
+                      "Data type is not supported by this operation!");
+
+        float bx = -beta_ * type_convert<float>(x);
+        y        = type_convert<Y>(x / (1.f + ck::math::exp(bx)));
     };
 
     float beta_ = 1.0f;
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
index 605f2569c6..2774214079 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp
@@ -137,13 +137,12 @@ struct ThreadwiseTensorSliceTransfer_v1r3
                 constexpr index_t src_offset = src_desc.CalculateOffset(
                     src_slice_origin_idx + idx_md + i * dst_scalar_step_in_vector);
 
-                SrcData v;
+                DstData v;
 
                 // apply element-wise operation
                 element_op_(v, src_buf[Number<src_offset>{}]);
 
-                // apply type convert
-                dst_vector.template AsType<DstData>()(i) = type_convert<DstData>(v);
+                dst_vector.template AsType<DstData>()(i) = v;
             });
 
             const bool is_dst_valid =
@@ -1289,13 +1288,13 @@ struct ThreadwiseTensorSliceTransfer_StaticToStatic
                 constexpr index_t dst_offset = dst_desc.CalculateOffset(
                     dst_slice_origin_idx + idx_md + i * dst_scalar_step_in_vector);
 
-                SrcData v;
+                DstData v;
 
                 // apply element-wise operation
                 element_op_(v, src_buf[Number<src_offset>{}]);
 
                 // apply type convert
-                dst_buf(Number<dst_offset>{}) = type_convert<DstData>(v);
+                dst_buf(Number<dst_offset>{}) = v;
             });
         });
     }
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
index 309b4afad8..95bd1e13d9 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
@@ -20,7 +20,8 @@ template <typename ADataType,
           typename AccDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CElementwiseOperation>
+          typename CElementwiseOperation,
+          typename ComputType = ADataType>
 struct ReferenceGemm : public device::BaseOperator
 {
     // Argument
@@ -64,8 +65,8 @@ struct ReferenceGemm : public device::BaseOperator
 
                 for(int k = 0; k < K; ++k)
                 {
-                    ADataType v_a;
-                    BDataType v_b;
+                    ComputType v_a;
+                    ComputType v_b;
 
                     // use PassThrough instead of ConvertBF16RTN for reference calculation
                     if constexpr(is_same_v<AElementwiseOperation,
diff --git a/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp b/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp
index 5fc92d1f8d..f9a5a995fe 100644
--- a/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_multiply_add_impl.hpp
@@ -83,8 +83,8 @@ bool profile_gemm_multiply_add_impl(int do_verification,
         d1_m_n.GenerateTensorValue(GeneratorTensor_2<D1DataType>{-1, 1});
         break;
     default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 0.2});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.1, 0.1});
         d0_m_n.GenerateTensorValue(GeneratorTensor_3<D0DataType>{0.0, 1.0});
         d1_m_n.GenerateTensorValue(GeneratorTensor_3<D1DataType>{0.0, 1.0});
     }

From 475188ca2ef2bfc931c2dc814c585e6248b9b463 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 13 Sep 2023 17:14:31 +0200
Subject: [PATCH 126/194] Add grouped conv bwd weight dl instances and new
 layout (#897)

* Add grouped conv bwd weight dl instances and new layout

* Add M and N padding

* Remove todo comment

* Enable grouped conv fwd dl k,c=1 generic instance

* Comment fixes
---
 .../grouped_conv_bwd_weight_dl_fp16.cpp       |  89 ++-
 .../run_grouped_conv_bwd_weight_example.inc   |  14 +-
 ... => device_grouped_conv_bwd_weight_dl.hpp} | 685 ++++++++++--------
 include/ck/utility/inner_product.hpp          |  12 +
 ...ce_grouped_conv_bwd_weight_dl_instance.hpp |  86 +++
 .../device_grouped_conv_fwd_dl_instance.hpp   |   4 +-
 .../grouped_convolution_backward_weight.hpp   | 304 ++++++++
 .../grouped_conv1d_bwd_weight/CMakeLists.txt  |  17 +-
 ...weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp |  45 ++
 ..._weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp |  45 ++
 ..._weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp |  45 ++
 ...weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp |  45 ++
 ..._weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp |  45 ++
 ..._weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp |  45 ++
 ...eight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp |   7 -
 ...weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp |   7 -
 ...weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp |   7 -
 .../grouped_conv2d_bwd_weight/CMakeLists.txt  |  16 +-
 ...ght_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |  46 ++
 ...ight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  46 ++
 ...ight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |  46 ++
 ...ght_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |  46 ++
 ...ight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  46 ++
 ...ight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |  46 ++
 ...ht_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |   1 -
 ...ght_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   1 -
 ...ght_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   1 -
 ...ht_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |   1 -
 ...ght_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |   1 -
 ...ght_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |   1 -
 .../grouped_conv3d_bwd_weight/CMakeLists.txt  |  17 +-
 ..._dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |  45 ++
 ...t_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |  45 ++
 ...t_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |  45 ++
 ..._dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |  46 ++
 ...t_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |  46 ++
 ...t_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |  46 ++
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |   7 -
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |   7 -
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |   7 -
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |   1 -
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |   1 -
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |   1 -
 .../test_grouped_convnd_bwd_weight.cpp        |  70 +-
 44 files changed, 1739 insertions(+), 445 deletions(-)
 rename include/ck/tensor_operation/gpu/device/impl/{device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp => device_grouped_conv_bwd_weight_dl.hpp} (63%)
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp

diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
index 375c309e1c..cb6d68e04a 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp"
 
 using InDataType  = F16;
 using WeiDataType = F16;
@@ -15,44 +15,55 @@ using WeiElementOp = PassThrough;
 using OutElementOp = PassThrough;
 
 template <ck::index_t NDimSpatial>
-using DeviceConvBwdWeightInstance =
-    ck::tensor_operation::device::DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl<
-        NDimSpatial,          // NDimSpatial
-        InDataType,           // InDataType
-        WeiDataType,          // WeiDataType
-        OutDataType,          // OutDataType
-        AccDataType,          // AccDataType
-        InElementOp,          // InElementwiseOperation
-        WeiElementOp,         // WeiElementwiseOperation
-        OutElementOp,         // OutElementwiseOperation
-        ConvBwdWeightDefault, // ConvBackwardWeightSpecialization
-        256,                  // BlockSize
-        128,                  // MPerBlock
-        128,                  // NPerBlock
-        16,                   // K0PerBlock
-        2,                    // K1
-        4,                    // M1PerThread
-        4,                    // N1PerThread
-        1,                    // KPerThread
-        S<8, 2>,              // M1N1ThreadClusterM1Xs
-        S<8, 2>,              // M1N1ThreadClusterN1Xs
-        S<1, 8, 1, 1, 2>,     // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
-        S<1, 2, 1, 128, 1>,   // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
-        S<0, 2, 3, 1, 4>,     // ABlockTransferThreadClusterArrangeOrder
-        S<0, 2, 3, 1, 4>,     // ABlockTransferSrcAccessOrder
-        S<1, 1, 1, 1, 1>,     // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
-        S<0, 2, 3, 1, 4>,     // ABlockTransferSrcVectorTensorContiguousDimOrder
-        S<1, 1, 1, 1, 1>,     // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
-        S<1, 1, 1, 8, 2>,     // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
-        S<1, 16, 1, 16, 1>,   // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
-        S<0, 1, 4, 2, 3>,     // BBlockTransferThreadClusterArrangeOrder
-        S<0, 1, 4, 2, 3>,     // BBlockTransferSrcAccessOrder
-        S<1, 1, 1, 8, 1>,     // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
-        S<0, 1, 4, 2, 3>,     // BBlockTransferSrcVectorTensorContiguousDimOrder
-        S<1, 1, 1, 1, 2>,     // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
-        S<0, 1, 2, 3, 4, 5>,  // CThreadTransferSrcDstAccessOrder
-        5,                    // CThreadTransferSrcDstVectorDim
-        4>;                   // CThreadTransferDstScalarPerVector
+using DeviceConvBwdWeightInstance = ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Dl<
+    NDimSpatial, // NDimSpatial
+    ck::tuple_element_t<NDimSpatial - 1,
+                        ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                  ck::tensor_layout::convolution::GNHWC,
+                                  ck::tensor_layout::convolution::GNDHWC>>, // InLayout
+    ck::tuple_element_t<NDimSpatial - 1,
+                        ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                  ck::tensor_layout::convolution::GKYXC,
+                                  ck::tensor_layout::convolution::GKZYXC>>, // WeiLayout
+    ck::tuple_element_t<NDimSpatial - 1,
+                        ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                  ck::tensor_layout::convolution::GNHWK,
+                                  ck::tensor_layout::convolution::GNDHWK>>, // OutLayout
+    InDataType,                                                             // InDataType
+    WeiDataType,                                                            // WeiDataType
+    OutDataType,                                                            // OutDataType
+    AccDataType,                                                            // AccDataType
+    InElementOp,          // InElementwiseOperation
+    WeiElementOp,         // WeiElementwiseOperation
+    OutElementOp,         // OutElementwiseOperation
+    ConvBwdWeightDefault, // ConvBackwardWeightSpecialization
+    256,                  // BlockSize
+    128,                  // MPerBlock
+    128,                  // NPerBlock
+    16,                   // K0PerBlock
+    2,                    // K1
+    4,                    // M1PerThread
+    4,                    // N1PerThread
+    1,                    // KPerThread
+    S<8, 2>,              // M1N1ThreadClusterM1Xs
+    S<8, 2>,              // M1N1ThreadClusterN1Xs
+    S<1, 8, 1, 1, 2>,     // ABlockTransferThreadSliceLengths_K0_M0_M1_K1
+    S<1, 2, 1, 128, 1>,   // ABlockTransferThreadClusterLengths_K0_M0_M1_K1
+    S<0, 2, 3, 1, 4>,     // ABlockTransferThreadClusterArrangeOrder
+    S<0, 2, 3, 1, 4>,     // ABlockTransferSrcAccessOrder
+    S<1, 1, 1, 1, 1>,     // ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1
+    S<0, 2, 3, 1, 4>,     // ABlockTransferSrcVectorTensorContiguousDimOrder
+    S<1, 1, 1, 1, 1>,     // ABlockTransferDstVectorTensorLengths_K0_M0_M1_K1
+    S<1, 1, 1, 8, 2>,     // BBlockTransferThreadSliceLengths_K0_N0_N1_K1
+    S<1, 16, 1, 16, 1>,   // BBlockTransferThreadClusterLengths_K0_N0_N1_K1
+    S<0, 1, 4, 2, 3>,     // BBlockTransferThreadClusterArrangeOrder
+    S<0, 1, 4, 2, 3>,     // BBlockTransferSrcAccessOrder
+    S<1, 1, 1, 8, 1>,     // BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1
+    S<0, 1, 4, 2, 3>,     // BBlockTransferSrcVectorTensorContiguousDimOrder
+    S<1, 1, 1, 1, 2>,     // BBlockTransferDstVectorTensorLengths_K0_N0_N1_K1
+    S<0, 1, 2, 3, 4, 5>,  // CThreadTransferSrcDstAccessOrder
+    5,                    // CThreadTransferSrcDstVectorDim
+    4>;                   // CThreadTransferDstScalarPerVector
 
 #include "run_grouped_conv_bwd_weight_example.inc"
 
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 29ce0324ab..34bca7dc86 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -14,20 +14,8 @@ template <ck::index_t NDimSpatial>
 bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
                                  const ck::utils::conv::ConvParam& conv_param)
 {
-    ck::index_t split_k;
-    // Set split_k = 2 for xdl op, split_k = 1 for dl
     // Dl op doesn't support split_k > 1
-    // TODO: Add Dl op split_k > 1 support
-    if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
-         ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
-         ck::get_device_name() == "gfx1102"))
-    {
-        split_k = 2;
-    }
-    else
-    {
-        split_k = 1;
-    }
+    constexpr ck::index_t split_k = 1;
 
     const auto in_g_n_c_wis_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
similarity index 63%
rename from include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
rename to include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
index 198751cdf3..014140c460 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_gnwc_gkxc_gnwk_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
@@ -14,6 +14,7 @@
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 
@@ -72,6 +73,9 @@ __global__ void
             const Block2CTileMap block_2_ctile_map,
             const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
 {
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx1030__) ||           \
+    defined(__gfx90a__) || defined(__gfx908__) || defined(__gfx940__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
     const index_t num_blocks_per_batch =
         __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
     const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
@@ -96,9 +100,27 @@ __global__ void
         block_2_ctile_map,
         integral_constant<bool, HasMainKBlockLoop>{},
         integral_constant<bool, HasDoubleTailKBlockLoop>{});
+#else
+    ignore = p_a_grid;
+    ignore = p_b_grid;
+    ignore = p_c_grid;
+    ignore = batch_count;
+    ignore = a_grid_desc_kbatch_k0_m0_m1_k1;
+    ignore = b_grid_desc_kbatch_k0_n0_n1_k1;
+    ignore = c_grid_desc_m0_m10_m11_n0_n10_n11;
+    ignore = block_2_ctile_map;
+    ignore = compute_ptr_offset_of_batch;
+
+    compute_ptr_offset_of_batch.GetAPtrOffset(0);
+    compute_ptr_offset_of_batch.GetBPtrOffset(0);
+    compute_ptr_offset_of_batch.GetCPtrOffset(0);
+#endif
 }
 
 template <ck::index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout,
           typename InDataType,
           typename WeiDataType,
           typename OutDataType,
@@ -134,29 +156,46 @@ template <ck::index_t NDimSpatial,
           typename CThreadTransferSrcDstAccessOrder,
           index_t CThreadTransferSrcDstVectorDim,
           index_t CThreadTransferDstScalarPerVector>
-struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
-    : public DeviceGroupedConvBwdWeight<
-          NDimSpatial,
-          ck::tuple_element_t<NDimSpatial - 1,
-                              ck::Tuple<ck::tensor_layout::convolution::GNWC,
-                                        ck::tensor_layout::convolution::GNHWC,
-                                        ck::tensor_layout::convolution::GNDHWC>>,
-          ck::tuple_element_t<NDimSpatial - 1,
-                              ck::Tuple<ck::tensor_layout::convolution::GKXC,
-                                        ck::tensor_layout::convolution::GKYXC,
-                                        ck::tensor_layout::convolution::GKZYXC>>,
-          ck::tuple_element_t<NDimSpatial - 1,
-                              ck::Tuple<ck::tensor_layout::convolution::GNWK,
-                                        ck::tensor_layout::convolution::GNHWK,
-                                        ck::tensor_layout::convolution::GNDHWK>>,
-          InDataType,
-          WeiDataType,
-          OutDataType,
-          InElementwiseOperation,
-          WeiElementwiseOperation,
-          OutElementwiseOperation>
+struct DeviceGroupedConvBwdWeight_Dl : public DeviceGroupedConvBwdWeight<NDimSpatial,
+                                                                         InLayout,
+                                                                         WeiLayout,
+                                                                         OutLayout,
+                                                                         InDataType,
+                                                                         WeiDataType,
+                                                                         OutDataType,
+                                                                         InElementwiseOperation,
+                                                                         WeiElementwiseOperation,
+                                                                         OutElementwiseOperation>
 {
-    using DeviceOp = DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl;
+    // 1d
+    static constexpr bool is_NWGK_GKXC_NWGC =
+        is_same_v<InLayout, tensor_layout::convolution::NWGC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::NWGK>;
+    static constexpr bool is_GNWK_GKXC_GNWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNWK>;
+    // 2d
+    static constexpr bool is_NHWGK_GKYXC_NHWGC =
+        is_same_v<InLayout, tensor_layout::convolution::NHWGC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::NHWGK>;
+    static constexpr bool is_GNHWK_GKYXC_GNHWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNHWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNHWK>;
+    // 3d
+    static constexpr bool is_NDHWGK_GKZYXC_NDHWGC =
+        is_same_v<InLayout, tensor_layout::convolution::NDHWGC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::NDHWGK>;
+    static constexpr bool is_GNDHWK_GKZYXC_GNDHWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNDHWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNDHWK>;
+
+    using DeviceOp = DeviceGroupedConvBwdWeight_Dl;
 
     using ADataType = OutDataType;
     using BDataType = InDataType;
@@ -176,6 +215,8 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     static constexpr auto I4 = Number<4>{};
     static constexpr auto I5 = Number<5>{};
 
+    static constexpr auto spatial_offset = I3;
+
     static constexpr auto K1Number     = Number<K1>{};
     static constexpr auto GemmK1Number = K1Number;
 
@@ -195,12 +236,12 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 1, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        const ck::index_t N,
-        const ck::index_t K,
-        const ck::index_t C,
-        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -209,90 +250,102 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     {
         using namespace ck;
 
-        const index_t Wi            = input_spatial_lengths[0];
-        const index_t Wo            = output_spatial_lengths[0];
-        const index_t X             = filter_spatial_lengths[0];
-        const index_t InLeftPadW    = input_left_pads[0];
-        const index_t InRightPadW   = input_right_pads[0];
-        const index_t ConvStrideW   = conv_filter_strides[0];
-        const index_t ConvDilationW = conv_filter_dilations[0];
+        const index_t N             = a_g_n_c_wis_lengths[I1];
+        const index_t K             = b_g_k_c_xs_lengths[I1];
+        const index_t C             = a_g_n_c_wis_lengths[I2];
+        const index_t Wi            = a_g_n_c_wis_lengths[spatial_offset];
+        const index_t Wo            = e_g_n_k_wos_lengths[spatial_offset];
+        const index_t X             = b_g_k_c_xs_lengths[spatial_offset];
+        const index_t InLeftPadW    = input_left_pads[I0];
+        const index_t InRightPadW   = input_right_pads[I0];
+        const index_t ConvStrideW   = conv_filter_strides[I0];
+        const index_t ConvDilationW = conv_filter_dilations[I0];
+
+        const auto InNStride  = a_g_n_c_wis_strides[I1];
+        const auto InCStride  = a_g_n_c_wis_strides[I2];
+        const auto InWStride  = a_g_n_c_wis_strides[spatial_offset];
+        const auto WeiKStride = b_g_k_c_xs_strides[I1];
+        const auto WeiCStride = b_g_k_c_xs_strides[I2];
+        const auto OutKStride = e_g_n_k_wos_strides[I2];
+        const auto OutWStride = e_g_n_k_wos_strides[spatial_offset];
 
         const index_t GemmKTotal = N * Wo;
-        const index_t GemmM      = K;
-        const index_t GemmN      = C * X;
-
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
             K0PerBlock;
-        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
 
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
             // A: output tensor
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Wo, K));
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Wo, K), make_tuple(OutWStride, OutKStride));
 
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto out_gemmkpad_gemmmpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    out_gemmktotal_gemmm_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, MPerBlock),
+                    Sequence<true, true>{});
 
             const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                out_gemmkpad_gemmmpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(out_gemmkpad_gemmmpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // B: input tensor
-            const auto in_gemmktotal_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Wi, C));
+            const auto in_gemmktotal_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Wi, C), make_tuple(InWStride, InCStride));
 
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto in_gemmkpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    in_gemmktotal_gemmn_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, NPerBlock),
+                    Sequence<true, true>{});
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                in_gemmkpad_gemmnpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(in_gemmkpad_gemmnpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // C: weights tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, X * C));
+            const auto wei_gemmm_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(K, X * C), make_tuple(WeiKStride, WeiCStride));
+
+            const auto wei_gemmmpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(wei_gemmm_gemmn_grid_desc,
+                                                                  make_tuple(MPerBlock, NPerBlock),
+                                                                  Sequence<true, true>{});
 
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_gemmmpad_gemmnpad_grid_desc);
         }
         else
         {
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Wo, K));
-            const auto in_n_wi_c_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Wi, C));
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Wo, K), make_tuple(OutWStride, OutKStride));
+            const auto in_n_wi_c_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N, Wi, C), make_tuple(InNStride, InWStride, InCStride));
 
             // A: output tensor
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto out_gemmkpad_gemmmpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    out_gemmktotal_gemmm_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, MPerBlock),
+                    Sequence<true, true>{});
 
             const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                out_gemmkpad_gemmmpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(out_gemmkpad_gemmmpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
@@ -321,38 +374,43 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                                             make_tuple(Sequence<1, 3>{}, Sequence<0, 2>{}),
                                             make_tuple(Sequence<1>{}, Sequence<0>{}));
 
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto in_gemmkpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    in_gemmktotal_gemmn_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, NPerBlock),
+                    Sequence<true, true>{});
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmN)),
+                in_gemmkpad_gemmnpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(in_gemmkpad_gemmnpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, X * C));
+            const auto wei_gemmm_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(K, X * C), make_tuple(WeiKStride, WeiCStride));
+
+            const auto wei_gemmmpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(wei_gemmm_gemmn_grid_desc,
+                                                                  make_tuple(MPerBlock, NPerBlock),
+                                                                  Sequence<true, true>{});
 
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_gemmmpad_gemmnpad_grid_desc);
         }
 
     } // function end
     template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        const ck::index_t N,
-        const ck::index_t K,
-        const ck::index_t C,
-        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -361,103 +419,111 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     {
         using namespace ck;
 
-        const index_t Hi = input_spatial_lengths[0];
-        const index_t Wi = input_spatial_lengths[1];
+        const index_t N  = a_g_n_c_wis_lengths[I1];
+        const index_t K  = b_g_k_c_xs_lengths[I1];
+        const index_t C  = a_g_n_c_wis_lengths[I2];
+        const index_t Hi = a_g_n_c_wis_lengths[spatial_offset];
+        const index_t Wi = a_g_n_c_wis_lengths[spatial_offset + I1];
+        const index_t Ho = e_g_n_k_wos_lengths[spatial_offset];
+        const index_t Wo = e_g_n_k_wos_lengths[spatial_offset + I1];
+        const index_t Y  = b_g_k_c_xs_lengths[spatial_offset];
+        const index_t X  = b_g_k_c_xs_lengths[spatial_offset + I1];
 
-        const index_t Ho = output_spatial_lengths[0];
-        const index_t Wo = output_spatial_lengths[1];
+        const index_t InLeftPadH    = input_left_pads[I0];
+        const index_t InLeftPadW    = input_left_pads[I1];
+        const index_t InRightPadH   = input_right_pads[I0];
+        const index_t InRightPadW   = input_right_pads[I1];
+        const index_t ConvStrideH   = conv_filter_strides[I0];
+        const index_t ConvStrideW   = conv_filter_strides[I1];
+        const index_t ConvDilationH = conv_filter_dilations[I0];
+        const index_t ConvDilationW = conv_filter_dilations[I1];
 
-        const index_t Y = filter_spatial_lengths[0];
-        const index_t X = filter_spatial_lengths[1];
-
-        const index_t InLeftPadH = input_left_pads[0];
-        const index_t InLeftPadW = input_left_pads[1];
-
-        const index_t InRightPadH = input_right_pads[0];
-        const index_t InRightPadW = input_right_pads[1];
-
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
-
-        const index_t ConvDilationH = conv_filter_dilations[0];
-        const index_t ConvDilationW = conv_filter_dilations[1];
+        const auto InNStride  = a_g_n_c_wis_strides[I1];
+        const auto InCStride  = a_g_n_c_wis_strides[I2];
+        const auto InHStride  = a_g_n_c_wis_strides[spatial_offset];
+        const auto InWStride  = a_g_n_c_wis_strides[spatial_offset + I1];
+        const auto WeiKStride = b_g_k_c_xs_strides[I1];
+        const auto WeiCStride = b_g_k_c_xs_strides[I2];
+        const auto OutKStride = e_g_n_k_wos_strides[I2];
+        const auto OutWStride = e_g_n_k_wos_strides[spatial_offset + I1];
 
         const index_t GemmKTotal = N * Ho * Wo;
-        const index_t GemmM      = K;
-        const index_t GemmN      = C * X * Y;
-
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
             K0PerBlock;
-        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
 
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
             // A: output tensor
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Ho * Wo, K), make_tuple(OutWStride, OutKStride));
 
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto out_gemmkpad_gemmmpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    out_gemmktotal_gemmm_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, MPerBlock),
+                    Sequence<true, true>{});
 
             const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                out_gemmkpad_gemmmpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(out_gemmkpad_gemmmpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // B: input tensor
-            const auto in_gemmktotal_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Hi * Wi, C));
+            const auto in_gemmktotal_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Hi * Wi, C), make_tuple(InWStride, InCStride));
 
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto in_gemmkpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    in_gemmktotal_gemmn_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, NPerBlock),
+                    Sequence<true, true>{});
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                in_gemmkpad_gemmnpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(in_gemmkpad_gemmnpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+            const auto wei_gemmm_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(K, Y * X * C), make_tuple(WeiKStride, WeiCStride));
+
+            const auto wei_gemmmpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(wei_gemmm_gemmn_grid_desc,
+                                                                  make_tuple(MPerBlock, NPerBlock),
+                                                                  Sequence<true, true>{});
 
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_gemmmpad_gemmnpad_grid_desc);
         }
         else
         {
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Ho * Wo, K));
-            const auto in_n_hi_wi_c_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Ho * Wo, K), make_tuple(OutWStride, OutKStride));
+            const auto in_n_hi_wi_c_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N, Hi, Wi, C), make_tuple(InNStride, InHStride, InWStride, InCStride));
 
             // A: output tensor
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto out_gemmkpad_gemmmpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    out_gemmktotal_gemmm_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, MPerBlock),
+                    Sequence<true, true>{});
 
             const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                out_gemmkpad_gemmmpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(out_gemmkpad_gemmmpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
@@ -488,39 +554,44 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                                             make_tuple(Sequence<1, 3, 5>{}, Sequence<0, 2, 4>{}),
                                             make_tuple(Sequence<1>{}, Sequence<0>{}));
 
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto in_gemmkpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    in_gemmktotal_gemmn_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, NPerBlock),
+                    Sequence<true, true>{});
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmN)),
+                in_gemmkpad_gemmnpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(in_gemmkpad_gemmnpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Y * X * C));
+            const auto wei_gemmm_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(K, Y * X * C), make_tuple(WeiKStride, WeiCStride));
+
+            const auto wei_gemmmpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(wei_gemmm_gemmn_grid_desc,
+                                                                  make_tuple(MPerBlock, NPerBlock),
+                                                                  Sequence<true, true>{});
 
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_gemmmpad_gemmnpad_grid_desc);
         }
 
     } // function end
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
-        const ck::index_t N,
-        const ck::index_t K,
-        const ck::index_t C,
-        const std::array<ck::index_t, NDimSpatial>& input_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& filter_spatial_lengths,
-        const std::array<ck::index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
         const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -529,110 +600,120 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     {
         using namespace ck;
 
-        const index_t Di = input_spatial_lengths[0];
-        const index_t Hi = input_spatial_lengths[1];
-        const index_t Wi = input_spatial_lengths[2];
+        const index_t N  = a_g_n_c_wis_lengths[I1];
+        const index_t K  = b_g_k_c_xs_lengths[I1];
+        const index_t C  = a_g_n_c_wis_lengths[I2];
+        const index_t Di = a_g_n_c_wis_lengths[spatial_offset + I0];
+        const index_t Hi = a_g_n_c_wis_lengths[spatial_offset + I1];
+        const index_t Wi = a_g_n_c_wis_lengths[spatial_offset + I2];
+        const index_t Do = e_g_n_k_wos_lengths[spatial_offset + I0];
+        const index_t Ho = e_g_n_k_wos_lengths[spatial_offset + I1];
+        const index_t Wo = e_g_n_k_wos_lengths[spatial_offset + I2];
+        const index_t Z  = b_g_k_c_xs_lengths[spatial_offset + I0];
+        const index_t Y  = b_g_k_c_xs_lengths[spatial_offset + I1];
+        const index_t X  = b_g_k_c_xs_lengths[spatial_offset + I2];
 
-        const index_t Do = output_spatial_lengths[0];
-        const index_t Ho = output_spatial_lengths[1];
-        const index_t Wo = output_spatial_lengths[2];
+        const index_t InLeftPadD    = input_left_pads[I0];
+        const index_t InLeftPadH    = input_left_pads[I1];
+        const index_t InLeftPadW    = input_left_pads[I2];
+        const index_t InRightPadD   = input_right_pads[I0];
+        const index_t InRightPadH   = input_right_pads[I1];
+        const index_t InRightPadW   = input_right_pads[I2];
+        const index_t ConvStrideD   = conv_filter_strides[I0];
+        const index_t ConvStrideH   = conv_filter_strides[I1];
+        const index_t ConvStrideW   = conv_filter_strides[I2];
+        const index_t ConvDilationD = conv_filter_dilations[I0];
+        const index_t ConvDilationH = conv_filter_dilations[I1];
+        const index_t ConvDilationW = conv_filter_dilations[I2];
 
-        const index_t Z = filter_spatial_lengths[0];
-        const index_t Y = filter_spatial_lengths[1];
-        const index_t X = filter_spatial_lengths[2];
-
-        const index_t InLeftPadD = input_left_pads[0];
-        const index_t InLeftPadH = input_left_pads[1];
-        const index_t InLeftPadW = input_left_pads[2];
-
-        const index_t InRightPadD = input_right_pads[0];
-        const index_t InRightPadH = input_right_pads[1];
-        const index_t InRightPadW = input_right_pads[2];
-
-        const index_t ConvStrideD = conv_filter_strides[0];
-        const index_t ConvStrideH = conv_filter_strides[1];
-        const index_t ConvStrideW = conv_filter_strides[2];
-
-        const index_t ConvDilationD = conv_filter_dilations[0];
-        const index_t ConvDilationH = conv_filter_dilations[1];
-        const index_t ConvDilationW = conv_filter_dilations[2];
+        const auto InNStride  = a_g_n_c_wis_strides[I1];
+        const auto InCStride  = a_g_n_c_wis_strides[I2];
+        const auto InDStride  = a_g_n_c_wis_strides[spatial_offset];
+        const auto InHStride  = a_g_n_c_wis_strides[spatial_offset + I1];
+        const auto InWStride  = a_g_n_c_wis_strides[spatial_offset + I2];
+        const auto WeiKStride = b_g_k_c_xs_strides[I1];
+        const auto WeiCStride = b_g_k_c_xs_strides[I2];
+        const auto OutKStride = e_g_n_k_wos_strides[I2];
+        const auto OutWStride = e_g_n_k_wos_strides[spatial_offset + I2];
 
         const index_t GemmKTotal = N * Do * Ho * Wo;
-        const index_t GemmM      = K;
-        const index_t GemmN      = C * Z * X * Y;
-
         const index_t GemmKBatch = batch_k;
         const index_t GemmK0 =
             math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock * GemmKBatch) *
             K0PerBlock;
-        const index_t GemmKPad = GemmKBatch * GemmK0 * GemmK1Number;
 
         if constexpr(ConvBackwardWeightSpecialization ==
                      ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
         {
             // A: output tensor
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Do * Ho * Wo, K), make_tuple(OutWStride, OutKStride));
 
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto out_gemmkpad_gemmmpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    out_gemmktotal_gemmm_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, MPerBlock),
+                    Sequence<true, true>{});
 
             const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                out_gemmkpad_gemmmpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(out_gemmkpad_gemmmpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // B: input tensor
-            const auto in_gemmktotal_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Di * Hi * Wi, C));
+            const auto in_gemmktotal_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Di * Hi * Wi, C), make_tuple(InWStride, InCStride));
 
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto in_gemmkpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    in_gemmktotal_gemmn_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, NPerBlock),
+                    Sequence<true, true>{});
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                in_gemmkpad_gemmnpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(in_gemmkpad_gemmnpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Z * Y * X * C));
+            const auto wei_gemmm_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(K, Z * Y * X * C), make_tuple(WeiKStride, WeiCStride));
+
+            const auto wei_gemmmpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(wei_gemmm_gemmn_grid_desc,
+                                                                  make_tuple(MPerBlock, NPerBlock),
+                                                                  Sequence<true, true>{});
 
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_gemmmpad_gemmnpad_grid_desc);
         }
         else
         {
-            const auto out_gemmktotal_gemmm_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N * Do * Ho * Wo, K));
-            const auto in_n_di_hi_wi_c_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
+            const auto out_gemmktotal_gemmm_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N * Do * Ho * Wo, K), make_tuple(OutWStride, OutKStride));
+            const auto in_n_di_hi_wi_c_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(N, Di, Hi, Wi, C),
+                make_tuple(InNStride, InDStride, InHStride, InWStride, InCStride));
 
             // A: output tensor
-            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
-                out_gemmktotal_gemmm_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmM)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto out_gemmkpad_gemmmpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    out_gemmktotal_gemmm_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, MPerBlock),
+                    Sequence<true, true>{});
 
             const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
-                out_gemmkpad_gemmm_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmM)),
+                out_gemmkpad_gemmmpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(out_gemmkpad_gemmmpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
@@ -672,27 +753,32 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                 make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}),
                 make_tuple(Sequence<1>{}, Sequence<0>{}));
 
-            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
-                in_gemmktotal_gemmn_grid_desc,
-                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
-                           make_pass_through_transform(GemmN)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto in_gemmkpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(
+                    in_gemmktotal_gemmn_grid_desc,
+                    make_tuple(GemmK1Number * K0PerBlock * GemmKBatch, NPerBlock),
+                    Sequence<true, true>{});
 
             const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
-                in_gemmkpad_gemmn_grid_desc,
-                make_tuple(make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
-                           make_pass_through_transform(GemmN)),
+                in_gemmkpad_gemmnpad_grid_desc,
+                make_tuple(
+                    make_unmerge_transform(make_tuple(GemmKBatch, GemmK0, GemmK1Number)),
+                    make_pass_through_transform(in_gemmkpad_gemmnpad_grid_desc.GetLength(I1))),
                 make_tuple(Sequence<0>{}, Sequence<1>{}),
                 make_tuple(Sequence<0, 1, 3>{}, Sequence<2>{}));
 
             // C: weight tensor
-            const auto wei_gemmm_gemmn_grid_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(K, Z * Y * X * C));
+            const auto wei_gemmm_gemmn_grid_desc = make_naive_tensor_descriptor(
+                make_tuple(K, Z * Y * X * C), make_tuple(WeiKStride, WeiCStride));
+
+            const auto wei_gemmmpad_gemmnpad_grid_desc =
+                ck::tensor_operation::device::PadTensorDescriptor(wei_gemmm_gemmn_grid_desc,
+                                                                  make_tuple(MPerBlock, NPerBlock),
+                                                                  Sequence<true, true>{});
 
             return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
                               in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
-                              wei_gemmm_gemmn_grid_desc);
+                              wei_gemmmpad_gemmnpad_grid_desc);
         }
 
     } // function end
@@ -701,22 +787,22 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
     static auto GetABCGridDesc()
     {
         return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<1>(
-            1, 1, 1, {1}, {1}, {1}, {1}, {1}, {1}, {1}, 1);
+            {1}, {1}, {1}, {1}, {1}, {1}, {1}, {1}, {1}, {1}, 1);
     }
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 2, bool>::type = false>
     static auto GetABCGridDesc()
     {
         return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<2>(
-            1, 1, 1, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, 1);
+            {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, {1, 1}, 1);
     }
 
     template <ck::index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
     static auto GetABCGridDesc()
     {
-        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<3>(1,
-                                                                  1,
-                                                                  1,
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<3>({1, 1, 1},
+                                                                  {1, 1, 1},
+                                                                  {1, 1, 1},
                                                                   {1, 1, 1},
                                                                   {1, 1, 1},
                                                                   {1, 1, 1},
@@ -785,11 +871,11 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                  WeiDataType* p_wei_grid,
                  const OutDataType* p_out_grid,
                  const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
-                 const std::array<index_t, NDimSpatial + 3>& /*a_g_n_c_wis_strides*/,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
                  const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
-                 const std::array<index_t, NDimSpatial + 3>& /*b_g_k_c_xs_strides*/,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
                  const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
-                 const std::array<index_t, NDimSpatial + 3>& /*e_g_n_k_wos_strides*/,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
                  const std::array<ck::index_t, NDimSpatial>& conv_filter_strides,
                  const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations,
                  const std::array<ck::index_t, NDimSpatial>& input_left_pads,
@@ -809,38 +895,24 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
               a_element_op_{out_element_op},
               b_element_op_{wei_element_op},
               c_element_op_{in_element_op},
-              Conv_G_{a_g_n_c_wis_lengths[0]},
-              Conv_N_{a_g_n_c_wis_lengths[1]},
-              Conv_K_{b_g_k_c_xs_lengths[1]},
-              Conv_C_{a_g_n_c_wis_lengths[2]},
-              input_spatial_lengths_{},
-              filter_spatial_lengths_{},
-              output_spatial_lengths_{},
+              Conv_G_{a_g_n_c_wis_lengths[I0]},
+              Conv_K_{b_g_k_c_xs_lengths[I1]},
+              Conv_C_{a_g_n_c_wis_lengths[I2]},
+              filter_lengths_{b_g_k_c_xs_lengths},
               conv_filter_strides_{conv_filter_strides},
               conv_filter_dilations_{conv_filter_dilations},
               input_left_pads_{input_left_pads},
               input_right_pads_{input_right_pads},
               k_batch_{split_k}
         {
-            constexpr index_t spatial_offset = 3;
-            std::copy(begin(a_g_n_c_wis_lengths) + spatial_offset,
-                      end(a_g_n_c_wis_lengths),
-                      begin(input_spatial_lengths_));
-            std::copy(begin(b_g_k_c_xs_lengths) + spatial_offset,
-                      end(b_g_k_c_xs_lengths),
-                      begin(filter_spatial_lengths_));
-            std::copy(begin(e_g_n_k_wos_lengths) + spatial_offset,
-                      end(e_g_n_k_wos_lengths),
-                      begin(output_spatial_lengths_));
-
             const auto descs =
                 DeviceOp::MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<NDimSpatial>(
-                    Conv_N_,
-                    Conv_K_,
-                    Conv_C_,
-                    input_spatial_lengths_,
-                    filter_spatial_lengths_,
-                    output_spatial_lengths_,
+                    a_g_n_c_wis_lengths, // input
+                    a_g_n_c_wis_strides,
+                    b_g_k_c_xs_lengths, // weight
+                    b_g_k_c_xs_strides,
+                    e_g_n_k_wos_lengths, // output
+                    e_g_n_k_wos_strides,
                     conv_filter_strides,
                     conv_filter_dilations,
                     input_left_pads,
@@ -863,24 +935,9 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
                 GridwiseGemm::MakeCBlockClusterAdaptor(c_grid_desc_m_n_, M01, N01, k_batch_);
 
             // A/B/C Batch Stride
-            compute_ptr_offset_of_batch_.BatchStrideA_ =
-                Conv_N_ * Conv_K_ *
-                std::accumulate(begin(output_spatial_lengths_),
-                                end(output_spatial_lengths_),
-                                index_t{1},
-                                std::multiplies<>{});
-            compute_ptr_offset_of_batch_.BatchStrideB_ =
-                Conv_N_ * Conv_C_ *
-                std::accumulate(begin(input_spatial_lengths_),
-                                end(input_spatial_lengths_),
-                                index_t{1},
-                                std::multiplies<>{});
-            compute_ptr_offset_of_batch_.BatchStrideC_ =
-                Conv_K_ * Conv_C_ *
-                std::accumulate(begin(filter_spatial_lengths_),
-                                end(filter_spatial_lengths_),
-                                index_t{1},
-                                std::multiplies<>{});
+            compute_ptr_offset_of_batch_.BatchStrideA_ = e_g_n_k_wos_strides[I0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = a_g_n_c_wis_strides[I0];
+            compute_ptr_offset_of_batch_.BatchStrideC_ = b_g_k_c_xs_strides[I0];
         }
 
         const ADataType* p_a_grid_;
@@ -908,13 +965,10 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
 
         // for checking IsSupportedArgument()
         const index_t Conv_G_;
-        const index_t Conv_N_;
         const index_t Conv_K_;
         const index_t Conv_C_;
 
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths_;
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths_;
+        std::array<ck::index_t, NDimSpatial + 3> filter_lengths_;
         const std::array<ck::index_t, NDimSpatial>& conv_filter_strides_;
         const std::array<ck::index_t, NDimSpatial>& conv_filter_dilations_;
         const std::array<ck::index_t, NDimSpatial>& input_left_pads_;
@@ -1036,10 +1090,14 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        // check device
-        if(!(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
-             ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
-             ck::get_device_name() == "gfx1102"))
+
+        // DL version only supports split_k equal to 1
+        if(arg.k_batch_ != 1)
+            return false;
+
+        if constexpr(!((NDimSpatial == 1 && (is_NWGK_GKXC_NWGC || is_GNWK_GKXC_GNWC)) ||
+                       (NDimSpatial == 2 && (is_NHWGK_GKYXC_NHWGC || is_GNHWK_GKYXC_GNHWC)) ||
+                       (NDimSpatial == 3 && (is_NDHWGK_GKZYXC_NDHWGC || is_GNDHWK_GKZYXC_GNDHWC))))
         {
             return false;
         }
@@ -1050,8 +1108,9 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
             // check if it's 1x1, stride=1 pad = 0 conv
             for(int i = 0; i < NDimSpatial; i++)
             {
-                if(!(arg.filter_spatial_lengths_[i] == 1 && arg.conv_filter_strides_[i] == 1 &&
-                     arg.input_left_pads_[i] == 0 && arg.input_right_pads_[i] == 0))
+                if(!(arg.filter_lengths_[spatial_offset + i] == 1 &&
+                     arg.conv_filter_strides_[i] == 1 && arg.input_left_pads_[i] == 0 &&
+                     arg.input_right_pads_[i] == 0))
                 {
                     return false;
                 }
@@ -1206,7 +1265,7 @@ struct DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceGroupedConvBwdWeightGnwcGkxcGnwk_Dl"
+        str << "DeviceGroupedConvBwdWeight_Dl"
             << "<"
             << BlockSize << ", "
             << MPerBlock << ", "
diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index b58b2b3319..16c139c008 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -72,6 +72,18 @@ inner_product<float4_t, float4_t, float>(const float4_t& a, const float4_t& b, f
                   c);
 }
 
+template <>
+__device__ void inner_product<bhalf_t, bhalf_t, float>(const bhalf_t& a, const bhalf_t& b, float& c)
+{
+    inner_product(type_convert<float>(a), type_convert<float>(b), c);
+}
+
+template <>
+__device__ void inner_product<half_t, half_t, float>(const half_t& a, const half_t& b, float& c)
+{
+    inner_product(type_convert<float>(a), type_convert<float>(b), c);
+}
+
 template <>
 __device__ void inner_product<half2_t, half2_t, float>(const half2_t& a, const half2_t& b, float& c)
 {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
new file mode 100644
index 0000000000..29fe298e42
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdWeightDefault =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
+
+static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_dl_f32_instances = std::tuple<
+    // clang-format off
+        //############################|        Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1|  M1Per|  N1Per|   KPer|  M1N1Thread|  M1N1Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|         ABlockTransfer|     ABlockTransfer|         ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|         BBlockTransfer|     BBlockTransfer|         BBlockTransfer|   CThreadTransfer| CThreadTransfer|    CThreadTransfer| 
+        //############################|        Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   | Thread| Thread| Thread| ClusterM1Xs| ClusterN1Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| SrcDstAccessOrder| SrcDstVectorDim| DstScalarPerVector|
+        //############################|    Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |       |       |       |            |            |       _K0_M0_M1_K1|         _K0_M0_M1_K1|   ArrangeOrder|               |           _K0_M0_M1_K1| ContiguousDimOrder|           _K0_M0_M1_K1|       _K0_N0_N1_K1|         _K0_N0_N1_K1|   ArrangeOrder|               |           _K0_N0_N1_K1| ContiguousDimOrder|           _K0_N0_N1_K1|                  |                |                   |
+        //############################|           |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |       |       |       |            |            |                   |                     |               |               |                       |                   |                       |                   |                     |               |               |                       |                   |                       |                  |                |                   |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Dl< NDimSpatial,  ALayout,   BLayout,   ELayout,    F32,     F32,     F32,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,    16,  1,      4,      4,      1,     S<8, 2>,     S<8, 2>,   S<1, 8, 1, 1, 1>,   S<1, 2, 1, 128, 1>, S<0, 2, 3, 1, 4>, S<0, 2, 3, 1, 4>,   S<1, 1, 1, 1, 1>,   S<0, 2, 3, 1, 4>,       S<1, 1, 1, 1, 1>,   S<1, 1, 1, 8, 1>,   S<1, 16, 1, 16, 1>, S<0, 1, 4, 2, 3>, S<0, 1, 4, 2, 3>,   S<1, 1, 1, 1, 1>,   S<0, 1, 4, 2, 3>,       S<1, 1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,             5,                   1>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_dl_f16_instances = std::tuple<
+    // clang-format off
+        //############################|        Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1|  M1Per|  N1Per|   KPer|  M1N1Thread|  M1N1Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|         ABlockTransfer|     ABlockTransfer|         ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|         BBlockTransfer|     BBlockTransfer|         BBlockTransfer|   CThreadTransfer| CThreadTransfer|    CThreadTransfer| 
+        //############################|        Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   | Thread| Thread| Thread| ClusterM1Xs| ClusterN1Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| SrcDstAccessOrder| SrcDstVectorDim| DstScalarPerVector|
+        //############################|    Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |       |       |       |            |            |       _K0_M0_M1_K1|         _K0_M0_M1_K1|   ArrangeOrder|               |           _K0_M0_M1_K1| ContiguousDimOrder|           _K0_M0_M1_K1|       _K0_N0_N1_K1|         _K0_N0_N1_K1|   ArrangeOrder|               |           _K0_N0_N1_K1| ContiguousDimOrder|           _K0_N0_N1_K1|                  |                |                   |
+        //############################|           |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |       |       |       |            |            |                   |                     |               |               |                       |                   |                       |                   |                     |               |               |                       |                   |                       |                  |                |                   |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Dl< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,    16,  1,      4,      4,      1,     S<8, 2>,     S<8, 2>,   S<1, 8, 1, 1, 1>,   S<1, 2, 1, 128, 1>, S<0, 2, 3, 1, 4>, S<0, 2, 3, 1, 4>,   S<1, 1, 1, 1, 1>,   S<0, 2, 3, 1, 4>,       S<1, 1, 1, 1, 1>,   S<1, 1, 1, 8, 1>,   S<1, 16, 1, 16, 1>, S<0, 1, 4, 2, 3>, S<0, 1, 4, 2, 3>,   S<1, 1, 1, 1, 1>,   S<0, 1, 4, 2, 3>,       S<1, 1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,             5,                   1>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_dl_bf16_instances = std::tuple<
+    // clang-format off
+        //############################|        Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1|  M1Per|  N1Per|   KPer|  M1N1Thread|  M1N1Thread|     ABlockTransfer|       ABlockTransfer| ABlockTransfer| ABlockTransfer|         ABlockTransfer|     ABlockTransfer|         ABlockTransfer|     BBlockTransfer|       BBlockTransfer| BBlockTransfer| BBlockTransfer|         BBlockTransfer|     BBlockTransfer|         BBlockTransfer|   CThreadTransfer| CThreadTransfer|    CThreadTransfer| 
+        //############################|        Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   | Thread| Thread| Thread| ClusterM1Xs| ClusterN1Xs| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| ThreadSliceLengths| ThreadClusterLengths|  ThreadCluster| SrcAccessOrder| SrcVectorTensorLengths|    SrcVectorTensor| DstVectorTensorLengths| SrcDstAccessOrder| SrcDstVectorDim| DstScalarPerVector|
+        //############################|    Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |       |       |       |            |            |       _K0_M0_M1_K1|         _K0_M0_M1_K1|   ArrangeOrder|               |           _K0_M0_M1_K1| ContiguousDimOrder|           _K0_M0_M1_K1|       _K0_N0_N1_K1|         _K0_N0_N1_K1|   ArrangeOrder|               |           _K0_N0_N1_K1| ContiguousDimOrder|           _K0_N0_N1_K1|                  |                |                   |
+        //############################|           |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |       |       |       |            |            |                   |                     |               |               |                       |                   |                       |                   |                     |               |               |                       |                   |                       |                  |                |                   |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Dl< NDimSpatial,  ALayout,   BLayout,   ELayout,   BF16,     F32,    BF16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,    16,  1,      4,      4,      1,     S<8, 2>,     S<8, 2>,   S<1, 8, 1, 1, 1>,   S<1, 2, 1, 128, 1>, S<0, 2, 3, 1, 4>, S<0, 2, 3, 1, 4>,   S<1, 1, 1, 1, 1>,   S<0, 2, 3, 1, 4>,       S<1, 1, 1, 1, 1>,   S<1, 1, 1, 8, 1>,   S<1, 16, 1, 16, 1>, S<0, 1, 4, 2, 3>, S<0, 1, 4, 2, 3>,   S<1, 1, 1, 1, 1>,   S<0, 1, 4, 2, 3>,       S<1, 1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,             5,                   1>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
index 86ff43e5cf..1ebd0c3d63 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
@@ -55,8 +55,8 @@ using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
         // generic instances
         // TODO: Change to ScalarPerVector = 1 when inner_product<half_t, half_t, float> will be supported
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,     4,     2,  2,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 2>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 1, 1, 2>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 2>,
-        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 2>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,     4,     2,  1,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 1>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 1, 1, 1>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
+        DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
 
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,       S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,               5,                 4>
     // clang-format on
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index cad3e1ace8..c7a694b4cc 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -17,6 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+// xdl
 // conv1d backward weight
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
@@ -200,6 +201,228 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 
+#ifdef DL_KERNELS
+// dl
+// conv1d backward weight
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+// conv2d backward weight
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+// conv3d backward weight
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
@@ -242,21 +465,54 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
+#endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
+#endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
             }
+            else if constexpr(is_same_v<InLayout, NWGC> && is_same_v<WeiLayout, GKXC> &&
+                              is_same_v<OutLayout, NWGK>)
+            {
+#ifdef DL_KERNELS
+                if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, float>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<WeiDataType, float> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+                }
+#endif
+            }
         }
         else if constexpr(NumDimSpatial == 2)
         {
@@ -266,12 +522,20 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                         op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                         op_ptrs);
                 }
@@ -279,6 +543,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
@@ -289,12 +557,20 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                         op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                         op_ptrs);
                 }
@@ -302,6 +578,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
@@ -315,12 +595,20 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                         op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                         op_ptrs);
                 }
@@ -328,6 +616,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
@@ -338,12 +630,20 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                         op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                         op_ptrs);
                 }
@@ -351,6 +651,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
+#ifdef DL_KERNELS
+                    add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+                        op_ptrs);
+#endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt
index 3808e0248f..660f2544ac 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt
@@ -1,5 +1,16 @@
-add_instance_library(device_grouped_conv1d_bwd_weight_instance
+set(GROUPED_CONV1D_BWD_WEIGHT
     device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
     device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
-    device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
-)
+    device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp)
+
+if(DL_KERNELS)
+    list(APPEND GROUPED_CONV1D_BWD_WEIGHT
+        device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
+        device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
+        device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
+        device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
+        device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
+        device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp)
+endif()
+
+add_instance_library(device_grouped_conv1d_bwd_weight_instance ${GROUPED_CONV1D_BWD_WEIGHT})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
new file mode 100644
index 0000000000..d7a15784a2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<1,
+                                                         GNWC,
+                                                         GKXC,
+                                                         GNWK,
+                                                         ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<1,
+                                                         GNWC,
+                                                         GKXC,
+                                                         GNWK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
new file mode 100644
index 0000000000..6f891ef94a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<1,
+                                                        GNWC,
+                                                        GKXC,
+                                                        GNWK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<1,
+                                                        GNWC,
+                                                        GKXC,
+                                                        GNWK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
new file mode 100644
index 0000000000..2db7fbdb9f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           GNWC,
+                                                           GKXC,
+                                                           GNWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<1,
+                                                        GNWC,
+                                                        GKXC,
+                                                        GNWK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<1,
+                                                        GNWC,
+                                                        GKXC,
+                                                        GNWK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..a92cb4285d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<1,
+                                                         NWGC,
+                                                         GKXC,
+                                                         NWGK,
+                                                         ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<1,
+                                                         NWGC,
+                                                         GKXC,
+                                                         NWGK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
new file mode 100644
index 0000000000..713d70f370
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<1,
+                                                        NWGC,
+                                                        GKXC,
+                                                        NWGK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<1,
+                                                        NWGC,
+                                                        GKXC,
+                                                        NWGK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
new file mode 100644
index 0000000000..05b29af74c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
+                                                           NWGC,
+                                                           GKXC,
+                                                           NWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<1,
+                                                        NWGC,
+                                                        GKXC,
+                                                        NWGK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<1,
+                                                        NWGC,
+                                                        GKXC,
+                                                        NWGK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
index 25ea4f48cf..f9368ab575 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
@@ -1,13 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
index 8065012f12..edba11efba 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
@@ -1,13 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
index c70a54c2dc..5f28994f09 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
@@ -1,13 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
index b7b9fc92d1..578e9029e9 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
@@ -1,9 +1,19 @@
-add_instance_library(device_grouped_conv2d_bwd_weight_instance
+set(GROUPED_CONV2D_BWD_WEIGHT
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
     device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
-    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
-)
+    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp)
 
+if(DL_KERNELS)
+    list(APPEND GROUPED_CONV2D_BWD_WEIGHT
+        device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+        device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+        device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+        device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+        device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+        device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp)
+endif()
+
+add_instance_library(device_grouped_conv2d_bwd_weight_instance ${GROUPED_CONV2D_BWD_WEIGHT})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
new file mode 100644
index 0000000000..37b465e6c4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<2,
+                                                         GNHWC,
+                                                         GKYXC,
+                                                         GNHWK,
+                                                         ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<2,
+                                                         GNHWC,
+                                                         GKYXC,
+                                                         GNHWK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
new file mode 100644
index 0000000000..64b38655e4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<2,
+                                                        GNHWC,
+                                                        GKYXC,
+                                                        GNHWK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<2,
+                                                        GNHWC,
+                                                        GKYXC,
+                                                        GNHWK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
new file mode 100644
index 0000000000..2cc6527dfb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           GNHWC,
+                                                           GKYXC,
+                                                           GNHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<2,
+                                                        GNHWC,
+                                                        GKYXC,
+                                                        GNHWK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<2,
+                                                        GNHWC,
+                                                        GKYXC,
+                                                        GNHWK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..cf3db83319
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<2,
+                                                         NHWGC,
+                                                         GKYXC,
+                                                         NHWGK,
+                                                         ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<2,
+                                                         NHWGC,
+                                                         GKYXC,
+                                                         NHWGK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..eb592b8372
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        NHWGK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        NHWGK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..ed7ab6f4a4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
+                                                           NHWGC,
+                                                           GKYXC,
+                                                           NHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        NHWGK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<2,
+                                                        NHWGC,
+                                                        GKYXC,
+                                                        NHWGK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
index cf39c86010..17f5ee4e26 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index d52f0b4d1c..abb90b7c71 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
index 62547a5014..2e0fef9cf8 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
index 1cb9991a86..614cc0a7e8 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
index e64d55c3bb..f6e1ada352 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
index 5fa4c9ba3c..384706414a 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
index 5118599b4f..6bd9d6e646 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
@@ -1,8 +1,19 @@
-add_instance_library(device_grouped_conv3d_bwd_weight_instance
+set(GROUPED_CONV3D_BWD_WEIGHT
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
-)
+    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp)
+
+if(DL_KERNELS)
+    list(APPEND GROUPED_CONV3D_BWD_WEIGHT
+        device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+        device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+        device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+        device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+        device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+        device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp)
+endif()
+
+add_instance_library(device_grouped_conv3d_bwd_weight_instance ${GROUPED_CONV3D_BWD_WEIGHT})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
new file mode 100644
index 0000000000..c9646d0856
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<3,
+                                                         GNDHWC,
+                                                         GKZYXC,
+                                                         GNDHWK,
+                                                         ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<3,
+                                                         GNDHWC,
+                                                         GKZYXC,
+                                                         GNDHWK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
new file mode 100644
index 0000000000..4eb54d0f90
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<3,
+                                                        GNDHWC,
+                                                        GKZYXC,
+                                                        GNDHWK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<3,
+                                                        GNDHWC,
+                                                        GKZYXC,
+                                                        GNDHWK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
new file mode 100644
index 0000000000..c22825ccec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -0,0 +1,45 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<3,
+                                                        GNDHWC,
+                                                        GKZYXC,
+                                                        GNDHWK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<3,
+                                                        GNDHWC,
+                                                        GKZYXC,
+                                                        GNDHWK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
new file mode 100644
index 0000000000..a37e6cbf3a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           BF16,
+                                                           F32,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<3,
+                                                         NDHWGC,
+                                                         GKZYXC,
+                                                         NDHWGK,
+                                                         ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_bf16_instances<3,
+                                                         NDHWGC,
+                                                         GKZYXC,
+                                                         NDHWGK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..b548c6fd50
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<3,
+                                                        NDHWGC,
+                                                        GKZYXC,
+                                                        NDHWGK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f16_instances<3,
+                                                        NDHWGC,
+                                                        GKZYXC,
+                                                        NDHWGK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
new file mode 100644
index 0000000000..10d708a3fe
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F32,
+                                                           F32,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<3,
+                                                        NDHWGC,
+                                                        GKZYXC,
+                                                        NDHWGK,
+                                                        ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_dl_f32_instances<3,
+                                                        NDHWGC,
+                                                        GKZYXC,
+                                                        NDHWGK,
+                                                        ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
index c8f456db83..91d80e4f73 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
@@ -1,13 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
index 099123ecbc..d03f0a7bac 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -1,13 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
index 0eda980b30..7c24cc8fd4 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
@@ -1,13 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
index 1e5c1946fb..a394e0d6f7 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
index e0b442bf24..eba721c7b8 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
index 4bb7948245..7dd289139c 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
@@ -2,7 +2,6 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 #include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
 
 namespace ck {
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
index ae15f6c0c9..bf95727fbc 100644
--- a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -14,6 +14,8 @@
 
 #include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
 
+using namespace ck::tensor_layout::convolution;
+
 template <typename Tuple>
 class TestGroupedConvndBwdWeight : public ::testing::Test
 {
@@ -27,28 +29,59 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
     using NDimSpatial = std::tuple_element_t<6, Tuple>;
 
     std::vector<ck::utils::conv::ConvParam> conv_params;
-    ck::index_t split_k{2};
+    std::vector<ck::index_t> split_ks{1, 2};
+
+    bool skip_case(const ck::utils::conv::ConvParam& params, const ck::index_t split_k)
+    {
+        // K or C are odd is supported only by DL kernel (only applies to fp16)
+        // DL kernel is only supported for split_k=1
+        if constexpr(std::is_same_v<InDataType, ck::half_t>)
+        {
+            if(split_k != 1 && (params.K_ % 2 != 0 || params.C_ % 2 != 0))
+            {
+                return true;
+            }
+        }
+
+        // 1d NWGC is only supported by DL kernel
+        // DL kernel is only supported for split_k=1
+        if constexpr(std::is_same_v<InLayout, NWGC> && std::is_same_v<OutLayout, NWGK>)
+        {
+            if(split_k != 1)
+            {
+                return true;
+            }
+        }
+
+        return false;
+    }
 
     void Run()
     {
         EXPECT_FALSE(conv_params.empty());
         bool pass = true;
 
-        for(auto& param : conv_params)
+        for(auto split_k : split_ks)
         {
-            pass = pass && ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial{},
-                                                                              InLayout,
-                                                                              WeiLayout,
-                                                                              OutLayout,
-                                                                              InDataType,
-                                                                              WeiDataType,
-                                                                              OutDataType>(
-                               true,  // do_verification
-                               1,     // init_method: integer value
-                               false, // do_log
-                               false, // time_kernel
-                               param,
-                               split_k);
+            for(auto& param : conv_params)
+            {
+                if(!skip_case(param, split_k))
+                {
+                    pass = pass && ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial{},
+                                                                                      InLayout,
+                                                                                      WeiLayout,
+                                                                                      OutLayout,
+                                                                                      InDataType,
+                                                                                      WeiDataType,
+                                                                                      OutDataType>(
+                                       true,  // do_verification
+                                       1,     // init_method: integer value
+                                       false, // do_log
+                                       false, // time_kernel
+                                       param,
+                                       split_k);
+                }
+            }
         }
         EXPECT_TRUE(pass);
     }
@@ -69,12 +102,13 @@ class TestGroupedConvndBwdWeight3d : public TestGroupedConvndBwdWeight<Tuple>
 {
 };
 
-using namespace ck::tensor_layout::convolution;
-
 using KernelTypes1d = ::testing::Types<
     std::tuple<float, float, float, GNWC, GKXC, GNWK, ck::Number<1>>,
     std::tuple<ck::half_t, ck::half_t, ck::half_t, GNWC, GKXC, GNWK, ck::Number<1>>,
-    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNWC, GKXC, GNWK, ck::Number<1>>>;
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNWC, GKXC, GNWK, ck::Number<1>>,
+    std::tuple<float, float, float, NWGC, GKXC, NWGK, ck::Number<1>>,
+    std::tuple<ck::half_t, ck::half_t, ck::half_t, NWGC, GKXC, NWGK, ck::Number<1>>,
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, NWGC, GKXC, NWGK, ck::Number<1>>>;
 using KernelTypes2d = ::testing::Types<
     std::tuple<float, float, float, GNHWC, GKYXC, GNHWK, ck::Number<2>>,
     std::tuple<ck::half_t, ck::half_t, ck::half_t, GNHWC, GKYXC, GNHWK, ck::Number<2>>,

From 5fe687fa270aeb9e387278a52ea4f09d997c4b6b Mon Sep 17 00:00:00 2001
From: Jun Liu <Liu.Jun@amd.com>
Date: Wed, 13 Sep 2023 14:38:12 -0700
Subject: [PATCH 127/194] [Cmake] Set cmake default build type Release and path
 to /opt/rocm (#914)

---
 CMakeLists.txt | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index a79af9cc32..e3b60347d5 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,5 +1,18 @@
 cmake_minimum_required(VERSION 3.14)
 
+# This has to be initialized before the project() command appears
+# Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D.  MSVC_IDE does not use CMAKE_BUILD_TYPE
+if( NOT MSVC_IDE AND NOT CMAKE_BUILD_TYPE )
+    set( CMAKE_BUILD_TYPE Release CACHE STRING "Choose the type of build, options are: None Debug Release RelWithDebInfo MinSizeRel." )
+endif()
+
+# Default installation path
+if(WIN32)
+    set(CMAKE_INSTALL_PREFIX "/opt/rocm/x86_64-w64-mingw32" CACHE PATH "")
+else()
+    set(CMAKE_INSTALL_PREFIX "/opt/rocm" CACHE PATH "")
+endif()
+
 set(version 1.1.0)
 # Check support for CUDA/HIP in Cmake
 project(composable_kernel VERSION ${version})

From 0d8efaa13d2cb93e6b1cd0ae702bf15778d1d4ec Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 14 Sep 2023 09:36:26 -0700
Subject: [PATCH 128/194] change the cmake update method (#918)

---
 Dockerfile | 11 ++---------
 1 file changed, 2 insertions(+), 9 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index e479268f48..b0954fdc75 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -73,15 +73,8 @@ RUN wget -qO /usr/local/bin/ninja.gz https://github.com/ninja-build/ninja/releas
 RUN gunzip /usr/local/bin/ninja.gz
 RUN chmod a+x /usr/local/bin/ninja
 RUN git clone https://github.com/nico/ninjatracing.git
-RUN apt purge --auto-remove -y cmake
-RUN apt update
-RUN apt install -y software-properties-common lsb-release
-RUN apt clean all
-RUN wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | gpg --dearmor - | tee /etc/apt/trusted.gpg.d/kitware.gpg >/dev/null
-RUN apt-add-repository "deb https://apt.kitware.com/ubuntu/ $(lsb_release -cs) main"
-RUN apt install -y kitware-archive-keyring
-RUN rm /etc/apt/trusted.gpg.d/kitware.gpg
-RUN apt install -y cmake
+# Update the cmake to the latest version
+RUN pip install --upgrade cmake
 
 # Setup ubsan environment to printstacktrace
 RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer

From f9d0eddb90d80b2f33da7a60194fac38dc4a19ba Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Thu, 14 Sep 2023 21:04:10 -0500
Subject: [PATCH 129/194] Add  fp16/fp8 support into Grouped gemm FixedNK
 (#874)

* move all arguments into device

* add b2c_tile_map

* add examples

* add SetDeviceKernelArgs

* dedicated fixed_nk solution

* init client api

* add grouped_gemm_bias example

* add a instance

* add instances

* formatting

* fixed cmake

* Update EnableCompilerWarnings.cmake

* Update cmake-ck-dev.sh

* clean; fixed comments

* fixed comment

* add instances for fp32 output

* add instances for fp32 output

* add fp32 out client example

* fixed CI

* init commit for kbatch

* add splitk gridwise

* format

* fixed

* clean deviceop

* clean code

* finish splitk

* fixed instances

* change m_loops to tile_loops

* add setkbatch

* clean code

* add splitK+bias

* add instances

* opt mk_nk instances

* clean examples

* fixed CI

* remove zero

* finished non-zero

* clean

* clean code

* optimized global_barrier

* fixed ci

* fixed CI

* instance and client

* removed AddBias

* format

* fixed CI

* fixed CI

* move 20_grouped_gemm to 21_grouped_gemm

* clean

* formatting

* clean

* clean

* fixed computeType

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 client_example/22_grouped_gemm/CMakeLists.txt |   8 +
 .../grouped_gemm_fixed_nk_fp16.cpp            | 238 +++++++++++++
 .../grouped_gemm_fixed_nk_fp8.cpp             | 238 +++++++++++++
 .../grouped_gemm_fixed_nk_i8.cpp              | 238 +++++++++++++
 example/15_grouped_gemm/CMakeLists.txt        |   5 +
 .../grouped_gemm_xdl_fixed_nk_fp8.cpp         | 330 ++++++++++++++++++
 .../impl/device_grouped_gemm_xdl_fixed_nk.hpp |   3 +
 .../element/unary_element_wise_operation.hpp  |  12 +
 ...se_gemm_multiple_d_xdl_splitk_cshuffle.hpp |  46 +--
 .../gpu/grouped_gemm_fixed_nk.hpp             | 190 ++++++++++
 .../gpu/grouped_gemm_fixed_nk/CMakeLists.txt  |  10 +
 ...fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp |  75 ++++
 ...fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp |  78 +++++
 ..._fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp |  75 ++++
 ..._fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp |  78 +++++
 ..._fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp |  75 ++++
 ..._fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp |  78 +++++
 17 files changed, 1749 insertions(+), 28 deletions(-)
 create mode 100644 client_example/22_grouped_gemm/CMakeLists.txt
 create mode 100644 client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp
 create mode 100644 client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp
 create mode 100644 client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp
 create mode 100644 example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp

diff --git a/client_example/22_grouped_gemm/CMakeLists.txt b/client_example/22_grouped_gemm/CMakeLists.txt
new file mode 100644
index 0000000000..05b9e1e29d
--- /dev/null
+++ b/client_example/22_grouped_gemm/CMakeLists.txt
@@ -0,0 +1,8 @@
+add_executable(client_grouped_gemm_fixed_nk_fp16 grouped_gemm_fixed_nk_fp16.cpp)
+target_link_libraries(client_grouped_gemm_fixed_nk_fp16 PRIVATE composable_kernel::device_operations)
+
+add_executable(client_grouped_gemm_fixed_nk_fp8 grouped_gemm_fixed_nk_fp8.cpp)
+target_link_libraries(client_grouped_gemm_fixed_nk_fp8 PRIVATE composable_kernel::device_operations)
+
+add_executable(client_grouped_gemm_fixed_nk_i8 grouped_gemm_fixed_nk_i8.cpp)
+target_link_libraries(client_grouped_gemm_fixed_nk_i8 PRIVATE composable_kernel::device_operations)
diff --git a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp
new file mode 100644
index 0000000000..3503ae8b24
--- /dev/null
+++ b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp
@@ -0,0 +1,238 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+#include <random>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp"
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType  = F16;
+using BDataType  = F16;
+using DsDataType = ck::Tuple<>;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideEs;
+
+    int sum_of_m = 0;
+
+    // Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+    Ms = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0};
+
+    int group_count = Ms.size();
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        Ns.push_back(768);
+        Ks.push_back(4608);
+
+        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
+        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
+        StrideEs.push_back(std::is_same<Row, ELayout>::value ? Ns[i] : Ms[i]);
+
+        sum_of_m += Ms[i];
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    std::vector<SimpleDeviceMem> a_dev_bufs, b_dev_bufs, e_dev_bufs;
+
+    a_dev_bufs.reserve(group_count);
+    b_dev_bufs.reserve(group_count);
+    e_dev_bufs.reserve(group_count);
+
+    std::vector<void*> p_e;
+
+    p_e.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GroupedGemmKernelArgument<1>>
+        grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        a_dev_bufs.emplace_back(sizeof(ADataType) *
+                                f_matrix_space_size(Ms[i], Ks[i], StrideAs[i], ALayout{}));
+        b_dev_bufs.emplace_back(sizeof(BDataType) *
+                                f_matrix_space_size(Ks[i], Ns[i], StrideBs[i], BLayout{}));
+        e_dev_bufs.emplace_back(sizeof(EDataType) *
+                                f_matrix_space_size(Ms[i], Ns[i], StrideEs[i], ELayout{}));
+
+        gemm_descs.push_back({sum_of_m, Ns[i], Ks[i], 1, StrideBs[i], 1, {0}});
+
+        p_e.push_back(e_dev_bufs[i].GetDeviceBuffer());
+
+        grouped_gemm_kernel_args_.push_back({a_dev_bufs[i].GetDeviceBuffer(),
+                                             b_dev_bufs[i].GetDeviceBuffer(),
+                                             {},
+                                             e_dev_bufs[i].GetDeviceBuffer(),
+                                             Ms[i],
+                                             Ns[i],
+                                             Ks[i],
+                                             StrideAs[i],
+                                             StrideBs[i],
+                                             {},
+                                             StrideEs[i]});
+    }
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                                            BLayout,
+                                                                            DsLayout,
+                                                                            ELayout,
+                                                                            ADataType,
+                                                                            BDataType,
+                                                                            DsDataType,
+                                                                            EDataType,
+                                                                            AElementOp,
+                                                                            BElementOp,
+                                                                            CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    std::vector<const void*> p_a = {}, p_b = {};
+    std::vector<std::array<const void*, 0>> p_ds = {};
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_e, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        SimpleDeviceMem grouped_gemm_kernel_args_dev(
+            op_ptr->GetDeviceKernelArgSize(argument_ptr.get()));
+
+        SimpleDeviceMem grouped_gemm_workspace_dev(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        hipGetErrorString(hipMemcpy(grouped_gemm_kernel_args_dev.GetDeviceBuffer(),
+                                    grouped_gemm_kernel_args_.data(),
+                                    op_ptr->GetDeviceKernelArgSize(argument_ptr.get()),
+                                    hipMemcpyHostToDevice));
+
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(),
+                                    grouped_gemm_workspace_dev.GetDeviceBuffer());
+
+        op_ptr->SetDeviceKernelArgs(argument_ptr.get(),
+                                    grouped_gemm_kernel_args_dev.GetDeviceBuffer());
+
+        op_ptr->SetKBatch(argument_ptr.get(), 32);
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = 0, num_btype = 0;
+            for(std::size_t j = 0; j < gemm_descs.size(); ++j)
+            {
+                flop += std::size_t(2) * Ms[j] * Ns[j] * Ks[j];
+
+                num_btype += sizeof(ADataType) * Ms[j] * Ks[j] + sizeof(BDataType) * Ks[j] * Ns[j] +
+                             sizeof(EDataType) * Ms[j] * Ns[j];
+            }
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp
new file mode 100644
index 0000000000..b288550b74
--- /dev/null
+++ b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp
@@ -0,0 +1,238 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+#include <random>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp"
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType  = F16;
+using BDataType  = F8;
+using DsDataType = ck::Tuple<>;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideEs;
+
+    int sum_of_m = 0;
+
+    Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    int group_count = Ms.size();
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        Ns.push_back(768);
+        Ks.push_back(4608);
+
+        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
+        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
+        StrideEs.push_back(std::is_same<Row, ELayout>::value ? Ns[i] : Ms[i]);
+
+        sum_of_m += Ms[i];
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    std::vector<SimpleDeviceMem> a_dev_bufs, b_dev_bufs, e_dev_bufs;
+
+    a_dev_bufs.reserve(group_count);
+    b_dev_bufs.reserve(group_count);
+    e_dev_bufs.reserve(group_count);
+
+    std::vector<void*> p_e;
+
+    p_e.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GroupedGemmKernelArgument<1>>
+        grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        a_dev_bufs.emplace_back(sizeof(ADataType) *
+                                f_matrix_space_size(Ms[i], Ks[i], StrideAs[i], ALayout{}));
+        b_dev_bufs.emplace_back(sizeof(BDataType) *
+                                f_matrix_space_size(Ks[i], Ns[i], StrideBs[i], BLayout{}));
+        e_dev_bufs.emplace_back(sizeof(EDataType) *
+                                f_matrix_space_size(Ms[i], Ns[i], StrideEs[i], ELayout{}));
+
+        gemm_descs.push_back({sum_of_m, Ns[i], Ks[i], 1, StrideBs[i], 1, {0}});
+
+        p_e.push_back(e_dev_bufs[i].GetDeviceBuffer());
+
+        grouped_gemm_kernel_args_.push_back({a_dev_bufs[i].GetDeviceBuffer(),
+                                             b_dev_bufs[i].GetDeviceBuffer(),
+                                             {},
+                                             e_dev_bufs[i].GetDeviceBuffer(),
+                                             Ms[i],
+                                             Ns[i],
+                                             Ks[i],
+                                             StrideAs[i],
+                                             StrideBs[i],
+                                             {},
+                                             StrideEs[i]});
+    }
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                                            BLayout,
+                                                                            DsLayout,
+                                                                            ELayout,
+                                                                            ADataType,
+                                                                            BDataType,
+                                                                            DsDataType,
+                                                                            EDataType,
+                                                                            AElementOp,
+                                                                            BElementOp,
+                                                                            CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    std::vector<const void*> p_a = {}, p_b = {};
+    std::vector<std::array<const void*, 0>> p_ds = {};
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_e, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        SimpleDeviceMem grouped_gemm_kernel_args_dev(
+            op_ptr->GetDeviceKernelArgSize(argument_ptr.get()));
+
+        SimpleDeviceMem grouped_gemm_workspace_dev(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        hipGetErrorString(hipMemcpy(grouped_gemm_kernel_args_dev.GetDeviceBuffer(),
+                                    grouped_gemm_kernel_args_.data(),
+                                    op_ptr->GetDeviceKernelArgSize(argument_ptr.get()),
+                                    hipMemcpyHostToDevice));
+
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(),
+                                    grouped_gemm_workspace_dev.GetDeviceBuffer());
+
+        op_ptr->SetDeviceKernelArgs(argument_ptr.get(),
+                                    grouped_gemm_kernel_args_dev.GetDeviceBuffer());
+
+        op_ptr->SetKBatch(argument_ptr.get(), 16);
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = 0, num_btype = 0;
+            for(std::size_t j = 0; j < gemm_descs.size(); ++j)
+            {
+                flop += std::size_t(2) * Ms[j] * Ns[j] * Ks[j];
+
+                num_btype += sizeof(ADataType) * Ms[j] * Ks[j] + sizeof(BDataType) * Ks[j] * Ns[j] +
+                             sizeof(EDataType) * Ms[j] * Ns[j];
+            }
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp
new file mode 100644
index 0000000000..c60daa3b36
--- /dev/null
+++ b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp
@@ -0,0 +1,238 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+#include <random>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp"
+
+using I8  = int8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType  = F16;
+using BDataType  = I8;
+using DsDataType = ck::Tuple<>;
+using EDataType  = F16;
+
+using ALayout  = Row;
+using BLayout  = Row;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::vector<int> Ms, Ns, Ks, StrideAs, StrideBs, StrideEs;
+
+    int sum_of_m = 0;
+
+    Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    int group_count = Ms.size();
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        Ns.push_back(768);
+        Ks.push_back(4608);
+
+        StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
+        StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
+        StrideEs.push_back(std::is_same<Row, ELayout>::value ? Ns[i] : Ms[i]);
+
+        sum_of_m += Ms[i];
+    }
+
+    auto f_matrix_space_size =
+        [](std::size_t nRow, std::size_t nCol, std::size_t stride, auto layout) {
+            using Layout = decltype(layout);
+
+            if constexpr(std::is_same<Layout, ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return (nRow - 1) * stride + nCol;
+            }
+            else
+            {
+                return (nCol - 1) * stride + nRow;
+            }
+        };
+
+    std::vector<SimpleDeviceMem> a_dev_bufs, b_dev_bufs, e_dev_bufs;
+
+    a_dev_bufs.reserve(group_count);
+    b_dev_bufs.reserve(group_count);
+    e_dev_bufs.reserve(group_count);
+
+    std::vector<void*> p_e;
+
+    p_e.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+
+    gemm_descs.reserve(group_count);
+
+    std::vector<ck::tensor_operation::device::GroupedGemmKernelArgument<1>>
+        grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; ++i)
+    {
+        a_dev_bufs.emplace_back(sizeof(ADataType) *
+                                f_matrix_space_size(Ms[i], Ks[i], StrideAs[i], ALayout{}));
+        b_dev_bufs.emplace_back(sizeof(BDataType) *
+                                f_matrix_space_size(Ks[i], Ns[i], StrideBs[i], BLayout{}));
+        e_dev_bufs.emplace_back(sizeof(EDataType) *
+                                f_matrix_space_size(Ms[i], Ns[i], StrideEs[i], ELayout{}));
+
+        gemm_descs.push_back({sum_of_m, Ns[i], Ks[i], 1, StrideBs[i], 1, {0}});
+
+        p_e.push_back(e_dev_bufs[i].GetDeviceBuffer());
+
+        grouped_gemm_kernel_args_.push_back({a_dev_bufs[i].GetDeviceBuffer(),
+                                             b_dev_bufs[i].GetDeviceBuffer(),
+                                             {},
+                                             e_dev_bufs[i].GetDeviceBuffer(),
+                                             Ms[i],
+                                             Ns[i],
+                                             Ks[i],
+                                             StrideAs[i],
+                                             StrideBs[i],
+                                             {},
+                                             StrideEs[i]});
+    }
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                                            BLayout,
+                                                                            DsLayout,
+                                                                            ELayout,
+                                                                            ADataType,
+                                                                            BDataType,
+                                                                            DsDataType,
+                                                                            EDataType,
+                                                                            AElementOp,
+                                                                            BElementOp,
+                                                                            CDEElementOp>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    const auto a_element_op   = AElementOp{};
+    const auto b_element_op   = BElementOp{};
+    const auto cde_element_op = CDEElementOp{};
+
+    std::string best_op_name;
+    bool found            = false;
+    int best_op_id        = -1;
+    float best_ave_time   = 0;
+    float best_tflops     = 0;
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    std::vector<const void*> p_a = {}, p_b = {};
+    std::vector<std::array<const void*, 0>> p_ds = {};
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr = op_ptrs[i];
+
+        auto argument_ptr = op_ptr->MakeArgumentPointer(
+            p_a, p_b, p_ds, p_e, gemm_descs, a_element_op, b_element_op, cde_element_op);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        SimpleDeviceMem grouped_gemm_kernel_args_dev(
+            op_ptr->GetDeviceKernelArgSize(argument_ptr.get()));
+
+        SimpleDeviceMem grouped_gemm_workspace_dev(op_ptr->GetWorkSpaceSize(argument_ptr.get()));
+
+        std::string op_name = op_ptr->GetTypeString();
+
+        hipGetErrorString(hipMemcpy(grouped_gemm_kernel_args_dev.GetDeviceBuffer(),
+                                    grouped_gemm_kernel_args_.data(),
+                                    op_ptr->GetDeviceKernelArgSize(argument_ptr.get()),
+                                    hipMemcpyHostToDevice));
+
+        op_ptr->SetWorkSpacePointer(argument_ptr.get(),
+                                    grouped_gemm_workspace_dev.GetDeviceBuffer());
+
+        op_ptr->SetDeviceKernelArgs(argument_ptr.get(),
+                                    grouped_gemm_kernel_args_dev.GetDeviceBuffer());
+
+        op_ptr->SetKBatch(argument_ptr.get(), 32);
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float ave_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop = 0, num_btype = 0;
+            for(std::size_t j = 0; j < gemm_descs.size(); ++j)
+            {
+                flop += std::size_t(2) * Ms[j] * Ns[j] * Ks[j];
+
+                num_btype += sizeof(ADataType) * Ms[j] * Ks[j] + sizeof(BDataType) * Ks[j] * Ns[j] +
+                             sizeof(EDataType) * Ms[j] * Ns[j];
+            }
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+            float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+            std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                found           = true;
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_tflops     = tflops;
+                best_ave_time   = ave_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << best_ave_time << " ms, " << best_tflops << " TFlops, "
+              << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    return 0;
+}
diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index 2f880af3cf..45e02126b6 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -25,6 +25,11 @@ if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
   add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
 endif()
+if(DTYPES MATCHES "f8" OR NOT DEFINED DTYPES)
+  add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
+endif()
+
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp
new file mode 100644
index 0000000000..1c50dc051b
--- /dev/null
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp
@@ -0,0 +1,330 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F8;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F16;
+
+using ALayout  = Row;
+using BLayout  = Col;
+using DsLayout = ck::Tuple<>;
+using ELayout  = Row;
+
+using AElementOp   = PassThrough;
+using BElementOp   = PassThrough;
+using CDEElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGroupedGemm_Xdl_Fixed_NK
+    // clang-format off
+//######| ALayout| BLayout| DsLayout| ELayout|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+//######|        |        |         |        |      Type|      Type|        Type|         DataType|       Type|      Type| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+//######|        |        |         |        |          |          |            |                 |           |          |   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+//######|        |        |         |        |          |          |            |                 |           |          |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        < ALayout, BLayout, DsLayout, ELayout, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType,  AElementOp,  BElementOp, CDEElementOp,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>;
+// clang-format on
+
+struct ProblemSize final
+{
+    std::vector<ck::index_t> Ms;
+    std::vector<ck::index_t> Ns;
+    std::vector<ck::index_t> Ks;
+
+    std::vector<ck::index_t> stride_As;
+    std::vector<ck::index_t> stride_Bs;
+    std::vector<ck::index_t> stride_Cs;
+
+    ck::index_t group_count;
+};
+
+struct ExecutionConfig final
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    int k_batch          = 1;
+    bool time_kernel     = false;
+};
+
+bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& config)
+{
+    auto group_count = problem_size.group_count;
+
+    // GEMM shape
+    std::vector<ck::tensor_operation::device::GemmDesc> gemm_descs;
+    std::vector<void*> p_Cs;
+
+    gemm_descs.reserve(group_count);
+
+    int sum_of_m = 0;
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    std::vector<Tensor<ADataType>> a_tensors;
+    std::vector<Tensor<BDataType>> b_tensors;
+    std::vector<Tensor<EDataType>> c_host_tensors;
+    std::vector<Tensor<EDataType>> c_device_tensors;
+
+    a_tensors.reserve(group_count);
+    b_tensors.reserve(group_count);
+    c_host_tensors.reserve(group_count);
+    c_device_tensors.reserve(group_count);
+
+    using DeviceMemPtr = std::unique_ptr<DeviceMem>;
+
+    std::vector<DeviceMemPtr> a_tensors_device, b_tensors_device, c_tensors_device;
+
+    a_tensors_device.reserve(group_count);
+    b_tensors_device.reserve(group_count);
+    c_tensors_device.reserve(group_count);
+
+    std::size_t flop = 0, num_btype = 0;
+
+    for(int i = 0; i < group_count; i++)
+    {
+        sum_of_m += problem_size.Ms[i];
+        a_tensors.push_back(Tensor<ADataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ks[i], problem_size.stride_As[i], ALayout{})));
+        b_tensors.push_back(Tensor<BDataType>(f_host_tensor_descriptor(
+            problem_size.Ks[i], problem_size.Ns[i], problem_size.stride_Bs[i], BLayout{})));
+        c_host_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ns[i], problem_size.stride_Cs[i], ELayout{})));
+        c_device_tensors.push_back(Tensor<EDataType>(f_host_tensor_descriptor(
+            problem_size.Ms[i], problem_size.Ns[i], problem_size.stride_Cs[i], ELayout{})));
+        std::cout << "gemm[" << i << "] a_m_k: " << a_tensors[i].mDesc
+                  << " b_k_n: " << b_tensors[i].mDesc << " c_m_n: " << c_device_tensors[i].mDesc
+                  << std::endl;
+
+        flop += std::size_t(2) * problem_size.Ms[i] * problem_size.Ks[i] * problem_size.Ns[i];
+        num_btype += sizeof(ADataType) * a_tensors[i].mDesc.GetElementSize() +
+                     sizeof(BDataType) * b_tensors[i].mDesc.GetElementSize() +
+                     sizeof(EDataType) * c_device_tensors[i].mDesc.GetElementSize();
+
+        switch(config.init_method)
+        {
+        case 0: break;
+        case 1:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+            break;
+        case 2:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+            break;
+        default:
+            a_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<0>{});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<1>{});
+        }
+    }
+
+    using GroupedGemmKernelArgument = ck::tensor_operation::device::GroupedGemmKernelArgument<>;
+
+    std::vector<GroupedGemmKernelArgument> grouped_gemm_kernel_args_;
+    grouped_gemm_kernel_args_.reserve(group_count);
+
+    for(int i = 0; i < group_count; i++)
+    {
+        a_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(ADataType) * sum_of_m * problem_size.Ks[i]));
+
+        b_tensors_device.emplace_back(std::make_unique<DeviceMem>(
+            sizeof(BDataType) * problem_size.Ns[i] * problem_size.Ks[i]));
+
+        c_tensors_device.emplace_back(
+            std::make_unique<DeviceMem>(sizeof(EDataType) * sum_of_m * problem_size.Ns[i]));
+
+        a_tensors_device[i]->ToDevice(a_tensors[i].mData.data(),
+                                      a_tensors[i].mDesc.GetElementSpaceSize() * sizeof(ADataType));
+        b_tensors_device[i]->ToDevice(b_tensors[i].mData.data(),
+                                      b_tensors[i].mDesc.GetElementSpaceSize() * sizeof(BDataType));
+        c_tensors_device[i]->SetZero();
+
+        p_Cs.push_back(c_tensors_device[i]->GetDeviceBuffer());
+
+        gemm_descs.push_back({sum_of_m,
+                              problem_size.Ns[i],
+                              problem_size.Ks[i],
+                              1,
+                              problem_size.stride_Bs[i],
+                              1,
+                              {}});
+
+        grouped_gemm_kernel_args_.push_back({a_tensors_device[i]->GetDeviceBuffer(),
+                                             b_tensors_device[i]->GetDeviceBuffer(),
+                                             {},
+                                             c_tensors_device[i]->GetDeviceBuffer(),
+                                             problem_size.Ms[i],
+                                             problem_size.Ns[i],
+                                             problem_size.Ks[i],
+                                             problem_size.stride_As[i],
+                                             problem_size.stride_Bs[i],
+                                             {},
+                                             problem_size.stride_Cs[i]});
+    }
+
+    auto a_element_op = AElementOp{};
+    auto b_element_op = BElementOp{};
+    auto c_element_op = CDEElementOp{};
+
+    auto gemm    = DeviceGemmInstance{};
+    auto invoker = gemm.MakeInvoker();
+
+    std::vector<const void*> p_As                = {};
+    std::vector<const void*> p_Bs                = {};
+    std::vector<std::array<const void*, 0>> p_Ds = {};
+
+    // do GEMM
+    auto argument = gemm.MakeArgument(
+        p_As, p_Bs, p_Ds, p_Cs, gemm_descs, a_element_op, b_element_op, c_element_op);
+
+    DeviceMem gemm_arg_dev_mem(gemm.GetDeviceKernelArgSize(&argument));
+    DeviceMem gemm_workspace_dev(gemm.GetWorkSpaceSize(&argument));
+
+    gemm.SetWorkSpacePointer(&argument, gemm_workspace_dev.GetDeviceBuffer());
+
+    hip_check_error(hipMemcpy(gemm_arg_dev_mem.GetDeviceBuffer(),
+                              grouped_gemm_kernel_args_.data(),
+                              gemm.GetDeviceKernelArgSize(&argument),
+                              hipMemcpyHostToDevice));
+
+    if(!gemm.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    gemm.SetDeviceKernelArgs(argument, gemm_arg_dev_mem.GetDeviceBuffer());
+    gemm.SetKBatch(argument, config.k_batch);
+
+    invoker.Run(argument, StreamConfig{nullptr, false});
+
+    if(config.time_kernel)
+    {
+        float ave_time   = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+        float tflops     = static_cast<float>(flop) / 1.E9 / ave_time;
+        float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                  << " GB/s, " << gemm.GetTypeString() << std::endl;
+    }
+
+    bool pass = true;
+    if(config.do_verification)
+    {
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                EDataType,
+                                                                                AccDataType,
+                                                                                AElementOp,
+                                                                                BElementOp,
+                                                                                CDEElementOp>;
+
+        for(std::size_t i = 0; i < gemm_descs.size(); i++)
+        {
+            c_tensors_device[i]->FromDevice(c_device_tensors[i].mData.data(),
+                                            c_device_tensors[i].mDesc.GetElementSize() *
+                                                sizeof(EDataType));
+            auto ref_gemm    = ReferenceGemmInstance{};
+            auto ref_invoker = ref_gemm.MakeInvoker();
+
+            auto ref_argument = ref_gemm.MakeArgument(a_tensors[i],
+                                                      b_tensors[i],
+                                                      c_host_tensors[i],
+                                                      a_element_op,
+                                                      b_element_op,
+                                                      c_element_op);
+
+            ref_invoker.Run(ref_argument);
+
+            pass &= ck::utils::check_err(c_device_tensors[i], c_host_tensors[i]);
+        }
+    }
+
+    return pass;
+}
+
+int main(int argc, char* argv[])
+{
+    ProblemSize problem_size;
+    ExecutionConfig config;
+
+    problem_size.group_count = 16;
+
+    problem_size.Ms = {
+        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
+
+    for(int i = 0; i < problem_size.group_count; i++)
+    {
+        problem_size.Ns.push_back(768);
+        problem_size.Ks.push_back(4608);
+
+        problem_size.stride_As.push_back(problem_size.Ks[i]);
+        problem_size.stride_Bs.push_back(problem_size.Ks[i]);
+        problem_size.stride_Cs.push_back(problem_size.Ns[i]);
+    }
+
+    if(argc == 5)
+    {
+        config.do_verification = std::stoi(argv[1]);
+        config.init_method     = std::stoi(argv[2]);
+        config.time_kernel     = std::stoi(argv[3]);
+        config.k_batch         = std::stoi(argv[4]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=n0, 1=yes)\n");
+        printf("arg4: k_batch (> 0)\n");
+        exit(0);
+    }
+
+    return !run_grouped_gemm(problem_size, config);
+}
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
index 8cea09ee54..56132f7a0f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
@@ -193,6 +193,7 @@ template <typename ALayout,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          typename ComputeType    = ADataType,
           LoopScheduler LoopSched = make_default_loop_scheduler()>
 struct DeviceGroupedGemm_Xdl_Fixed_NK : public DeviceGroupedGemmFixedNK<ALayout,
                                                                         BLayout,
@@ -217,6 +218,8 @@ struct DeviceGroupedGemm_Xdl_Fixed_NK : public DeviceGroupedGemmFixedNK<ALayout,
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_splitk_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
+        BDataType,
+        ComputeType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 69a6540a0c..59cd54a122 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -75,12 +75,24 @@ struct PassThrough
         y = type_convert<bhalf_t>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<float, half_t>(float& y, const half_t& x) const
+    {
+        y = type_convert<float>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<int8_t, int8_t>(int8_t& y, const int8_t& x) const
     {
         y = x;
     }
 
+    template <>
+    __host__ __device__ void operator()<half_t, int8_t>(half_t& y, const int8_t& x) const
+    {
+        y = type_convert<half_t>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<int8_t, int32_t>(int8_t& y, const int32_t& x) const
     {
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
index e22391293e..a3343bd3a0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
@@ -29,7 +29,9 @@ namespace ck {
 //   E = cde_op(C, D0, D1, ...)
 // Assume:
 //   D0, D1, ... and E have the same layout
-template <typename ABDataType, // FIXME: don't assume A/B have same datatype
+template <typename ADataType,
+          typename BDataType,
+          typename ComputeType,
           typename AccDataType,
           typename CShuffleDataType,
           typename DsDataType,
@@ -96,17 +98,6 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
     using GridwiseGemmPipe = remove_cvref_t<
         decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
-    // denorm test fix, required to work around fp16 mfma issue
-    // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
-    // when mfma if fixed, remove this section and update
-    // ABDataTypeAdjusted -> ABDataType throughout this file
-#if CK_WORKAROUND_DENORM_FIX
-    using ABDataTypeAdjusted =
-        conditional_t<is_same_v<ABDataType, ck::half_t>, ck::bhalf_t, ABDataType>;
-#else
-    using ABDataTypeAdjusted = ABDataType;
-#endif
-
     __host__ __device__ static constexpr auto GetABlockDescriptor_KBatch_AK0PerBlock_MPerBlock_AK1()
     {
         // A matrix in LDS memory, dst of blockwise copy
@@ -196,7 +187,7 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
             c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
 
         return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
-                             sizeof(ABDataType),
+                             sizeof(ComputeType),
                          c_block_size * sizeof(CShuffleDataType));
     }
 
@@ -401,8 +392,8 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
         // check tensor size: cannot be larger than 2GB each
         constexpr long_index_t TwoGB = (long_index_t{1} << 31);
 
-        if(!(a_grid_desc_kbatch_ak0_m_ak1.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
-             b_grid_desc_kbatch_bk0_n_bk1.GetElementSpaceSize() * sizeof(ABDataType) <= TwoGB &&
+        if(!(a_grid_desc_kbatch_ak0_m_ak1.GetElementSpaceSize() * sizeof(ADataType) <= TwoGB &&
+             b_grid_desc_kbatch_bk0_n_bk1.GetElementSpaceSize() * sizeof(BDataType) <= TwoGB &&
              e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
         {
             return false;
@@ -470,8 +461,8 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
               typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
               typename CDEElementwiseOperation_,
               typename Block2ETileMap>
-    __device__ static void Run(const ABDataType* __restrict__ p_a_grid,
-                               const ABDataType* __restrict__ p_b_grid,
+    __device__ static void Run(const ADataType* __restrict__ p_a_grid,
+                               const BDataType* __restrict__ p_b_grid,
                                DsGridPointer p_ds_grid,
                                EDataType* __restrict__ p_e_grid,
                                void* __restrict__ p_shared,
@@ -538,8 +529,8 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
                                                 Sequence<1, AK0PerBlock, MPerBlock, AK1>,
                                                 ABlockTransferThreadClusterLengths_KBatch_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
-                                                ABDataType,
-                                                ABDataTypeAdjusted,
+                                                ADataType,
+                                                ComputeType,
                                                 decltype(a_grid_desc_kbatch_ak0_m_ak1),
                                                 decltype(a_block_desc_kbatch_ak0_m_ak1),
                                                 ABlockTransferSrcAccessOrder,
@@ -569,8 +560,8 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
                                                 Sequence<1, BK0PerBlock, NPerBlock, BK1>,
                                                 BBlockTransferThreadClusterLengths_KBatch_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
-                                                ABDataType,
-                                                ABDataTypeAdjusted,
+                                                BDataType,
+                                                ComputeType,
                                                 decltype(b_grid_desc_kbatch_bk0_n_bk1),
                                                 decltype(b_block_desc_kbatch_bk0_n_bk1),
                                                 BBlockTransferSrcAccessOrder,
@@ -606,11 +597,11 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
         // sanity check
         constexpr index_t KPack =
             math::max(math::lcm(AK1, BK1),
-                      MfmaSelector<ABDataTypeAdjusted, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+                      MfmaSelector<ComputeType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ABDataTypeAdjusted,
+            ComputeType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -683,11 +674,10 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ABDataTypeAdjusted*>(p_shared),
-            a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+            static_cast<ComputeType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ABDataTypeAdjusted*>(p_shared) + a_block_space_size_aligned,
+            static_cast<ComputeType*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(0, KPerBlock / AK1, 0, 0);
@@ -999,8 +989,8 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
                                const index_t KBatch,
                                const Block2ETileMap& block_2_etile_map)
     {
-        const auto p_a_grid = reinterpret_cast<const ABDataType*>(p_a_grid_);
-        const auto p_b_grid = reinterpret_cast<const ABDataType*>(p_b_grid_);
+        const auto p_a_grid = reinterpret_cast<const ADataType*>(p_a_grid_);
+        const auto p_b_grid = reinterpret_cast<const BDataType*>(p_b_grid_);
         const auto p_e_grid = reinterpret_cast<EDataType*>(p_e_grid_);
 
         using DsGridDesc_M_N =
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp
new file mode 100644
index 0000000000..e8c368cb38
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_gemm_fixed_nk.hpp
@@ -0,0 +1,190 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_gemm_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// fp16_output
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+// fp8_inputB
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         F8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+// i8_inputB
+void add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         I8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         Empty_Tuple,
+                                                         Row,
+                                                         F16,
+                                                         I8,
+                                                         Empty_Tuple,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances);
+
+template <typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          typename ADataType,
+          typename BDataType,
+          typename EDataType>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceGroupedGemmFixedNK<ALayout,
+                                                           BLayout,
+                                                           Empty_Tuple,
+                                                           ELayout,
+                                                           ADataType,
+                                                           BDataType,
+                                                           Empty_Tuple,
+                                                           EDataType,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>
+{
+    using DeviceOp = DeviceGroupedGemmFixedNK<ALayout,
+                                              BLayout,
+                                              Empty_Tuple,
+                                              ELayout,
+                                              ADataType,
+                                              BDataType,
+                                              Empty_Tuple,
+                                              EDataType,
+                                              PassThrough,
+                                              PassThrough,
+                                              PassThrough>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        // fp16_output
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        // fp8_input
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        // i8_input
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, int8_t> &&
+                     is_same_v<EDataType, half_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instances(op_ptrs);
+            }
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                         is_same_v<ELayout, Row>)
+            {
+                add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instances(op_ptrs);
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
new file mode 100644
index 0000000000..45f8130c57
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
@@ -0,0 +1,10 @@
+add_instance_library(device_grouped_gemm_fixed_nk_instance
+   device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
+
+   device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
+
+   device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
+   device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..584d2be17e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using DsDataType = ck::Tuple<>;
+
+using DsLayout = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..9f0e47fe53
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using DsDataType = ck::Tuple<>;
+
+using DsLayout = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    64,   8,   8,   32,   32,    4,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    64,   8,   8,   32,   32,    1,    4,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,   F16,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F16,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..885ca8c632
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using DsDataType = ck::Tuple<>;
+
+using DsLayout = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F8,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..9f1ee423ee
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using DsDataType = ck::Tuple<>;
+
+using DsLayout = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    64,   8,   8,   32,   32,    4,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    64,   8,   8,   32,   32,    1,    4,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    F8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         F8,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..fb1ba51fec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
@@ -0,0 +1,75 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I8  = int8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using DsDataType = ck::Tuple<>;
+
+using DsLayout = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2, S< 1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Row,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2, S< 1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Row,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         I8,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..23b88c2a4a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using I8  = int8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using DsDataType = ck::Tuple<>;
+
+using DsLayout = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
+    // clang-format off
+        //############################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //############################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //############################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //############################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    64,   8,   8,   32,   32,    4,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    64,   8,   8,   32,   32,    2,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    64,   8,   8,   32,   32,    1,    4,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    64,   8,   8,   32,   32,    2,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl_Fixed_NK<    Row,    Col,    DsLayout,    Row,   F16,    I8,     F32,      F32,  DsDataType,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    // clang-format on
+    >;
+
+void add_device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<DeviceGroupedGemmFixedNK<Row,
+                                                         Col,
+                                                         DsLayout,
+                                                         Row,
+                                                         F16,
+                                                         I8,
+                                                         DsDataType,
+                                                         F16,
+                                                         PassThrough,
+                                                         PassThrough,
+                                                         PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances, device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_irregular_tile_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From bc2d0583d38228bd4bfc51a3d88d31f2e4388d5c Mon Sep 17 00:00:00 2001
From: Bartlomiej Kocot <barkocot@amd.com>
Date: Wed, 13 Sep 2023 15:34:25 +0000
Subject: [PATCH 130/194] Stylistic improvements for grouped convolution code

Remove unnecessary ignoring

Update test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
---
 .../gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp     | 4 ----
 .../grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp  | 1 -
 .../test_grouped_convnd_bwd_weight.cpp                        | 4 ++--
 3 files changed, 2 insertions(+), 7 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
index 014140c460..ef57e6e4d2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
@@ -110,10 +110,6 @@ __global__ void
     ignore = c_grid_desc_m0_m10_m11_n0_n10_n11;
     ignore = block_2_ctile_map;
     ignore = compute_ptr_offset_of_batch;
-
-    compute_ptr_offset_of_batch.GetAPtrOffset(0);
-    compute_ptr_offset_of_batch.GetBPtrOffset(0);
-    compute_ptr_offset_of_batch.GetCPtrOffset(0);
 #endif
 }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
index 1ebd0c3d63..abe4bb1d32 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_dl_instance.hpp
@@ -54,7 +54,6 @@ using device_grouped_conv2d_fwd_dl_f16_instances = std::tuple<
            // ########################################|            |       |        |            |        |        |         |          |          |          |   Operation|    Operation|     Operation| Specialization|                  |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder| Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
            // ########################################|            |       |        |            |        |        |         |          |          |          |            |             |              |               |                  |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                    |                   |                     |               |               |                    |                   |                    |                    |                |                   |
         // generic instances
-        // TODO: Change to ScalarPerVector = 1 when inner_product<half_t, half_t, float> will be supported
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,     8,    16,     4,     2,  1,          1,          2,      1,       S<4, 2>,       S<1, 1>,      S<2, 1, 2, 1>,        S<1, 1, 8, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 1, 1, 1>,        S<2, 1, 4, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
         DeviceGroupedConvFwdDlMultipleD_NHWC_KYXC_NHWK<           2,    F16,     F16,  DsDatatype,     F16,     F32, InLayout, WeiLayout,  DsLayout, OutLayout, PassThrough,  PassThrough,  CDEElementOp,       ConvSpec,    GemmMNKPadding,   256,   128,   128,    16,  1,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<8, 1, 1, 1>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<1, 1, 1, 1>,      S<1, 2, 0, 3>,       S<1, 1, 1, 1>, S<0, 1, 2, 3, 4, 5>,               5,                 1>,
 
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
index bf95727fbc..2132020320 100644
--- a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -33,8 +33,8 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
 
     bool skip_case(const ck::utils::conv::ConvParam& params, const ck::index_t split_k)
     {
-        // K or C are odd is supported only by DL kernel (only applies to fp16)
-        // DL kernel is only supported for split_k=1
+        // Odd K or C values are supported only by DL kernel (only applies to fp16)
+        // DL kernel currently supports only `split_k=1`
         if constexpr(std::is_same_v<InDataType, ck::half_t>)
         {
             if(split_k != 1 && (params.K_ % 2 != 0 || params.C_ % 2 != 0))

From f17af2e9ed7e9a9625e415927c8cb6b3efa1b172 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Sun, 17 Sep 2023 20:56:27 -0500
Subject: [PATCH 131/194] Add native conversions fp8<->fp32 (#908)

* Add native conversions

* Add bf8 conversions
---
 include/ck/utility/f8_utils.hpp     |   5 +-
 include/ck/utility/type_convert.hpp | 113 ++++++++++++++++++++++++++--
 2 files changed, 110 insertions(+), 8 deletions(-)

diff --git a/include/ck/utility/f8_utils.hpp b/include/ck/utility/f8_utils.hpp
index 5fbebb708d..217b339b66 100644
--- a/include/ck/utility/f8_utils.hpp
+++ b/include/ck/utility/f8_utils.hpp
@@ -5,6 +5,8 @@
 
 #include "ck/utility/data_type.hpp"
 
+// these conversions are disabled if native conversions available
+#if !defined(__gfx940__) && !defined(__gfx941__) && !defined(__gfx942__)
 #if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
 namespace ck {
 
@@ -242,4 +244,5 @@ __host__ __device__ Y cast_from_f8(X x)
 }
 
 } // namespace ck::utils
-#endif
+#endif // #if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+#endif // #if !defined(__gfx940__) && !defined(__gfx941__) && !defined(__gfx942__)
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 5c5447f94e..70619ee0a5 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -85,6 +85,19 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_
 template <>
 inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival       = __builtin_amdgcn_cvt_pk_fp8_f32(val.fval, val.fval, ival, false); // false -> WORD0
+    val.i32val = ival;
+    return val.i8val[0];
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -92,20 +105,33 @@ inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
     return utils::
         cast_to_f8<float, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(x,
                                                                                                rng);
+#endif
 }
 
 // convert fp8 to fp32
 template <>
 inline __host__ __device__ float type_convert<float, f8_t>(f8_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    float fval;
+    uint32_t i32val = static_cast<uint32_t>(x);
+    fval            = __builtin_amdgcn_cvt_f32_fp8(i32val, 0);
+    // asm volatile("v_cvt_f32_fp8 %0, %1 src0_sel:BYTE_0" : "=v"(fval) : "v"(i32val));
+    return fval;
+#else
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<f8_t, float, negative_zero_nan>(x);
+#endif
 }
 
 // convert fp16 to fp8
 template <>
 inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return type_convert<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -113,14 +139,20 @@ inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 
 // convert fp8 to fp16
 template <>
 inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // use native conversion to float and convert to fp16
+    return type_convert<half_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<f8_t, half_t, negative_zero_nan>(x);
+#endif
 }
 #endif
 
@@ -129,6 +161,19 @@ inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
 template <>
 inline __host__ __device__ bf8_t type_convert<bf8_t, float>(float x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival       = __builtin_amdgcn_cvt_pk_bf8_f32(val.fval, val.fval, ival, false); // false -> WORD0
+    val.i32val = ival;
+    return val.i8val[0];
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -136,20 +181,33 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, float>(float x)
     return utils::
         cast_to_f8<float, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 
 // convert bf8 to fp32
 template <>
 inline __host__ __device__ float type_convert<float, bf8_t>(bf8_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    float fval;
+    uint32_t i32val = static_cast<uint32_t>(x);
+    fval            = __builtin_amdgcn_cvt_f32_bf8(i32val, 0);
+    // asm volatile("v_cvt_f32_bf8 %0, %1 src0_sel:BYTE_0" : "=v"(fval) : "v"(i32val));
+    return fval;
+#else
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<bf8_t, float, negative_zero_nan>(x);
+#endif
 }
 
 // convert fp16 to bf8
 template <>
 inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return type_convert<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -157,14 +215,20 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 
 // convert bf8 to fp16
 template <>
 inline __host__ __device__ half_t type_convert<half_t, bf8_t>(bf8_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // use native conversion to float and convert to fp16
+    return type_convert<half_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<bf8_t, half_t, negative_zero_nan>(x);
+#endif
 }
 #endif
 
@@ -234,30 +298,47 @@ __host__ __device__ constexpr Y f8_convert_sr(X x);
 template <>
 inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
 {
+    constexpr int seed = 42;
+    uint32_t rng       = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival          = __builtin_amdgcn_cvt_sr_fp8_f32(val.fval, rng, ival, 0); // 0 pos
+    val.i32val    = ival;
+    return val.i8val[0]; // little endian
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
-    constexpr int seed               = 42;
-    // as thread id is not available on host, use 0 for prn generation
-    uint32_t rng = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
     return utils::
         cast_to_f8<float, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(x,
                                                                                                rng);
+#endif
 }
 
 // convert fp16 to fp8 with stochastic rounding
 template <>
 inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return f8_convert_sr<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
     constexpr int seed               = 42;
-    // as thread id is not available on host, use 0 for prn generation
     uint32_t rng = prand_generator<half_t, seed>(reinterpret_cast<uintptr_t>(&x), x);
     return utils::
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 #endif
 
@@ -266,21 +347,38 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
 template <>
 inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, float>(float x)
 {
+    constexpr int seed = 42;
+    uint32_t rng       = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    union
+    {
+        float fval;
+        uint32_t i32val;
+        uint8_t i8val[4]; // not endian independent
+    } val;
+    val.fval      = x;
+    uint32_t ival = 0;
+    ival          = __builtin_amdgcn_cvt_sr_bf8_f32(val.fval, rng, ival, 0); // 0 pos
+    val.i32val    = ival;
+    return val.i8val[0]; // little endian
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
-    constexpr int seed               = 42;
-    // as thread id is not available on host, use 0 for prn generation
-    uint32_t rng = prand_generator<float, seed>(reinterpret_cast<uintptr_t>(&x), x);
     return utils::
         cast_to_f8<float, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 
 // convert fp16 to bf8 with stochastic rounding
 template <>
 inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
 {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    // convert to float and use native converion
+    return f8_convert_sr<f8_t>(type_convert<float>(x));
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
@@ -290,6 +388,7 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#endif
 }
 #endif
 

From 63cd4592487f9e1ba32ed7c62990faee375c504a Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Mon, 18 Sep 2023 14:08:23 +0200
Subject: [PATCH 132/194] Fix DL GEMM instances with too large vector size
 (#901)

* Fix vector lengths of DL GEMM instances with padding
* Add checks for correctness of vector lenghts in DL GEMM
---
 .../gpu/device/impl/device_gemm_dl.hpp        | 51 +++++++++++++++++++
 ...16_f16_f16_km_kn_mn_irregular_instance.cpp | 28 +++++-----
 ...16_f16_f16_km_nk_mn_irregular_instance.cpp | 28 +++++-----
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp | 28 +++++-----
 ...l_i8_i8_i8_km_kn_mn_irregular_instance.cpp | 44 ++++++++--------
 ...l_i8_i8_i8_km_nk_mn_irregular_instance.cpp | 44 ++++++++--------
 ...l_i8_i8_i8_mk_kn_mn_irregular_instance.cpp | 44 ++++++++--------
 7 files changed, 159 insertions(+), 108 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
index d9f8324bd2..514aa4452e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp
@@ -273,6 +273,9 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
               block_2_ctile_map_{},
               M01_{M01},
               N01_{N01},
+              M_raw_{M},
+              N_raw_{N},
+              K_raw_{K},
               a_element_op_{a_element_op},
               b_element_op_{b_element_op},
               c_element_op_{c_element_op}
@@ -314,6 +317,10 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
         index_t M01_;
         index_t N01_;
 
+        index_t M_raw_;
+        index_t N_raw_;
+        index_t K_raw_;
+
         // TODO: unused since gridwise_gemm_dl_v1r3 does NOT support prologue for the time being.
         AElementwiseOperation a_element_op_;
         BElementwiseOperation b_element_op_;
@@ -485,6 +492,50 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        // Make sure that the M, N, K dimensions before padding are divisible by respective vector
+        // lengths.
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, ALayout>::value)
+        {
+            constexpr auto A_K_vec_length =
+                ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1::At(I0) *
+                ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1::At(I3);
+            if(arg.K_raw_ % A_K_vec_length != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            constexpr auto A_M_vec_lenght =
+                ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1::At(I1) *
+                ABlockTransferSrcVectorTensorLengths_K0_M0_M1_K1::At(I2);
+            if(arg.M_raw_ % A_M_vec_lenght != 0)
+            {
+                return false;
+            }
+        }
+
+        if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+        {
+            constexpr auto B_N_vec_lenght =
+                BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1::At(I1) *
+                BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1::At(I2);
+            if(arg.N_raw_ % B_N_vec_lenght != 0)
+            {
+                return false;
+            }
+        }
+        else
+        {
+            constexpr auto B_K_vec_length =
+                BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1::At(I0) *
+                BBlockTransferSrcVectorTensorLengths_K0_N0_N1_K1::At(I3);
+            if(arg.K_raw_ % B_K_vec_length != 0)
+            {
+                return false;
+            }
+        }
+
         if(ck::get_device_name() == "gfx906" || ck::get_device_name() == "gfx1030" ||
            ck::get_device_name() == "gfx1100" || ck::get_device_name() == "gfx1101" ||
            ck::get_device_name() == "gfx1102")
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
index d444e29aa3..8e39b47f0c 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instances = std::tuple<
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
         // MPerBlock=128, NPerBlock=128
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=16, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=16
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=8
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=8
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,     S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
index f8f8a0bd33..c18d8ae737 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instances = std::tuple<
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
         // MPerBlock=128, NPerBlock=128       
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=64       
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=16, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=16
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=8
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=8
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index 840a4fabe1..1959f22136 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -35,26 +35,26 @@ using device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instances = std::tuple<
         // #########|      |      |      |        |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|      |      |      |        |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
         // MPerBlock=128, NPerBlock=128
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<8, 2>,       S<8, 2>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,   128,    16,  2,          4,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 2>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  2,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 2>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>,      S<2, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=16, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    16,    64,    16,  2,          1,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 1, 2>,      S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,       S<4, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=16
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    16,    16,  2,          4,          1,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<16, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=64
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<4, 1>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=8
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<4, 1>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=8
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<   F16,   F16,   F16,     F32,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
index d1eb8edf97..f88d9a9a60 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
@@ -32,36 +32,36 @@ using device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instances = std::tuple<
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                 |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                 |                   |
         // MPerBlock=128, NPerBlock=128
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=128, NPerBlock=64
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=128
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=64
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=32, NPerBlock=32
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=16, NPerBlock=16
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    16,    16,    16,    16,  4,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 4>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 4, 4>,       S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=64
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 4, 4>,      S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,     8,    64,    32,  4,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 4>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 4, 4>,      S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=8
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,      S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,    64,    64,     8,    32,  4,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,      S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=8
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl<  int8_t,  int8_t,  int8_t,    int32_t,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,     8,     8,     8,     4,  4,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 4>,        S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<1, 1, 4, 4>,       S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
index 3de5458e05..ba30ac8edf 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
@@ -32,36 +32,36 @@ using device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instances = std::tuple<
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder| Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                    |                    |                |                   |
         // MPerBlock=128, NPerBlock=128       
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=128, NPerBlock=64
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=128
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=64
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=32, NPerBlock=32
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<2, 1, 4, 4>,        S<4, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,      S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=16, NPerBlock=16
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 4, 2>,       S<4, 1, 4, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=64
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<8, 1, 4, 2>,      S<4, 1, 16, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=8
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<8, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=8
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t,  int8_t,  int8_t,    int32_t,     Col,     Col,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<1, 1, 4, 2>,       S<4, 1, 2, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,        S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>,      S<4, 1, 1, 2>,       S<1, 1, 8, 1>,   S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 2>,      S<1, 2, 0, 3>,        S<1, 1, 1, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
index 745a4bb31d..ad810276fc 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
@@ -32,36 +32,36 @@ using device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instances = std::tuple<
         // #########|       |        |        |           |        |        |        |   Operation|   Operation|   Operation|               |      |      |      |      |   |           |           |       |              |              |        K0_M0_M1_K1|          K0_M0_M1_K1|   ArrangeOrder|          Order| Lengths_K0_M0_M1_K1| ContiguousDimOrder|  Lengths_K0_M0_M1_K1|        K0_N0_N1_K1|          K0_N0_N1_K1|   ArrangeOrder|          Order| Lengths_K0_N0_N1_K1| ContiguousDimOrder|  Lengths_K0_N0_N1_K1|               Order|                |                   |
         // #########|       |        |        |           |        |        |        |            |            |            |               |      |      |      |      |   |           |           |       |              |              |                   |                     |               |               |                    |                   |                     |                   |                     |               |               |                    |                   |                     |                    |                |                   |
         // MPerBlock=128, NPerBlock=128
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<8, 2>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<4, 4>,       S<4, 2>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   128,   128,   128,    16,  4,          4,          8,      1,       S<2, 8>,       S<2, 4>,      S<8, 1, 2, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 8, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=128, NPerBlock=64
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,   128,    64,    16,  4,          4,          2,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,      S<2, 1, 128, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=128
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<4, 4>,       S<4, 4>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,   256,    64,   128,    16,  4,          2,          4,      1,       S<2, 8>,       S<2, 8>,      S<8, 1, 1, 4>,       S<2, 1, 64, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<8, 1, 32, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=64
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<2, 4>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<8, 1>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,    64,     8,  4,          4,          4,      1,       S<4, 2>,       S<8, 1>,      S<4, 1, 2, 4>,       S<2, 1, 32, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=32, NPerBlock=32
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          2,          4,      1,       S<4, 2>,       S<2, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<4, 2>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    32,    32,    32,     8,  4,          4,          2,      1,       S<2, 2>,       S<2, 4>,      S<4, 1, 2, 4>,       S<2, 1, 16, 1>,  S<1, 2, 0, 3>,  S<1, 2, 0, 3>,       S<4, 1, 1, 4>,      S<1, 2, 0, 3>,        S<1, 1, 1, 4>,      S<2, 1, 4, 4>,       S<4, 1, 8, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 4>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=16, NPerBlock=16
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<2, 2>,       S<2, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    16,    16,    16,    16,  2,          2,          2,      1,       S<4, 1>,       S<4, 1>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<4, 1, 4, 2>,        S<4, 1, 4, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=64
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,     8,    64,    32,  2,          1,          2,      1,       S<2, 2>,       S<8, 2>,      S<4, 1, 1, 2>,        S<8, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=64, NPerBlock=8
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,    64,    64,     8,    32,  2,          2,          1,      1,       S<8, 2>,       S<2, 2>,      S<8, 1, 4, 2>,       S<4, 1, 16, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,       S<32, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
         // MPerBlock=8, NPerBlock=8
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
-        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 4, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<4, 1>,       S<2, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          1,          2,      1,       S<1, 4>,       S<1, 2>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<2, 1>,       S<4, 1>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>,
+        DeviceGemmDl< int8_t, int8_t, int8_t,      int32_t,     Row,     Row,     Row, PassThrough, PassThrough, PassThrough,  GemmMNPadding,     8,     8,     8,     4,  2,          2,          1,      1,       S<1, 2>,       S<1, 4>,      S<4, 1, 1, 2>,        S<1, 1, 8, 1>,  S<0, 3, 1, 2>,  S<0, 3, 1, 2>,       S<4, 1, 1, 2>,      S<0, 3, 1, 2>,        S<1, 1, 1, 2>,      S<1, 1, 4, 2>,        S<4, 1, 2, 1>,   S<0, 3, 1, 2>,  S<0, 3, 1, 2>,      S<1, 1, 1, 1>,      S<0, 3, 1, 2>,        S<1, 1, 4, 2>, S<0, 1, 2, 3, 4, 5>,                5,                  1>
     // clang-format on
     >;
 

From 5a4416c8a7b6e28257a580d212373bd54a4aae3d Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 19 Sep 2023 09:17:39 -0700
Subject: [PATCH 133/194] fix the ckprofiler package build in a loop (#926)

---
 CMakeLists.txt | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index e3b60347d5..32defea756 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -105,6 +105,7 @@ if(NOT PROFILER_ONLY)
         TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
 else()
     add_definitions(-DPROFILER_ONLY)
+    set(GPU_TARGETS "" CACHE STRING "" FORCE)
     if(GPU_TARGETS)
         message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx9, gfx10, or gfx11")
     endif()
@@ -117,11 +118,12 @@ else()
     else()
         message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx9, gfx10, or gfx11")
     endif()
+    set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
 endif()
 
 message("Supported GPU_TARGETS= ${DEFAULT_GPU_TARGETS}")
 
-set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " ")
+set(AMDGPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
 
 if(GPU_TARGETS)
     message("Building CK for the following targets: ${GPU_TARGETS}")

From 718065ebd2509251291a3c3adeb7d0cef2ecb8dc Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 19 Sep 2023 09:35:45 -0700
Subject: [PATCH 134/194] update to rocm5.7 by default (#925)

* update to rocm5.7 by default

* fix jenkinsfile syntax
---
 Dockerfile  | 23 +++++------------------
 Jenkinsfile | 10 +++++-----
 2 files changed, 10 insertions(+), 23 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index b0954fdc75..4ba13649e4 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -16,25 +16,12 @@ RUN apt-get install -y --allow-unauthenticated apt-utils wget gnupg2 curl
 ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
 RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
 
-RUN wget https://repo.radeon.com/amdgpu-install/5.6/ubuntu/focal/amdgpu-install_5.6.50600-1_all.deb  --no-check-certificate
-RUN apt-get update && \
-DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
-    ./amdgpu-install_5.6.50600-1_all.deb
+RUN wget https://repo.radeon.com/amdgpu-install/5.7/ubuntu/focal/amdgpu-install_5.7.50700-1_all.deb  --no-check-certificate
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated ./amdgpu-install_5.7.50700-1_all.deb
 
-RUN if [ "$ROCMVERSION" != "5.7" ]; then \
-        wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
-        sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
-        sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'; \
-    elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "" ] || [ "$compiler_version" = "amd-stg-open" ]; then \
-        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
-        apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
-        amdgpu-repo --amdgpu-build=1609671 --rocm-build=compute-rocm-npi-mi300/1354; \
-    elif [ "$ROCMVERSION" = "5.7" ] && [ "$compiler_version" = "rc1" ]; then \
-        sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amdgpu-install-internal_5.7-20.04-1_all.deb" && \
-        apt update && apt-get install -y ./amdgpu-install-internal_5.7-20.04-1_all.deb && \
-        sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.7 rel-19 > /etc/apt/sources.list.d/rocm-build.list' && \
-        amdgpu-repo --amdgpu-build=1637781; \
-    fi
+RUN wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
+    sh -c "echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] $DEB_ROCM_REPO focal main > /etc/apt/sources.list.d/rocm.list" && \
+    sh -c 'echo deb [arch=amd64 signed-by=/etc/apt/trusted.gpg.d/rocm-keyring.gpg] https://repo.radeon.com/amdgpu/$ROCMVERSION/ubuntu focal main > /etc/apt/sources.list.d/amdgpu.list'
 
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
 RUN amdgpu-install -y --usecase=rocm --no-dkms
diff --git a/Jenkinsfile b/Jenkinsfile
index c1a28a74ed..0fdadf8a87 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -33,7 +33,7 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.ROCMVERSION != "5.7"){
+    if (params.ROCMVERSION != "6.0"){
        if (params.COMPILER_VERSION == "") {
            img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}"
        }
@@ -613,8 +613,8 @@ def process_results(Map conf=[:]){
 }
 
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
-CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=rc1
-                                              0 21 * * * % ROCMVERSION=5.6;COMPILER_VERSION=;COMPILER_COMMIT=
+CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=
+                                              0 21 * * * % ROCMVERSION=5.7;COMPILER_VERSION=;COMPILER_COMMIT=
                                               0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
 
 pipeline {
@@ -632,8 +632,8 @@ pipeline {
             description: "Force building docker image (default: false), set to true if docker image needs to be updated.")
         string(
             name: 'ROCMVERSION', 
-            defaultValue: '5.6', 
-            description: 'Specify which ROCM version to use: 5.6 (default).')
+            defaultValue: '5.7', 
+            description: 'Specify which ROCM version to use: 5.7 (default).')
         string(
             name: 'COMPILER_VERSION', 
             defaultValue: '', 

From 58817bf967d3b7b314233f6331f0430cf40651ae Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 19 Sep 2023 18:50:58 -0700
Subject: [PATCH 135/194] fix the building of the amd-stg-open compiler (#927)

---
 Dockerfile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Dockerfile b/Dockerfile
index 4ba13649e4..4f227d0362 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -101,7 +101,7 @@ RUN sh -c "echo compiler commit = '$compiler_commit'"
 RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && mkdir build && cd build && \
-        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
+        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
         make -j 8 ; \
     else echo "using the release compiler"; \
     fi
@@ -109,7 +109,7 @@ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" = "" ];
 RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
-        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
+        cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld" -DLLVM_ENABLE_RUNTIMES="compiler-rt" ../llvm && \
         make -j 8 ; \
     else echo "using the release compiler"; \
     fi

From bba085d2b58da0b222e784b8a335c3a24acbacc9 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 20 Sep 2023 22:15:56 -0700
Subject: [PATCH 136/194] Refactoring cmake files to build data types
 separately. (#932)

* refactor cmake files for the tests

* refactor cmake files for examples

* fix cmake for gemm example

* fix the cmake file for all examples

* add splitting by data types in gemm_splitk instance header

* rename test to reflect only dl instances are used

* clean up CI workspace, update cmake for instances

* change the jenkinsfile syntax

* build all instances except DL on gfx11

* move workspace cleanup after stages

* clean up workspace after every stage

* isolate data types in grouped_conv_fwd header

* isolate dl instances for grouped_conv2d_fwd

* fix syntax

* fix cmake and batchnorm instances

* fix typo

* fix reduction instances

* fix grouped_conv headers

* fix syntax

* replace parsing logic for instances, replace bfp16 with bf16

* fix the client examples build

* clean up DTYPES from instances cmake files

* update the parsing logic in cmake files

* make an exception for reduction kernels

* update few remaining cmake files to handle DTYPES

* fix syntax

* fix cmake conflicts

* replace f8 with fp8 test name

* resolve conflicts for dpp instances
---
 CMakeLists.txt                                |  15 +-
 example/01_gemm/CMakeLists.txt                |  82 +++----
 example/02_gemm_bilinear/CMakeLists.txt       |   2 -
 example/03_gemm_bias_relu/CMakeLists.txt      |   2 -
 .../04_gemm_add_add_fastgelu/CMakeLists.txt   |  16 +-
 example/09_convnd_fwd/CMakeLists.txt          |  26 +--
 .../CMakeLists.txt                            |  32 +--
 example/13_pool2d_fwd/CMakeLists.txt          |   8 +-
 example/14_gemm_quantization/CMakeLists.txt   |   7 +-
 example/15_grouped_gemm/CMakeLists.txt        |  54 +++--
 ...dl_bfp16.cpp => grouped_gemm_xdl_bf16.cpp} |   0
 .../CMakeLists.txt                            |  43 ++--
 example/17_convnd_bwd_data/CMakeLists.txt     |  13 +-
 example/18_batched_gemm_reduce/CMakeLists.txt |   2 -
 .../20_grouped_conv_bwd_weight/CMakeLists.txt |  20 +-
 example/21_gemm_layernorm/CMakeLists.txt      |   3 +-
 example/22_cgemm/CMakeLists.txt               |  16 +-
 example/24_batched_gemm/CMakeLists.txt        |  22 +-
 ...dl_bfp16.cpp => batched_gemm_xdl_bf16.cpp} |   0
 example/25_gemm_bias_e_permute/CMakeLists.txt |   6 +-
 example/26_contraction/CMakeLists.txt         |  12 +-
 example/27_layernorm/CMakeLists.txt           |   6 +-
 .../CMakeLists.txt                            |   4 +-
 .../CMakeLists.txt                            |   8 +-
 .../CMakeLists.txt                            |  32 +--
 example/31_batched_gemm_gemm/CMakeLists.txt   |  18 +-
 .../CMakeLists.txt                            |  35 +--
 example/35_splitK_gemm/CMakeLists.txt         |  23 +-
 ...xdl_bfp16.cpp => splitK_gemm_xdl_bf16.cpp} |   0
 .../CMakeLists.txt                            |   4 +-
 .../CMakeLists.txt                            |  11 +-
 example/39_permute/CMakeLists.txt             |  15 +-
 .../40_conv2d_fwd_quantization/CMakeLists.txt |   4 -
 .../41_grouped_conv_conv_fwd/CMakeLists.txt   |  16 +-
 example/42_groupnorm/CMakeLists.txt           |   8 +-
 .../CMakeLists.txt                            |   8 +-
 example/44_elementwise_permute/CMakeLists.txt |   6 +-
 example/46_gemm_add_multiply/CMakeLists.txt   |   8 +-
 example/48_pool3d_fwd/CMakeLists.txt          |   4 +-
 example/49_maxpool2d_bwd/CMakeLists.txt       |  12 +-
 example/50_put_element/CMakeLists.txt         |   4 +-
 example/CMakeLists.txt                        | 114 +++++++++-
 .../gpu/batchnorm_backward.hpp                |  50 +++--
 .../gpu/batchnorm_forward.hpp                 |  44 ++--
 .../gpu/batchnorm_infer.hpp                   |  44 ++--
 .../gpu/gemm_splitk.hpp                       |  17 +-
 .../gpu/grouped_convolution_backward_data.hpp |  60 ++++--
 .../grouped_convolution_backward_weight.hpp   | 133 +++++++++---
 .../gpu/grouped_convolution_forward.hpp       | 175 ++++++++++-----
 .../gpu/reduce/device_reduce_instance.hpp     |  59 +++---
 .../gpu/CMakeLists.txt                        | 102 +++++++--
 .../gpu/avg_pool3d_bwd/CMakeLists.txt         |  12 +-
 .../gpu/batched_gemm/CMakeLists.txt           |  22 +-
 .../CMakeLists.txt                            |   2 -
 .../batched_gemm_bias_permute/CMakeLists.txt  |   3 +-
 .../gpu/batched_gemm_gemm/CMakeLists.txt      |   2 -
 .../gpu/batched_gemm_multi_d/CMakeLists.txt   |  44 ++--
 .../gpu/batched_gemm_reduce/CMakeLists.txt    |   2 -
 .../batched_gemm_softmax_gemm/CMakeLists.txt  |   2 -
 .../CMakeLists.txt                            |  14 +-
 .../gpu/contraction_bilinear/CMakeLists.txt   |  27 ++-
 .../gpu/contraction_scale/CMakeLists.txt      |  26 ++-
 .../gpu/conv2d_bwd_data/CMakeLists.txt        |  29 +--
 .../gpu/conv2d_fwd/CMakeLists.txt             |  20 +-
 .../elementwise_normalization/CMakeLists.txt  |   2 -
 .../gpu/gemm/CMakeLists.txt                   | 200 ++++++++----------
 .../gpu/gemm_add_add_fastgelu/CMakeLists.txt  |   2 -
 .../gpu/gemm_add_fastgelu/CMakeLists.txt      |   2 -
 .../CMakeLists.txt                            |   2 -
 .../gpu/gemm_bilinear/CMakeLists.txt          |   2 -
 .../gpu/gemm_fastgelu/CMakeLists.txt          |   2 -
 .../gpu/gemm_multiply_add/CMakeLists.txt      |  15 +-
 .../gpu/gemm_splitk/CMakeLists.txt            |  40 ++--
 .../gpu/gemm_streamk/CMakeLists.txt           |   2 -
 .../gpu/grouped_gemm/CMakeLists.txt           |   2 -
 .../gpu/grouped_gemm_fastgelu/CMakeLists.txt  |   2 -
 .../gpu/max_pool_bwd/CMakeLists.txt           |  12 +-
 .../gpu/normalization/CMakeLists.txt          |  13 +-
 .../gpu/pool3d_fwd/CMakeLists.txt             |  18 +-
 .../gpu/quantization/CMakeLists.txt           |  24 +--
 .../gpu/softmax/CMakeLists.txt                |  11 +-
 test/CMakeLists.txt                           | 121 +++++++++--
 test/batched_gemm/CMakeLists.txt              |  28 ++-
 test/batched_gemm_gemm/CMakeLists.txt         |   8 +-
 test/batched_gemm_multi_d/CMakeLists.txt      |   4 +-
 ...d.cpp => test_batched_gemm_multi_d_dl.cpp} |   0
 test/batched_gemm_reduce/CMakeLists.txt       |   7 +-
 test/batched_gemm_softmax_gemm/CMakeLists.txt |  14 +-
 .../CMakeLists.txt                            |  41 ++--
 test/data_type/CMakeLists.txt                 |  15 +-
 test/data_type/{f8.cpp => fp8.cpp}            |   0
 test/elementwise_normalization/CMakeLists.txt |   6 +-
 test/gemm/CMakeLists.txt                      |  30 ++-
 test/gemm_layernorm/CMakeLists.txt            |  10 +-
 test/gemm_reduce/CMakeLists.txt               |   7 +-
 test/grouped_gemm/CMakeLists.txt              |   2 -
 test/normalization/CMakeLists.txt             |  24 ++-
 test/reduce/CMakeLists.txt                    |   6 +-
 98 files changed, 1279 insertions(+), 1005 deletions(-)
 rename example/15_grouped_gemm/{grouped_gemm_xdl_bfp16.cpp => grouped_gemm_xdl_bf16.cpp} (100%)
 rename example/24_batched_gemm/{batched_gemm_xdl_bfp16.cpp => batched_gemm_xdl_bf16.cpp} (100%)
 rename example/35_splitK_gemm/{splitK_gemm_xdl_bfp16.cpp => splitK_gemm_xdl_bf16.cpp} (100%)
 rename test/batched_gemm_multi_d/{test_batched_gemm_multi_d.cpp => test_batched_gemm_multi_d_dl.cpp} (100%)
 rename test/data_type/{f8.cpp => fp8.cpp} (100%)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 32defea756..95ef2df7e9 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,4 +1,5 @@
 cmake_minimum_required(VERSION 3.14)
+cmake_policy(SET CMP0140 NEW)
 
 # This has to be initialized before the project() command appears
 # Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D.  MSVC_IDE does not use CMAKE_BUILD_TYPE
@@ -383,31 +384,31 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
     set(cmake_instance)
     file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
     set(add_inst 0)
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8")
+    if(("${cmake_instance}" MATCHES "fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
         #message("fp8 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf8\" " AND DTYPES MATCHES "bf8")
+    if(("${cmake_instance}" MATCHES "bf8" OR "${cmake_instance}" MATCHES "_b8") AND DTYPES MATCHES "bf8")
         #message("bf8 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
+    if(("${cmake_instance}" MATCHES "fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
         #message("fp16 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32")
+    if(("${cmake_instance}" MATCHES "fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
         #message("fp32 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64")
+    if(("${cmake_instance}" MATCHES "fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
         #message("fp64 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16")
+    if(("${cmake_instance}" MATCHES "bf16" OR "${cmake_instance}" MATCHES "_b16") AND DTYPES MATCHES "bf16")
         #message("bf16 instance found!")
         set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8")
+    if(("${cmake_instance}" MATCHES "int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
         #message("int8 instance found!")
         set(add_inst 1)
     endif()
diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 5574d09001..2351d3f105 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -1,51 +1,54 @@
-if(DL_KERNELS)
-  add_custom_target(example_gemm_dl)
+add_custom_target(example_gemm_dl)
 
-  add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
-  add_dependencies(example_gemm_dl example_gemm_dl_fp32)
-  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
+add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_dl example_gemm_dl_fp32)
+endif()
+add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-    add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
-  endif()
-  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+endif()
+add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
+add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_dl example_gemm_dl_int8)
-  endif()
-
-  if(USE_BITINT_EXTENSION_INT4)
+endif()
+if(USE_BITINT_EXTENSION_INT4)
     add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
     add_dependencies(example_gemm_dl example_gemm_dl_int4)
-  endif(USE_BITINT_EXTENSION_INT4)
-endif()
+endif(USE_BITINT_EXTENSION_INT4)
 
 add_custom_target(example_gemm_xdl)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
-  add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
-  add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
-
-  if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
+endif()
+add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
+endif()
+add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
+endif()
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
     add_custom_target(example_gemm_wmma)
     add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
-    add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
-  endif()
-
+    if(result EQUAL 0)
+      add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
+    endif()
 endif()
 
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
+add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
 
   add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
   add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
 endif()
 
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
+add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
 endif()
 
@@ -54,22 +57,23 @@ if(USE_BITINT_EXTENSION_INT4)
   add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
 endif(USE_BITINT_EXTENSION_INT4)
 
-if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
-  # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-  add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
+# FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
+add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
 endif()
 
 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
 
-if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
-  if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-    add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+
+if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
   endif()
 endif()
 
-if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
-  add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
+add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
 endif()
diff --git a/example/02_gemm_bilinear/CMakeLists.txt b/example/02_gemm_bilinear/CMakeLists.txt
index 52e6330523..d82c42d5a9 100644
--- a/example/02_gemm_bilinear/CMakeLists.txt
+++ b/example/02_gemm_bilinear/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list1 gfx1100 gfx1101 gfx1102)
 list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
@@ -19,4 +18,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()
diff --git a/example/03_gemm_bias_relu/CMakeLists.txt b/example/03_gemm_bias_relu/CMakeLists.txt
index a247a052cb..2f5cba924d 100644
--- a/example/03_gemm_bias_relu/CMakeLists.txt
+++ b/example/03_gemm_bias_relu/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()
diff --git a/example/04_gemm_add_add_fastgelu/CMakeLists.txt b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
index 15ec62c89f..9fe833dda0 100644
--- a/example/04_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
@@ -3,24 +3,24 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
     add_custom_target(example_gemm_add_add_fastgelu_xdl)
-    if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
     endif()
-    if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
     endif()
-    if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
     endif()
     if(USE_BITINT_EXTENSION_INT4)
        add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
        add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
     endif(USE_BITINT_EXTENSION_INT4)
-    if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+    add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+    if(result EQUAL 0)
       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
     endif()
    set(target 1)
diff --git a/example/09_convnd_fwd/CMakeLists.txt b/example/09_convnd_fwd/CMakeLists.txt
index 1af1e6c858..f9903bfe03 100644
--- a/example/09_convnd_fwd/CMakeLists.txt
+++ b/example/09_convnd_fwd/CMakeLists.txt
@@ -2,34 +2,16 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-    if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_fp32 convnd_fwd_xdl_fp32.cpp)
-    endif()
-    if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_fp16 convnd_fwd_xdl_fp16.cpp)
-    endif()
-    if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_bf16 convnd_fwd_xdl_bf16.cpp)
-    endif()
-    if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
       add_example_executable(example_convnd_fwd_xdl_int8 convnd_fwd_xdl_int8.cpp)
-    endif()
-    # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
-    if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+      # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
       add_example_executable_no_testing(example_convnd_fwd_xdl_fp64 convnd_fwd_xdl_fp64.cpp)
-    endif()
    set(target 1)
  endif()
 endforeach()
 
-if(DL_KERNELS)
-  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
-  endif()
-  if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
-  endif()
-  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
-  endif()
-endif()
+add_example_executable(example_convnd_fwd_dl_fp16 convnd_fwd_dl_fp16.cpp)
+add_example_executable(example_convnd_fwd_dl_fp32 convnd_fwd_dl_fp32.cpp)
+add_example_executable(example_convnd_fwd_dl_int8 convnd_fwd_dl_int8.cpp)
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
index e7d941ae6b..150d146e31 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
@@ -2,27 +2,27 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(example_convnd_fwd_reduce_xdl)
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+  add_custom_target(example_convnd_fwd_reduce_xdl)
+  add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
-   endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+  endif()
+  add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
-   endif()
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+  endif()
+  add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
-   endif()
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+  endif()
+  add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
-   endif()
-   if(USE_BITINT_EXTENSION_INT4)
+  endif()
+  if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
       add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
-   endif(USE_BITINT_EXTENSION_INT4)
-   set(target 1)
+  endif(USE_BITINT_EXTENSION_INT4)
+  set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/example/13_pool2d_fwd/CMakeLists.txt b/example/13_pool2d_fwd/CMakeLists.txt
index d0f356757b..e2a923cded 100644
--- a/example/13_pool2d_fwd/CMakeLists.txt
+++ b/example/13_pool2d_fwd/CMakeLists.txt
@@ -1,6 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
-endif()
+add_example_executable(example_pool2d_fwd_fp16 pool2d_fwd_fp16.cpp)
+add_example_executable(example_pool2d_fwd_fp32 pool2d_fwd_fp32.cpp)
diff --git a/example/14_gemm_quantization/CMakeLists.txt b/example/14_gemm_quantization/CMakeLists.txt
index 3b3ad80dd8..9793e8b8a0 100644
--- a/example/14_gemm_quantization/CMakeLists.txt
+++ b/example/14_gemm_quantization/CMakeLists.txt
@@ -1,9 +1,5 @@
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 # dlops
-if(DL_KERNELS)
-  add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
-endif()
-
+add_example_executable(example_gemm_dl_quantization_int8 gemm_dl_quantization_int8.cpp)
 # xdlops
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
@@ -14,4 +10,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()
\ No newline at end of file
diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index 45e02126b6..f2c76b76f2 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -1,36 +1,44 @@
 add_custom_target(example_grouped_gemm_xdl)
-
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
+add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
+if(result EQUAL 0)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
 endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
-  add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
-  add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
-  add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
-  add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
-  add_dependencies(example_grouped_gemm_xdl 
-                   example_grouped_gemm_xdl_fp16
-                   example_grouped_gemm_multiple_d_dl_fp16
-                   example_grouped_gemm_xdl_splitk_fp16
-                   example_grouped_gemm_xdl_fixed_nk_fp16
-                   example_grouped_gemm_xdl_fixed_nk_bias_fp16)
+add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
 endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_bfp16 grouped_gemm_xdl_bfp16.cpp)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bfp16)
+add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
+if(result EQUAL 0)
+    add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
 endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
+add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_bf16 grouped_gemm_xdl_bf16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
+endif()
+add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
 endif()
-if(DTYPES MATCHES "f8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
+add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
 endif()
 
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
+  if(result EQUAL 0)
+    add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
+  endif()
 endif()
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_bf16.cpp
similarity index 100%
rename from example/15_grouped_gemm/grouped_gemm_xdl_bfp16.cpp
rename to example/15_grouped_gemm/grouped_gemm_xdl_bf16.cpp
diff --git a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
index 00786d34a3..67cf5666d7 100644
--- a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+++ b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
@@ -6,30 +6,43 @@ foreach(gpu IN LISTS GPU_TARGETS)
    add_custom_target(example_gemm_reduce_xdl_max)
    add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
    add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
-    add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
-    add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+   add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
+   endif()
+   add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
+   endif()
+   add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
    endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
-    add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
+
+   add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
+   endif()
+   add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
    endif()
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
-    add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
+
+   add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
+   endif()
+   add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
    endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
-    add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
+
+   add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
+   endif()
+   add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
    endif()
    
@@ -40,7 +53,9 @@ foreach(gpu IN LISTS GPU_TARGETS)
 
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
-      add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
+      if(result EQUAL 0)
+        add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
+      endif()
    endif()
    set(target 1)
  endif()
diff --git a/example/17_convnd_bwd_data/CMakeLists.txt b/example/17_convnd_bwd_data/CMakeLists.txt
index e187bd4337..7c6d10d8a0 100644
--- a/example/17_convnd_bwd_data/CMakeLists.txt
+++ b/example/17_convnd_bwd_data/CMakeLists.txt
@@ -1,15 +1,16 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_example_executable(example_convnd_bwd_data_xdl_fp16 convnd_bwd_data_xdl_fp16.cpp)
-   target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
+   if(result EQUAL 0)
+      target_link_libraries(example_convnd_bwd_data_xdl_fp16 PRIVATE utility)
+   endif()
    set(target 1)
  endif()
 endforeach()
-  if(DL_KERNELS)
-    add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
-    target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
-  endif()
+
+add_example_executable(example_convnd_bwd_data_dl_fp16 convnd_bwd_data_dl_fp16.cpp)
+if(result EQUAL 0)
+  target_link_libraries(example_convnd_bwd_data_dl_fp16 PRIVATE utility)
 endif()
diff --git a/example/18_batched_gemm_reduce/CMakeLists.txt b/example/18_batched_gemm_reduce/CMakeLists.txt
index a1bb398af0..94ed129dc0 100644
--- a/example/18_batched_gemm_reduce/CMakeLists.txt
+++ b/example/18_batched_gemm_reduce/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -7,4 +6,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index d649567ed2..ab8ed5c2bf 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -3,22 +3,20 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_weight)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
    endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
+   add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
+   if(result EQUAL 0)
     add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
    endif()
    set(target 1)
  endif()
 endforeach()
 
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  if(DL_KERNELS)
-    add_custom_target(example_grouped_conv_bwd_weight_dl)
-    add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
-    add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
-  endif()
-endif()
\ No newline at end of file
+add_custom_target(example_grouped_conv_bwd_weight_dl)
+add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
+endif()
diff --git a/example/21_gemm_layernorm/CMakeLists.txt b/example/21_gemm_layernorm/CMakeLists.txt
index 6a6735efd6..e231bc619b 100644
--- a/example/21_gemm_layernorm/CMakeLists.txt
+++ b/example/21_gemm_layernorm/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -10,4 +9,4 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()
+
diff --git a/example/22_cgemm/CMakeLists.txt b/example/22_cgemm/CMakeLists.txt
index 854f07fda6..f401f7187f 100644
--- a/example/22_cgemm/CMakeLists.txt
+++ b/example/22_cgemm/CMakeLists.txt
@@ -1,19 +1,19 @@
 add_custom_target(example_cgemm_xdl)
 
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
+add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
+if(result EQUAL 0)
   add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
 endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
+add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
+if(result EQUAL 0)
   add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
 endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
 add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp)
-add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
+if(result EQUAL 0)
+  add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
 endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
+add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
 endif()
 if(USE_BITINT_EXTENSION_INT4)
diff --git a/example/24_batched_gemm/CMakeLists.txt b/example/24_batched_gemm/CMakeLists.txt
index 48a3b58ff5..c80540de7d 100644
--- a/example/24_batched_gemm/CMakeLists.txt
+++ b/example/24_batched_gemm/CMakeLists.txt
@@ -1,21 +1,23 @@
 add_custom_target(example_batched_gemm_xdl)
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-  add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
+add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
+if(result EQUAL 0)
   add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32)
 endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
+add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
+if(result EQUAL 0)
   add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16)
 endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-  add_example_executable(example_batched_gemm_xdl_bfp16 batched_gemm_xdl_bfp16.cpp)
-  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bfp16)
+add_example_executable(example_batched_gemm_xdl_bf16 batched_gemm_xdl_bf16.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bf16)
 endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-  add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
+add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
+if(result EQUAL 0)
   add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8)
 endif()
 if(USE_BITINT_EXTENSION_INT4)
   add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp)
-  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
+  if(result EQUAL 0)
+    add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
+  endif()
 endif()
diff --git a/example/24_batched_gemm/batched_gemm_xdl_bfp16.cpp b/example/24_batched_gemm/batched_gemm_xdl_bf16.cpp
similarity index 100%
rename from example/24_batched_gemm/batched_gemm_xdl_bfp16.cpp
rename to example/24_batched_gemm/batched_gemm_xdl_bf16.cpp
diff --git a/example/25_gemm_bias_e_permute/CMakeLists.txt b/example/25_gemm_bias_e_permute/CMakeLists.txt
index eb274b2338..cbc3c007bc 100644
--- a/example/25_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/25_gemm_bias_e_permute/CMakeLists.txt
@@ -1,4 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp)
-    add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp)
-endif()
+add_example_executable(example_gemm_bias_e_permute_g1m3n2k1_xdl_fp16 gemm_bias_e_permute_g1m3n2k1_xdl_fp16.cpp)
+add_example_executable(example_gemm_bias_e_permute_g1m2n3k1_xdl_fp16 gemm_bias_e_permute_g1m2n3k1_xdl_fp16.cpp)
diff --git a/example/26_contraction/CMakeLists.txt b/example/26_contraction/CMakeLists.txt
index 6cab88b13e..167a9f147d 100644
--- a/example/26_contraction/CMakeLists.txt
+++ b/example/26_contraction/CMakeLists.txt
@@ -1,8 +1,4 @@
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
-    add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
-endif()
-if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
-    add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
-    add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
-endif()
+add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
+add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
+add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
+add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
diff --git a/example/27_layernorm/CMakeLists.txt b/example/27_layernorm/CMakeLists.txt
index 9cb2cd0766..94c23ce774 100644
--- a/example/27_layernorm/CMakeLists.txt
+++ b/example/27_layernorm/CMakeLists.txt
@@ -1,4 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
-    add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
-endif()
+add_example_executable(example_layernorm_fp16 layernorm_fp16.cpp)
+add_example_executable(example_layernorm_splitk_fp16 layernorm_splitk_fp16.cpp)
diff --git a/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt b/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
index 2fda1f62a9..44ab16894c 100644
--- a/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/28_grouped_gemm_bias_e_permute/CMakeLists.txt
@@ -1,3 +1 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
-endif()
+add_example_executable(example_grouped_gemm_bias_e_permute_xdl_fp16 grouped_gemm_bias_e_permute_xdl_fp16.cpp)
diff --git a/example/29_batched_gemm_bias_e_permute/CMakeLists.txt b/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
index 09c3e6c608..32a87dd200 100644
--- a/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/29_batched_gemm_bias_e_permute/CMakeLists.txt
@@ -1,7 +1,5 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
+add_example_executable(example_batched_gemm_bias_e_permute_xdl_fp16 batched_gemm_bias_e_permute_xdl_fp16.cpp)
 
-    if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
-        add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
-    endif()
+if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
+    add_example_executable(example_batched_gemm_bias_e_permute_wmma_fp16 batched_gemm_bias_e_permute_wmma_fp16.cpp)
 endif()
diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index e37413c098..4ab5bd75f2 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -5,27 +5,31 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
    add_custom_target(example_grouped_conv_fwd_multiple_d)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
+   if(result EQUAL 0)
       add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
-      add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
+   endif()
+   add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
+   if(result EQUAL 0)
       add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
    endif()
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
+   if(result EQUAL 0)
       add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
    endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
+   if(result EQUAL 0)
       add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
    endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
+   if(result EQUAL 0)
       add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
    endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
-      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
+      if(result EQUAL 0)
+         add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
+      endif()
    endif() # USE_BITINT_EXTENSION_INT4
 
    set(target 1)
@@ -35,12 +39,8 @@ endforeach()
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
-   endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
-   endif()
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
    set(target 1)
  endif()
 endforeach()
diff --git a/example/31_batched_gemm_gemm/CMakeLists.txt b/example/31_batched_gemm_gemm/CMakeLists.txt
index 2074520f8c..93f16c945f 100644
--- a/example/31_batched_gemm_gemm/CMakeLists.txt
+++ b/example/31_batched_gemm_gemm/CMakeLists.txt
@@ -1,17 +1,11 @@
 list(APPEND gpu_list1 gfx908 gfx90a gfx940 gfx941 gfx942)
-list(APPEND gpu_list2 gfx908 gfx90a)
+
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-      add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
-   endif()
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
-   endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
-   endif()
+   add_example_executable(example_batched_gemm_gemm_xdl_fp32 batched_gemm_gemm_xdl_fp32.cpp)
+   add_example_executable(example_batched_gemm_gemm_xdl_fp16 batched_gemm_gemm_xdl_fp16.cpp)
+   add_example_executable(example_batched_gemm_gemm_xdl_bf16 batched_gemm_gemm_xdl_bf16.cpp)
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_batched_gemm_gemm_xdl_int4 batched_gemm_gemm_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
@@ -20,7 +14,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
 endforeach()
 
 if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
-   endif()
+   add_example_executable(example_batched_gemm_gemm_xdl_int8 batched_gemm_gemm_xdl_int8.cpp)
 endif()
diff --git a/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt b/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
index 0463bf6bd3..25eb44ae59 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
+++ b/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
@@ -1,24 +1,31 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
-    add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
-    add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
-    add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
-    add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
-    add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
-endif()
-
 add_custom_target(example_gemm_scale_softmax_gemm)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+
+add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
+endif()
+add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
+endif()
+add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
+endif()
+add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
+endif()
+add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
 endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
+add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
+endif()
+add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
 endif()
+
diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index 251a9b93c5..8970a57648 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -3,25 +3,28 @@ set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_splitK_gemm_xdl)
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
+
+  add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
    endif()
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
+  add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
    endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_splitK_gemm_xdl_bfp16 splitK_gemm_xdl_bfp16.cpp)
-    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bfp16)
+  add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
+  if(result EQUAL 0)
+    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16)
    endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
+  add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
+  if(result EQUAL 0)
     add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
    endif()
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
-      add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
+      if(result EQUAL 0)
+        add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
+      endif()
    endif()
    set(target 1)
  endif()
diff --git a/example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp b/example/35_splitK_gemm/splitK_gemm_xdl_bf16.cpp
similarity index 100%
rename from example/35_splitK_gemm/splitK_gemm_xdl_bfp16.cpp
rename to example/35_splitK_gemm/splitK_gemm_xdl_bf16.cpp
diff --git a/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt b/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
index 36bb5720d5..a9be3a7108 100644
--- a/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
+++ b/example/37_batched_gemm_add_add_relu_gemm_add/CMakeLists.txt
@@ -1,3 +1 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
-endif()
+add_example_executable(example_batched_gemm_add_add_relu_gemm_add_xdl_fp16 batched_gemm_add_add_relu_gemm_add_xdl_fp16.cpp)
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index 3821f8aaca..1675e3663a 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -1,15 +1,16 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_data)
    add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
+   if(result EQUAL 0)
+      add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
+   endif()
    add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
-
-   add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
-   add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+   if(result EQUAL 0)
+     add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+   endif()
    set(target 1)
  endif()
 endforeach()
-endif()
diff --git a/example/39_permute/CMakeLists.txt b/example/39_permute/CMakeLists.txt
index 5b43de9725..bcf47b4926 100644
--- a/example/39_permute/CMakeLists.txt
+++ b/example/39_permute/CMakeLists.txt
@@ -1,11 +1,14 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_custom_target(example_permute)
-
-    add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
-    add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
-    add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
+add_custom_target(example_permute)
 
+add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_permute example_permute_1xHxW_fp16)
+endif()
+add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_permute example_permute_NxHxW_fp16)
+endif()
+add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
+if(result EQUAL 0)
     add_dependencies(example_permute example_permute_HxWx4_fp16)
 endif()
diff --git a/example/40_conv2d_fwd_quantization/CMakeLists.txt b/example/40_conv2d_fwd_quantization/CMakeLists.txt
index 55464957ac..2d804cafe9 100644
--- a/example/40_conv2d_fwd_quantization/CMakeLists.txt
+++ b/example/40_conv2d_fwd_quantization/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -11,7 +10,6 @@ foreach(gpu IN LISTS GPU_TARGETS)
  endif()
 endforeach()
 
-  if(DL_KERNELS)
   # Conv perlayer quantization
   add_example_executable(example_conv2d_fwd_dl_perlayer_quantization_int8 conv2d_fwd_dl_perlayer_quantization_int8.cpp)
   # Conv perchannel quantization
@@ -24,5 +22,3 @@ endforeach()
   add_example_executable(example_conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8 conv2d_fwd_dl_bias_tanh_perlayer_quantization_int8.cpp)
   # Conv + bias + tanh perchannel quantization
   add_example_executable(example_conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8 conv2d_fwd_dl_bias_tanh_perchannel_quantization_int8.cpp)
-  endif()
-endif()
\ No newline at end of file
diff --git a/example/41_grouped_conv_conv_fwd/CMakeLists.txt b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
index 085d90a9e9..ae251e88d2 100644
--- a/example/41_grouped_conv_conv_fwd/CMakeLists.txt
+++ b/example/41_grouped_conv_conv_fwd/CMakeLists.txt
@@ -3,15 +3,9 @@ list(APPEND gpu_list2 gfx908 gfx90a)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
-   endif()
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
-   endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
-   endif()
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp32 grouped_conv_conv_fwd_xdl_fp32.cpp)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_fp16 grouped_conv_conv_fwd_xdl_fp16.cpp)
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_bf16 grouped_conv_conv_fwd_xdl_bf16.cpp)
    if(USE_BITINT_EXTENSION_INT4)
       add_example_executable(example_grouped_conv_conv_fwd_xdl_int4 grouped_conv_conv_fwd_xdl_int4.cpp)
    endif(USE_BITINT_EXTENSION_INT4)
@@ -20,7 +14,5 @@ foreach(gpu IN LISTS GPU_TARGETS)
 endforeach()
 
 if(NOT GPU_TARGETS MATCHES "gfx94" AND NOT GPU_TARGETS MATCHES "gfx1")
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
-   endif()
+   add_example_executable(example_grouped_conv_conv_fwd_xdl_int8 grouped_conv_conv_fwd_xdl_int8.cpp)
 endif()
diff --git a/example/42_groupnorm/CMakeLists.txt b/example/42_groupnorm/CMakeLists.txt
index bc2246a2bf..e8c306ac58 100644
--- a/example/42_groupnorm/CMakeLists.txt
+++ b/example/42_groupnorm/CMakeLists.txt
@@ -1,5 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
-    add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
-    add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
-endif()
+add_example_executable(example_groupnorm_sigmoid_mul_fp16 groupnorm_sigmoid_mul_fp16.cpp)
+add_example_executable(example_groupnorm_splitk_fp16 groupnorm_splitk_fp16.cpp)
+add_example_executable(example_groupnorm_swish_fp16 groupnorm_swish_fp16.cpp)
diff --git a/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt b/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt
index 7e070f5357..c29f18f162 100644
--- a/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt
+++ b/example/43_splitk_gemm_bias_e_permute/CMakeLists.txt
@@ -1,6 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
-endif()
+add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp16 splitk_gemm_bias_e_permute_xdl_fp16.cpp)
+add_example_executable(example_splitk_gemm_bias_e_permute_xdl_fp32 splitk_gemm_bias_e_permute_xdl_fp32.cpp)
diff --git a/example/44_elementwise_permute/CMakeLists.txt b/example/44_elementwise_permute/CMakeLists.txt
index 877a820315..0e0091a986 100644
--- a/example/44_elementwise_permute/CMakeLists.txt
+++ b/example/44_elementwise_permute/CMakeLists.txt
@@ -1,4 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
-    add_example_executable(example_elementwise_permute_4D_fp16_2d elementwise_permute_4D_fp16_2d.cpp)
-endif()
+add_example_executable(example_elementwise_permute_4D_fp16 elementwise_permute_4D_fp16.cpp)
+add_example_executable(example_elementwise_permute_4D_fp16_2d elementwise_permute_4D_fp16_2d.cpp)
diff --git a/example/46_gemm_add_multiply/CMakeLists.txt b/example/46_gemm_add_multiply/CMakeLists.txt
index cf7d81f895..bfe057e8da 100644
--- a/example/46_gemm_add_multiply/CMakeLists.txt
+++ b/example/46_gemm_add_multiply/CMakeLists.txt
@@ -1,6 +1,2 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    if(DL_KERNELS)
-        add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
-    endif()
-    add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
-endif()
+add_example_executable(example_gemm_add_multiply_dl_fp16 gemm_add_multiply_dl_fp16.cpp)
+add_example_executable(example_gemm_add_multiply_xdl_fp16 gemm_add_multiply_xdl_fp16.cpp)
diff --git a/example/48_pool3d_fwd/CMakeLists.txt b/example/48_pool3d_fwd/CMakeLists.txt
index f821f2c97b..492cb4d37e 100644
--- a/example/48_pool3d_fwd/CMakeLists.txt
+++ b/example/48_pool3d_fwd/CMakeLists.txt
@@ -1,3 +1 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
-endif()
+add_example_executable(example_pool3d_fwd_fp16 pool3d_fwd_fp16.cpp)
diff --git a/example/49_maxpool2d_bwd/CMakeLists.txt b/example/49_maxpool2d_bwd/CMakeLists.txt
index fe98b7c997..b29cf9ccbc 100644
--- a/example/49_maxpool2d_bwd/CMakeLists.txt
+++ b/example/49_maxpool2d_bwd/CMakeLists.txt
@@ -1,9 +1,3 @@
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
-endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
-endif()
+add_example_executable(example_maxpool2d_bwd_bf16 maxpool2d_bwd_bf16.cpp)
+add_example_executable(example_maxpool2d_bwd_fp16 maxpool2d_bwd_fp16.cpp)
+add_example_executable(example_maxpool2d_bwd_fp32 maxpool2d_bwd_fp32.cpp)
diff --git a/example/50_put_element/CMakeLists.txt b/example/50_put_element/CMakeLists.txt
index eca410008f..1b0020ebcf 100644
--- a/example/50_put_element/CMakeLists.txt
+++ b/example/50_put_element/CMakeLists.txt
@@ -1,3 +1 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
-endif()
+add_example_executable(example_put_element_fp16 put_element_fp16.cpp)
diff --git a/example/CMakeLists.txt b/example/CMakeLists.txt
index 1fdd2f6d14..d030c53453 100644
--- a/example/CMakeLists.txt
+++ b/example/CMakeLists.txt
@@ -7,20 +7,114 @@ add_custom_target(examples)
 
 function(add_example_executable EXAMPLE_NAME FILE_NAME)
     message("adding example ${EXAMPLE_NAME}")
-    add_executable(${EXAMPLE_NAME} ${FILE_NAME})
-    target_link_libraries(${EXAMPLE_NAME} PRIVATE utility)
-    add_test(NAME ${EXAMPLE_NAME} COMMAND $<TARGET_FILE:${EXAMPLE_NAME}> ${ARGN})
-    add_dependencies(examples ${EXAMPLE_NAME})
-    add_dependencies(check ${EXAMPLE_NAME})
-    rocm_install(TARGETS ${EXAMPLE_NAME} COMPONENT examples)
+    set(result 1)
+    if(DEFINED DTYPES)
+    foreach(source IN LISTS FILE_NAME)
+        set(test 0)
+        foreach(type IN LISTS DTYPES)
+            if(type MATCHES "fp16")
+                set(type1 "_f16")
+            elseif(type MATCHES "fp32")
+                set(type1 "_f32")
+            elseif(type MATCHES "fp8")
+                set(type1 "_f8")
+            elseif(type MATCHES "bf16")
+                set(type1 "_b16")
+            elseif(type MATCHES "fp64")
+                set(type1 "_f64")
+            elseif(type MATCHES "int8")
+                set(type1 "_i8")
+            endif()
+            if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
+                #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                set(test 0)
+                break()
+            elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                NOT(source MATCHES type OR source MATCHES type1))
+                    #if filename contains a type which doesn't match any selected type, mark it for removal
+                    set(test 1)
+            endif()
+        endforeach()
+        if(test EQUAL 1)
+            message("removing example source file ${source} ")
+            list(REMOVE_ITEM FILE_NAME "${source}")
+        endif()
+    endforeach()
+    endif()
+    foreach(source IN LISTS FILE_NAME)
+        if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
+            message("removing dl example ${source} ")
+            list(REMOVE_ITEM FILE_NAME "${source}")
+        endif()
+    endforeach()
+    #only continue if there are some source files left on the list
+    if(FILE_NAME)
+        add_executable(${EXAMPLE_NAME} ${FILE_NAME})
+        target_link_libraries(${EXAMPLE_NAME} PRIVATE utility)
+        add_test(NAME ${EXAMPLE_NAME} COMMAND $<TARGET_FILE:${EXAMPLE_NAME}> ${ARGN})
+        add_dependencies(examples ${EXAMPLE_NAME})
+        add_dependencies(check ${EXAMPLE_NAME})
+        rocm_install(TARGETS ${EXAMPLE_NAME} COMPONENT examples)
+        set(result 0)
+    endif()
+    #message("add_example returns ${result}")
+    return(PROPAGATE result)
 endfunction(add_example_executable EXAMPLE_NAME)
 
 function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
     message("adding example ${EXAMPLE_NAME}")
-    add_executable(${EXAMPLE_NAME} ${FILE_NAME})
-    target_link_libraries(${EXAMPLE_NAME} PRIVATE utility)
-    add_dependencies(examples ${EXAMPLE_NAME})
-    rocm_install(TARGETS ${EXAMPLE_NAME} COMPONENT examples)
+    set(result 1)
+    if(DEFINED DTYPES)
+    foreach(source IN LISTS FILE_NAME)
+        set(test 0)
+        foreach(type IN LISTS DTYPES)
+                if(type MATCHES "fp16")
+                    set(type1 "_f16")
+                elseif(type MATCHES "fp32")
+                    set(type1 "_f32")
+                elseif(type MATCHES "fp8")
+                    set(type1 "_f8")
+                elseif(type MATCHES "bf16")
+                    set(type1 "_b16")
+                elseif(type MATCHES "fp64")
+                    set(type1 "_f64")
+                elseif(type MATCHES "int8")
+                    set(type1 "_i8")
+                endif()
+                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
+                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                    set(test 0)
+                    break()
+                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                  NOT(source MATCHES type OR source MATCHES type1))
+                    #if filename contains a type which doesn't match any selected type, mark it for removal
+                    set(test 1)
+                endif()
+        endforeach()
+        if(test EQUAL 1)
+            message("removing example ${source} ")
+            list(REMOVE_ITEM FILE_NAME "${source}")
+        endif()    
+    endforeach()
+    endif()
+    foreach(source IN LISTS FILE_NAME)
+        if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
+            message("removing dl example ${source} ")
+            list(REMOVE_ITEM FILE_NAME "${source}")
+        endif()
+    endforeach()
+    #only continue if there are some source files left on the list
+    if(FILE_NAME)
+        add_executable(${EXAMPLE_NAME} ${FILE_NAME})
+        target_link_libraries(${EXAMPLE_NAME} PRIVATE utility)
+        add_dependencies(examples ${EXAMPLE_NAME})
+        rocm_install(TARGETS ${EXAMPLE_NAME} COMPONENT examples)
+        set(result 0)
+    endif()
+    #message("add_example returns ${result}")
+    return(PROPAGATE result)
 endfunction(add_example_executable_no_testing EXAMPLE_NAME)
 
 # add all example subdir
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
index 0e1f6f04e8..efb76df256 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_backward.hpp
@@ -16,26 +16,26 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// FP16
+#ifdef CK_ENABLE_FP16
 void add_device_batchnorm_backward_rank_4_3_f16_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<F16, F32, F32, F32, F16, F32, F32, PassThrough, 4, 3>>>&);
-
-// FP32
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_batchnorm_backward_rank_4_3_f32_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<F32, F32, F32, F32, F32, F32, F32, PassThrough, 4, 3>>>&);
-
-// BF16
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_batchnorm_backward_rank_4_3_bf16_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<BF16, F32, F32, F32, BF16, F32, F32, PassThrough, 4, 3>>>&);
-
-// FP64
+#endif
+#ifdef CK_ENABLE_FP64
 void add_device_batchnorm_backward_rank_4_3_f64_instances(
     std::vector<std::unique_ptr<
         DeviceBatchNormBwd<F64, F64, F64, F64, F64, F64, F64, PassThrough, 4, 3>>>&);
-
+#endif
 template <typename XDataType,
           typename DxDataType,
           typename DyDataType,
@@ -72,7 +72,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<DxDataType, F32> &&
                      is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
                      is_same_v<ScaleDataType, F16> && is_same_v<DscaleDbiasDataType, F32> &&
@@ -83,37 +83,43 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_backward_rank_4_3_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<DxDataType, F32> &&
-                          is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
-                          is_same_v<ScaleDataType, F32> && is_same_v<DscaleDbiasDataType, F32> &&
-                          is_same_v<MeanVarDataType, F32>)
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<DxDataType, F32> &&
+                     is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
+                     is_same_v<ScaleDataType, F32> && is_same_v<DscaleDbiasDataType, F32> &&
+                     is_same_v<MeanVarDataType, F32>)
         {
             if constexpr(Rank == 4 && NumReduceDim == 3 && is_same_v<DyElementwiseOp, PassThrough>)
             {
                 add_device_batchnorm_backward_rank_4_3_f32_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<DxDataType, F32> &&
-                          is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
-                          is_same_v<ScaleDataType, BF16> && is_same_v<DscaleDbiasDataType, F32> &&
-                          is_same_v<MeanVarDataType, F32>)
+#endif
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<XDataType, BF16> && is_same_v<DxDataType, F32> &&
+                     is_same_v<DyDataType, F32> && is_same_v<AccDataType, F32> &&
+                     is_same_v<ScaleDataType, BF16> && is_same_v<DscaleDbiasDataType, F32> &&
+                     is_same_v<MeanVarDataType, F32>)
         {
             if constexpr(Rank == 4 && NumReduceDim == 3 && is_same_v<DyElementwiseOp, PassThrough>)
             {
                 add_device_batchnorm_backward_rank_4_3_bf16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F64> && is_same_v<DxDataType, F64> &&
-                          is_same_v<DyDataType, F64> && is_same_v<AccDataType, F64> &&
-                          is_same_v<ScaleDataType, F64> && is_same_v<DscaleDbiasDataType, F64> &&
-                          is_same_v<MeanVarDataType, F64>)
+#endif
+#ifdef CK_ENABLE_FP64
+        if constexpr(is_same_v<XDataType, F64> && is_same_v<DxDataType, F64> &&
+                     is_same_v<DyDataType, F64> && is_same_v<AccDataType, F64> &&
+                     is_same_v<ScaleDataType, F64> && is_same_v<DscaleDbiasDataType, F64> &&
+                     is_same_v<MeanVarDataType, F64>)
         {
             if constexpr(Rank == 4 && NumReduceDim == 3 && is_same_v<DyElementwiseOp, PassThrough>)
             {
                 add_device_batchnorm_backward_rank_4_3_f64_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
index 8fd1c7665d..08e35d722c 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_forward.hpp
@@ -16,26 +16,26 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// FP16
+#ifdef CK_ENABLE_FP16
 void add_device_batchnorm_forward_rank_4_3_f16_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<F16, F16, F32, F16, F16, F32, PassThrough, 4, 3>>>&);
-
-// FP32
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_batchnorm_forward_rank_4_3_f32_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<F32, F32, F32, F32, F32, F32, PassThrough, 4, 3>>>&);
-
-// BF16
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_batchnorm_forward_rank_4_3_bf16_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<BF16, BF16, F32, BF16, BF16, F32, PassThrough, 4, 3>>>&);
-
-// FP64
+#endif
+#ifdef CK_ENABLE_FP64
 void add_device_batchnorm_forward_rank_4_3_f64_instances(
     std::vector<
         std::unique_ptr<DeviceBatchNormFwd<F64, F64, F64, F64, F64, F64, PassThrough, 4, 3>>>&);
-
+#endif
 template <typename XDataType,
           typename YDataType,
           typename AccDataType,
@@ -69,7 +69,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<YDataType, F16> &&
                      is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, F16> &&
                      is_same_v<BiasDataType, F16> && is_same_v<MeanVarDataType, F32>)
@@ -79,34 +79,40 @@ struct DeviceOperationInstanceFactory<
                 add_device_batchnorm_forward_rank_4_3_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
-                          is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, F32> &&
-                          is_same_v<BiasDataType, F32> && is_same_v<MeanVarDataType, F32>)
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
+                     is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, F32> &&
+                     is_same_v<BiasDataType, F32> && is_same_v<MeanVarDataType, F32>)
         {
             if constexpr(Rank == 4 && NumReduceDim == 3 && is_same_v<YElementwiseOp, PassThrough>)
             {
                 add_device_batchnorm_forward_rank_4_3_f32_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
-                          is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, BF16> &&
-                          is_same_v<BiasDataType, BF16> && is_same_v<MeanVarDataType, F32>)
+#endif
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
+                     is_same_v<AccDataType, F32> && is_same_v<ScaleDataType, BF16> &&
+                     is_same_v<BiasDataType, BF16> && is_same_v<MeanVarDataType, F32>)
         {
             if constexpr(Rank == 4 && NumReduceDim == 3 && is_same_v<YElementwiseOp, PassThrough>)
             {
                 add_device_batchnorm_forward_rank_4_3_bf16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
-                          is_same_v<AccDataType, F64> && is_same_v<ScaleDataType, F64> &&
-                          is_same_v<BiasDataType, F64> && is_same_v<MeanVarDataType, F64>)
+#endif
+#ifdef CK_ENABLE_FP64
+        if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
+                     is_same_v<AccDataType, F64> && is_same_v<ScaleDataType, F64> &&
+                     is_same_v<BiasDataType, F64> && is_same_v<MeanVarDataType, F64>)
         {
             if constexpr(Rank == 4 && NumReduceDim == 3 && is_same_v<YElementwiseOp, PassThrough>)
             {
                 add_device_batchnorm_forward_rank_4_3_f64_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
index f6f4df7e2e..97c216b4ad 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/batchnorm_infer.hpp
@@ -16,38 +16,38 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// FP16
+#ifdef CK_ENABLE_FP16
 void add_device_batchnorm_infer_rank_4_f16_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<F16, F32, F32, F16, F16>,
         ck::Tuple<F16>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
-// FP32
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_batchnorm_infer_rank_4_f32_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<F32, F32, F32, F32, F32>,
         ck::Tuple<F32>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
-// BF16
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_batchnorm_infer_rank_4_bf16_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<BF16, F32, F32, BF16, BF16>,
         ck::Tuple<BF16>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
-// FP64
+#endif
+#ifdef CK_ENABLE_FP64
 void add_device_batchnorm_infer_rank_4_f64_instances(
     std::vector<std::unique_ptr<ck::tensor_operation::device::DeviceElementwise<
         ck::Tuple<F64, F64, F64, F64, F64>,
         ck::Tuple<F64>,
         ck::tensor_operation::element_wise::NormalizeInInfer,
         4>>>&);
-
+#endif
 template <typename XDataType,
           typename YDataType,
           typename ScaleDataType,
@@ -69,7 +69,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<YDataType, F16> &&
                      is_same_v<ScaleDataType, F16> && is_same_v<BiasDataType, F16> &&
                      is_same_v<MeanVarDataType, F32>)
@@ -79,34 +79,40 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceElemen
                 add_device_batchnorm_infer_rank_4_f16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
-                          is_same_v<ScaleDataType, F32> && is_same_v<BiasDataType, F32> &&
-                          is_same_v<MeanVarDataType, F32>)
+#endif
+#ifdef CK_ENABLE_FP32
+        if constexpr(is_same_v<XDataType, F32> && is_same_v<YDataType, F32> &&
+                     is_same_v<ScaleDataType, F32> && is_same_v<BiasDataType, F32> &&
+                     is_same_v<MeanVarDataType, F32>)
         {
             if constexpr(Rank == 4)
             {
                 add_device_batchnorm_infer_rank_4_f32_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
-                          is_same_v<ScaleDataType, BF16> && is_same_v<BiasDataType, BF16> &&
-                          is_same_v<MeanVarDataType, F32>)
+#endif
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<XDataType, BF16> && is_same_v<YDataType, BF16> &&
+                     is_same_v<ScaleDataType, BF16> && is_same_v<BiasDataType, BF16> &&
+                     is_same_v<MeanVarDataType, F32>)
         {
             if constexpr(Rank == 4)
             {
                 add_device_batchnorm_infer_rank_4_bf16_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
-                          is_same_v<ScaleDataType, F64> && is_same_v<BiasDataType, F64> &&
-                          is_same_v<MeanVarDataType, F64>)
+#endif
+#ifdef CK_ENABLE_FP64
+        if constexpr(is_same_v<XDataType, F64> && is_same_v<YDataType, F64> &&
+                     is_same_v<ScaleDataType, F64> && is_same_v<BiasDataType, F64> &&
+                     is_same_v<MeanVarDataType, F64>)
         {
             if constexpr(Rank == 4)
             {
                 add_device_batchnorm_infer_rank_4_f64_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index d239cd7706..4cba0875d9 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -36,7 +36,8 @@ void add_device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
@@ -56,8 +57,8 @@ void add_device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F32, F32, F32, PassThrough, PassThrough, PassThrough>>>&
         instances);
-
-#if defined CK_ENABLE_FP8
+#endif
+#if(defined(CK_ENABLE_FP16) || defined(CK_ENABLE_FP8))
 void add_device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Col, Row, Row, F8, F16, F16, PassThrough, PassThrough, PassThrough>>>&
@@ -129,7 +130,7 @@ struct DeviceOperationInstanceFactory<
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
                      is_same_v<CDataType, float>)
         {
@@ -154,6 +155,8 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances(op_ptrs);
             }
         }
+#endif
+#ifdef CK_ENABLE_FP16
         else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -178,7 +181,8 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
-#if defined CK_ENABLE_FP8
+#endif
+#if(defined(CK_ENABLE_FP16) || defined(CK_ENABLE_FP8))
         else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
                           is_same_v<CDataType, half_t>)
         {
@@ -228,7 +232,6 @@ struct DeviceOperationInstanceFactory<
             }
         }
 #endif
-
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index 58c9064535..2b49e17187 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -16,6 +16,7 @@ namespace device {
 namespace instance {
 
 // conv2d backward data
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -29,7 +30,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -43,7 +45,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   GNHWK,
@@ -57,7 +60,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
@@ -71,7 +75,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
@@ -85,7 +90,8 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
                                                                   NHWGK,
@@ -99,8 +105,9 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
 // conv3d backward data
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   GNDHWK,
@@ -114,7 +121,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   GNDHWK,
@@ -128,7 +136,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   GNDHWK,
@@ -142,7 +151,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   NDHWGK,
@@ -156,7 +166,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   NDHWGK,
@@ -170,7 +181,8 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
                                                                   NDHWGK,
@@ -184,7 +196,7 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
-
+#endif
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -230,42 +242,54 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
                          is_same_v<OutLayout, GNHWK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
                               is_same_v<OutLayout, NHWGK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
         else if constexpr(NumDimSpatial == 3)
@@ -274,46 +298,58 @@ struct DeviceOperationInstanceFactory<
             if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
                          is_same_v<OutLayout, GNDHWK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
                               is_same_v<OutLayout, NDHWGK>)
             {
+#ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, BF16> && is_same_v<WeiDataType, BF16> &&
                                   is_same_v<OutDataType, BF16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index c7a694b4cc..c0a5b4bfe5 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -19,6 +19,7 @@ namespace instance {
 
 // xdl
 // conv1d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -30,7 +31,8 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_insta
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -42,7 +44,8 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -54,8 +57,9 @@ void add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
 // conv2d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -67,7 +71,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -79,7 +84,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -91,7 +97,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -103,7 +110,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_in
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -115,7 +123,8 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -127,8 +136,9 @@ void add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
 // conv3d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -140,7 +150,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -152,7 +163,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -164,7 +176,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -176,7 +189,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -188,7 +202,8 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -200,10 +215,12 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+#endif
 
 #ifdef DL_KERNELS
 // dl
 // conv1d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -215,7 +232,8 @@ void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instan
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -227,7 +245,8 @@ void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            GNWC,
@@ -239,7 +258,8 @@ void add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            NWGC,
@@ -251,7 +271,8 @@ void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instan
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            NWGC,
@@ -263,7 +284,8 @@ void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<1,
                                                            NWGC,
@@ -275,8 +297,9 @@ void add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
 // conv2d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -288,7 +311,8 @@ void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_ins
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -300,7 +324,8 @@ void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            GNHWC,
@@ -312,7 +337,8 @@ void add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -324,7 +350,8 @@ void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_ins
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -336,7 +363,8 @@ void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<2,
                                                            NHWGC,
@@ -348,8 +376,9 @@ void add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
 // conv3d backward weight
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -361,7 +390,8 @@ void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -373,7 +403,8 @@ void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            GNDHWC,
@@ -385,7 +416,8 @@ void add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -397,7 +429,8 @@ void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -409,7 +442,8 @@ void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
                                                            NDHWGC,
@@ -422,6 +456,7 @@ void add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 #endif
+#endif
 
 template <ck::index_t NumDimSpatial,
           typename InLayout,
@@ -462,6 +497,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InLayout, GNWC> && is_same_v<WeiLayout, GKXC> &&
                          is_same_v<OutLayout, GNWK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
@@ -470,6 +506,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
@@ -478,6 +516,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
@@ -489,21 +529,27 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NWGC> && is_same_v<WeiLayout, GKXC> &&
                               is_same_v<OutLayout, NWGK>)
             {
 #ifdef DL_KERNELS
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
@@ -511,6 +557,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
 #endif
             }
         }
@@ -519,6 +566,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
                          is_same_v<OutLayout, GNHWK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
@@ -529,6 +577,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
@@ -539,6 +589,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
@@ -550,10 +602,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
                               is_same_v<OutLayout, NHWGK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
@@ -564,6 +618,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
@@ -574,6 +630,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
@@ -585,6 +643,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
         else if constexpr(NumDimSpatial == 3)
@@ -592,6 +651,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
                          is_same_v<OutLayout, GNDHWK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
@@ -602,6 +662,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
@@ -612,6 +674,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
@@ -623,10 +687,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
                               is_same_v<OutLayout, NDHWGK>)
             {
+#ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                              is_same_v<OutDataType, float>)
                 {
@@ -637,6 +703,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t>)
                 {
@@ -647,6 +715,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_BF16
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<WeiDataType, float> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
@@ -658,6 +728,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
             }
         }
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index 576b9d8983..7fc88c33a6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -16,7 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_BF16
 // grouped conv1d forward, GNWC/GKXC/GNWK
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
@@ -31,7 +31,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -45,7 +46,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -59,7 +61,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_INT8
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
                                                               GNWC,
@@ -73,7 +76,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 // grouped conv2d forward, GNHWC/GKYXC/GNHWK
 void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -88,7 +92,8 @@ void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -102,7 +107,8 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -116,7 +122,9 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef DL_KERNELS
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -130,7 +138,8 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -144,7 +153,9 @@ void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -158,6 +169,8 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+
+#ifdef DL_KERNELS
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -171,7 +184,9 @@ void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#endif
+#ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
@@ -185,6 +200,8 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -199,7 +216,9 @@ void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+#endif
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
+#ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -213,7 +232,8 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -227,7 +247,8 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -241,7 +262,8 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 // grouped conv3d forward, GNDHWC/GKZYXC/GNDHWK
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -256,7 +278,8 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -270,7 +293,8 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -284,7 +308,8 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_INT8
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               GNDHWC,
@@ -298,7 +323,8 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_BF16
 // grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -313,7 +339,8 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -327,7 +354,8 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -341,7 +369,8 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_INT8
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -355,6 +384,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
 template <ck::index_t NumDimSpatial,
           typename InLayout,
@@ -397,127 +427,168 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
         if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC> &&
                      is_same_v<WeiLayout, GKXC> && is_same_v<OutLayout, GNWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+#endif
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
+#endif
+#ifdef CK_ENABLE_INT8
             else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
                               is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
                           is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+#ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+#endif
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BDF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
                           is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(op_ptrs);
             }
+#endif
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
                           is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+#endif
+#ifdef CK_ENABLE_FP16
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+#endif
+#ifdef CK_ENABLE_BF16
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(op_ptrs);
             }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+#endif
+#ifdef CK_ENABLE_INT8
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(op_ptrs);
             }
+#endif
         }
 
         return op_ptrs;
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
index 9930b1a6fa..4c22b047f3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance.hpp
@@ -2,13 +2,20 @@
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #pragma once
-
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_min.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_max.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f16_f16_amax.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f16_f32_f16_norm2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f32_f32_norm2.hpp"
@@ -18,39 +25,10 @@
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f32_f64_f32_norm2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f16_f32_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f32_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f32_f64_f32_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_min.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_max.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f16_f16_amax.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_add.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_avg.hpp"
-#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f16_f32_f16_norm2.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f32_f32_norm2.hpp"
@@ -60,17 +38,38 @@
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f32_f64_f32_norm2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_avg.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_norm2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_min.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_max.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_f64_f64_f64_amax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_f64_f64_f64_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_norm2.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_min.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_max.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_f64_f64_f64_amax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_min.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_max.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i8_i8_amax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_i8_i32_i8_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_min.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_max.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i8_i8_amax.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_i8_i32_i8_avg.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_avg.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_norm2.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_min.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_max.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_blockwise_b16_f32_b16_amax.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_add.hpp"
+#include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_multiblock_atomic_add_b16_f32_f32_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_add.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_avg.hpp"
 #include "ck/library/tensor_operation_instance/gpu/reduce/device_reduce_instance_threadwise_b16_f32_b16_norm2.hpp"
diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
index 1d54a141b7..a0478c9f09 100644
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
@@ -1,9 +1,57 @@
 function(add_instance_library INSTANCE_NAME)
     message("adding instance ${INSTANCE_NAME}")
-    add_library(${INSTANCE_NAME} OBJECT ${ARGN})
-    target_compile_features(${INSTANCE_NAME} PUBLIC)
-    set_target_properties(${INSTANCE_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON)
-    clang_tidy_check(${INSTANCE_NAME})
+    set(result 1)
+    if(DEFINED DTYPES)
+      foreach(source IN LISTS ARGN)
+        set(test 0)
+        foreach(type IN LISTS DTYPES)
+                if(type MATCHES "fp16")
+                        set(type1 "_f16")
+                elseif(type MATCHES "fp32")
+                        set(type1 "_f32")
+                elseif(type MATCHES "fp8")
+                        set(type1 "_f8")
+                elseif(type MATCHES "bf16")
+                        set(type1 "_b16")
+                elseif(type MATCHES "fp64")
+                        set(type1 "_f64")
+                elseif(type MATCHES "int8")
+                        set(type1 "_i8")
+                endif()
+                #make an exception for reduction kernels
+                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}" OR "${source}" MATCHES "device_reduce_instance")
+                        #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                        set(test 0)
+                        break()
+                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                  NOT(source MATCHES type OR source MATCHES type1))
+                        #if filename contains a type which doesn't match any selected type, mark it for removal
+                        set(test 1)
+                endif()
+        endforeach()
+        if(test EQUAL 1)
+                message("removing instance ${source} ")
+                list(REMOVE_ITEM ARGN "${source}")
+        endif()
+      endforeach()
+    endif()
+    foreach(source IN LISTS ARGN)
+        if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
+                message("removing dl instance ${source} ")
+                list(REMOVE_ITEM ARGN "${source}")
+        endif()
+    endforeach()
+    #only continue if there are some source files left on the list
+    if(ARGN)
+        add_library(${INSTANCE_NAME} OBJECT ${ARGN})
+        target_compile_features(${INSTANCE_NAME} PUBLIC)
+        set_target_properties(${INSTANCE_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON)
+        clang_tidy_check(${INSTANCE_NAME})
+        set(result 0)
+    endif()
+    #message("add_instance_library returns ${result}")
+    return(PROPAGATE result)
 endfunction(add_instance_library INSTANCE_NAME)
 
 
@@ -15,33 +63,49 @@ IF(IS_DIRECTORY "${subdir_path}")
     set(cmake_instance)
     file(READ "${subdir_path}/CMakeLists.txt" cmake_instance)
     set(add_inst 0)
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp8\" " AND DTYPES MATCHES "fp8")
-            #message("fp8 instance found!")
+    if(("${cmake_instance}" MATCHES "_fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
+            message("fp8 instance found!")
             set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp16\"" AND DTYPES MATCHES "fp16")
-            #message("fp16 instance found!")
+    if(("${cmake_instance}" MATCHES "_fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
+            message("fp16 instance found!")
             set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp32\"" AND DTYPES MATCHES "fp32")
-            #message("fp32 instance found!")
+    if(("${cmake_instance}" MATCHES "_fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
+            message("fp32 instance found!")
             set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"fp64\"" AND DTYPES MATCHES "fp64")
-            #message("fp64 instance found!")
+    if(("${cmake_instance}" MATCHES "_fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
+            message("fp64 instance found!")
             set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"bf16\"" AND DTYPES MATCHES "bf16")
-            #message("bf16 instance found!")
+    if("${cmake_instance}" MATCHES "_bf16" AND DTYPES MATCHES "bf16")
+            message("bf16 instance found!")
             set(add_inst 1)
     endif()
-    if("${cmake_instance}" MATCHES "DTYPES MATCHES \"int8\"" AND DTYPES MATCHES "int8")
-            #message("int8 instance found!")
+    if(("${cmake_instance}" MATCHES "_int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
+            message("int8 instance found!")
             set(add_inst 1)
     endif()
-    if(NOT "${cmake_instance}" MATCHES "DTYPES" OR NOT DEFINED DTYPES)
-            #message("instance should be built for all types!")
-            set(add_inst 1)
+    if(NOT "${cmake_instance}" MATCHES "_fp8" OR
+                   NOT "${cmake_instance}" MATCHES "_f8" OR
+                   NOT "${cmake_instance}" MATCHES "_fp16" OR
+                   NOT "${cmake_instance}" MATCHES "_f16" OR
+                   NOT "${cmake_instance}" MATCHES "_fp32" OR
+                   NOT "${cmake_instance}" MATCHES "_f32" OR
+                   NOT "${cmake_instance}" MATCHES "_fp64" OR
+                   NOT "${cmake_instance}" MATCHES "_f64" OR
+                   NOT "${cmake_instance}" MATCHES "_bf16" OR
+                   NOT "${cmake_instance}" MATCHES "_int8" OR
+                   NOT "${cmake_instance}" MATCHES "_i8" OR
+                   NOT "${cmake_instance}" MATCHES "_int4" OR
+                   NOT DEFINED DTYPES)
+             message("instance should be built for all types!")
+             set(add_inst 1)
+     endif()
+    if("${cmake_instance}" MATCHES "quantization" AND DEFINED DTYPES AND NOT DTYPES MATCHES "int8")
+           message("quantization instances will not be built!")
+           set(add_inst 0)
     endif()
     if("${cmake_instance}" MATCHES "ONLY DL_KERNELS" AND NOT DEFINED DL_KERNELS)
 	    message("Found only dl instances, but DL_KERNELS is not set. Skipping.")
diff --git a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt
index ec079e3ba3..084714b707 100644
--- a/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/avg_pool3d_bwd/CMakeLists.txt
@@ -1,11 +1,5 @@
 set(DEVICE_AVGPOOL_BWD_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_f16_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_f32_instance.cpp)
-endif()
+list(APPEND DEVICE_AVGPOOL_BWD_INSTANCES device_avg_pool3d_bwd_ndhwc_f16_instance.cpp
+                                         device_avg_pool3d_bwd_ndhwc_bf16_instance.cpp
+                                         device_avg_pool3d_bwd_ndhwc_f32_instance.cpp)
 add_instance_library(device_avg_pool3d_bwd_instance ${DEVICE_AVGPOOL_BWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt
index 5b342595d8..69b6ddc754 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/CMakeLists.txt
@@ -1,26 +1,18 @@
 set(BATCHED_GEMM_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
                                       device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
                                       device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
-                                      device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gkn_gmn_instance.cpp
                                       device_batched_gemm_xdl_bf16_bf16_bf16_gmk_gnk_gmn_instance.cpp
                                       device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gkn_gmn_instance.cpp
-                                      device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_bf16_bf16_bf16_gkm_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f32_f32_f32_gmk_gkn_gmn_instance.cpp
                                       device_batched_gemm_xdl_f32_f32_f32_gmk_gnk_gmn_instance.cpp
                                       device_batched_gemm_xdl_f32_f32_f32_gkm_gkn_gmn_instance.cpp
-                                      device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp)
-endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   list(APPEND BATCHED_GEMM_INSTANCES device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
+                                      device_batched_gemm_xdl_f32_f32_f32_gkm_gnk_gmn_instance.cpp
+                                      device_batched_gemm_xdl_int8_int8_int8_gmk_gkn_gmn_instance.cpp
                                       device_batched_gemm_xdl_int8_int8_int8_gmk_gnk_gmn_instance.cpp
                                       device_batched_gemm_xdl_int8_int8_int8_gkm_gkn_gmn_instance.cpp
                                       device_batched_gemm_xdl_int8_int8_int8_gkm_gnk_gmn_instance.cpp)
-endif()
 add_instance_library(device_batched_gemm_instance ${BATCHED_GEMM_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt
index 6710035ec6..d0e9b265af 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_add_relu_gemm_add/CMakeLists.txt
@@ -1,6 +1,4 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_add_relu_gemm_add_instance
     device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
     device_batched_gemm_add_relu_gemm_add_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
 )
-endif()
\ No newline at end of file
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
index b0f37e68fd..cd9c95c066 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_bias_permute/CMakeLists.txt
@@ -1,5 +1,4 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_bias_permute_instance
     device_batched_gemm_bias_permute_m2_n3_k1_xdl_c_shuffle_f16_f16_f16_f16_instance.cpp
 )
-endif()
+
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
index cdb1a53901..865a31e79a 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_gemm/CMakeLists.txt
@@ -1,6 +1,4 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_gemm_instance
     device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
     device_batched_gemm_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gon_gmo_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
index 444c93b118..b874bc50ee 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_multi_d/CMakeLists.txt
@@ -1,25 +1,21 @@
 # ONLY DL_KERNELS
-if(DL_KERNELS)
-  set(BATCHED_GEMM_MULTID_INSTANCES)
-  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp)
-  endif()
-  if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp)
-    list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp)
-  endif()
-  add_instance_library(device_batched_gemm_multi_d_instance ${BATCHED_GEMM_MULTID_INSTANCES})
-endif()
+set(BATCHED_GEMM_MULTID_INSTANCES)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gkn_gmn_irregular_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gmk_gnk_gmn_irregular_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gkn_gmn_irregular_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_f16_f16_f16_gkm_gnk_gmn_irregular_instance.cpp)
+
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gkn_gmn_irregular_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gmk_gnk_gmn_irregular_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gkn_gmn_irregular_instance.cpp)
+list(APPEND BATCHED_GEMM_MULTID_INSTANCES device_batched_gemm_multi_d_dl_i8_i8_i8_gkm_gnk_gmn_irregular_instance.cpp)
+
+add_instance_library(device_batched_gemm_multi_d_instance ${BATCHED_GEMM_MULTID_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
index 728e35fc3b..28226fabac 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_reduce/CMakeLists.txt
@@ -1,8 +1,6 @@
-if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_reduce_instance
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gkn_gmn_instance.cpp
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gmk_gnk_gmn_instance.cpp
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gkn_gmn_instance.cpp
     device_batched_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_gkm_gnk_gmn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
index 5ac55655d4..6244477e16 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -1,5 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_batched_gemm_softmax_gemm_instance
     device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index d5110e4505..3fd4e03449 100644
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -1,11 +1,7 @@
 set(DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp)
-    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp)
-    list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp)
-endif()
+list(APPEND DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES
+    device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+    device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+    device_batched_gemm_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+    device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp)
 add_instance_library(device_batched_gemm_softmax_gemm_permute_instance ${DEVICE_BATCHED_GEMM_SOFTMAX_GEMM_PERMUTE_INSTANCES})
-
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
index 1db6985f61..ee9c419e1a 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
@@ -1,17 +1,14 @@
 set(DEVICE_CONTRACTION_BILINEAR_INSTANCES)
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    #float
-    list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
-                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
-                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
-                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
-    #double
-    list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
-                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
-                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
-                                                      device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
-endif()
-add_instance_library(device_contraction_bilinear_instance ${DEVICE_CONTRACTION_BILINEAR_INSTANCES})
+#float
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
 
+#double
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
+
+add_instance_library(device_contraction_bilinear_instance ${DEVICE_CONTRACTION_BILINEAR_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
index aed8bef2a8..673fae3373 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
@@ -1,17 +1,15 @@
 set(DEVICE_CONTRACTION_SCALE_INSTANCES)
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    #float
-    list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
-                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
-                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
-                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
-    #double
-    list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
-                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
-                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
-                                                   device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp)
-endif()
+#float
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp)
+
+#double
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp)
+
 add_instance_library(device_contraction_scale_instance ${DEVICE_CONTRACTION_SCALE_INSTANCES})
 
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
index 54ef9cc7a6..49dfc01fd9 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/conv2d_bwd_data/CMakeLists.txt
@@ -1,23 +1,10 @@
 set(CONV2D_BWD_DATA_INSTANCES)
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp)
-   if(DL_KERNELS)
-      list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp)
-   endif()
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp)
-   if(DL_KERNELS)
-      list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp)
-   endif()
-endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp)
-   if(DL_KERNELS)
-      list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp)
-   endif()
-endif()
+list(APPEND CONV2D_BWD_DATA_INSTANCES device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+                                      device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instance.cpp
+                                      device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+                                      device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+                                      device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f16_instance.cpp
+                                      device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_int8_instance.cpp
+                                      device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_int8_instance.cpp)
+
 add_instance_library(device_conv2d_bwd_data_instance ${CONV2D_BWD_DATA_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
index 96ecc95658..ba0ca32517 100644
--- a/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/conv2d_fwd/CMakeLists.txt
@@ -1,16 +1,8 @@
 set(DEVICE_CONV2D_FWD_INSTANCES)
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp)
-   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp)
-endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp)
-endif()
-   
+list(APPEND DEVICE_CONV2D_FWD_INSTANCES device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instance.cpp
+                                        device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instance.cpp
+                                        device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instance.cpp
+                                        device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instance.cpp
+                                        device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instance.cpp)
+
 add_instance_library(device_conv2d_fwd_instance ${DEVICE_CONV2D_FWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
index ea53b82be5..0c7cc2cd31 100644
--- a/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/elementwise_normalization/CMakeLists.txt
@@ -1,5 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_elementwise_normalization_instance
     device_elementwise_normalization_f16_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 6c76ee843a..22cf1a462d 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -1,113 +1,99 @@
 set(GEMM_INSTANCES)
-if(DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
-    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp)
-    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp)
-    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp)
-    list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp)
-   if(DL_KERNELS)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp)
-   endif()
-endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-   if(DL_KERNELS)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp)
-   endif()
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp)
-endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-   if(DL_KERNELS)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp)
-     list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp)
-   endif()
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp)
-   list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp)
-endif()
+list(APPEND GEMM_INSTANCES device_gemm_xdl_f64_f64_f64_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_f64_f64_f64_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_f64_f64_f64_km_kn_mn_instance.cpp
+                           device_gemm_xdl_f64_f64_f64_km_nk_mn_instance.cpp)
+
+list(APPEND GEMM_INSTANCES device_gemm_xdl_f32_f32_f32_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_f32_f32_f32_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_f32_f32_f32_km_kn_mn_instance.cpp
+                           device_gemm_xdl_f32_f32_f32_km_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f32_f32_f32_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f32_f32_f32_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f32_f32_f32_km_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f32_f32_f32_km_nk_mn_instance.cpp
+                           device_gemm_dl_f32_f32_f32_mk_kn_mn_instance.cpp
+                           device_gemm_dl_f32_f32_f32_mk_nk_mn_instance.cpp
+                           device_gemm_dl_f32_f32_f32_km_kn_mn_instance.cpp
+                           device_gemm_dl_f32_f32_f32_km_nk_mn_instance.cpp)
+
+list(APPEND GEMM_INSTANCES device_gemm_dl_f16_f16_f16_mk_kn_mn_instance.cpp
+                           device_gemm_dl_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+                           device_gemm_dl_f16_f16_f16_mk_nk_mn_instance.cpp
+                           device_gemm_dl_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+                           device_gemm_dl_f16_f16_f16_km_kn_mn_instance.cpp
+                           device_gemm_dl_f16_f16_f16_km_kn_mn_irregular_instance.cpp
+                           device_gemm_dl_f16_f16_f16_km_nk_mn_instance.cpp
+                           device_gemm_dl_f16_f16_f16_km_nk_mn_irregular_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_km_kn_mn_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_km_nk_mn_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_mk_kn_mn_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_mk_nk_mn_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_km_kn_mn_irregular_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_km_nk_mn_irregular_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+                           device_gemm_dpp_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+                           device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_2_stage_f16_f16_f16_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_add_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_add_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/km_nk_mn_irregular_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_add_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_kn_mn_irregular_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_add_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_interwave_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v1_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_default_pipeline_v2_instance.cpp
+                           device_gemm_xdl_f16_f16_f16/mk_nk_mn_irregular_interwave_pipeline_v1_instance.cpp)
+
+list(APPEND GEMM_INSTANCES device_gemm_dl_i8_i8_i8_mk_kn_mn_instance.cpp
+                           device_gemm_dl_i8_i8_i8_mk_kn_mn_irregular_instance.cpp
+                           device_gemm_dl_i8_i8_i8_mk_nk_mn_instance.cpp
+                           device_gemm_dl_i8_i8_i8_mk_nk_mn_irregular_instance.cpp
+                           device_gemm_dl_i8_i8_i8_km_kn_mn_instance.cpp
+                           device_gemm_dl_i8_i8_i8_km_kn_mn_irregular_instance.cpp
+                           device_gemm_dl_i8_i8_i8_km_nk_mn_instance.cpp
+                           device_gemm_dl_i8_i8_i8_km_nk_mn_irregular_instance.cpp
+                           device_gemm_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp)
+
+list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp)
 
 add_instance_library(device_gemm_instance ${GEMM_INSTANCES})
 
+set(ENABLE_PIPELINE_V2_OPT OFF)
 
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  set(ENABLE_PIPELINE_V2_OPT OFF)
-
-  if (ENABLE_PIPELINE_V2_OPT)
+if (ENABLE_PIPELINE_V2_OPT)
     set(MAX_ILP_OPTS
        -mllvm
        -amdgpu-enable-max-ilp-scheduling-strategy
@@ -137,5 +123,5 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
        COMPILE_OPTIONS "${MAX_ILP_OPTS}"
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
-  endif(ENABLE_PIPELINE_V2_OPT)
-endif(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+endif(ENABLE_PIPELINE_V2_OPT)
+
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
index 9028829fe9..bbf81a5fa2 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_add_fastgelu/CMakeLists.txt
@@ -1,8 +1,6 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_add_fastgelu_instance
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_kn_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_km_nk_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
    device_gemm_add_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
index 1085966807..0beb10e379 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_fastgelu/CMakeLists.txt
@@ -1,8 +1,6 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_fastgelu_instance
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_kn_mn_mn_instance.cpp
    device_gemm_add_fastgelu_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
index 6079f90130..97693a2566 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_add_relu_add_layernorm/CMakeLists.txt
@@ -1,8 +1,6 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_add_relu_add_layernorm_instance
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_kn_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_km_nk_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_kn_mn_mn_mn_instance.cpp
    device_gemm_add_relu_add_xdl_c_shuffle_layernorm_f16_mk_nk_mn_mn_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
index aef8fe86dd..426edeed74 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_bilinear/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_bilinear_instance
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instance.cpp
    device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_nk_mn_mn_instance.cpp
@@ -9,4 +8,3 @@ add_instance_library(device_gemm_bilinear_instance
    device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instance.cpp
    device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_nk_mn_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
index 772373dcb0..17d27ab150 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_fastgelu/CMakeLists.txt
@@ -1,8 +1,6 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_fastgelu_instance
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_kn_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_km_nk_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
    device_gemm_fastgelu_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
index 36bd6a4aa3..6cbd7528e7 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_multiply_add/CMakeLists.txt
@@ -1,13 +1,6 @@
 set(GEMM_MULTIPLY_ADD_INSTANCES)
-
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp)
-  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp)
-endif()
-
-if((DTYPES MATCHES "fp16" AND DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
-  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp)
-  list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp)
-endif()
-
+list(APPEND GEMM_MULTIPLY_ADD_INSTANCES device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_kn_mn_mn_mn_instance.cpp
+                                        device_gemm_multiply_add_xdl_c_shuffle_f16_f16_f16_f16_f16_mk_nk_mn_mn_mn_instance.cpp
+                                        device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_kn_mn_mn_mn_instance.cpp
+                                        device_gemm_multiply_add_xdl_c_shuffle_f16_f8_f32_f32_f16_mk_nk_mn_mn_mn_instance.cpp)
 add_instance_library(device_gemm_multiply_add_instance ${GEMM_MULTIPLY_ADD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
index 043b28a1be..837e040eac 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
@@ -1,28 +1,20 @@
 set(GEMM_SPLITK_INSTANCES)
 
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp)
-endif()
-
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp)
-endif()
-
-if((DTYPES MATCHES "fp16" AND DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp)
-  list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp)
-endif()
+list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f32_f32_f32_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp)
 
 add_instance_library(device_gemm_splitk_instance ${GEMM_SPLITK_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
index d9baf3f062..8dd0112a6b 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_streamk/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_gemm_streamk_instance
    # device_gemm_xdl_streamk_f32_f32_f32_mk_kn_mn_instance.cpp
    # device_gemm_xdl_streamk_f32_f32_f32_mk_nk_mn_instance.cpp
@@ -9,4 +8,3 @@ add_instance_library(device_gemm_streamk_instance
    # device_gemm_xdl_streamk_f16_f16_f16_km_kn_mn_instance.cpp
    # device_gemm_xdl_streamk_f16_f16_f16_km_nk_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
index f1553e3d5b..b973b70aac 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
    device_grouped_gemm_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -9,4 +8,3 @@ add_instance_library(device_grouped_gemm_instance
    device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
    device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
index a45bf36399..648f2146cb 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fastgelu/CMakeLists.txt
@@ -1,8 +1,6 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_instance_library(device_grouped_gemm_fastgelu_instance
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_mk_nk_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_kn_mn_instance.cpp
    device_grouped_gemm_fastgelu_xdl_f16_f16_f16_km_nk_mn_instance.cpp
 )
-endif()
diff --git a/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
index 820a63480e..d12864398e 100644
--- a/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/max_pool_bwd/CMakeLists.txt
@@ -1,11 +1,5 @@
 set(DEVICE_MAXPOOL_BWD_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f16_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_bf16_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f32_instance.cpp)
-endif()
+list(APPEND DEVICE_MAXPOOL_BWD_INSTANCES device_max_pool_bwd_f16_instance.cpp
+                                         device_max_pool_bwd_bf16_instance.cpp
+                                         device_max_pool_bwd_f32_instance.cpp)
 add_instance_library(device_max_pool_bwd_instance ${DEVICE_MAXPOOL_BWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
index 4892a9c183..19818a523b 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt
@@ -1,15 +1,14 @@
 set(DEVICE_NORMALIZATION_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f16_instance.cpp
+
+list(APPEND DEVICE_NORMALIZATION_INSTANCES 
+    device_layernorm2d_f16_instance.cpp
     device_layernorm4d_f16_instance.cpp
     device_groupnorm_f16_instance.cpp
     device_groupnorm_swish_f16_instance.cpp
-    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_NORMALIZATION_INSTANCES device_layernorm2d_f32_instance.cpp
+    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
+    device_layernorm2d_f32_instance.cpp
     device_layernorm4d_f32_instance.cpp
     device_groupnorm_f32_instance.cpp
     device_groupnorm_swish_f32_instance.cpp)
-endif()
+
 add_instance_library(device_normalization_instance ${DEVICE_NORMALIZATION_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
index 63bbe75465..533ce89dfd 100644
--- a/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/pool3d_fwd/CMakeLists.txt
@@ -1,14 +1,8 @@
 set(DEVICE_POOL3D_FWD_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
-                                          device_max_pool3d_fwd_ndhwc_f16_instance.cpp)
-endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp
-                                          device_max_pool3d_fwd_ndhwc_bf16_instance.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
-                                          device_max_pool3d_fwd_ndhwc_f32_instance.cpp)
-endif()
+list(APPEND DEVICE_POOL3D_FWD_INSTANCES device_avg_pool3d_fwd_ndhwc_f16_instance.cpp
+                                        device_max_pool3d_fwd_ndhwc_f16_instance.cpp
+                                        device_avg_pool3d_fwd_ndhwc_f32_instance.cpp
+                                        device_max_pool3d_fwd_ndhwc_f32_instance.cpp
+                                        device_avg_pool3d_fwd_ndhwc_bf16_instance.cpp
+                                        device_max_pool3d_fwd_ndhwc_bf16_instance.cpp)
 add_instance_library(device_pool3d_fwd_instance ${DEVICE_POOL3D_FWD_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
index 00eb6ff1c1..c22a6e9e96 100644
--- a/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/quantization/CMakeLists.txt
@@ -1,5 +1,3 @@
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-
 set(CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp)
 set(CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_xdl_perchannel_quantization_int8_instance.cpp)
 set(CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_xdl_bias_perlayer_quantization_int8_instance.cpp)
@@ -10,17 +8,16 @@ set(GEMM_QUANT_SRC
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
     gemm/device_gemm_quantization_xdl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp
 )
-if(DL_KERNELS)
-    list(APPEND CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp)
-    list(APPEND CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp)
-    list(APPEND CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp)
-    list(APPEND CONV2D_BIAS_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp)
-    list(APPEND GEMM_QUANT_SRC
-        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
-        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
-        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
-        gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
-endif()
+
+list(APPEND CONV2D_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_perlayer_quantization_int8_instance.cpp)
+list(APPEND CONV2D_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_perchannel_quantization_int8_instance.cpp)
+list(APPEND CONV2D_BIAS_PERLAYER_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perlayer_quantization_int8_instance.cpp)
+list(APPEND CONV2D_BIAS_PERCHANNEL_QUANT_SRC conv2d_fwd/device_conv2d_dl_bias_perchannel_quantization_int8_instance.cpp)
+list(APPEND GEMM_QUANT_SRC
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_kn_mn_instance.cpp
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_km_nk_mn_instance.cpp
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_kn_mn_instance.cpp
+    gemm/device_gemm_quantization_dl_c_shuffle_i8_i8_i8_mk_nk_mn_instance.cpp)
 
 add_instance_library(device_quantization_instance
     ${CONV2D_PERLAYER_QUANT_SRC}
@@ -29,4 +26,3 @@ add_instance_library(device_quantization_instance
     ${CONV2D_BIAS_PERCHANNEL_QUANT_SRC}
     ${GEMM_QUANT_SRC}
 )
-endif()
\ No newline at end of file
diff --git a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
index ba0197477f..dbe3764115 100644
--- a/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/softmax/CMakeLists.txt
@@ -1,20 +1,17 @@
 set(DEVICE_SOFTMAX_INSTANCES)
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_SOFTMAX_INSTANCES device_softmax_f16_f16_instance_rank3_reduce1.cpp
+list(APPEND DEVICE_SOFTMAX_INSTANCES 
+    device_softmax_f16_f16_instance_rank3_reduce1.cpp
     device_softmax_f16_f16_instance_rank3_reduce2.cpp
     device_softmax_f16_f16_instance_rank3_reduce3.cpp
     device_softmax_f16_f16_instance_rank4_reduce1.cpp
     device_softmax_f16_f16_instance_rank4_reduce2.cpp
     device_softmax_f16_f16_instance_rank4_reduce3.cpp
-    device_softmax_f16_f16_instance_rank4_reduce4.cpp)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-    list(APPEND DEVICE_SOFTMAX_INSTANCES device_softmax_f32_f32_instance_rank3_reduce1.cpp
+    device_softmax_f16_f16_instance_rank4_reduce4.cpp
+    device_softmax_f32_f32_instance_rank3_reduce1.cpp
     device_softmax_f32_f32_instance_rank3_reduce2.cpp
     device_softmax_f32_f32_instance_rank3_reduce3.cpp
     device_softmax_f32_f32_instance_rank4_reduce1.cpp
     device_softmax_f32_f32_instance_rank4_reduce2.cpp
     device_softmax_f32_f32_instance_rank4_reduce3.cpp
     device_softmax_f32_f32_instance_rank4_reduce4.cpp)
-endif()
 add_instance_library(device_softmax_instance ${DEVICE_SOFTMAX_INSTANCES})
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 8fddd60858..368965aa8a 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -9,26 +9,121 @@ add_custom_target(tests)
 
 function(add_test_executable TEST_NAME)
     message("adding test ${TEST_NAME}")
-    add_executable(${TEST_NAME} ${ARGN})
-    add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}>)
-    add_dependencies(tests ${TEST_NAME})
-    add_dependencies(check ${TEST_NAME})
-    rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
+    set(result 1)
+    if(DEFINED DTYPES)
+    foreach(source IN LISTS ARGN)
+        set(test 0)
+        foreach(type IN LISTS DTYPES)
+            if(type MATCHES "fp16")
+                set(type1 "_f16")
+            elseif(type MATCHES "fp32")
+                set(type1 "_f32")
+            elseif(type MATCHES "fp8")
+                set(type1 "_f8")
+            elseif(type MATCHES "bf16")
+                set(type1 "_b16")
+            elseif(type MATCHES "fp64")
+                set(type1 "_f64")
+            elseif(type MATCHES "int8")
+                set(type1 "_i8")
+            endif()
+            if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
+                #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                set(test 0)
+                break()
+            elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                NOT(source MATCHES type OR source MATCHES type1))
+                    #if filename contains a type which doesn't match any selected type, mark it for removal
+                    set(test 1)
+            endif()
+        endforeach()
+        if(test EQUAL 1)
+            message("removing test ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
+        endif()
+      endforeach()
+      endif()
+      foreach(source IN LISTS ARGN)
+        if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
+            message("removing dl test ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
+        endif()
+    endforeach()
+
+    #only continue if there are some source files left on the list
+    if(ARGN)
+        add_executable(${TEST_NAME} ${ARGN})
+        add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}>)
+        add_dependencies(tests ${TEST_NAME})
+        add_dependencies(check ${TEST_NAME})
+        rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
+        set(result 0)
+    endif()
+    #message("add_test returns ${result}")
+    return(PROPAGATE result)
 endfunction(add_test_executable TEST_NAME)
 
 include(GoogleTest)
 
 function(add_gtest_executable TEST_NAME)
     message("adding gtest ${TEST_NAME}")
-    add_executable(${TEST_NAME} ${ARGN})
-    add_dependencies(tests ${TEST_NAME})
-    add_dependencies(check ${TEST_NAME})
+    set(result 1)
+    if(DEFINED DTYPES)
+    foreach(source IN LISTS ARGN)
+        set(test 0)
+        foreach(type IN LISTS DTYPES)
+            if(type MATCHES "fp16")
+                set(type1 "_f16")
+            elseif(type MATCHES "fp32")
+                set(type1 "_f32")
+            elseif(type MATCHES "fp8")
+                set(type1 "_f8")
+            elseif(type MATCHES "bf16")
+                set(type1 "_b16")
+            elseif(type MATCHES "fp64")
+                set(type1 "_f64")
+            elseif(type MATCHES "int8")
+                set(type1 "_i8")
+            endif()
+            if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
+                #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                set(test 0)
+                break()
+            elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                NOT(source MATCHES type OR source MATCHES type1))
+                    #if filename contains a type which doesn't match any selected type, mark it for removal
+                    set(test 1)
+            endif()
+        endforeach()
+        if(test EQUAL 1)
+            message("removing gtest ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
+        endif()
+    endforeach()
+    endif()
+    foreach(source IN LISTS ARGN)
+        if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
+            message("removing dl test ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
+        endif()
+    endforeach()
+    #only continue if there are some source files left on the list
+    if(ARGN)
+        add_executable(${TEST_NAME} ${ARGN})
+        add_dependencies(tests ${TEST_NAME})
+        add_dependencies(check ${TEST_NAME})
 
-    # suppress gtest warnings
-    target_compile_options(${TEST_NAME} PRIVATE -Wno-global-constructors -Wno-undef)
-    target_link_libraries(${TEST_NAME} PRIVATE gtest_main)
-    add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}>)
-    rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
+        # suppress gtest warnings
+        target_compile_options(${TEST_NAME} PRIVATE -Wno-global-constructors -Wno-undef)
+        target_link_libraries(${TEST_NAME} PRIVATE gtest_main)
+        add_test(NAME ${TEST_NAME} COMMAND $<TARGET_FILE:${TEST_NAME}>)
+        rocm_install(TARGETS ${TEST_NAME} COMPONENT tests)
+        set(result 0)
+    endif()
+    #message("add_gtest returns ${result}")
+    return(PROPAGATE result)
 endfunction(add_gtest_executable TEST_NAME)
 
 add_subdirectory(magic_number_division)
diff --git a/test/batched_gemm/CMakeLists.txt b/test/batched_gemm/CMakeLists.txt
index 8a3269e90f..d169c135ca 100644
--- a/test/batched_gemm/CMakeLists.txt
+++ b/test/batched_gemm/CMakeLists.txt
@@ -2,25 +2,21 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
-      target_link_libraries(test_batched_gemm_fp16 PRIVATE utility)
-      target_link_libraries(test_batched_gemm_fp16 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
+   if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_fp16 PRIVATE utility device_batched_gemm_instance)
    endif()
-   if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-      add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
-      target_link_libraries(test_batched_gemm_fp32 PRIVATE utility)
-      target_link_libraries(test_batched_gemm_fp32 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
+   if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_fp32 PRIVATE utility device_batched_gemm_instance)
    endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-      add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
-      target_link_libraries(test_batched_gemm_bf16 PRIVATE utility)
-      target_link_libraries(test_batched_gemm_bf16 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
+   if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_bf16 PRIVATE utility device_batched_gemm_instance)
    endif()
-   if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
-      add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
-      target_link_libraries(test_batched_gemm_int8 PRIVATE utility)
-      target_link_libraries(test_batched_gemm_int8 PRIVATE device_batched_gemm_instance)
+   add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
+   if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_int8 PRIVATE utility device_batched_gemm_instance)
    endif()
    set(target 1)
  endif()
diff --git a/test/batched_gemm_gemm/CMakeLists.txt b/test/batched_gemm_gemm/CMakeLists.txt
index 404e74f3bc..03f1d3a4eb 100644
--- a/test/batched_gemm_gemm/CMakeLists.txt
+++ b/test/batched_gemm_gemm/CMakeLists.txt
@@ -2,12 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-      add_custom_target(test_batched_gemm_gemm)
-      add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
+    add_custom_target(test_batched_gemm_gemm)
+    add_gtest_executable(test_batched_gemm_gemm_fp16 test_batched_gemm_gemm_fp16.cpp)
+    if(result EQUAL 0)
       target_link_libraries(test_batched_gemm_gemm_fp16 PRIVATE utility device_batched_gemm_gemm_instance)
       add_dependencies(test_batched_gemm_gemm test_batched_gemm_gemm_fp16)
       set(target 1)
-   endif()
+    endif()
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_multi_d/CMakeLists.txt b/test/batched_gemm_multi_d/CMakeLists.txt
index 825f0dd235..d5e4c4fbe8 100644
--- a/test/batched_gemm_multi_d/CMakeLists.txt
+++ b/test/batched_gemm_multi_d/CMakeLists.txt
@@ -1,4 +1,4 @@
-if(DL_KERNELS)
-    add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d.cpp)
+add_gtest_executable(test_batched_gemm_multi_d test_batched_gemm_multi_d_dl.cpp)
+if(result EQUAL 0)
     target_link_libraries(test_batched_gemm_multi_d PRIVATE utility device_batched_gemm_multi_d_instance)
 endif()
diff --git a/test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp b/test/batched_gemm_multi_d/test_batched_gemm_multi_d_dl.cpp
similarity index 100%
rename from test/batched_gemm_multi_d/test_batched_gemm_multi_d.cpp
rename to test/batched_gemm_multi_d/test_batched_gemm_multi_d_dl.cpp
diff --git a/test/batched_gemm_reduce/CMakeLists.txt b/test/batched_gemm_reduce/CMakeLists.txt
index af95a50eab..32c6ee85d1 100644
--- a/test/batched_gemm_reduce/CMakeLists.txt
+++ b/test/batched_gemm_reduce/CMakeLists.txt
@@ -2,10 +2,9 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-     add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
-     target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility)
-     target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE device_batched_gemm_reduce_instance)
+    add_test_executable(test_batched_gemm_reduce_fp16 batched_gemm_reduce_fp16.cpp)
+    if(result EQUAL 0)
+     target_link_libraries(test_batched_gemm_reduce_fp16 PRIVATE utility device_batched_gemm_reduce_instance)
      set(target 1)
    endif()
  endif()
diff --git a/test/batched_gemm_softmax_gemm/CMakeLists.txt b/test/batched_gemm_softmax_gemm/CMakeLists.txt
index c49175a2eb..c011a6a3c5 100644
--- a/test/batched_gemm_softmax_gemm/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm/CMakeLists.txt
@@ -2,12 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-     add_custom_target(test_batched_gemm_softmax_gemm)
-     add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
-     target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
-     add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
-     set(target 1)
-   endif()
+    add_custom_target(test_batched_gemm_softmax_gemm)
+    add_gtest_executable(test_batched_gemm_softmax_gemm_fp16 test_batched_gemm_softmax_gemm_fp16.cpp)
+    if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_softmax_gemm_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_instance)
+      add_dependencies(test_batched_gemm_softmax_gemm test_batched_gemm_softmax_gemm_fp16)
+      set(target 1)
+    endif()
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
index ae28daa80e..3164863eef 100644
--- a/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
+++ b/test/batched_gemm_softmax_gemm_permute/CMakeLists.txt
@@ -2,25 +2,28 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-     add_custom_target(test_batched_gemm_softmax_gemm_permute)
-   endif()
-   if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-     add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
-     add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
-     target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-     target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
-     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
-   endif()
-   if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
-     add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
-     add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
-     target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-     target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
-     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
-     add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
-   endif()
+    add_custom_target(test_batched_gemm_softmax_gemm_permute)
+    add_gtest_executable(test_batched_gemm_softmax_gemm_permute_fp16 test_batched_gemm_softmax_gemm_permute_fp16.cpp)
+    if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+      add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_fp16)
+    endif()
+    add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_fp16 test_batched_gemm_bias_softmax_gemm_permute_fp16.cpp)
+    if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_fp16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+      add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_fp16)
+    endif()
+   
+    add_gtest_executable(test_batched_gemm_softmax_gemm_permute_bf16 test_batched_gemm_softmax_gemm_permute_bf16.cpp)
+    if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+      add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_softmax_gemm_permute_bf16)
+    endif()
+    add_gtest_executable(test_batched_gemm_bias_softmax_gemm_permute_bf16 test_batched_gemm_bias_softmax_gemm_permute_bf16.cpp)
+    if(result EQUAL 0)
+      target_link_libraries(test_batched_gemm_bias_softmax_gemm_permute_bf16 PRIVATE utility device_batched_gemm_softmax_gemm_permute_instance)
+      add_dependencies(test_batched_gemm_softmax_gemm_permute test_batched_gemm_bias_softmax_gemm_permute_bf16)
+    endif()
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/data_type/CMakeLists.txt b/test/data_type/CMakeLists.txt
index baf0174556..b1606d2a75 100644
--- a/test/data_type/CMakeLists.txt
+++ b/test/data_type/CMakeLists.txt
@@ -1,14 +1,15 @@
 if (USE_BITINT_EXTENSION_INT4)
   add_gtest_executable(test_int4 int4.cpp)
-  target_link_libraries(test_int4 PRIVATE utility)
+  if(result EQUAL 0)
+    target_link_libraries(test_int4 PRIVATE utility)
+  endif()
 endif()
 
-if(DTYPES MATCHES "fp8" OR NOT DEFINED DTYPES)
-  add_gtest_executable(test_f8 f8.cpp)
-  target_link_libraries(test_f8 PRIVATE utility)
+add_gtest_executable(test_fp8 fp8.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_fp8 PRIVATE utility)
 endif()
-
-if(DTYPES MATCHES "bf8" OR NOT DEFINED DTYPES)
-  add_gtest_executable(test_bf8 bf8.cpp)
+add_gtest_executable(test_bf8 bf8.cpp)
+if(result EQUAL 0)
   target_link_libraries(test_bf8 PRIVATE utility)
 endif()
diff --git a/test/data_type/f8.cpp b/test/data_type/fp8.cpp
similarity index 100%
rename from test/data_type/f8.cpp
rename to test/data_type/fp8.cpp
diff --git a/test/elementwise_normalization/CMakeLists.txt b/test/elementwise_normalization/CMakeLists.txt
index 74a3e4999e..aed67901b5 100644
--- a/test/elementwise_normalization/CMakeLists.txt
+++ b/test/elementwise_normalization/CMakeLists.txt
@@ -1,6 +1,6 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_custom_target(test_elementwise_normalization)
-  add_gtest_executable(test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp)
+add_custom_target(test_elementwise_normalization)
+add_gtest_executable(test_elementwise_layernorm_fp16 test_elementwise_layernorm_fp16.cpp)
+if(result EQUAL 0)
   target_link_libraries(test_elementwise_layernorm_fp16 PRIVATE utility device_elementwise_normalization_instance)
   add_dependencies(test_elementwise_normalization test_elementwise_layernorm_fp16)
 endif()
\ No newline at end of file
diff --git a/test/gemm/CMakeLists.txt b/test/gemm/CMakeLists.txt
index 7a8836bfef..f88a134041 100644
--- a/test/gemm/CMakeLists.txt
+++ b/test/gemm/CMakeLists.txt
@@ -1,30 +1,28 @@
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
 add_test_executable(test_gemm_fp32 gemm_fp32.cpp)
-target_link_libraries(test_gemm_fp32 PRIVATE utility)
-target_link_libraries(test_gemm_fp32 PRIVATE device_gemm_instance)
+if(result EQUAL 0)
+    target_link_libraries(test_gemm_fp32 PRIVATE utility device_gemm_instance)
 endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 add_test_executable(test_gemm_fp16 gemm_fp16.cpp)
-target_link_libraries(test_gemm_fp16 PRIVATE utility)
-target_link_libraries(test_gemm_fp16 PRIVATE device_gemm_instance)
-add_library(gemm_standalone_xdl_fp16_instances STATIC
+if(result EQUAL 0)
+    target_link_libraries(test_gemm_fp16 PRIVATE utility device_gemm_instance)
+    add_library(gemm_standalone_xdl_fp16_instances STATIC
     instance/gemm_f16_nn_instance.cpp
     instance/gemm_f16_nt_instance.cpp
     instance/gemm_f16_tn_instance.cpp
     instance/gemm_wavelet_f16_tn_instance.cpp
     instance/gemm_f16_tt_instance.cpp
-)
+    )
+endif()
 add_test_executable(test_gemm_standalone_xdl_fp16 gemm_standalone_xdl_fp16.cpp)
-target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility)
-target_include_directories(test_gemm_standalone_xdl_fp16 PRIVATE instance/)
+if(result EQUAL 0)
+    target_link_libraries(test_gemm_standalone_xdl_fp16 PRIVATE gemm_standalone_xdl_fp16_instances utility)
+    target_include_directories(test_gemm_standalone_xdl_fp16 PRIVATE instance/)
 endif()
-if(DTYPES MATCHES "bf16" OR NOT DEFINED DTYPES)
 add_test_executable(test_gemm_bf16 gemm_bf16.cpp)
-target_link_libraries(test_gemm_bf16 PRIVATE utility)
-target_link_libraries(test_gemm_bf16 PRIVATE device_gemm_instance)
+if(result EQUAL 0)
+    target_link_libraries(test_gemm_bf16 PRIVATE utility device_gemm_instance)
 endif()
-if(DTYPES MATCHES "int8" OR NOT DEFINED DTYPES)
 add_test_executable(test_gemm_int8 gemm_int8.cpp)
-target_link_libraries(test_gemm_int8 PRIVATE utility)
-target_link_libraries(test_gemm_int8 PRIVATE device_gemm_instance)
+if(result EQUAL 0)
+    target_link_libraries(test_gemm_int8 PRIVATE utility device_gemm_instance)
 endif()
\ No newline at end of file
diff --git a/test/gemm_layernorm/CMakeLists.txt b/test/gemm_layernorm/CMakeLists.txt
index ba0a99b67f..bfc4404bd8 100644
--- a/test/gemm_layernorm/CMakeLists.txt
+++ b/test/gemm_layernorm/CMakeLists.txt
@@ -2,12 +2,12 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-  if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
    add_custom_target(test_gemm_layernorm)
    add_gtest_executable(test_gemm_add_relu_add_layernorm_fp16 test_gemm_add_relu_add_layernorm_fp16.cpp)
-   target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
-   add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
-   set(target 1)
-  endif()
+   if(result EQUAL 0)
+     target_link_libraries(test_gemm_add_relu_add_layernorm_fp16 PRIVATE utility device_gemm_add_relu_add_layernorm_instance)
+     add_dependencies(test_gemm_layernorm test_gemm_add_relu_add_layernorm_fp16)
+     set(target 1)
+   endif()
  endif()
 endforeach()
diff --git a/test/gemm_reduce/CMakeLists.txt b/test/gemm_reduce/CMakeLists.txt
index 43c8d60745..42a53c3048 100644
--- a/test/gemm_reduce/CMakeLists.txt
+++ b/test/gemm_reduce/CMakeLists.txt
@@ -1,5 +1,4 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_test_executable(test_gemm_reduce_fp16 gemm_reduce_fp16.cpp)
-  target_link_libraries(test_gemm_reduce_fp16 PRIVATE utility)
-  target_link_libraries(test_gemm_reduce_fp16 PRIVATE device_gemm_reduce_instance)
+add_test_executable(test_gemm_reduce_fp16 gemm_reduce_fp16.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_gemm_reduce_fp16 PRIVATE utility device_gemm_reduce_instance)
 endif()
\ No newline at end of file
diff --git a/test/grouped_gemm/CMakeLists.txt b/test/grouped_gemm/CMakeLists.txt
index 476d953ed8..8c57b667e2 100644
--- a/test/grouped_gemm/CMakeLists.txt
+++ b/test/grouped_gemm/CMakeLists.txt
@@ -1,4 +1,3 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
@@ -13,4 +12,3 @@ foreach(gpu IN LISTS GPU_TARGETS)
    set(target 1)
  endif()
 endforeach()
-endif()
diff --git a/test/normalization/CMakeLists.txt b/test/normalization/CMakeLists.txt
index 2beda4dd74..fe4b1ad82e 100644
--- a/test/normalization/CMakeLists.txt
+++ b/test/normalization/CMakeLists.txt
@@ -1,19 +1,21 @@
-if(DTYPES MATCHES "fp16" OR DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-  add_custom_target(test_normalization)
-endif()
-if(DTYPES MATCHES "fp32" OR NOT DEFINED DTYPES)
-  add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
-  add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
+add_custom_target(test_normalization)
+add_gtest_executable(test_layernorm2d_fp32 test_layernorm2d_fp32.cpp)
+if(result EQUAL 0)
   target_link_libraries(test_layernorm2d_fp32 PRIVATE utility device_normalization_instance)
-  target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
   add_dependencies(test_normalization test_layernorm2d_fp32)
+endif()
+add_gtest_executable(test_groupnorm_fp32 test_groupnorm_fp32.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_groupnorm_fp32 PRIVATE utility device_normalization_instance)
   add_dependencies(test_normalization test_groupnorm_fp32)
 endif()
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
-  add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
+add_gtest_executable(test_layernorm2d_fp16 test_layernorm2d_fp16.cpp)
+if(result EQUAL 0)
   target_link_libraries(test_layernorm2d_fp16 PRIVATE utility device_normalization_instance)
-  target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
   add_dependencies(test_normalization test_layernorm2d_fp16)
+endif()
+add_gtest_executable(test_groupnorm_fp16 test_groupnorm_fp16.cpp)
+if(result EQUAL 0)
+  target_link_libraries(test_groupnorm_fp16 PRIVATE utility device_normalization_instance)
   add_dependencies(test_normalization test_groupnorm_fp16)
 endif()
diff --git a/test/reduce/CMakeLists.txt b/test/reduce/CMakeLists.txt
index fb436165ea..fabaf7cc54 100644
--- a/test/reduce/CMakeLists.txt
+++ b/test/reduce/CMakeLists.txt
@@ -1,7 +1,5 @@
 add_test_executable(test_reduce_no_index reduce_no_index.cpp)
 add_test_executable(test_reduce_with_index reduce_with_index.cpp)
-target_link_libraries(test_reduce_no_index PRIVATE utility)
-target_link_libraries(test_reduce_no_index PRIVATE device_reduce_instance)
-target_link_libraries(test_reduce_with_index PRIVATE utility)
-target_link_libraries(test_reduce_with_index PRIVATE device_reduce_instance)
+target_link_libraries(test_reduce_no_index PRIVATE utility device_reduce_instance)
+target_link_libraries(test_reduce_with_index PRIVATE utility device_reduce_instance)
 

From ede64ae9db8eb052329a7fa7560cc660cfd6be09 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Fri, 22 Sep 2023 10:08:45 -0500
Subject: [PATCH 137/194] Update naming (#937)

---
 .../gpu/gemm_splitk/CMakeLists.txt               | 16 ++++++++--------
 ...xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp} |  0
 ...xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp} |  0
 ...xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp} |  0
 ...xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp} |  0
 ...xdl_splitk_fp8_f16_f16_km_kn_mn_instance.cpp} |  0
 ...xdl_splitk_fp8_f16_f16_km_nk_mn_instance.cpp} |  0
 ...xdl_splitk_fp8_f16_f16_mk_kn_mn_instance.cpp} |  0
 ...xdl_splitk_fp8_f16_f16_mk_nk_mn_instance.cpp} |  0
 9 files changed, 8 insertions(+), 8 deletions(-)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp => device_gemm_xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp => device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp => device_gemm_xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp => device_gemm_xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp => device_gemm_xdl_splitk_fp8_f16_f16_km_kn_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp => device_gemm_xdl_splitk_fp8_f16_f16_km_nk_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp => device_gemm_xdl_splitk_fp8_f16_f16_mk_kn_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/gemm_splitk/{device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp => device_gemm_xdl_splitk_fp8_f16_f16_mk_nk_mn_instance.cpp} (100%)

diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
index 837e040eac..36c0b50e8e 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
@@ -8,13 +8,13 @@ list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_in
                                   device_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
                                   device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
                                   device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp)
+                                  device_gemm_xdl_splitk_fp8_f16_f16_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_fp8_f16_f16_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_fp8_f16_f16_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_fp8_f16_f16_km_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp)
 
 add_instance_library(device_gemm_splitk_instance ${GEMM_SPLITK_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_kn_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_kn_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_km_kn_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_kn_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_km_kn_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_km_nk_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_km_nk_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_mk_kn_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_kn_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_mk_kn_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_mk_nk_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f8_f16_f16_mk_nk_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_fp8_f16_f16_mk_nk_mn_instance.cpp

From c95538325b49a9a12c761f8783b0b0f8c3161f2a Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Sat, 23 Sep 2023 18:56:31 +0200
Subject: [PATCH 138/194] Add 3d grouped conv fwd wmma instances (#935)

* Add 3d grouped conv fwd wmma instances

* Refactor fwd conv tests

* Split wmma instances for each specialization

* Minor stylistic fixes
---
 ...evice_grouped_conv2d_fwd_wmma_instance.hpp | 134 -----
 .../device_grouped_conv_fwd_wmma_instance.hpp | 136 +++++
 .../gpu/grouped_convolution_forward.hpp       | 465 +++++++++++++++++-
 .../gpu/grouped_conv1d_fwd/CMakeLists.txt     |   8 +-
 ...d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp |   0
 ...1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp |   0
 ...1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp |   0
 ...d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp |   0
 .../gpu/grouped_conv2d_fwd/CMakeLists.txt     |  42 +-
 ...wd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  66 ---
 ...fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp |  66 ---
 ..._fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   0
 ..._fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   0
 ..._fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |   0
 ..._fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |   0
 ...a_gnhwc_gkyxc_gnhwk_f16_1x1p0_instance.cpp |  40 ++
 ...gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp |  40 ++
 ...wd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  40 ++
 ...ma_gnhwc_gkyxc_gnhwk_f16_oddc_instance.cpp |  40 ++
 ...ma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instance.cpp |  40 ++
 ..._gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp |  40 ++
 ...fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp |  40 ++
 ...mma_gnhwc_gkyxc_gnhwk_i8_oddc_instance.cpp |  40 ++
 ...a_nhwgc_gkyxc_nhwgk_f16_1x1p0_instance.cpp |  40 ++
 ...nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp |  40 ++
 ...wd_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  40 ++
 ...ma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp |  40 ++
 ...ma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instance.cpp |  40 ++
 ..._nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp |  40 ++
 ...fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp |  40 ++
 ...mma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp |  40 ++
 ...wd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |   0
 ...fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   2 +-
 ...fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   0
 ...wd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |   0
 ...fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |   0
 ...fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |   0
 .../gpu/grouped_conv3d_fwd/CMakeLists.txt     |  36 +-
 ...ndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp |  41 ++
 ...hwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp |  41 ++
 ...wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |  41 ++
 ...gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp |  41 ++
 ...gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp |  41 ++
 ...dhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp |  41 ++
 ..._wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp |  41 ++
 ..._gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp |  41 ++
 ...dhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp |  41 ++
 ...wgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp |  41 ++
 ...wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |  41 ++
 ...ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp |  41 ++
 ...ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp |  41 ++
 ...hwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp |  41 ++
 ..._wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp |  41 ++
 ..._ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp |  41 ++
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |   0
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |   0
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |   0
 ...xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp |   0
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |   0
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |   0
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |   0
 ...xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp |   0
 test/grouped_convnd_fwd/CMakeLists.txt        |   2 +-
 .../grouped_convnd_fwd/grouped_convnd_fwd.cpp | 284 -----------
 .../test_grouped_convnd_fwd.cpp               | 133 +++++
 65 files changed, 2092 insertions(+), 578 deletions(-)
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/{ => xdl}/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/{ => xdl}/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/{ => xdl}/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/{ => xdl}/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp (100%)
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => dl}/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => dl}/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => dl}/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => dl}/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp (100%)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => xdl}/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => xdl}/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp (98%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => xdl}/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => xdl}/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => xdl}/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/{ => xdl}/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp (100%)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/{ => xdl}/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp (100%)
 delete mode 100644 test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
 create mode 100644 test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp

diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp
deleted file mode 100644
index c9cf0f8e1d..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp
+++ /dev/null
@@ -1,134 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using BF16 = ck::bhalf_t;
-using F16  = ck::half_t;
-using F32  = float;
-using I8   = int8_t;
-using I32  = int32_t;
-
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using NHWGC = ck::tensor_layout::convolution::NHWGC;
-using GNHWC = ck::tensor_layout::convolution::GNHWC;
-
-using GKYXC = ck::tensor_layout::convolution::GKYXC;
-
-using NHWGK = ck::tensor_layout::convolution::NHWGK;
-using GNHWK = ck::tensor_layout::convolution::GNHWK;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-static constexpr auto ConvFwdDefault =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
-
-static constexpr auto ConvFwd1x1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
-
-static constexpr auto ConvFwd1x1S1P0 =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
-
-static constexpr auto ConvFwdOddC =
-    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          typename DsDatatype,
-          typename CDEElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-using device_grouped_conv2d_fwd_wmma_f16_instances = std::tuple<
-    // clang-format off
-        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        // blocksize=256
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  8,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  8,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  8,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  8,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        // blocksize=128
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  8,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  8,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
-        // blocksize=64
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  8,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  8,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  8,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        // blocksize=32
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  8,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  8,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  8,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  8,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
-    // clang-format on
-    >;
-
-template <typename ALayout,
-          typename BLayout,
-          typename DsLayout,
-          typename ELayout,
-          typename DsDatatype,
-          typename CDEElementOp,
-          ConvolutionForwardSpecialization ConvSpec>
-using device_grouped_conv2d_fwd_wmma_i8_instances = std::tuple<
-    // clang-format off
-        //########################################|  NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //########################################| Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //########################################|        |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //########################################|        |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        // blocksize=256
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  16,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  16,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        // blocksize=128
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  16,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  16,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
-        // blocksize=64
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  16,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  16,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  16,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
-        // blocksize=32
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  16,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  16,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
-        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<       2, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  16,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>
-    // clang-format on
-    >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp
new file mode 100644
index 0000000000..f925397832
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp
@@ -0,0 +1,136 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+using I8   = int8_t;
+using I32  = int32_t;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvFwdDefault =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto ConvFwd1x1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Pad0;
+
+static constexpr auto ConvFwd1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0;
+
+static constexpr auto ConvFwdOddC =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::OddC;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_wmma_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|   Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  8,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  8,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  8,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  8,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  8,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  8,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  8,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  8,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  8,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  8,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  8,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  8,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16, DsDatatype,    F16,     F32,      F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  8,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_wmma_i8_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|         Ds|  EData| AccData| CShuffle|            A|           B|          CDE|    ConvForward|           GEMM| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|   Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|   DataType|   Type|    Type| DataType|  Elementwise| Elementwise|  Elementwise| Specialization| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |    Operation|   Operation|    Operation|               |               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|          |        |        |         |        |      |      |           |       |        |         |             |            |             |               |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        //generic instance
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,               1,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,               1,              16,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     4,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,    64,   256,     4,  16,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   256,    64,     4,  16,    16,   16,       8,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   256,   128,   128,     8,  16,    16,   16,       4,       2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,    64,     8,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     4,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    64,   128,     8,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     4,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   128,    64,     8,  16,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,    32,   256,     4,  16,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,   128,   256,    32,     4,  16,    16,   16,       8,       1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    64,     4,  16,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    64,    32,     4,  16,    16,   16,       2,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,    32,     8,  16,    16,   16,       1,       2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    64,    32,   128,     4,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    64,     4,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    64,    16,     4,  16,    16,   16,       4,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    32,    32,     4,  16,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvFwdMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  I8,   I8, DsDatatype,    I8,     I32,      I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, GemmMNKPadding,    32,    16,    16,     4,  16,    16,   16,       1,       1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index 7fc88c33a6..e6dbd349dd 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -170,6 +170,48 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
 #ifdef DL_KERNELS
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -200,6 +242,48 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 #endif
 #if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
 void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
@@ -234,6 +318,62 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                               PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               NHWGC,
@@ -248,6 +388,63 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -293,6 +490,62 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
@@ -323,6 +576,62 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_BF16
 // grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
@@ -354,6 +663,62 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
@@ -384,6 +749,62 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
 #endif
 
 template <ck::index_t NumDimSpatial,
@@ -478,6 +899,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
 #endif
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_BF16
@@ -492,6 +916,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                               is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(op_ptrs);
             }
 #endif
         }
@@ -516,14 +943,28 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #ifdef DL_KERNELS
                 add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
 #endif
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instances(op_ptrs);
             }
 #endif
-#ifdef CK_ENABLE_BDF16
+#ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(op_ptrs);
             }
+#endif
+#ifdef CK_ENABLE_INT8
+            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                              is_same_v<OutDataType, int8_t>)
+            {
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(op_ptrs);
+                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(op_ptrs);
+            }
 #endif
         }
         else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
@@ -541,6 +982,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_BF16
@@ -555,6 +1002,11 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(op_ptrs);
             }
 #endif
         }
@@ -573,6 +1025,12 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_BF16
@@ -587,6 +1045,11 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                          is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instances(op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+                    op_ptrs);
+                add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(op_ptrs);
             }
 #endif
         }
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/CMakeLists.txt
index 1d90593e37..f51a484bb5 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/CMakeLists.txt
@@ -1,6 +1,6 @@
 add_instance_library(device_grouped_conv1d_fwd_instance
-   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
-   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
-   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
-   device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
+   xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
+   xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+   xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+   xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_fwd/xdl/device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
index 6b0ad99d69..1542d611f7 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/CMakeLists.txt
@@ -1,21 +1,37 @@
 add_instance_library(device_grouped_conv2d_fwd_instance
    #xdl
    # GNHWC, GKYXC, GNHWK
-   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
-   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
-   device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+   xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+   xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
    # NHWGC, GKYXC, NHWGK
-   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
-   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-   device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+   xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+   xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
    #dl
    # GNHWC, GKYXC, GNHWK
-   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
-   device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-   # WMMA
-   device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
-   device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
+   dl/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   dl/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
    # NHWGC, GKYXC, NHWGK
-   device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-   device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+   dl/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   dl/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+   # WMMA
+   # GNHWC, GKYXC, GNHWK
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instance.cpp
+   # NHWGC, GKYXC, NHWGK
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp
+   wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
deleted file mode 100644
index 3904592fca..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ /dev/null
@@ -1,66 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwdDefault>{});
-
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwd1x1P0>{});
-
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwd1x1S1P0>{});
-
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_f16_instances<GNHWC,
-                                                                                GKYXC,
-                                                                                Empty_Tuple,
-                                                                                GNHWK,
-                                                                                Empty_Tuple,
-                                                                                PassThrough,
-                                                                                ConvFwdOddC>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
deleted file mode 100644
index 0cded93d67..0000000000
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
+++ /dev/null
@@ -1,66 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv2d_fwd_wmma_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances)
-{
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdDefault>{});
-
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1P0>{});
-
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwd1x1S1P0>{});
-
-    add_device_operation_instances(instances,
-                                   device_grouped_conv2d_fwd_wmma_i8_instances<GNHWC,
-                                                                               GKYXC,
-                                                                               Empty_Tuple,
-                                                                               GNHWK,
-                                                                               Empty_Tuple,
-                                                                               PassThrough,
-                                                                               ConvFwdOddC>{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/dl/device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instance.cpp
new file mode 100644
index 0000000000..2aa6608c84
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..f928d525a5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
new file mode 100644
index 0000000000..94cf439997
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instance.cpp
new file mode 100644
index 0000000000..f519a19369
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              GNHWC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              GNHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instance.cpp
new file mode 100644
index 0000000000..926b7227c3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..b72c4cf5d2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
new file mode 100644
index 0000000000..108602e133
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instance.cpp
new file mode 100644
index 0000000000..b503542876
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             GNHWC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             GNHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instance.cpp
new file mode 100644
index 0000000000..8316debe57
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..9b90db14cf
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..cd69e5aa24
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp
new file mode 100644
index 0000000000..78d7711a4b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<2,
+                                                                              NHWGC,
+                                                                              GKYXC,
+                                                                              Empty_Tuple,
+                                                                              NHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instance.cpp
new file mode 100644
index 0000000000..193d35dc61
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..c3a336127e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
new file mode 100644
index 0000000000..3e57cd4963
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp
new file mode 100644
index 0000000000..75f8ff032d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/wmma/device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, hi, wi, g, c] * wei[g, k, y, x, c] = out[n, ho, wo, g, k]
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<2,
+                                                                             NHWGC,
+                                                                             GKYXC,
+                                                                             Empty_Tuple,
+                                                                             NHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
similarity index 98%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
index debdb05b4d..cbdce0009e 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -8,7 +8,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-// Compilation parameters for in[g, n, hi ,wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
+// Compilation parameters for in[g, n, hi, wi, c] * wei[g, k, y, x, c] = out[g, n, ho, wo, k]
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
                                                               GNHWC,
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_fwd/xdl/device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
index cd209dbf9e..728e2d1c34 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
@@ -1,11 +1,31 @@
 add_instance_library(device_grouped_conv3d_fwd_instance
-   device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
-   device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
-   device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
-   device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
 
-   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
-   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
-   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-   device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
+
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
+
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp
+
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
+
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp
new file mode 100644
index 0000000000..d955944829
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..05e42a6cd4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
new file mode 100644
index 0000000000..85138786b3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp
new file mode 100644
index 0000000000..ae1931afa5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              GNDHWC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              GNDHWK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp
new file mode 100644
index 0000000000..fd2b8b2ecc
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..4de33c3575
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
new file mode 100644
index 0000000000..9a6856de62
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp
new file mode 100644
index 0000000000..ca40e163d0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[g, n, di, hi, wi, c] * wei[g, k, z, y, x, c] = out[g, n, do, ho,
+// wo, k]
+void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              GNDHWC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              GNDHWK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             GNDHWC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             GNDHWK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp
new file mode 100644
index 0000000000..4357f583b5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..199747eb9f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..1c3f20bd9d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp
new file mode 100644
index 0000000000..a3cb037090
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_f16_instances<3,
+                                                                              NDHWGC,
+                                                                              GKZYXC,
+                                                                              Empty_Tuple,
+                                                                              NDHWGK,
+                                                                              Empty_Tuple,
+                                                                              PassThrough,
+                                                                              ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp
new file mode 100644
index 0000000000..721fa653f5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..459af039d7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
new file mode 100644
index 0000000000..e02557e1aa
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp
new file mode 100644
index 0000000000..e48e3675af
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp
@@ -0,0 +1,41 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for in[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = out[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_fwd_wmma_i8_instances<3,
+                                                                             NDHWGC,
+                                                                             GKZYXC,
+                                                                             Empty_Tuple,
+                                                                             NDHWGK,
+                                                                             Empty_Tuple,
+                                                                             PassThrough,
+                                                                             ConvFwdOddC>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
diff --git a/test/grouped_convnd_fwd/CMakeLists.txt b/test/grouped_convnd_fwd/CMakeLists.txt
index 38da884734..869bd77d3d 100644
--- a/test/grouped_convnd_fwd/CMakeLists.txt
+++ b/test/grouped_convnd_fwd/CMakeLists.txt
@@ -1,3 +1,3 @@
-add_gtest_executable(test_grouped_convnd_fwd grouped_convnd_fwd.cpp)
+add_gtest_executable(test_grouped_convnd_fwd test_grouped_convnd_fwd.cpp)
 target_link_libraries(test_grouped_convnd_fwd PRIVATE utility device_grouped_conv1d_fwd_instance device_grouped_conv2d_fwd_instance device_grouped_conv3d_fwd_instance)
 
diff --git a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp b/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
deleted file mode 100644
index c856255ea3..0000000000
--- a/test/grouped_convnd_fwd/grouped_convnd_fwd.cpp
+++ /dev/null
@@ -1,284 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-#include <iostream>
-#include <initializer_list>
-#include <vector>
-#include <gtest/gtest.h>
-
-#include "profiler/profile_grouped_conv_fwd_impl.hpp"
-
-class TestGroupedConvNdFwd : public ::testing::Test
-{
-    protected:
-    std::vector<ck::utils::conv::ConvParam> conv_params;
-};
-
-// 1d GNWC/GKXC/GNWK
-TEST_F(TestGroupedConvNdFwd, GroupedConv1dFwdGNWC)
-{
-    conv_params.clear();
-    conv_params.push_back({1, 2, 128, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
-    conv_params.push_back({1, 2, 128, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
-    conv_params.push_back({1, 2, 128, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
-    conv_params.push_back({1, 1, 1, 1, 32, {3}, {32}, {1}, {1}, {1}, {1}});
-    conv_params.push_back({1, 1, 1, 64, 3, {3}, {32}, {1}, {1}, {1}, {1}});
-
-    for(auto& param : conv_params)
-    {
-        bool pass;
-
-        // fp32
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
-                                                           ck::tensor_layout::convolution::GNWC,
-                                                           ck::tensor_layout::convolution::GKXC,
-                                                           ck::tensor_layout::convolution::GNWK,
-                                                           float,
-                                                           float,
-                                                           float>(true,  // do_verification
-                                                                  1,     // init_method
-                                                                  false, // do_log
-                                                                  false, // time_kernel
-                                                                  param);
-
-        EXPECT_TRUE(pass);
-
-        // fp16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
-                                                           ck::tensor_layout::convolution::GNWC,
-                                                           ck::tensor_layout::convolution::GKXC,
-                                                           ck::tensor_layout::convolution::GNWK,
-                                                           ck::half_t,
-                                                           ck::half_t,
-                                                           ck::half_t>(true,  // do_verification
-                                                                       1,     // init_method
-                                                                       false, // do_log
-                                                                       false, // time_kernel
-                                                                       param);
-
-        EXPECT_TRUE(pass);
-
-        // bf16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
-                                                           ck::tensor_layout::convolution::GNWC,
-                                                           ck::tensor_layout::convolution::GKXC,
-                                                           ck::tensor_layout::convolution::GNWK,
-                                                           ck::bhalf_t,
-                                                           ck::bhalf_t,
-                                                           ck::bhalf_t>(true,  // do_verification
-                                                                        1,     // init_method
-                                                                        false, // do_log
-                                                                        false, // time_kernel
-                                                                        param);
-
-        EXPECT_TRUE(pass);
-
-        // int8
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<1,
-                                                           ck::tensor_layout::convolution::GNWC,
-                                                           ck::tensor_layout::convolution::GKXC,
-                                                           ck::tensor_layout::convolution::GNWK,
-                                                           int8_t,
-                                                           int8_t,
-                                                           int8_t>(true,  // do_verification
-                                                                   1,     // init_method
-                                                                   false, // do_log
-                                                                   false, // time_kernel
-                                                                   param);
-
-        EXPECT_TRUE(pass);
-    }
-}
-
-// 2d GNHWC/GKYXC/GNHWK
-TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdGNHWC)
-{
-    conv_params.clear();
-    conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
-    conv_params.push_back({2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
-    conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-
-    for(auto& param : conv_params)
-    {
-        bool pass;
-
-        // fp32
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
-                                                           ck::tensor_layout::convolution::GNHWC,
-                                                           ck::tensor_layout::convolution::GKYXC,
-                                                           ck::tensor_layout::convolution::GNHWK,
-                                                           float,
-                                                           float,
-                                                           float>(true,  // do_verification
-                                                                  1,     // init_method
-                                                                  false, // do_log
-                                                                  false, // time_kernel
-                                                                  param);
-
-        EXPECT_TRUE(pass);
-
-        // fp16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
-                                                           ck::tensor_layout::convolution::GNHWC,
-                                                           ck::tensor_layout::convolution::GKYXC,
-                                                           ck::tensor_layout::convolution::GNHWK,
-                                                           ck::half_t,
-                                                           ck::half_t,
-                                                           ck::half_t>(true,  // do_verification
-                                                                       1,     // init_method
-                                                                       false, // do_log
-                                                                       false, // time_kernel
-                                                                       param);
-
-        EXPECT_TRUE(pass);
-
-        // bf16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
-                                                           ck::tensor_layout::convolution::GNHWC,
-                                                           ck::tensor_layout::convolution::GKYXC,
-                                                           ck::tensor_layout::convolution::GNHWK,
-                                                           ck::bhalf_t,
-                                                           ck::bhalf_t,
-                                                           ck::bhalf_t>(true,  // do_verification
-                                                                        1,     // init_method
-                                                                        false, // do_log
-                                                                        false, // time_kernel
-                                                                        param);
-
-        EXPECT_TRUE(pass);
-
-        // int8
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
-                                                           ck::tensor_layout::convolution::GNHWC,
-                                                           ck::tensor_layout::convolution::GKYXC,
-                                                           ck::tensor_layout::convolution::GNHWK,
-                                                           int8_t,
-                                                           int8_t,
-                                                           int8_t>(true,  // do_verification
-                                                                   1,     // init_method
-                                                                   false, // do_log
-                                                                   false, // time_kernel
-                                                                   param);
-
-        EXPECT_TRUE(pass);
-    }
-}
-
-// 3d GNDHWC/GKZYXC/GNDHWK
-TEST_F(TestGroupedConvNdFwd, GroupedConv3dFwdGNDHWC)
-{
-    conv_params.clear();
-    conv_params.push_back(
-        {3, 2, 128, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    conv_params.push_back(
-        {3, 2, 128, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-    conv_params.push_back(
-        {3, 2, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    conv_params.push_back(
-        {3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-    this->conv_params.push_back(
-        {3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-    conv_params.push_back(
-        {3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-
-    for(auto& param : conv_params)
-    {
-        bool pass;
-
-        // fp32
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
-                                                           ck::tensor_layout::convolution::GNDHWC,
-                                                           ck::tensor_layout::convolution::GKZYXC,
-                                                           ck::tensor_layout::convolution::GNDHWK,
-                                                           float,
-                                                           float,
-                                                           float>(true,  // do_verification
-                                                                  1,     // init_method
-                                                                  false, // do_log
-                                                                  false, // time_kernel
-                                                                  param);
-
-        EXPECT_TRUE(pass);
-
-        // fp16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
-                                                           ck::tensor_layout::convolution::GNDHWC,
-                                                           ck::tensor_layout::convolution::GKZYXC,
-                                                           ck::tensor_layout::convolution::GNDHWK,
-                                                           ck::half_t,
-                                                           ck::half_t,
-                                                           ck::half_t>(true,  // do_verification
-                                                                       1,     // init_method
-                                                                       false, // do_log
-                                                                       false, // time_kernel
-                                                                       param);
-
-        EXPECT_TRUE(pass);
-
-        // bf16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
-                                                           ck::tensor_layout::convolution::GNDHWC,
-                                                           ck::tensor_layout::convolution::GKZYXC,
-                                                           ck::tensor_layout::convolution::GNDHWK,
-                                                           ck::bhalf_t,
-                                                           ck::bhalf_t,
-                                                           ck::bhalf_t>(true,  // do_verification
-                                                                        1,     // init_method
-                                                                        false, // do_log
-                                                                        false, // time_kernel
-                                                                        param);
-
-        EXPECT_TRUE(pass);
-
-        // int8
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<3,
-                                                           ck::tensor_layout::convolution::GNDHWC,
-                                                           ck::tensor_layout::convolution::GKZYXC,
-                                                           ck::tensor_layout::convolution::GNDHWK,
-                                                           int8_t,
-                                                           int8_t,
-                                                           int8_t>(true,  // do_verification
-                                                                   1,     // init_method
-                                                                   false, // do_log
-                                                                   false, // time_kernel
-                                                                   param);
-
-        EXPECT_TRUE(pass);
-    }
-}
-
-// 2d NHWGC/KYXGC/NHWGK
-TEST_F(TestGroupedConvNdFwd, GroupedConv2dFwdNHWGC)
-{
-    conv_params.clear();
-    conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
-    conv_params.push_back({2, 2, 128, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    conv_params.push_back({2, 2, 128, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
-    conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-
-    for(auto& param : conv_params)
-    {
-        bool pass;
-
-        // fp16
-        pass = ck::profiler::profile_grouped_conv_fwd_impl<2,
-                                                           ck::tensor_layout::convolution::NHWGC,
-                                                           ck::tensor_layout::convolution::GKYXC,
-                                                           ck::tensor_layout::convolution::NHWGK,
-                                                           ck::half_t,
-                                                           ck::half_t,
-                                                           ck::half_t>(true,  // do_verification
-                                                                       1,     // init_method
-                                                                       false, // do_log
-                                                                       false, // time_kernel
-                                                                       param);
-
-        EXPECT_TRUE(pass);
-    }
-}
diff --git a/test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp b/test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp
new file mode 100644
index 0000000000..dde8313f94
--- /dev/null
+++ b/test/grouped_convnd_fwd/test_grouped_convnd_fwd.cpp
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <vector>
+#include <gtest/gtest.h>
+
+#include "profiler/profile_grouped_conv_fwd_impl.hpp"
+
+template <typename Tuple>
+class TestGroupedConvndFwd : public ::testing::Test
+{
+    protected:
+    using DataType  = std::tuple_element_t<0, Tuple>;
+    using InLayout  = std::tuple_element_t<1, Tuple>;
+    using WeiLayout = std::tuple_element_t<2, Tuple>;
+    using OutLayout = std::tuple_element_t<3, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+
+    template <ck::index_t NDimSpatial>
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_grouped_conv_fwd_impl<NDimSpatial,
+                                                                       InLayout,
+                                                                       WeiLayout,
+                                                                       OutLayout,
+                                                                       DataType,
+                                                                       DataType,
+                                                                       DataType>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+using namespace ck::tensor_layout::convolution;
+
+using KernelTypes1d = ::testing::Types<std::tuple<float, GNWC, GKXC, GNWK>,
+                                       std::tuple<ck::half_t, GNWC, GKXC, GNWK>,
+                                       std::tuple<ck::bhalf_t, GNWC, GKXC, GNWK>,
+                                       std::tuple<int8_t, GNWC, GKXC, GNWK>>;
+
+using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWC, GKYXC, GNHWK>,
+                                       std::tuple<ck::half_t, GNHWC, GKYXC, GNHWK>,
+                                       std::tuple<ck::bhalf_t, GNHWC, GKYXC, GNHWK>,
+                                       std::tuple<int8_t, GNHWC, GKYXC, GNHWK>,
+                                       std::tuple<float, NHWGC, GKYXC, NHWGK>,
+                                       std::tuple<ck::half_t, NHWGC, GKYXC, NHWGK>,
+                                       std::tuple<ck::bhalf_t, NHWGC, GKYXC, NHWGK>,
+                                       std::tuple<int8_t, NHWGC, GKYXC, NHWGK>>;
+
+using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWC, GKZYXC, GNDHWK>,
+                                       std::tuple<ck::half_t, GNDHWC, GKZYXC, GNDHWK>,
+                                       std::tuple<ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK>,
+                                       std::tuple<int8_t, GNDHWC, GKZYXC, GNDHWK>,
+                                       std::tuple<float, NDHWGC, GKZYXC, NDHWGK>,
+                                       std::tuple<ck::half_t, NDHWGC, GKZYXC, NDHWGK>,
+                                       std::tuple<ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK>,
+                                       std::tuple<int8_t, NDHWGC, GKZYXC, NDHWGK>>;
+
+template <typename Tuple>
+class TestGroupedConvndFwd1d : public TestGroupedConvndFwd<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndFwd2d : public TestGroupedConvndFwd<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndFwd3d : public TestGroupedConvndFwd<Tuple>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndFwd1d, KernelTypes1d);
+TYPED_TEST_SUITE(TestGroupedConvndFwd2d, KernelTypes2d);
+TYPED_TEST_SUITE(TestGroupedConvndFwd3d, KernelTypes3d);
+
+TYPED_TEST(TestGroupedConvndFwd1d, Test1D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back({1, 2, 32, 128, 256, {1}, {14}, {2}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 2, 32, 128, 256, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 32, 128, 256, {1}, {3}, {1}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 1, 1, 1, 32, {3}, {32}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 1, 64, 3, {3}, {32}, {1}, {1}, {1}, {1}});
+    this->template Run<1>();
+}
+
+TYPED_TEST(TestGroupedConvndFwd2d, Test2D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {2, 2, 32, 128, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 2, 32, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 2, 32, 128, 256, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back({2, 1, 1, 1, 32, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 1, 64, 3, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 1, 1, 1, {3, 3}, {32, 32}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->template Run<2>();
+}
+
+TYPED_TEST(TestGroupedConvndFwd3d, Test3D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {3, 2, 32, 128, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 2, 32, 128, 256, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 2, 32, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 1, 1, 1, 32, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 64, 3, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 1, 1, 1, {3, 3, 3}, {32, 32, 32}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->template Run<3>();
+}

From 2ea75bd6d7f0a061f78e4bb007a840a394a74ba9 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 26 Sep 2023 08:39:11 -0700
Subject: [PATCH 139/194] Resolve some data type issues and cmake policy.
 (#940)

* split the types in gemm_bilinear instances, add condition to cmake policy

* fix syntax

* split the data types in batchnorm examples

* fix the batchnorm_bwd test

* fix types in the batchnorm_bwd test
---
 CMakeLists.txt                                |  5 ++++-
 .../gpu/gemm_bilinear.hpp                     | 20 ++++++++++--------
 test/batchnorm/batchnorm_bwd_rank_4.cpp       | 21 +++++++++++++++----
 test/batchnorm/batchnorm_fwd_rank_4.cpp       | 21 +++++++++++++++----
 test/batchnorm/batchnorm_infer_rank_4.cpp     | 21 +++++++++++++++----
 5 files changed, 66 insertions(+), 22 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 95ef2df7e9..e5c82b9705 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,5 +1,8 @@
 cmake_minimum_required(VERSION 3.14)
-cmake_policy(SET CMP0140 NEW)
+if(POLICY CMP0140)
+  # policies CMP0140 not known to CMake until 3.25
+  cmake_policy(SET CMP0140 NEW)
+endif()
 
 # This has to be initialized before the project() command appears
 # Set the default of CMAKE_BUILD_TYPE to be release, unless user specifies with -D.  MSVC_IDE does not use CMAKE_BUILD_TYPE
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
index 387499e584..1a518a5302 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_bilinear.hpp
@@ -11,12 +11,12 @@
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-#ifdef CK_ENABLE_FP16
+
 namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
-
+#ifdef CK_ENABLE_FP16
 void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_km_kn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Col,
                                                     Row,
@@ -68,7 +68,8 @@ void add_device_gemm_bilinear_xdl_c_shuffle_f16_f16_f16_f16_mk_nk_mn_mn_instance
                                                     PassThrough,
                                                     PassThrough,
                                                     Bilinear>>>& instances);
-
+#endif
+#ifdef CK_ENABLE_INT8
 void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_mk_kn_mn_mn_instances(
     std::vector<std::unique_ptr<DeviceGemmMultipleD<Row,
                                                     Row,
@@ -120,7 +121,7 @@ void add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(
                                                     PassThrough,
                                                     PassThrough,
                                                     Bilinear>>>& instances);
-
+#endif
 // GEMM + Bilinear
 template <typename ALayout,
           typename BLayout,
@@ -158,7 +159,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
+#ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
                      is_same_v<DDataType, half_t> && is_same_v<EDataType, half_t>)
         {
@@ -187,8 +188,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                     op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
-                          is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
+#endif
+#ifdef CK_ENABLE_INT8
+        if constexpr(is_same_v<ADataType, std::int8_t> && is_same_v<BDataType, std::int8_t> &&
+                     is_same_v<DDataType, std::int8_t> && is_same_v<EDataType, std::int8_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<DLayout, Row> && is_same_v<ELayout, Row>)
@@ -211,7 +214,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
                 add_device_gemm_bilinear_wmma_c_shuffle_i8_i8_i8_i8_km_nk_mn_mn_instances(op_ptrs);
             }
         }
-
+#endif
         return op_ptrs;
     }
 };
@@ -220,4 +223,3 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGemmMu
 } // namespace device
 } // namespace tensor_operation
 } // namespace ck
-#endif
diff --git a/test/batchnorm/batchnorm_bwd_rank_4.cpp b/test/batchnorm/batchnorm_bwd_rank_4.cpp
index a4696cf2a3..cc514261e6 100644
--- a/test/batchnorm/batchnorm_bwd_rank_4.cpp
+++ b/test/batchnorm/batchnorm_bwd_rank_4.cpp
@@ -70,10 +70,23 @@ class TestBatchNormBwdRank4 : public ::testing::Test
     }
 };
 
-using KernelTypes = ::testing::Types<std::tuple<F16, F32, F32, F32, F16, F32, F32>,
-                                     std::tuple<F32, F32, F32, F32, F32, F32, F32>,
-                                     std::tuple<BF16, F32, F32, F32, BF16, F32, F32>,
-                                     std::tuple<F64, F64, F64, F64, F64, F64, F64>>;
+using KernelTypes = ::testing::Types<
+#ifdef CK_ENABLE_FP16
+    std::tuple<F16, F32, F32, F32, F16, F32, F32>
+#endif
+#ifdef CK_ENABLE_FP32
+    ,
+    std::tuple<F32, F32, F32, F32, F32, F32, F32>
+#endif
+#ifdef CK_ENABLE_BF16
+    ,
+    std::tuple<BF16, F32, F32, F32, BF16, F32, F32>
+#endif
+#ifdef CK_ENABLE_FP64
+    ,
+    std::tuple<F64, F64, F64, F64, F64, F64, F64>
+#endif
+    >;
 
 TYPED_TEST_SUITE(TestBatchNormBwdRank4, KernelTypes);
 
diff --git a/test/batchnorm/batchnorm_fwd_rank_4.cpp b/test/batchnorm/batchnorm_fwd_rank_4.cpp
index 9b6fbd0f66..6bf635f0cd 100644
--- a/test/batchnorm/batchnorm_fwd_rank_4.cpp
+++ b/test/batchnorm/batchnorm_fwd_rank_4.cpp
@@ -87,10 +87,23 @@ class TestBatchNormFwdRank4 : public ::testing::Test
     }
 };
 
-using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, F32>,
-                                     std::tuple<F32, F32, F32, F32, F32, F32>,
-                                     std::tuple<BF16, BF16, F32, BF16, BF16, F32>,
-                                     std::tuple<F64, F64, F64, F64, F64, F64>>;
+using KernelTypes = ::testing::Types<
+#ifdef CK_ENABLE_FP16
+    std::tuple<F16, F16, F32, F16, F16, F32>
+#endif
+#ifdef CK_ENABLE_FP32
+    ,
+    std::tuple<F32, F32, F32, F32, F32, F32>
+#endif
+#ifdef CK_ENABLE_BF16
+    ,
+    std::tuple<BF16, BF16, F32, BF16, BF16, F32>
+#endif
+#ifdef CK_ENABLE_FP64
+    ,
+    std::tuple<F64, F64, F64, F64, F64, F64>
+#endif
+    >;
 
 TYPED_TEST_SUITE(TestBatchNormFwdRank4, KernelTypes);
 
diff --git a/test/batchnorm/batchnorm_infer_rank_4.cpp b/test/batchnorm/batchnorm_infer_rank_4.cpp
index ecb4043b36..0165192acf 100644
--- a/test/batchnorm/batchnorm_infer_rank_4.cpp
+++ b/test/batchnorm/batchnorm_infer_rank_4.cpp
@@ -67,10 +67,23 @@ class TestBatchNormInferRank4 : public ::testing::Test
     }
 };
 
-using KernelTypes = ::testing::Types<std::tuple<F16, F16, F32, F16, F16, F32>,
-                                     std::tuple<F32, F32, F32, F32, F32, F32>,
-                                     std::tuple<BF16, BF16, F32, BF16, BF16, F32>,
-                                     std::tuple<F64, F64, F64, F64, F64, F64>>;
+using KernelTypes = ::testing::Types<
+#ifdef CK_ENABLE_FP16
+    std::tuple<F16, F16, F32, F16, F16, F32>
+#endif
+#ifdef CK_ENABLE_FP32
+    ,
+    std::tuple<F32, F32, F32, F32, F32, F32>
+#endif
+#ifdef CK_ENABLE_BF16
+    ,
+    std::tuple<BF16, BF16, F32, BF16, BF16, F32>
+#endif
+#ifdef CK_ENABLE_FP64
+    ,
+    std::tuple<F64, F64, F64, F64, F64, F64>
+#endif
+    >;
 
 TYPED_TEST_SUITE(TestBatchNormInferRank4, KernelTypes);
 

From 0b296a2722e6769ddaadd93f305086f3928f28fb Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 26 Sep 2023 11:22:31 -0700
Subject: [PATCH 140/194] split ckProfiler gfx9 package into gfx90 and gfx94
 (#946)

---
 CMakeLists.txt | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index e5c82b9705..7b8e43b9df 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -111,16 +111,18 @@ else()
     add_definitions(-DPROFILER_ONLY)
     set(GPU_TARGETS "" CACHE STRING "" FORCE)
     if(GPU_TARGETS)
-        message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx9, gfx10, or gfx11")
+        message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx90, gfx94, gfx10, or gfx11")
     endif()
-    if(GPU_ARCH MATCHES "gfx9")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942")
+    if(GPU_ARCH MATCHES "gfx90")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx900;gfx906;gfx908;gfx90a")
+    elseif(GPU_ARCH MATCHES "gfx94")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx940;gfx941;gfx942")
     elseif(GPU_ARCH MATCHES "gfx10")
         rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1030")
     elseif(GPU_ARCH MATCHES "gfx11")
         rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1100;gfx1101;gfx1102")
     else()
-        message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx9, gfx10, or gfx11")
+        message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx90, gfx94, gfx10, or gfx11")
     endif()
     set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
 endif()

From 94bfa50256166d7fec4dd59c90e25d1a81b6531b Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Tue, 26 Sep 2023 14:59:33 -0500
Subject: [PATCH 141/194] Add fp8 gemm instances (#920)

* Add fp8 gemm instances

* Update instance naming
---
 .../tensor_operation_instance/gpu/gemm.hpp    | 43 ++++++++++++
 .../gpu/gemm/CMakeLists.txt                   |  5 ++
 ..._shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp | 69 +++++++++++++++++++
 ..._shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp | 69 +++++++++++++++++++
 ..._shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp | 69 +++++++++++++++++++
 ..._shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp | 66 ++++++++++++++++++
 .../include/profiler/profile_gemm_impl.hpp    |  6 ++
 profiler/src/profile_gemm.cpp                 | 24 ++++++-
 8 files changed, 350 insertions(+), 1 deletion(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp

diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
index 3ceea58c1a..7128a10fa1 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm.hpp
@@ -312,6 +312,23 @@ void add_device_gemm_xdl_f64_f64_f64_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F64, F64, F64, PassThrough, PassThrough, PassThrough>>>&
         instances);
 #endif
+#ifdef CK_ENABLE_FP8
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances);
+#endif
 template <typename ALayout,
           typename BLayout,
           typename CLayout,
@@ -505,6 +522,32 @@ struct DeviceOperationInstanceFactory<
 #endif
             }
         }
+#endif
+#ifdef CK_ENABLE_FP8
+        else if constexpr(is_same_v<ADataType, ck::f8_t> && is_same_v<BDataType, ck::f8_t> &&
+                          is_same_v<CDataType, ck::f8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(op_ptrs);
+            }
+        }
 #endif
         return op_ptrs;
     }
diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 22cf1a462d..89820775c3 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -89,6 +89,11 @@ list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_bf16_bf16_bf16_mk_kn_mn_ins
                            device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_kn_mn_instance.cpp
                            device_gemm_xdl_c_shuffle_bf16_bf16_bf16_km_nk_mn_instance.cpp)
 
+list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp
+                           device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp)
+
 add_instance_library(device_gemm_instance ${GEMM_INSTANCES})
 
 set(ENABLE_PIPELINE_V2_OPT OFF)
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000..9739046d3c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_kn_mn_instance.cpp
@@ -0,0 +1,69 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#ifdef CK_ENABLE_FP8
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F32 = float;
+using F8  = f8_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[k, n] = c[m, n]
+using device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances =
+    std::tuple<
+        // clang-format off
+        //#####################| ALayout| BLayout| CLayout|  AData|  BData|  CData| AccData| CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################|        |        |        |   Type|   Type|   Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################|        |        |        |       |       |       |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################|        |        |        |       |       |       |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,   4,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,   4,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,   4,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,   4,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,   4,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_c_shuffle_f8_f8_f8_km_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000..810c1b87cb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_km_nk_mn_instance.cpp
@@ -0,0 +1,69 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#ifdef CK_ENABLE_FP8
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F32 = float;
+using F8  = f8_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[k, m] * b[n, k] = c[m, n]
+using device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances =
+    std::tuple<
+        // clang-format off
+        //#####################| ALayout| BLayout| CLayout|  AData|  BData|  CData| AccData| CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################|        |        |        |   Type|   Type|   Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################|        |        |        |       |       |       |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################|        |        |        |       |       |       |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,   4,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,   4,  16,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,   4,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,   4,  16,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,   4,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,   4,  16,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,   4,  16,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,   4,  16,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Col,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,             16,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Col, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_c_shuffle_f8_f8_f8_km_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..e1aac45134
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
@@ -0,0 +1,69 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#ifdef CK_ENABLE_FP8
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F32 = float;
+using F8  = f8_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances =
+    std::tuple<
+        // clang-format off
+        //#####################| ALayout| BLayout| CLayout|  AData|  BData|  CData| AccData| CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################|        |        |        |   Type|   Type|   Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################|        |        |        |       |       |       |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################|        |        |        |       |       |       |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
+#endif
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..5c557312aa
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#ifdef CK_ENABLE_FP8
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F32 = float;
+using F8  = f8_t;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances =
+    std::tuple<
+        // clang-format off
+        //#####################| ALayout| BLayout| CLayout|  AData|  BData|  CData| AccData| CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################|        |        |        |   Type|   Type|   Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################|        |        |        |       |       |       |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################|        |        |        |       |       |       |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>
+        // clang-format on
+        >;
+
+void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemm<Row, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
+#endif
diff --git a/profiler/include/profiler/profile_gemm_impl.hpp b/profiler/include/profiler/profile_gemm_impl.hpp
index eaab5dbcc2..ccebb020c4 100644
--- a/profiler/include/profiler/profile_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_impl.hpp
@@ -223,6 +223,12 @@ int profile_gemm_impl(int do_verification,
     {
         std::cout << "Best Perf for datatype = int8";
     }
+#if defined CK_ENABLE_FP8
+    else if constexpr(is_same<CDataType, f8_t>::value)
+    {
+        std::cout << "Best Perf for datatype = fp8";
+    }
+#endif
 
     if constexpr(is_same<ALayout, tensor_layout::gemm::RowMajor>::value)
     {
diff --git a/profiler/src/profile_gemm.cpp b/profiler/src/profile_gemm.cpp
index 9ca7fc4c88..df243c96d6 100644
--- a/profiler/src/profile_gemm.cpp
+++ b/profiler/src/profile_gemm.cpp
@@ -23,6 +23,7 @@ enum struct GemmDataType
     F16_F16_F16,    // 1
     BF16_BF16_BF16, // 2
     INT8_INT8_INT8, // 3
+    F8_F8_F8,       // 4
 };
 
 #define OP_NAME "gemm"
@@ -31,7 +32,7 @@ enum struct GemmDataType
 static void print_helper_msg()
 {
     std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
-              << "arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8)\n"
+              << "arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8; 4: fp8)\n"
               << "arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n"
               << "                     1: A[m, k] * B[n, k] = C[m, n];\n"
               << "                     2: A[k, m] * B[k, n] = C[m, n];\n"
@@ -76,6 +77,9 @@ int profile_gemm(int argc, char* argv[])
     using INT8  = int8_t;
     using INT32 = int32_t;
 #endif
+#ifdef CK_ENABLE_FP8
+    using F8 = ck::f8_t;
+#endif
 
     using Row = ck::tensor_layout::gemm::RowMajor;
     using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -194,6 +198,24 @@ int profile_gemm(int argc, char* argv[])
     {
         return profile(Col{}, Col{}, Row{}, INT8{}, INT8{}, INT32{}, INT8{});
     }
+#endif
+#ifdef CK_ENABLE_FP8
+    else if(data_type == GemmDataType::F8_F8_F8 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F8{}, F8{}, F32{}, F8{});
+    }
+    else if(data_type == GemmDataType::F8_F8_F8 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F8{}, F8{}, F32{}, F8{});
+    }
+    else if(data_type == GemmDataType::F8_F8_F8 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F8{}, F8{}, F32{}, F8{});
+    }
+    else if(data_type == GemmDataType::F8_F8_F8 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F8{}, F8{}, F32{}, F8{});
+    }
 #endif
     else
     {

From 48ba6e8a69947c67b0b4e5021178fe3e2f2c638e Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 26 Sep 2023 18:40:00 -0500
Subject: [PATCH 142/194] Fixed Gemmv2r3 kpad (#938)

* added kpad support into v2r3

* add generic instances

* fixed comments

* fixed mnk padding

* Update device_batched_gemm_xdl.hpp

* fixed kpad

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../gpu/grid/gridwise_gemm_pipeline_v2.hpp               | 4 ++--
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp               | 9 +++------
 2 files changed, 5 insertions(+), 8 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
index d3d7d5af85..25e1cebdbe 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp
@@ -9,13 +9,13 @@ namespace ck {
 
 struct GridwiseGemmPipeline_v2
 {
-    __host__ __device__ static constexpr bool IsSupported(index_t num_loop)
+    __host__ __device__ static constexpr bool IsSupported(const index_t num_loop)
     {
         // TODO: improve applicability
         return num_loop % 2 == 0;
     }
 
-    __host__ __device__ static constexpr bool CalculateHasMainLoop(index_t num_loop)
+    __host__ __device__ static constexpr bool CalculateHasMainLoop(const index_t num_loop)
     {
         return (num_loop / 2) > 1;
     }
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index b6c146ae61..e941f96553 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -175,7 +175,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         return math::integer_divide_ceil(N, NPerBlock) * NPerBlock;
     }
 
-    __host__ static auto CalculateK0(index_t K) { return math::integer_divide_floor(K, K1Value); }
+    __host__ static auto CalculateK0(index_t K) { return math::integer_divide_ceil(K, K1Value); }
 
     // Argument
     struct Problem
@@ -369,9 +369,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
                       "Invalid tuning param!");
 
         // check gridwise gemm pipeline
-        const index_t K0      = problem.K / K1Value;
-        const auto num_k_loop = K0 / K0PerBlock;
-
+        const auto num_k_loop = math::integer_divide_ceil(problem.K0, K0PerBlock);
         if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {
             return false;
@@ -1026,8 +1024,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext
         }
 
         // check gridwise gemm pipeline
-        const index_t K0      = problem.K / K1;
-        const auto num_k_loop = K0 / K0PerBlock;
+        const auto num_k_loop = math::integer_divide_ceil(problem.K0, K0PerBlock);
 
         if(!GridwiseGemmPipe::IsSupported(num_k_loop))
         {

From 420b5a038270b9107c4663dc61de4a9e5b4dd99c Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Tue, 26 Sep 2023 17:43:09 -0700
Subject: [PATCH 143/194] Use lower case for ckprofiler package. (#948)

* split ckProfiler gfx9 package into gfx90 and gfx94

* use lower case for package names
---
 CMakeLists.txt | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7b8e43b9df..f7a9d37bba 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -446,14 +446,14 @@ if(NOT DEFINED INSTANCES_ONLY)
 
    rocm_package_setup_component(profiler
         LIBRARY_NAME composablekernel
-        PACKAGE_NAME ckProfiler
+        PACKAGE_NAME ckprofiler
    )
    add_subdirectory(profiler)
   else()
     #When building PROFILER_ONLY, label the package with GPU_ARCH
     rocm_package_setup_component(profiler
        LIBRARY_NAME composablekernel
-       PACKAGE_NAME ckProfiler_${GPU_ARCH}
+       PACKAGE_NAME ckprofiler_${GPU_ARCH}
     )
     add_subdirectory(profiler)
   endif()

From 11676c7e49110b917734b164a070879ad67ea35d Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 26 Sep 2023 21:16:23 -0500
Subject: [PATCH 144/194] Add multiple A/B support (#906)

* add gridwise_multi_abd

* move element_op into RunRead

* merge element_wise op with data read

* add multiABD example

* allow packed elementwise_op

* changed example

* clean

* clean

* add is_detected

* fix

* minor fix

* add scaleAdd_vec4 example

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/60_gemm_multiABD/CMakeLists.txt       |   10 +
 .../gemm_multiABD_xdl_fp16.cpp                |  361 ++++++
 ...hread_group_tensor_slice_transfer_v7r2.hpp |  214 ++++
 .../gpu/device/device_gemm_multiple_abd.hpp   |   60 +
 .../device_gemm_multiple_abd_xdl_cshuffle.hpp |  766 ++++++++++++
 ...ridwise_gemm_multiple_abd_xdl_cshuffle.hpp | 1033 +++++++++++++++++
 .../threadwise_tensor_slice_transfer_v7r2.hpp |  420 +++++++
 include/ck/utility/is_detected.hpp            |   34 +
 include/ck/utility/tuple.hpp                  |    2 +
 9 files changed, 2900 insertions(+)
 create mode 100644 example/60_gemm_multiABD/CMakeLists.txt
 create mode 100644 example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7r2.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_gemm_multiple_abd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
 create mode 100644 include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
 create mode 100644 include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp
 create mode 100644 include/ck/utility/is_detected.hpp

diff --git a/example/60_gemm_multiABD/CMakeLists.txt b/example/60_gemm_multiABD/CMakeLists.txt
new file mode 100644
index 0000000000..9e2f70649f
--- /dev/null
+++ b/example/60_gemm_multiABD/CMakeLists.txt
@@ -0,0 +1,10 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+     add_example_executable(example_gemm_multiABD_xdl_fp16 gemm_multiABD_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
+endif()
diff --git a/example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp b/example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp
new file mode 100644
index 0000000000..23d41d7cc4
--- /dev/null
+++ b/example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp
@@ -0,0 +1,361 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F16;
+using EDataType        = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using DLayout = Row;
+using ELayout = Row;
+
+struct AddScale
+{
+    static constexpr auto I0 = ck::Number<0>{};
+    static constexpr auto I1 = ck::Number<1>{};
+    static constexpr auto I2 = ck::Number<2>{};
+    static constexpr auto I3 = ck::Number<3>{};
+
+    __host__ __device__ constexpr void
+    operator()(ck::half4_t& a, const ck::half4_t& a0, const ck::half4_t& a1) const
+    {
+        const auto a0_v_t = ck::vector_type<ck::half_t, 4>{a0};
+        const auto a1_v_t = ck::vector_type<ck::half_t, 4>{a1};
+
+        auto r_v_t = ck::vector_type<ck::half_t, 4>{};
+
+        r_v_t.AsType<ck::half_t>()(I0) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I0] + a1_v_t.AsType<ck::half_t>()[I0]);
+        r_v_t.AsType<ck::half_t>()(I1) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I1] + a1_v_t.AsType<ck::half_t>()[I1]);
+        r_v_t.AsType<ck::half_t>()(I2) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I2] + a1_v_t.AsType<ck::half_t>()[I2]);
+        r_v_t.AsType<ck::half_t>()(I3) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I3] + a1_v_t.AsType<ck::half_t>()[I3]);
+
+        a = r_v_t.AsType<ck::half4_t>()[I0];
+    }
+
+    __host__ __device__ constexpr void
+    operator()(ck::half_t& a, const ck::half_t& a0, const ck::half_t& a1) const
+    {
+        a = scale * (a0 + a1);
+    }
+
+    static constexpr ck::index_t vec_len = 4;
+
+    float scale = 1.0;
+};
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
+        ck::half_t& e, const float& c, const ck::half_t& d) const
+    {
+        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
+    };
+
+    float alpha_;
+    float beta_;
+};
+
+using AElementOp   = AddScale;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleABD_Xdl_CShuffle<
+    ck::Tuple<ALayout, ALayout>,
+    ck::Tuple<BLayout>,
+    ck::Tuple<DLayout>,
+    ELayout,
+    ck::Tuple<ADataType, ADataType>,
+    ck::Tuple<BDataType>,
+    AccDataType,
+    CShuffleDataType,
+    ck::Tuple<DDataType>,
+    EDataType,
+    AElementOp,
+    BElementOp,
+    CDEElementOp,
+    GemmSpec,
+    1,
+    256,
+    256,
+    128,
+    32,
+    8,
+    8,
+    32,
+    32,
+    4,
+    2,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    1,
+    1,
+    S<1, 32, 1, 8>,
+    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideD = 4096;
+    ck::index_t StrideE = 4096;
+
+    float alpha = 1.0f;
+    float beta  = 1.0f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 6)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        alpha = std::stof(argv[4]);
+        beta  = std::stof(argv[5]);
+    }
+    else if(argc == 13)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        StrideA = std::stoi(argv[7]);
+        StrideB = std::stoi(argv[8]);
+        StrideD = std::stoi(argv[9]);
+        StrideE = std::stoi(argv[10]);
+
+        alpha = std::stof(argv[11]);
+        beta  = std::stof(argv[12]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
+               "beta\n");
+        exit(0);
+    }
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<ADataType> a1_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
+    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
+        break;
+    default:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem a0_device_buf(sizeof(ADataType) * a0_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem a1_device_buf(sizeof(ADataType) * a1_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_m_k.mData.data());
+    a1_device_buf.ToDevice(a1_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d_device_buf.ToDevice(d_m_n.mData.data());
+    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
+
+    auto a_element_op   = AElementOp{0.2};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
+                                                          a1_device_buf.GetDeviceBuffer()},
+                               std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
+                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                               e_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               std::array<ck::index_t, 2>{StrideA, StrideA},
+                               std::array<ck::index_t, 1>{StrideB},
+                               std::array<ck::index_t, 1>{StrideD},
+                               StrideE,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_m_n({M, N});
+
+        Tensor<ADataType> a_m_k({M, K});
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int k = 0; k < K; ++k)
+            {
+                a_element_op(a_m_k(m, k), a0_m_k(m, k), a1_m_k(m, k));
+            }
+        }
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                PassThrough,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7r2.hpp b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7r2.hpp
new file mode 100644
index 0000000000..1a9bb3213a
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7r2.hpp
@@ -0,0 +1,214 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_description/cluster_descriptor.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp"
+#include "ck/utility/is_detected.hpp"
+
+namespace ck {
+
+// Thread-group level multi-source, multi-destination tensor slice data movement
+// Assume:
+//   1. All sources and destinations are DynamicBuffer
+//   2. Same VectorDim and ScalerPerVector for all sources and destinations
+//   3. DstInMemOps are per destination tensor
+//   4. ThreadTransferSrcResetCoordinateAfterRunFlags are per source tensor
+//   5. ThreadTransferDstResetCoordinateAfterRunFlags are per destination tensor
+//
+// Does following things to avoid scratch memory issue
+//   1. Pass tensor descritpors by reference (or tuple of references)
+//   2. Does not keep reference to tensor descriptor
+//   3. Does not construct new tensor coordinate when call Run()
+template <typename ThreadGroup,
+          typename SrcDatas,
+          typename DstDatas,
+          typename SrcDescs,
+          typename DstDescs,
+          typename ElementwiseOperation,
+          typename DstInMemOps, // Sequence<InMemoryDataOperationEnum ...>
+          typename SliceLengths,
+          typename ThreadClusterLengths,
+          typename ThreadClusterArrangeOrder,
+          typename SrcDimAccessOrder,
+          typename DstDimAccessOrder,
+          index_t SrcVectorDim,
+          index_t DstVectorDim,
+          index_t SrcScalarPerVector,
+          index_t DstScalarPerVector,
+          typename ThreadTransferSrcResetCoordinateAfterRunFlags,
+          typename ThreadTransferDstResetCoordinateAfterRunFlags>
+struct ThreadGroupTensorSliceTransfer_v7r2
+{
+    static constexpr index_t nDim =
+        remove_cvref_t<tuple_element_t<0, SrcDescs>>::GetNumOfDimension();
+
+    static constexpr index_t nSrc = remove_cvref_t<SrcDescs>::Size();
+    static constexpr index_t nDst = remove_cvref_t<DstDescs>::Size();
+
+    using Index = MultiIndex<nDim>;
+
+    static constexpr auto thread_slice_lengths = SliceLengths{} / ThreadClusterLengths{};
+
+    __device__ constexpr ThreadGroupTensorSliceTransfer_v7r2(
+        const SrcDescs& src_descs,
+        const StaticallyIndexedArray<Index, nSrc>& src_block_slice_origins,
+        const DstDescs& dst_descs,
+        const StaticallyIndexedArray<Index, nDst>& dst_block_slice_origins,
+        const ElementwiseOperation& element_op)
+        : threadwise_transfer_(src_descs,
+                               StaticallyIndexedArray<Index, nSrc>{},
+                               dst_descs,
+                               StaticallyIndexedArray<Index, nDst>{},
+                               element_op)
+    {
+        static_assert(nSrc == SrcDatas::Size() && nSrc == SrcDescs::Size() &&
+                          nSrc == ThreadTransferSrcResetCoordinateAfterRunFlags::Size() &&
+                          nDst == DstDatas::Size() && nDst == DstDescs::Size() &&
+                          nDst == ThreadTransferDstResetCoordinateAfterRunFlags::Size(),
+                      "wrong!");
+
+        static_for<0, nSrc, 1>{}([&](auto i) {
+            static_assert(
+                nDim == remove_cvref_t<tuple_element_t<i.value, SrcDescs>>::GetNumOfDimension(),
+                "wrong!");
+        });
+
+        static_for<0, nDst, 1>{}([&](auto i) {
+            static_assert(
+                nDim == remove_cvref_t<tuple_element_t<i.value, DstDescs>>::GetNumOfDimension(),
+                "wrong!");
+        });
+
+        static_assert(nDim == ThreadClusterLengths::Size() &&
+                          nDim == ThreadClusterArrangeOrder::Size() &&
+                          nDim == SrcDimAccessOrder::Size() && nDim == DstDimAccessOrder::Size(),
+                      "wrong! nDim not consistent");
+
+        static_assert(
+            is_same<SliceLengths, decltype(thread_slice_lengths * ThreadClusterLengths{})>{},
+            "wrong! threads should be mapped to cover entire slicing window");
+
+        static_assert(ThreadGroup::GetNumOfThread() >= thread_cluster_desc_.GetElementSize(),
+                      "wrong! ThreadGroup::GetNumOfThread() too small");
+
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            const auto thread_cluster_idx = thread_cluster_desc_.CalculateBottomIndex(
+                make_multi_index(get_thread_local_1d_id()));
+
+            const auto thread_data_idx_begin = thread_cluster_idx * thread_slice_lengths;
+
+            const auto src_thread_slice_origins = generate_tuple(
+                [&](auto i) { return src_block_slice_origins[i] + thread_data_idx_begin; },
+                Number<nSrc>{});
+
+            const auto dst_thread_slice_origins = generate_tuple(
+                [&](auto i) { return dst_block_slice_origins[i] + thread_data_idx_begin; },
+                Number<nDst>{});
+
+            threadwise_transfer_.SetSrcSliceOrigins(src_descs, src_thread_slice_origins);
+            threadwise_transfer_.SetDstSliceOrigins(dst_descs, dst_thread_slice_origins);
+        }
+    }
+
+    template <typename SrcBuffers>
+    __device__ void RunRead(const SrcDescs& src_descs, const SrcBuffers& src_bufs)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.RunRead(src_descs, src_bufs);
+        }
+    }
+
+    template <typename T>
+    using is_tuple = decltype(std::declval<T&>().IsTuple());
+
+    template <typename DstBuffers>
+    __device__ void RunWrite(const DstDescs& dst_descs, DstBuffers dst_bufs)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            if constexpr(is_detected<is_tuple, decltype(dst_bufs)>::value)
+                threadwise_transfer_.RunWrite(dst_descs, dst_bufs);
+            else
+                threadwise_transfer_.RunWrite(dst_descs, tie(dst_bufs));
+        }
+    }
+
+    template <typename SrcBuffers, typename DstBuffers>
+    __device__ void Run(const SrcDescs& src_descs,
+                        const SrcBuffers& src_bufs,
+                        const DstDescs& dst_descs,
+                        DstBuffers dst_bufs)
+    {
+        RunRead(src_descs, src_bufs);
+        RunWrite(dst_descs, dst_bufs);
+    }
+
+    template <index_t ISrc>
+    __device__ void
+    MoveSrcSliceWindow(const SrcDescs& src_descs, Number<ISrc> iSrc, const Index& step)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.MoveSrcSliceWindow(src_descs, iSrc, step);
+        }
+    }
+
+    __device__ void MoveSrcSliceWindow(const SrcDescs& src_descs, const Index& step)
+    {
+        static_for<0, SrcDescs::Size(), 1>{}(
+            [&](auto i) { MoveSrcSliceWindow(src_descs, i, step); });
+    }
+
+    template <index_t IDst>
+    __device__ void
+    MoveDstSliceWindow(const DstDescs& dst_descs, Number<IDst> iDst, const Index& step)
+    {
+        if(ThreadGroup::GetNumOfThread() == thread_cluster_desc_.GetElementSize() or
+           ThreadGroup::GetThreadId() < thread_cluster_desc_.GetElementSize())
+        {
+            threadwise_transfer_.MoveDstSliceWindow(dst_descs, iDst, step);
+        }
+    }
+
+    __device__ void MoveDstSliceWindow(const DstDescs& dst_descs, const Index& step)
+    {
+        static_for<0, DstDescs::Size(), 1>{}(
+            [&](auto i) { MoveDstSliceWindow(dst_descs, i, step); });
+    }
+
+    private:
+    static constexpr auto thread_cluster_desc_ =
+        make_cluster_descriptor(ThreadClusterLengths{}, ThreadClusterArrangeOrder{});
+
+    using ThreadwiseTransfer =
+        ThreadwiseTensorSliceTransfer_v7r2<SrcDatas,
+                                           DstDatas,
+                                           SrcDescs,
+                                           DstDescs,
+                                           ElementwiseOperation,
+                                           DstInMemOps,
+                                           decltype(thread_slice_lengths),
+                                           SrcDimAccessOrder,
+                                           DstDimAccessOrder,
+                                           SrcVectorDim,
+                                           DstVectorDim,
+                                           SrcScalarPerVector,
+                                           DstScalarPerVector,
+                                           ThreadTransferSrcResetCoordinateAfterRunFlags,
+                                           ThreadTransferDstResetCoordinateAfterRunFlags>;
+
+    ThreadwiseTransfer threadwise_transfer_;
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_multiple_abd.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_abd.hpp
new file mode 100644
index 0000000000..cbb9fadc6d
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_multiple_abd.hpp
@@ -0,0 +1,60 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A0[M, K], B0[K, N],
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <typename AsLayout,
+          typename BsLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename AsDataType,
+          typename BsDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceGemmMultipleABD : public BaseOperator
+{
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
+                        std::array<const void*, NumBTensor> p_bs,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        ck::index_t M,
+                        ck::index_t N,
+                        ck::index_t K,
+                        std::array<ck::index_t, NumATensor> StrideAs,
+                        std::array<ck::index_t, NumBTensor> StrideBs,
+                        std::array<ck::index_t, NumDTensor> StrideDs,
+                        ck::index_t StrideE,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
new file mode 100644
index 0000000000..827a341a50
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp
@@ -0,0 +1,766 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_gemm_multiple_abd.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm,
+          typename AsPointer,
+          typename BsPointer,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AsGridDesc_AK0_M_AK1,
+          typename BsGridDesc_BK0_N_BK1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename Block2ETileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_gemm_multiple_abd_xdl_cshuffle(
+            AsPointer p_as_grid,
+            BsPointer p_bs_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1,
+            const BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const Block2ETileMap block_2_etile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_as_grid,
+                                                  p_bs_grid,
+                                                  p_ds_grid,
+                                                  p_e_grid,
+                                                  p_shared,
+                                                  a_element_op,
+                                                  b_element_op,
+                                                  cde_element_op,
+                                                  as_grid_desc_ak0_m_ak1,
+                                                  bs_grid_desc_bk0_n_bk1,
+                                                  ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  block_2_etile_map);
+#else
+    ignore = p_as_grid;
+    ignore = p_bs_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = as_grid_desc_ak0_m_ak1;
+    ignore = bs_grid_desc_bk0_n_bk1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = block_2_etile_map;
+#endif
+}
+
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <typename AsLayout,
+          typename BsLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename AsDataType,
+          typename BsDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1,
+          index_t BK1,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct DeviceGemmMultipleABD_Xdl_CShuffle : public DeviceGemmMultipleABD<AsLayout,
+                                                                         BsLayout,
+                                                                         DsLayout,
+                                                                         ELayout,
+                                                                         AsDataType,
+                                                                         BsDataType,
+                                                                         DsDataType,
+                                                                         EDataType,
+                                                                         AElementwiseOperation,
+                                                                         BElementwiseOperation,
+                                                                         CDEElementwiseOperation>
+{
+    using DeviceOp = DeviceGemmMultipleABD_Xdl_CShuffle;
+
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+#if 0
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpec, index_t, index_t, index_t>{MPerBlock, NPerBlock, KPerBlock};
+
+    static auto MakeAGridDescriptor_M_K(index_t MRaw, index_t KRaw, index_t StrideAs)
+    {
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, AsLayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(StrideAs, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, AsLayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(I1, StrideAs));
+            }
+        }();
+
+        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+    }
+
+    static auto MakeBGridDescriptor_N_K(index_t KRaw, index_t NRaw, index_t StrideBs)
+    {
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BsLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(I1, StrideBs));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BsLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(StrideBs, I1));
+            }
+        }();
+
+        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+    }
+
+    template <typename ELay>
+    static auto MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
+    {
+        const auto e_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELay>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    static auto MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                                         const std::array<index_t, NumDTensor>& NRaws,
+                                         const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return DeviceOp::MakeEGridDescriptor_M_N<DLayout>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+#endif
+    using ComputeDataType = EDataType;
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemmMultipleABD_xdl_cshuffle<
+        AsDataType,
+        BsDataType,
+        ComputeDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        NumGemmKPrefetchStage,
+        BlockSize,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        AK1,
+        BK1,
+        MPerXDL,
+        NPerXDL,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEBlockTransferScalarPerVector_NPerBlock,
+        LoopSched,
+        PipelineVer>;
+
+    // desc for problem definition
+    using AsGridDesc_M_K =
+        remove_cvref_t<decltype(GridwiseGemm::template MakeAsGridDescriptor_M_K<AsLayout, GemmSpec>(
+            {}, {}, {}))>;
+    using BsGridDesc_N_K =
+        remove_cvref_t<decltype(GridwiseGemm::template MakeBsGridDescriptor_N_K<BsLayout, GemmSpec>(
+            {}, {}, {}))>;
+    using DsGridDesc_M_N =
+        remove_cvref_t<decltype(GridwiseGemm::template MakeDsGridDescriptor_M_N<DsLayout, GemmSpec>(
+            {}, {}, {}))>;
+    using EGridDesc_M_N =
+        decltype(GridwiseGemm::template MakeEGridDescriptor_M_N<ELayout, GemmSpec>(1, 1, 1));
+
+    // desc for blockwise copy
+    using AsGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(AsGridDesc_M_K{}))>;
+    using BsGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(BsGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
+
+    // block-to-e-tile map
+    using Block2ETileMap =
+        remove_cvref_t<decltype(GridwiseGemm::MakeBlock2ETileMap(EGridDesc_M_N{}))>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(std::array<const void*, NumATensor> p_as_grid,
+                 std::array<const void*, NumBTensor> p_bs_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 index_t MRaw,
+                 index_t NRaw,
+                 index_t KRaw,
+                 std::array<index_t, NumATensor> StrideAs,
+                 std::array<index_t, NumBTensor> StrideBs,
+                 std::array<index_t, NumDTensor> StrideDs,
+                 index_t StrideE,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_as_grid_{},
+              p_bs_grid_{},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              as_grid_desc_m_k_{},
+              bs_grid_desc_n_k_{},
+              ds_grid_desc_m_n_{},
+              e_grid_desc_m_n_{GridwiseGemm::template MakeEGridDescriptor_M_N<ELayout, GemmSpec>(
+                  MRaw, NRaw, StrideE)},
+              as_grid_desc_ak0_m_ak1_{},
+              bs_grid_desc_bk0_n_bk1_{},
+              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              e_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              block_2_etile_map_{GridwiseGemm::MakeBlock2ETileMap(e_grid_desc_m_n_)},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op},
+              MRaw_{MRaw},
+              NRaw_{NRaw},
+              KRaw_{KRaw}
+        {
+            // populate pointer, desc for As
+            static_for<0, NumATensor, 1>{}([&](auto i) {
+                using ALayout   = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
+                using ADataType = remove_cvref_t<tuple_element_t<i.value, AsDataType>>;
+
+                // A pointer
+                p_as_grid_(i) = static_cast<const ADataType*>(p_as_grid[i]);
+
+                // A desc
+                as_grid_desc_m_k_(i) =
+                    GridwiseGemm::template MakeAGridDescriptor_M_K<ALayout, GemmSpec>(
+                        MRaw, KRaw, StrideAs[i]);
+            });
+
+            // populate pointer, desc for Bs
+            static_for<0, NumBTensor, 1>{}([&](auto i) {
+                using BLayout   = remove_cvref_t<tuple_element_t<i.value, BsLayout>>;
+                using BDataType = remove_cvref_t<tuple_element_t<i.value, BsDataType>>;
+
+                // B pointer
+                p_bs_grid_(i) = static_cast<const BDataType*>(p_bs_grid[i]);
+
+                // B desc
+                bs_grid_desc_n_k_(i) =
+                    GridwiseGemm::template MakeBGridDescriptor_N_K<BLayout, GemmSpec>(
+                        KRaw, NRaw, StrideBs[i]);
+            });
+
+            // populate pointer, desc for Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    GridwiseGemm::template MakeEGridDescriptor_M_N<DLayout, GemmSpec>(
+                        MRaw, NRaw, StrideDs[i]);
+            });
+
+            // populate desc for Ds/E
+            if(GridwiseGemm::CheckValidity(as_grid_desc_m_k_,
+                                           bs_grid_desc_n_k_,
+                                           ds_grid_desc_m_n_,
+                                           e_grid_desc_m_n_,
+                                           block_2_etile_map_))
+            {
+                as_grid_desc_ak0_m_ak1_ =
+                    GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_);
+
+                bs_grid_desc_bk0_n_bk1_ =
+                    GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_);
+
+                ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        ds_grid_desc_m_n_);
+
+                e_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        e_grid_desc_m_n_);
+            }
+        }
+
+        void Print() const
+        {
+            // std::cout << "A[M, K]: " << as_grid_desc_m_k_ << std::endl;
+            // std::cout << "B[N, K]: " << bs_grid_desc_n_k_ << std::endl;
+            // static_for<0, NumDTensor, 1>{}(
+            //[&](auto i) { std::cout << "Ds[M, N]: " << ds_grid_desc_m_n_[i] << std::endl; });
+            // std::cout << "E[M, N]: " << e_grid_desc_m_n_ << std::endl;
+        }
+
+        //  private:
+        // pointers
+        typename GridwiseGemm::AsGridPointer p_as_grid_;
+        typename GridwiseGemm::BsGridPointer p_bs_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // tensor descriptors for problem definiton
+        AsGridDesc_M_K as_grid_desc_m_k_;
+        BsGridDesc_N_K bs_grid_desc_n_k_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        // tensor descriptors for block/thread-wise copy
+        AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1_;
+        BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1_;
+        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
+        EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock e_grid_desc_mblock_mperblock_nblock_nperblock_;
+
+        // block-to-e-tile map
+        Block2ETileMap block_2_etile_map_;
+
+        // element-wise op
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        // for checking vector load/store
+        index_t MRaw_;
+        index_t NRaw_;
+        index_t KRaw_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(!GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
+                                            arg.bs_grid_desc_n_k_,
+                                            arg.ds_grid_desc_m_n_,
+                                            arg.e_grid_desc_m_n_,
+                                            arg.block_2_etile_map_))
+            {
+                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_gemm_multiple_abd_xdl_cshuffle<
+                    GridwiseGemm,
+                    typename GridwiseGemm::AsGridPointer,
+                    typename GridwiseGemm::BsGridPointer,
+                    typename GridwiseGemm::DsGridPointer,
+                    EDataType,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    DeviceOp::AsGridDesc_AK0_M_AK1,
+                    DeviceOp::BsGridDesc_BK0_N_BK1,
+                    DeviceOp::DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    DeviceOp::EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    DeviceOp::Block2ETileMap,
+                    has_main_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_as_grid_,
+                                              arg.p_bs_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.as_grid_desc_ak0_m_ak1_,
+                                              arg.bs_grid_desc_bk0_n_bk1_,
+                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.block_2_etile_map_);
+            };
+
+            const auto K = arg.as_grid_desc_m_k_[I0].GetLength(I1);
+
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
+        // check vector load/store
+        {
+            using Row = ck::tensor_layout::gemm::RowMajor;
+            using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+            bool all_valid = true;
+
+            static_for<0, NumATensor, 1>{}([&](auto i) {
+                using ALayout = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
+                // check vector load of A
+                if constexpr(is_same_v<ALayout, Row> && ABlockTransferSrcVectorDim == 2)
+                {
+                    if(arg.KRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                    {
+                        all_valid = false;
+                    }
+                }
+                else if constexpr(is_same_v<ALayout, Col> && ABlockTransferSrcVectorDim == 1)
+                {
+                    // FIXME: not rigorous
+                    if(arg.MRaw_ % ABlockTransferSrcScalarPerVector != 0)
+                    {
+                        all_valid = false;
+                    }
+                }
+                else
+                {
+                    all_valid = false;
+                }
+            });
+
+            static_for<0, NumBTensor, 1>{}([&](auto i) {
+                using BLayout = remove_cvref_t<tuple_element_t<i.value, BsLayout>>;
+                // check vector laod of B
+                if constexpr(is_same_v<BLayout, Col> && BBlockTransferSrcVectorDim == 2)
+                {
+                    if(arg.KRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                    {
+                        all_valid = false;
+                    }
+                }
+                else if constexpr(is_same_v<BLayout, Row> && BBlockTransferSrcVectorDim == 1)
+                {
+                    // FIXME: not rigorous
+                    if(arg.NRaw_ % BBlockTransferSrcScalarPerVector != 0)
+                    {
+                        all_valid = false;
+                    }
+                }
+                else
+                {
+                    all_valid = false;
+                }
+            });
+
+            // check vector load of Ds
+            // only support RowMajor for now
+
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                if constexpr(!is_same_v<DLayout, Row>)
+                {
+                    all_valid = false;
+                }
+            });
+
+            if(!all_valid)
+            {
+                return false;
+            }
+
+            // check vector store of E
+            // only support RowMajor for now
+            if constexpr(is_same_v<ELayout, Row>)
+            {
+                if(arg.NRaw_ % CDEBlockTransferScalarPerVector_NPerBlock != 0)
+                {
+                    return false;
+                }
+            }
+            else
+            {
+                return false;
+            }
+        }
+
+        return GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
+                                           arg.bs_grid_desc_n_k_,
+                                           arg.ds_grid_desc_m_n_,
+                                           arg.e_grid_desc_m_n_,
+                                           arg.block_2_etile_map_);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::array<const void*, NumATensor> p_as,
+                             std::array<const void*, NumBTensor> p_bs,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             index_t MRaw,
+                             index_t NRaw,
+                             index_t KRaw,
+                             std::array<index_t, NumATensor> StrideAs,
+                             std::array<index_t, NumBTensor> StrideBs,
+                             std::array<index_t, NumDTensor> StrideDs,
+                             index_t StrideE,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_as,
+                        p_bs,
+                        p_ds,
+                        p_e,
+                        MRaw,
+                        NRaw,
+                        KRaw,
+                        StrideAs,
+                        StrideBs,
+                        StrideDs,
+                        StrideE,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
+                        std::array<const void*, NumBTensor> p_bs,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        index_t MRaw,
+                        index_t NRaw,
+                        index_t KRaw,
+                        std::array<ck::index_t, NumATensor> StrideAs,
+                        std::array<ck::index_t, NumBTensor> StrideBs,
+                        std::array<ck::index_t, NumDTensor> StrideDs,
+                        index_t StrideE,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_as,
+                                          p_bs,
+                                          p_ds,
+                                          p_e,
+                                          MRaw,
+                                          NRaw,
+                                          KRaw,
+                                          StrideAs,
+                                          StrideBs,
+                                          StrideDs,
+                                          StrideE,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceGemmMultipleABD_Xdl_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
+            << ">"
+            << " LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
new file mode 100644
index 0000000000..d98cbea166
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
@@ -0,0 +1,1033 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/multi_index_transform_helper.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp"
+#include "ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp"
+#include "ck/tensor_operation/gpu/block/thread_group_tensor_slice_transfer_v7r2.hpp"
+#include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+
+namespace ck {
+
+// GEMM:
+//   input : A0[M, K], A1[M, K]
+//   input : B0[N, K], B1[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <typename AsDataType,
+          typename BsDataType,
+          typename ComputeDataType_,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          InMemoryDataOperationEnum EGlobalMemoryDataOperation,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1Value,
+          index_t BK1Value,
+          index_t MPerXdl,
+          index_t NPerXdl,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool AThreadTransferSrcResetCoordinateAfterRun,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BThreadTransferSrcResetCoordinateAfterRun,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched,
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct GridwiseGemmMultipleABD_xdl_cshuffle
+{
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    using GemmSpecialization = ck::tensor_operation::device::GemmSpecialization;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+    static constexpr auto I6 = Number<6>{};
+    static constexpr auto I7 = Number<7>{};
+
+    // K1 should be Number<...>
+    static constexpr auto AK1         = Number<AK1Value>{};
+    static constexpr auto BK1         = Number<BK1Value>{};
+    static constexpr auto AK0PerBlock = Number<KPerBlock / AK1Value>{};
+    static constexpr auto BK0PerBlock = Number<KPerBlock / BK1Value>{};
+
+    using ThisThreadBlock = ThisThreadBlock<BlockSize>;
+
+    using GridwiseGemmPipe = remove_cvref_t<
+        decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
+
+#if CK_WORKAROUND_DENORM_FIX
+    using ComputeDataType =
+        conditional_t<is_same_v<ComputeDataType_, ck::half_t>, ck::bhalf_t, ComputeDataType_>;
+#else
+    using ComputeDataType = ComputeDataType_;
+#endif
+
+    __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
+    {
+        // A matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(AK0PerBlock, Number<MPerBlock>{}, AK1),
+            make_tuple(Number<MPerBlock + ABlockLdsExtraM>{} * AK1, AK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1()
+    {
+        // B matrix in LDS memory, dst of blockwise copy
+        return make_naive_tensor_descriptor(
+            make_tuple(BK0PerBlock, Number<NPerBlock>{}, BK1),
+            make_tuple(Number<NPerBlock + BBlockLdsExtraN>{} * BK1, BK1, I1));
+    }
+
+    __host__ __device__ static constexpr auto
+    GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
+    {
+        constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+        constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            make_naive_tensor_descriptor_packed(
+                make_tuple(I1,
+                           Number<CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl>{},
+                           I1,
+                           Number<CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>{}));
+
+        return c_shuffle_block_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    static constexpr auto MakeAsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using ADataType = remove_cvref_t<tuple_element_t<i.value, AsDataType>>;
+
+                return static_cast<const ADataType*>(nullptr);
+            },
+            Number<NumATensor>{});
+    }
+
+    static constexpr auto MakeBsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using BDataType = remove_cvref_t<tuple_element_t<i.value, BsDataType>>;
+
+                return static_cast<const BDataType*>(nullptr);
+            },
+            Number<NumBTensor>{});
+    }
+
+    // ck::Tuple<const D0DataType*, const D1DataType*, ...>
+    static constexpr auto MakeDsGridPointer()
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                return static_cast<const DDataType*>(nullptr);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __host__ __device__ static constexpr index_t GetSharedMemoryNumberOfByte()
+    {
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        constexpr auto b_block_space_size_aligned = math::integer_least_multiple(
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize(), max_lds_align);
+
+        // LDS allocation for C shuffle in LDS
+        constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+            GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+        constexpr auto c_block_size =
+            c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
+
+        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
+                             sizeof(ComputeDataType),
+                         c_block_size * sizeof(CShuffleDataType));
+    }
+
+    // A desc for source in blockwise copy
+    template <typename AGridDesc_M_K>
+    __host__ __device__ static constexpr auto
+    MakeAGridDescriptor_AK0_M_AK1(const AGridDesc_M_K& a_grid_desc_m_k)
+    {
+        const auto M = a_grid_desc_m_k.GetLength(I0);
+        const auto K = a_grid_desc_m_k.GetLength(I1);
+
+        const auto AK0 = K / AK1;
+
+        return transform_tensor_descriptor(a_grid_desc_m_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(AK0, AK1)),
+                                                      make_pass_through_transform(M)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    template <typename AsGridDesc_M_K>
+    __host__ __device__ static constexpr auto
+    MakeAsGridDescriptor_AK0_M_AK1(const AsGridDesc_M_K& as_grid_desc_m_k)
+    {
+        return generate_tuple(
+            [&](auto i) { return MakeAGridDescriptor_AK0_M_AK1(as_grid_desc_m_k[i]); },
+            Number<NumATensor>{});
+    }
+
+    // B desc for source in blockwise copy
+    template <typename BGridDesc_N_K>
+    __host__ __device__ static constexpr auto
+    MakeBGridDescriptor_BK0_N_BK1(const BGridDesc_N_K& b_grid_desc_n_k)
+    {
+        const auto N = b_grid_desc_n_k.GetLength(I0);
+        const auto K = b_grid_desc_n_k.GetLength(I1);
+
+        const auto BK0 = K / BK1;
+
+        return transform_tensor_descriptor(b_grid_desc_n_k,
+                                           make_tuple(make_unmerge_transform(make_tuple(BK0, BK1)),
+                                                      make_pass_through_transform(N)),
+                                           make_tuple(Sequence<1>{}, Sequence<0>{}),
+                                           make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+    }
+
+    template <typename BsGridDesc_N_K>
+    __host__ __device__ static constexpr auto
+    MakeBsGridDescriptor_BK0_N_BK1(const BsGridDesc_N_K& bs_grid_desc_n_k)
+    {
+        return generate_tuple(
+            [&](auto i) { return MakeBGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k[i]); },
+            Number<NumBTensor>{});
+    }
+
+    // E desc for destination in blockwise copy
+    template <typename EGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        const auto M = e_grid_desc_m_n.GetLength(I0);
+        const auto N = e_grid_desc_m_n.GetLength(I1);
+
+        const auto MBlock = M / MPerBlock;
+        const auto NBlock = N / NPerBlock;
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock = transform_tensor_descriptor(
+            e_grid_desc_m_n,
+            make_tuple(make_unmerge_transform(make_tuple(MBlock, Number<MPerBlock>{})),
+                       make_unmerge_transform(make_tuple(NBlock, Number<NPerBlock>{}))),
+            make_tuple(Sequence<0>{}, Sequence<1>{}),
+            make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
+
+        return e_grid_desc_mblock_mperblock_nblock_nperblock;
+    }
+
+    // Ds desc for source in blockwise copy
+    template <typename DsGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(const DsGridDesc_M_N& ds_grid_desc_m_n)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // return block_id to E matrix tile idx (m0, n0) mapping
+    template <typename EGridDesc_M_N>
+    __host__ __device__ static constexpr auto
+    MakeBlock2ETileMap(const EGridDesc_M_N& e_grid_desc_m_n)
+    {
+        return BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, NPerBlock, EGridDesc_M_N>(
+            e_grid_desc_m_n);
+    }
+
+    // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    template <typename AsGridDesc_M_K,
+              typename BsGridDesc_N_K,
+              typename DsGridDesc_M_N,
+              typename EGridDesc_M_N,
+              typename Block2ETileMap>
+    __host__ __device__ static constexpr bool CheckValidity(const AsGridDesc_M_K& as_grid_desc_m_k,
+                                                            const BsGridDesc_N_K& bs_grid_desc_n_k,
+                                                            const DsGridDesc_M_N& ds_grid_desc_m_n,
+                                                            const EGridDesc_M_N& e_grid_desc_m_n,
+                                                            const Block2ETileMap& block_2_etile_map)
+    {
+        static_assert((MPerBlock % (MPerXdl * MXdlPerWave) == 0) &&
+                          (NPerBlock % (NXdlPerWave * NPerXdl)) == 0,
+                      "Invalid tuning param!");
+        static_assert(KPerBlock % AK1Value == 0 && KPerBlock % BK1Value == 0,
+                      "KPerBlock must be divisible by AK1Value and BK1Value!");
+
+        const auto M  = as_grid_desc_m_k[I0].GetLength(I0);
+        const auto N  = bs_grid_desc_n_k[I0].GetLength(I0);
+        const auto AK = as_grid_desc_m_k[I0].GetLength(I1);
+        const auto BK = bs_grid_desc_n_k[I0].GetLength(I1);
+
+        // check consistency of desc
+        if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1) && AK == BK))
+        {
+            return false;
+        }
+
+        constexpr long_index_t TwoGB = (long_index_t{1} << 31);
+
+        bool valid = true;
+        static_for<0, NumATensor, 1>{}([&](auto i) {
+            using ADataType = remove_cvref_t<tuple_element_t<i.value, AsDataType>>;
+            valid =
+                valid && (as_grid_desc_m_k[i].GetElementSpaceSize() * sizeof(ADataType) <= TwoGB);
+            valid = valid && (M == as_grid_desc_m_k[i].GetLength(I0) &&
+                              AK == as_grid_desc_m_k[i].GetLength(I1));
+        });
+
+        static_for<0, NumBTensor, 1>{}([&](auto i) {
+            using BDataType = remove_cvref_t<tuple_element_t<i.value, BsDataType>>;
+            valid =
+                valid && (bs_grid_desc_n_k[i].GetElementSpaceSize() * sizeof(BDataType) <= TwoGB);
+            valid = valid && (N == bs_grid_desc_n_k[i].GetLength(I0) &&
+                              BK == bs_grid_desc_n_k[i].GetLength(I1));
+        });
+
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            valid = valid && (M == ds_grid_desc_m_n[i].GetLength(I0) &&
+                              N == ds_grid_desc_m_n[i].GetLength(I1));
+        });
+
+        if(!valid)
+        {
+            return false;
+        }
+
+        // check tile size
+        if(!(M % MPerBlock == 0 && N % NPerBlock == 0 && AK % KPerBlock == 0))
+        {
+            return false;
+        }
+
+        // check gridwise gemm pipeline
+        const auto num_k_loop = AK / KPerBlock;
+
+        if(!GridwiseGemmPipe::IsSupported(num_k_loop))
+        {
+            return false;
+        }
+
+        // check block-to-E-tile
+        if(!block_2_etile_map.CheckValidity(e_grid_desc_m_n))
+        {
+            return false;
+        }
+
+        // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
+        // check tensor size: cannot be larger than 2GB each
+
+        if(!(e_grid_desc_m_n.GetElementSpaceSize() * sizeof(EDataType) <= TwoGB))
+        {
+            return false;
+        }
+
+        return true;
+    }
+
+    __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
+    {
+        const index_t num_loop = K / KPerBlock;
+
+        return GridwiseGemmPipe::CalculateHasMainLoop(num_loop);
+    }
+
+    using AsGridPointer = decltype(MakeAsGridPointer());
+    using BsGridPointer = decltype(MakeBsGridPointer());
+    using DsGridPointer = decltype(MakeDsGridPointer());
+
+    template <typename ALayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeAGridDescriptor_M_K(index_t MRaw, index_t KRaw, index_t StrideA)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+
+        const auto a_grid_desc_mraw_kraw = [&]() {
+            if constexpr(is_same_v<tensor_layout::gemm::RowMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(StrideA, I1));
+            }
+            else if constexpr(is_same_v<tensor_layout::gemm::ColumnMajor, ALayout>)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, KRaw),
+                                                    make_tuple(I1, StrideA));
+            }
+        }();
+
+        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+    }
+
+    template <typename AsLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeAsGridDescriptor_M_K(const std::array<index_t, NumATensor>& MRaws,
+                             const std::array<index_t, NumATensor>& KRaws,
+                             const std::array<index_t, NumATensor>& AsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using ALayout = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
+
+                return MakeEGridDescriptor_M_N<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
+            },
+            Number<NumATensor>{});
+    }
+
+    template <typename BLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeBGridDescriptor_N_K(index_t KRaw, index_t NRaw, index_t StrideB)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+
+        const auto b_grid_desc_nraw_kraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(I1, StrideB));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(NRaw, KRaw),
+                                                    make_tuple(StrideB, I1));
+            }
+        }();
+
+        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+    }
+
+    template <typename BsLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeBsGridDescriptor_N_K(const std::array<index_t, NumBTensor>& KRaws,
+                             const std::array<index_t, NumBTensor>& NRaws,
+                             const std::array<index_t, NumBTensor>& BsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using BLayout = remove_cvref_t<tuple_element_t<i.value, BsLayout>>;
+
+                return MakeBGridDescriptor_N_K<BLayout, GemmSpec>(KRaws[i], NRaws[i], BsStride[i]);
+            },
+            Number<NumBTensor>{});
+    }
+
+    template <typename ELayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeEGridDescriptor_M_N(index_t MRaw, index_t NRaw, index_t StrideE)
+    {
+        constexpr auto matrix_padder =
+            ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+                MPerBlock, NPerBlock, KPerBlock};
+        const auto e_grid_desc_mraw_nraw = [&]() {
+            if constexpr(is_same<tensor_layout::gemm::RowMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(StrideE, I1));
+            }
+            else if constexpr(is_same<tensor_layout::gemm::ColumnMajor, ELayout>::value)
+            {
+                return make_naive_tensor_descriptor(make_tuple(MRaw, NRaw),
+                                                    make_tuple(I1, StrideE));
+            }
+        }();
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    template <typename DsLayout, GemmSpecialization GemmSpec>
+    __host__ __device__ static auto
+    MakeDsGridDescriptor_M_N(const std::array<index_t, NumDTensor>& MRaws,
+                             const std::array<index_t, NumDTensor>& NRaws,
+                             const std::array<index_t, NumDTensor>& DsStride)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return MakeEGridDescriptor_M_N<DLayout, GemmSpec>(MRaws[i], NRaws[i], DsStride[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    __device__ __host__ static constexpr auto GetMPerBlock() { return MPerBlock; }
+
+    template <bool HasMainKBlockLoop,
+              typename AsGridDesc_AK0_M_AK1,
+              typename BsGridDesc_BK0_N_BK1,
+              typename DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+              typename Block2ETileMap>
+    __device__ static void Run(AsGridPointer p_as_grid,
+                               BsGridPointer p_bs_grid,
+                               DsGridPointer p_ds_grid,
+                               EDataType* __restrict__ p_e_grid,
+                               void* __restrict__ p_shared,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation& cde_element_op,
+                               const AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1,
+                               const BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1,
+                               const DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock&
+                                   e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        const auto as_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_as_grid[i], as_grid_desc_ak0_m_ak1[i].GetElementSpaceSize());
+            },
+            Number<NumATensor>{});
+
+        const auto bs_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_bs_grid[i], bs_grid_desc_bk0_n_bk1[i].GetElementSpaceSize());
+            },
+            Number<NumBTensor>{});
+
+        const auto ds_grid_buf = generate_tuple(
+            [&](auto i) {
+                return make_dynamic_buffer<AddressSpaceEnum::Global>(
+                    p_ds_grid[i],
+                    ds_grid_desc_mblock_mperblock_nblock_nperblock[i].GetElementSpaceSize());
+            },
+            Number<NumDTensor>{});
+
+        auto e_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_e_grid, e_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+        // divide block work by [M, N]
+        const auto block_work_idx =
+            block_2_etile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+
+        if(!block_2_etile_map.ValidCTileIndex(
+               block_work_idx,
+               make_tuple(e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I0),
+                          e_grid_desc_mblock_mperblock_nblock_nperblock.GetLength(I2))))
+        {
+            return;
+        }
+
+        // HACK: this force m/n_block_data_idx_on_grid into SGPR
+        const index_t m_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I0] * MPerBlock);
+
+        const index_t n_block_data_idx_on_grid =
+            __builtin_amdgcn_readfirstlane(block_work_idx[I1] * NPerBlock);
+
+        // lds max alignment
+        constexpr auto max_lds_align = math::lcm(AK1, BK1);
+
+        // A matrix in LDS memory, dst of blockwise copy
+        constexpr auto a_block_desc_ak0_m_ak1 = GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1();
+
+        // B matrix in LDS memory, dst of blockwise copy
+        constexpr auto b_block_desc_bk0_n_bk1 = GetBBlockDescriptor_BK0PerBlock_NPerBlock_BK1();
+
+        const auto idx_as_block_begin =
+            generate_tuple([&](auto) { return make_multi_index(0, m_block_data_idx_on_grid, 0); },
+                           Number<NumATensor>{});
+
+        auto a_blockwise_copy = ThreadGroupTensorSliceTransfer_v7r2<
+            ThisThreadBlock,
+            AsDataType,
+            Tuple<ComputeDataType>,
+            decltype(as_grid_desc_ak0_m_ak1),
+            decltype(tie(a_block_desc_ak0_m_ak1)),
+            AElementwiseOperation,
+            Sequence<static_cast<index_t>(InMemoryDataOperationEnum::Set)>,
+            Sequence<AK0PerBlock, MPerBlock, AK1>,
+            ABlockTransferThreadClusterLengths_AK0_M_AK1,
+            ABlockTransferThreadClusterArrangeOrder,
+            ABlockTransferSrcAccessOrder,
+            Sequence<1, 0, 2>,
+            ABlockTransferSrcVectorDim,
+            2,
+            ABlockTransferSrcScalarPerVector,
+            ABlockTransferDstScalarPerVector_AK1,
+            uniform_sequence_gen_t<NumATensor, false>,
+            Sequence<true>>{as_grid_desc_ak0_m_ak1,
+                            idx_as_block_begin,
+                            tie(a_block_desc_ak0_m_ak1),
+                            make_tuple(make_multi_index(0, 0, 0)),
+                            a_element_op};
+
+        const auto idx_bs_block_begin =
+            generate_tuple([&](auto) { return make_multi_index(0, n_block_data_idx_on_grid, 0); },
+                           Number<NumBTensor>{});
+
+        auto b_blockwise_copy = ThreadGroupTensorSliceTransfer_v7r2<
+            ThisThreadBlock,
+            BsDataType,
+            Tuple<ComputeDataType>,
+            decltype(bs_grid_desc_bk0_n_bk1),
+            decltype(tie(b_block_desc_bk0_n_bk1)),
+            BElementwiseOperation,
+            Sequence<static_cast<index_t>(InMemoryDataOperationEnum::Set)>,
+            Sequence<BK0PerBlock, NPerBlock, BK1>,
+            BBlockTransferThreadClusterLengths_BK0_N_BK1,
+            BBlockTransferThreadClusterArrangeOrder,
+            BBlockTransferSrcAccessOrder,
+            Sequence<1, 0, 2>,
+            BBlockTransferSrcVectorDim,
+            2,
+            BBlockTransferSrcScalarPerVector,
+            BBlockTransferDstScalarPerVector_BK1,
+            uniform_sequence_gen_t<NumBTensor, false>,
+            Sequence<true>>{bs_grid_desc_bk0_n_bk1,
+                            idx_bs_block_begin,
+                            tie(b_block_desc_bk0_n_bk1),
+                            make_tuple(make_multi_index(0, 0, 0)),
+                            b_element_op};
+
+        // GEMM definition
+        //   c_mtx += transpose(a_mtx) * b_mtx
+        //     a_mtx[K0PerBlock, MPerBlock] is in LDS
+        //     b_mtx[K0PerBlock, NPerBlock] is in LDS
+        //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
+        //       register
+        // sanity check
+        constexpr index_t KPack =
+            math::max(math::lcm(AK1, BK1),
+                      MfmaSelector<ComputeDataType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+
+        auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
+            BlockSize,
+            ComputeDataType,
+            AccDataType,
+            decltype(a_block_desc_ak0_m_ak1),
+            decltype(b_block_desc_bk0_n_bk1),
+            MPerXdl,
+            NPerXdl,
+            MXdlPerWave,
+            NXdlPerWave,
+            KPack,
+            LoopSched>();
+
+        auto c_thread_buf = blockwise_gemm.GetCThreadBuffer();
+
+        // LDS allocation for A and B: be careful of alignment
+        constexpr auto a_block_space_size_aligned = math::integer_least_multiple(
+            a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
+
+        auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ComputeDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+
+        auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+            static_cast<ComputeDataType*>(p_shared) + a_block_space_size_aligned,
+            b_block_desc_bk0_n_bk1.GetElementSpaceSize());
+
+        constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
+        constexpr auto b_block_slice_copy_step = make_multi_index(KPerBlock / BK1, 0, 0);
+
+        const index_t num_k_block_main_loop = __builtin_amdgcn_readfirstlane(
+            (as_grid_desc_ak0_m_ak1[I0].GetLength(I0) * as_grid_desc_ak0_m_ak1[I0].GetLength(I2)) /
+            KPerBlock);
+
+        // gridwise GEMM pipeline
+        const auto gridwise_gemm_pipeline =
+            GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>();
+
+        gridwise_gemm_pipeline.template Run<HasMainKBlockLoop>(as_grid_desc_ak0_m_ak1,
+                                                               a_block_desc_ak0_m_ak1,
+                                                               a_blockwise_copy,
+                                                               as_grid_buf,
+                                                               a_block_buf,
+                                                               a_block_slice_copy_step,
+                                                               bs_grid_desc_bk0_n_bk1,
+                                                               b_block_desc_bk0_n_bk1,
+                                                               b_blockwise_copy,
+                                                               bs_grid_buf,
+                                                               b_block_buf,
+                                                               b_block_slice_copy_step,
+                                                               blockwise_gemm,
+                                                               c_thread_buf,
+                                                               num_k_block_main_loop);
+
+        // shuffle C and write out
+        {
+            static_assert(MXdlPerWave % CShuffleMXdlPerWavePerShuffle == 0 &&
+                              NXdlPerWave % CShuffleNXdlPerWavePerShuffle == 0,
+                          "wrong!");
+
+            constexpr index_t MWave = MPerBlock / (MXdlPerWave * MPerXdl);
+            constexpr index_t NWave = NPerBlock / (NXdlPerWave * NPerXdl);
+
+            // TODO: hacky, fix it!
+            constexpr auto c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2 =
+                blockwise_gemm.GetCThreadDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            // TODO: hacky, fix it!
+            // c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp is only used to get lengths
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp =
+                blockwise_gemm.GetCBlockDescriptor_M0_N0_M1_N1_M2_M3_M4_N2();
+
+            constexpr auto M0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I0);
+            constexpr auto N0 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I1);
+            constexpr auto M1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I2);
+            constexpr auto N1 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I3);
+            constexpr auto M2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I4);
+            constexpr auto M3 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I5);
+            constexpr auto M4 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I6);
+            constexpr auto N2 = c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2_tmp.GetLength(I7);
+
+            constexpr auto c_shuffle_block_desc_mblock_mperblock_nblock_nperblock =
+                GetCShuffleBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock();
+
+            auto c_shuffle_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
+                static_cast<CShuffleDataType*>(p_shared),
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
+
+            constexpr auto c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2 = transform_tensor_descriptor(
+                c_shuffle_block_desc_mblock_mperblock_nblock_nperblock,
+                make_tuple(
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleMXdlPerWavePerShuffle>{}, // M0 (MXdlPerWave) per shuffle
+                        M1,                                      // M1 = MWave
+                        M2,                                      // M2 * M3 * M4 = MPerXdl
+                        M3,
+                        M4)),
+                    make_freeze_transform(I0),
+                    make_unmerge_transform(make_tuple(
+                        Number<CShuffleNXdlPerWavePerShuffle>{}, // N0 (NXdlPerWave) per shuffle
+                        N1,                                      // N1 = NWave
+                        N2))),                                   // N2 = NPerXdl
+                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
+                make_tuple(
+                    Sequence<>{}, Sequence<0, 2, 4, 5, 6>{}, Sequence<>{}, Sequence<1, 3, 7>{}));
+
+            // calculate origin of thread output tensor on global memory
+            //     blockwise GEMM c matrix starting index
+            const auto c_thread_mtx_on_block =
+                blockwise_gemm.CalculateCThreadOriginDataIndex(I0, I0, I0, I0);
+
+            const index_t m_thread_data_on_block = c_thread_mtx_on_block[I0];
+            const index_t n_thread_data_on_block = c_thread_mtx_on_block[I1];
+
+            const auto m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(M0, M1, M2, M3, M4))),
+                    make_tuple(Sequence<0, 1, 2, 3, 4>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto m_thread_data_on_block_idx =
+                m_thread_data_on_block_to_m0_m1_m2_m3_m4_adaptor.CalculateBottomIndex(
+                    make_multi_index(m_thread_data_on_block));
+
+            const auto n_thread_data_on_block_to_n0_n1_n2_adaptor =
+                make_single_stage_tensor_adaptor(
+                    make_tuple(make_merge_transform(make_tuple(N0, N1, N2))),
+                    make_tuple(Sequence<0, 1, 2>{}),
+                    make_tuple(Sequence<0>{}));
+
+            const auto n_thread_data_on_block_idx =
+                n_thread_data_on_block_to_n0_n1_n2_adaptor.CalculateBottomIndex(
+                    make_multi_index(n_thread_data_on_block));
+
+            // shuffle: threadwise copy C from VGPR to LDS
+            auto c_thread_copy_vgpr_to_lds =
+                ThreadwiseTensorSliceTransfer_v1r3<AccDataType,
+                                                   CShuffleDataType,
+                                                   decltype(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   decltype(c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2),
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   Sequence<CShuffleMXdlPerWavePerShuffle,
+                                                            CShuffleNXdlPerWavePerShuffle,
+                                                            I1,
+                                                            I1,
+                                                            M2,
+                                                            I1,
+                                                            M4,
+                                                            I1>,
+                                                   Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                                   7,
+                                                   1,
+                                                   InMemoryDataOperationEnum::Set,
+                                                   1,
+                                                   true>{
+                    c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                    make_multi_index(0,
+                                     0,
+                                     m_thread_data_on_block_idx[I1],
+                                     n_thread_data_on_block_idx[I1],
+                                     m_thread_data_on_block_idx[I2],
+                                     m_thread_data_on_block_idx[I3],
+                                     m_thread_data_on_block_idx[I4],
+                                     n_thread_data_on_block_idx[I2]),
+                    ck::tensor_operation::element_wise::PassThrough{}};
+
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_desc_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_desc_mblock_mperblock_nblock_nperblock),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_desc_mblock_mperblock_nblock_nperblock[i]; },
+                    Number<NumDTensor>{}));
+
+            // tuple of reference to C/Ds tensor descriptors
+            const auto c_ds_buf_refs = concat_tuple_of_reference(
+                tie(c_shuffle_block_buf),
+                generate_tie(
+                    [&](auto i) -> const auto& // return type should be reference
+                    { return ds_grid_buf[i]; },
+                    Number<NumDTensor>{}));
+
+            // tuple of starting index of C/Ds blockwise copy
+            const auto idx_c_ds_block_begin = container_concat(
+                make_tuple(make_multi_index(0, 0, 0, 0)),
+                generate_tuple(
+                    [&](auto) {
+                        return make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0);
+                    },
+                    Number<NumDTensor>{}));
+
+            // blockwise copy C/D/E between LDS and global
+            auto cde_block_copy_lds_and_global = ThreadGroupTensorSliceTransfer_v7r2<
+                ThisThreadBlock,
+                decltype(container_concat(make_tuple(CShuffleDataType{}), DsDataType{})),
+                Tuple<EDataType>,
+                decltype(c_ds_desc_refs),
+                decltype(tie(e_grid_desc_mblock_mperblock_nblock_nperblock)),
+                CDEElementwiseOperation,
+                Sequence<static_cast<index_t>(EGlobalMemoryDataOperation)>, // FIXME: make Sequence
+                                                                            // support arbitray type
+                Sequence<1,
+                         CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                         1,
+                         CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>, // BlockSliceLengths,
+                CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+                Sequence<0, 1, 2, 3>, // typename ThreadClusterArrangeOrder,
+                Sequence<0, 1, 2, 3>, // typename SrcDimAccessOrder,
+                Sequence<0, 1, 2, 3>, // typename DstDimAccessOrder,
+                3,                    // index_t SrcVectorDim,
+                3,                    // index_t DstVectorDim,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+                CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+                sequence_merge_t<
+                    Sequence<true>,
+                    uniform_sequence_gen_t<NumDTensor,
+                                           false>>, // ThreadTransferSrcResetCoordinateAfterRunFlags
+                Sequence<false>>                    // ThreadTransferDstResetCoordinateAfterRunFlags
+                {c_ds_desc_refs,
+                 idx_c_ds_block_begin,
+                 tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                 make_tuple(make_multi_index(block_work_idx[I0], 0, block_work_idx[I1], 0)),
+                 cde_element_op};
+
+            // space filling curve for threadwise C in VGPR before shuffle
+            constexpr auto sfc_c_vgpr =
+                SpaceFillingCurve<Sequence<MXdlPerWave, NXdlPerWave, 1, 1, M2, 1, M4, 1>,
+                                  Sequence<0, 1, 2, 3, 4, 5, 6, 7>,
+                                  Sequence<CShuffleMXdlPerWavePerShuffle,
+                                           CShuffleNXdlPerWavePerShuffle,
+                                           1,
+                                           1,
+                                           M2,
+                                           1,
+                                           M4,
+                                           1>>{};
+
+            // space filling curve for shuffled blockwise C/D/E
+            constexpr auto sfc_cde_block =
+                SpaceFillingCurve<Sequence<1, MPerBlock, 1, NPerBlock>,
+                                  Sequence<0, 2, 1, 3>,
+                                  Sequence<1,
+                                           CShuffleMXdlPerWavePerShuffle * MWave * MPerXdl,
+                                           1,
+                                           CShuffleNXdlPerWavePerShuffle * NWave * NPerXdl>>{};
+
+            constexpr index_t num_access = sfc_c_vgpr.GetNumOfAccess();
+
+            static_assert(num_access == sfc_cde_block.GetNumOfAccess(), "wrong!");
+
+            static_for<0, num_access, 1>{}([&](auto access_id) {
+                // make sure it's safe to write to LDS
+                block_sync_lds();
+
+                // each thread write its data from VGPR to LDS
+                c_thread_copy_vgpr_to_lds.Run(c_thread_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              sfc_c_vgpr.GetIndexTupleOfNumber(access_id),
+                                              c_thread_buf,
+                                              c_block_desc_m0_n0_m1_n1_m2_m3_m4_n2,
+                                              c_shuffle_block_buf);
+
+                // make sure it's safe to read from LDS
+                block_sync_lds();
+
+                // each block copy its data from LDS to global
+                cde_block_copy_lds_and_global.Run(
+                    c_ds_desc_refs,
+                    c_ds_buf_refs,
+                    tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                    tie(e_grid_buf));
+
+                if constexpr(access_id < num_access - 1)
+                {
+                    constexpr auto cde_lds_and_global_step =
+                        sfc_cde_block.GetForwardStep(access_id);
+
+                    // move on Ds
+                    static_for<0, NumDTensor, 1>{}([&](auto i) {
+                        cde_block_copy_lds_and_global.MoveSrcSliceWindow(
+                            c_ds_desc_refs, i + I1, cde_lds_and_global_step);
+                    });
+
+                    // move on E
+                    cde_block_copy_lds_and_global.MoveDstSliceWindow(
+                        tie(e_grid_desc_mblock_mperblock_nblock_nperblock),
+                        I0,
+                        cde_lds_and_global_step);
+                }
+            });
+        }
+    }
+
+    template <bool HasMainKBlockLoop,
+              GemmSpecialization GemmSpec,
+              typename AsLayout,
+              typename BsLayout,
+              typename DsLayout,
+              typename ELayout,
+              typename Block2ETileMap>
+    __device__ static void Run(AsGridPointer p_as_grid,
+                               BsGridPointer p_bs_grid,
+                               DsGridPointer p_ds_grid,
+                               void* __restrict__ p_e_grid_,
+                               void* __restrict__ p_shared,
+                               const AElementwiseOperation& a_element_op,
+                               const BElementwiseOperation& b_element_op,
+                               const CDEElementwiseOperation& cde_element_op,
+                               const index_t M,
+                               const index_t N,
+                               const index_t K,
+                               const std::array<index_t, NumATensor> StrideAs,
+                               const std::array<index_t, NumBTensor> StrideBs,
+                               const std::array<index_t, NumDTensor> StrideDs,
+                               const index_t StrideE,
+                               const Block2ETileMap& block_2_etile_map)
+    {
+        using AsGridDesc_M_K =
+            remove_cvref_t<decltype(MakeAsGridDescriptor_M_K<AsLayout, GemmSpec>({}, {}, {}))>;
+        using BsGridDesc_N_K =
+            remove_cvref_t<decltype(MakeBsGridDescriptor_N_K<BsLayout, GemmSpec>({}, {}, {}))>;
+        using DsGridDesc_M_N =
+            remove_cvref_t<decltype(MakeDsGridDescriptor_M_N<DsLayout, GemmSpec>({}, {}, {}))>;
+
+        const auto p_e_grid = reinterpret_cast<EDataType*>(p_e_grid_);
+
+        AsGridDesc_M_K as_grid_desc_m_k;
+        BsGridDesc_N_K bs_grid_desc_n_k;
+        DsGridDesc_M_N ds_grid_desc_m_n;
+
+        static_for<0, NumATensor, 1>{}([&](auto j) {
+            using ALayout = remove_cvref_t<tuple_element_t<j.value, AsLayout>>;
+
+            as_grid_desc_m_k(j) = MakeAGridDescriptor_M_K<ALayout, GemmSpec>(M, K, StrideAs[j]);
+        });
+
+        static_for<0, NumBTensor, 1>{}([&](auto j) {
+            using BLayout = remove_cvref_t<tuple_element_t<j.value, BsLayout>>;
+
+            bs_grid_desc_n_k(j) = MakeBGridDescriptor_N_K<BLayout, GemmSpec>(N, K, StrideBs[j]);
+        });
+
+        static_for<0, NumDTensor, 1>{}([&](auto j) {
+            using DLayout = remove_cvref_t<tuple_element_t<j.value, DsLayout>>;
+
+            ds_grid_desc_m_n(j) = MakeEGridDescriptor_M_N<DLayout, GemmSpec>(M, N, StrideDs[j]);
+        });
+
+        const auto e_grid_desc_m_n = MakeEGridDescriptor_M_N<ELayout, GemmSpec>(M, N, StrideE);
+
+        // tensor descriptors for block/thread-wise copy
+        const auto as_grid_desc_ak0_m_ak1 = MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k);
+
+        const auto bs_grid_desc_bk0_n_bk1 = MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k);
+
+        const auto ds_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(ds_grid_desc_m_n);
+
+        const auto e_grid_desc_mblock_mperblock_nblock_nperblock =
+            MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(e_grid_desc_m_n);
+
+        Run<HasMainKBlockLoop>(p_as_grid,
+                               p_bs_grid,
+                               p_ds_grid,
+                               p_e_grid,
+                               p_shared,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op,
+                               as_grid_desc_ak0_m_ak1,
+                               bs_grid_desc_bk0_n_bk1,
+                               ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                               e_grid_desc_mblock_mperblock_nblock_nperblock,
+                               block_2_etile_map);
+    }
+};
+
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp
new file mode 100644
index 0000000000..003a2b96d9
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp
@@ -0,0 +1,420 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_description/tensor_space_filling_curve.hpp"
+#include "ck/utility/is_detected.hpp"
+
+namespace ck {
+
+// Thread-level multi-source, multi-destination tensor slice data movement
+// Assume:
+//   1. All sources and destinations are DynamicBuffer
+//   2. Same VectorDim and ScalerPerVector for all sources and destinations
+//   3. DstInMemOps are per destination tensor
+//   4. ThreadTransferSrcResetCoordinateAfterRunFlags are per source tensor
+//   5. ThreadTransferDstResetCoordinateAfterRunFlags are per destination tensor
+//   6. Does not need to know src_descs and dst_descs at compile-time
+//   7. Does not need to know src_slice_origins and dst_slice_origins at compile-time,
+//
+// Does following things to avoid scratch memory issue
+//   1. Use StaticallyIndexedArray or vector_type instead of C array for thread buffer
+//   2. Pass tensor descritpors by reference (or tuple of references)
+//   3. Does not keep reference to tensor descriptor
+//   4. Does not construct new tensor coordinate when call Run()
+template <typename SrcDatas,
+          typename DstDatas,
+          typename SrcDescs,
+          typename DstDescs,
+          typename ElementwiseOperation,
+          typename DstInMemOps, // Sequence<InMemoryDataOperationEnum ...>
+          typename SliceLengths,
+          typename SrcDimAccessOrder,
+          typename DstDimAccessOrder,
+          index_t SrcVectorDim,
+          index_t DstVectorDim,
+          index_t SrcScalarPerVector,
+          index_t DstScalarPerVector,
+          typename SrcResetCoordinateAfterRunFlags, // Sequence<bool ...>
+          typename DstResetCoordinateAfterRunFlags> // Sequence<bool ...>
+struct ThreadwiseTensorSliceTransfer_v7r2
+{
+    static constexpr auto I0 = Number<0>{};
+
+    static constexpr index_t nDim = SliceLengths::Size();
+
+    static constexpr index_t nSrc = SrcDescs::Size();
+    static constexpr index_t nDst = DstDescs::Size();
+
+    using Index = MultiIndex<nDim>;
+
+    // return a tuple of coordiantes for a tuple of tensor
+    template <typename Descs,
+              typename Indices,
+              enable_if_t<Descs::Size() == Indices::Size(), bool> = false>
+    static constexpr auto MakeCoordinates(const Descs& descs, const Indices& indices)
+    {
+        return generate_tuple([&](auto i) { return make_tensor_coordinate(descs[i], indices[i]); },
+                              Number<Descs::Size()>{});
+    }
+
+    using SrcCoords = decltype(MakeCoordinates(SrcDescs{}, StaticallyIndexedArray<Index, nSrc>{}));
+    using DstCoords = decltype(MakeCoordinates(DstDescs{}, StaticallyIndexedArray<Index, nDst>{}));
+
+    // scalar per access on each dim
+    // FIXME: don't use lambda_scalar_per_access
+    static constexpr auto src_scalar_per_access = generate_sequence(
+        detail::lambda_scalar_per_access<SrcVectorDim, SrcScalarPerVector>{}, Number<nDim>{});
+
+    using SrcSpaceFillingCurve = SpaceFillingCurve<SliceLengths,
+                                                   SrcDimAccessOrder,
+                                                   remove_cv_t<decltype(src_scalar_per_access)>>;
+
+    static constexpr auto dst_scalar_per_access = generate_sequence(
+        detail::lambda_scalar_per_access<DstVectorDim, DstScalarPerVector>{}, Number<nDim>{});
+
+    using DstSpaceFillingCurve = SpaceFillingCurve<SliceLengths,
+                                                   DstDimAccessOrder,
+                                                   remove_cv_t<decltype(dst_scalar_per_access)>>;
+
+    __device__ constexpr ThreadwiseTensorSliceTransfer_v7r2(
+        const SrcDescs& src_descs,
+        const StaticallyIndexedArray<Index, nSrc>& src_slice_origins,
+        const DstDescs& dst_descs,
+        const StaticallyIndexedArray<Index, nDst>& dst_slice_origins,
+        const ElementwiseOperation& element_op)
+        : src_coords_(MakeCoordinates(src_descs, src_slice_origins)),
+          dst_coords_(MakeCoordinates(dst_descs, dst_slice_origins)),
+          element_op_(element_op)
+    {
+        static_assert(SliceLengths::At(Number<SrcVectorDim>{}) % SrcScalarPerVector == 0,
+                      "wrong! cannot evenly divide");
+
+        static_assert(SliceLengths::At(Number<DstVectorDim>{}) % DstScalarPerVector == 0,
+                      "wrong! cannot evenly divide");
+    }
+
+    template <typename Indices, enable_if_t<SrcDescs::Size() == Indices::Size(), bool> = false>
+    __device__ void SetSrcSliceOrigins(const SrcDescs& src_descs,
+                                       const Indices& src_slice_origin_idxs)
+    {
+        static_for<0, nSrc, 1>{}([&](auto i) {
+            src_coords_(i) = make_tensor_coordinate(src_descs[i], src_slice_origin_idxs[i]);
+        });
+    }
+
+    template <typename Indices, enable_if_t<DstDescs::Size() == Indices::Size(), bool> = false>
+    __device__ void SetDstSliceOrigins(const DstDescs& dst_descs,
+                                       const Indices& dst_slice_origin_idxs)
+    {
+        static_for<0, nDst, 1>{}([&](auto i) {
+            dst_coords_(i) = make_tensor_coordinate(dst_descs[i], dst_slice_origin_idxs[i]);
+        });
+    }
+
+    template <typename DataTypes, index_t ScalarPerVector>
+    __device__ static auto generate_vectors()
+    {
+        auto data_types = DataTypes{};
+
+        constexpr index_t num = data_types.Size();
+
+        return generate_tuple(
+            [&](auto i) {
+                using DataType = remove_cvref_t<decltype(data_types[i])>;
+
+                return vector_type_maker_t<DataType, ScalarPerVector>{};
+            },
+            Number<num>{});
+    }
+
+    template <typename T>
+    using has_vec_len = decltype(std::declval<T&>().vec_len);
+
+    // SrcDescs: Tuple<const SrcDesc0&, const SrcDesc1&, ...>
+    // SrcBuffers: Tuple<const SrcBuffer0&, const SrcBuffer1&, ...>
+    template <typename SrcBuffers,
+              enable_if_t<SrcDescs::Size() == SrcBuffers::Size(), bool> = false>
+    __device__ void RunRead(const SrcDescs& src_descs, const SrcBuffers& src_bufs)
+    {
+        // loop over space-filling curve
+        static_for<0, num_access, 1>{}([&](auto iAccess) {
+            auto src_vectors = generate_vectors<SrcDatas, SrcScalarPerVector>();
+            auto dst_vectors = generate_vectors<DstDatas, DstScalarPerVector>();
+
+            // copy data from src_bufs into src_vectors
+            static_for<0, nSrc, 1>{}([&](auto i) {
+                using src_vector_t = typename remove_cvref_t<decltype(src_vectors[i])>::type;
+
+                const bool is_src_valid =
+                    coordinate_has_valid_offset_assuming_visible_index_is_valid(src_descs[i],
+                                                                                src_coords_[i]);
+
+                src_vectors(i).template AsType<src_vector_t>()(I0) =
+                    src_bufs[i].template Get<src_vector_t>(src_coords_[i].GetOffset(),
+                                                           is_src_valid);
+            });
+
+            if constexpr(is_detected<has_vec_len, decltype(element_op_)>::value)
+            {
+                constexpr auto elem_op_vec_len = decltype(element_op_)::vec_len;
+
+                static_assert(is_same<remove_cvref_t<decltype(elem_op_vec_len)>, index_t>::value,
+                              "vec_len in element_op_ type is not index_t");
+
+                static_assert(elem_op_vec_len == 1 || elem_op_vec_len == 2 ||
+                                  elem_op_vec_len == 4 || elem_op_vec_len == 8,
+                              "vec_len in element_op_ must be 1, 2, 4, 8");
+
+                static_assert(SrcScalarPerVector % elem_op_vec_len == 0,
+                              "vec_len in element_op_ cannot be divided by SrcScalarPerVector!");
+
+                // apply pointwise function
+                static_for<0, SrcScalarPerVector / elem_op_vec_len, 1>{}([&](auto i) {
+                    // get reference to src data
+                    const auto src_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iSrc) -> const auto& {
+                            using SrcData = remove_cvref_t<tuple_element_t<iSrc.value, SrcDatas>>;
+
+                            using elem_op_vec_t =
+                                typename vector_type<SrcData, elem_op_vec_len>::type;
+
+                            return src_vectors[iSrc].template AsType<elem_op_vec_t>()[i];
+                        },
+                        Number<nSrc>{});
+
+                    // get reference to dst data
+                    auto dst_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iDst) -> auto& {
+                            using DstData = remove_cvref_t<tuple_element_t<iDst.value, DstDatas>>;
+
+                            using elem_op_vec_t =
+                                typename vector_type<DstData, elem_op_vec_len>::type;
+
+                            return dst_vectors(iDst).template AsType<elem_op_vec_t>()(i);
+                        },
+                        Number<nDst>{});
+
+                    // apply pointwise function
+                    // pointwise function signature:
+                    // element_op_(dst_data_refs[I0],
+                    //             dst_data_refs[I1],
+                    //             ...,
+                    //             src_data_refs[I0],
+                    //             src_data_refs[I1],
+                    //             ...)
+                    unpack2(element_op_, dst_data_refs, src_data_refs);
+                });
+            }
+            else
+            {
+                // apply pointwise function
+                static_for<0, SrcScalarPerVector, 1>{}([&](auto i) {
+                    // get reference to src data
+                    const auto src_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iSrc) -> const auto& {
+                            using SrcData = remove_cvref_t<tuple_element_t<iSrc.value, SrcDatas>>;
+
+                            return src_vectors[iSrc].template AsType<SrcData>()[i];
+                        },
+                        Number<nSrc>{});
+
+                    // get reference to dst data
+                    auto dst_data_refs = generate_tie(
+                        // return type should be lvalue
+                        [&](auto iDst) -> auto& {
+                            using DstData = remove_cvref_t<tuple_element_t<iDst.value, DstDatas>>;
+
+                            return dst_vectors(iDst).template AsType<DstData>()(i);
+                        },
+                        Number<nDst>{});
+
+                    // apply pointwise function
+                    // pointwise function signature:
+                    // element_op_(dst_data_refs[I0],
+                    //             dst_data_refs[I1],
+                    //             ...,
+                    //             src_data_refs[I0],
+                    //             src_data_refs[I1],
+                    //             ...)
+                    unpack2(element_op_, dst_data_refs, src_data_refs);
+                });
+            }
+
+            dst_vectors_tuple_(iAccess) = dst_vectors;
+
+            // move coordinate
+            if constexpr(iAccess.value != num_access - 1)
+            {
+                constexpr auto forward_step = SrcSpaceFillingCurve::GetForwardStep(iAccess);
+
+                static_for<0, nSrc, 1>{}([&](auto i) {
+                    move_tensor_coordinate(src_descs[i],
+                                           src_coords_(i),
+                                           make_tensor_coordinate_step(src_descs[i], forward_step));
+                });
+            }
+        });
+
+        // move coordinate back to slice origin (or not)
+        static_for<0, nSrc, 1>{}([&](auto i) {
+            if constexpr(SrcResetCoordinateAfterRunFlags::At(i))
+            {
+                const auto src_reset_step =
+                    make_tensor_coordinate_step(src_descs[i], GetSrcCoordinateResetStep());
+
+                move_tensor_coordinate(src_descs[i], src_coords_(i), src_reset_step);
+            }
+        });
+    }
+
+    // DstDescs: Tuple<const DstDesc0&, const DstDesc1&, ...>
+    // DstBuffers: Tuple<const DstBuffer0&, const DstBuffer1&, ...>
+    template <typename DstBuffers,
+              enable_if_t<DstDescs::Size() == DstBuffers::Size(), bool> = false>
+    __device__ void RunWrite(const DstDescs& dst_descs, DstBuffers dst_bufs)
+    {
+        // loop over space-filling curve
+        static_for<0, num_access, 1>{}([&](auto iAccess) {
+            auto dst_vectors = dst_vectors_tuple_[iAccess];
+
+            // copy data from buf_vectors into dst_bufs
+            static_for<0, nDst, 1>{}([&](auto i) {
+                using dst_vector_t = typename remove_cvref_t<decltype(dst_vectors[i])>::type;
+
+                const bool is_dst_valid =
+                    coordinate_has_valid_offset_assuming_visible_index_is_valid(dst_descs[i],
+                                                                                dst_coords_[i]);
+
+                constexpr InMemoryDataOperationEnum DstInMemOp =
+                    static_cast<InMemoryDataOperationEnum>(DstInMemOps::At(i.value));
+
+                dst_bufs(i).template Update<DstInMemOp, dst_vector_t>(
+                    dst_coords_[i].GetOffset(),
+                    is_dst_valid,
+                    dst_vectors[i].template AsType<dst_vector_t>()[I0]);
+            });
+
+            // move coordinate
+            if constexpr(iAccess.value != num_access - 1)
+            {
+                constexpr auto forward_step = DstSpaceFillingCurve::GetForwardStep(iAccess);
+
+                static_for<0, nDst, 1>{}([&](auto i) {
+                    move_tensor_coordinate(dst_descs[i],
+                                           dst_coords_(i),
+                                           make_tensor_coordinate_step(dst_descs[i], forward_step));
+                });
+            }
+        });
+
+        static_for<0, nDst, 1>{}([&](auto i) {
+            if constexpr(DstResetCoordinateAfterRunFlags::At(i))
+            {
+                const auto dst_reset_step =
+                    make_tensor_coordinate_step(dst_descs[i], GetDstCoordinateResetStep());
+
+                move_tensor_coordinate(dst_descs[i], dst_coords_(i), dst_reset_step);
+            }
+        });
+    }
+
+    // SrcDescs: Tuple<const SrcDesc0&, const SrcDesc1&, ...>
+    // SrcBuffers: Tuple<const SrcBuffer0&, const SrcBuffer1&, ...>
+    // DstDescs: Tuple<const DstDesc0&, const DstDesc1&, ...>
+    // DstBuffers: Tuple<const DstBuffer0&, const DstBuffer1&, ...>
+    template <typename SrcBuffers,
+              typename DstBuffers,
+              enable_if_t<SrcDescs::Size() == SrcBuffers::Size() &&
+                              DstDescs::Size() == DstBuffers::Size(),
+                          bool> = false>
+    __device__ void Run(const SrcDescs& src_descs,
+                        const SrcBuffers& src_bufs,
+                        const DstDescs& dst_descs,
+                        DstBuffers dst_bufs)
+    {
+        RunRead(src_descs, src_bufs);
+        RunWrite(dst_descs, dst_bufs);
+    }
+
+    __device__ static constexpr auto GetSrcCoordinateResetStep()
+    {
+        if constexpr(num_access == 0)
+        {
+            return typename SrcSpaceFillingCurve::Index{};
+        }
+        else
+        {
+            return SrcSpaceFillingCurve::GetStepBetween(Number<num_access - 1>{}, Number<0>{});
+        }
+    }
+
+    __device__ static constexpr auto GetDstCoordinateResetStep()
+    {
+        if constexpr(num_access == 0)
+        {
+            return typename DstSpaceFillingCurve::Index{};
+        }
+        else
+        {
+            return DstSpaceFillingCurve::GetStepBetween(Number<num_access - 1>{}, Number<0>{});
+        }
+    }
+
+    // src_slice_origin_step_idx need to be known at compile-time, for performance reason
+    template <index_t ISrc>
+    __device__ void MoveSrcSliceWindow(const SrcDescs& src_descs,
+                                       Number<ISrc> iSrc,
+                                       const Index& src_slice_origin_step_idx)
+    {
+        // if src coord was not reset by RunRead(), then need to adjust the step here
+        const auto adjusted_step_idx =
+            SrcResetCoordinateAfterRunFlags::At(iSrc)
+                ? src_slice_origin_step_idx
+                : src_slice_origin_step_idx + GetSrcCoordinateResetStep();
+
+        // is it OK to construct a new step every time?
+        const auto adjusted_step = make_tensor_coordinate_step(src_descs[iSrc], adjusted_step_idx);
+
+        move_tensor_coordinate(src_descs[iSrc], src_coords_(iSrc), adjusted_step);
+    }
+
+    // dst_slice_origin_step_idx need to be known at compile-time, for performance reason
+    template <index_t IDst>
+    __device__ void MoveDstSliceWindow(const DstDescs& dst_descs,
+                                       Number<IDst> iDst,
+                                       const Index& dst_slice_origin_step_idx)
+    {
+        // if dst coord was not reset by Run(), then need to adjust the step here
+        const auto adjusted_step_idx =
+            DstResetCoordinateAfterRunFlags::At(iDst)
+                ? dst_slice_origin_step_idx
+                : dst_slice_origin_step_idx + GetDstCoordinateResetStep();
+
+        // is it OK to construct a new step every time?
+        const auto adjusted_step = make_tensor_coordinate_step(dst_descs[iDst], adjusted_step_idx);
+
+        move_tensor_coordinate(dst_descs[iDst], dst_coords_(iDst), adjusted_step);
+    }
+
+    private:
+    using SrcVectorsType = decltype(generate_vectors<SrcDatas, SrcScalarPerVector>());
+    using DstVectorsType = decltype(generate_vectors<DstDatas, DstScalarPerVector>());
+
+    static constexpr auto num_access = SrcSpaceFillingCurve::GetNumOfAccess();
+
+    StaticallyIndexedArray<DstVectorsType, num_access> dst_vectors_tuple_;
+
+    SrcCoords src_coords_;
+    DstCoords dst_coords_;
+    const ElementwiseOperation element_op_;
+};
+
+} // namespace ck
diff --git a/include/ck/utility/is_detected.hpp b/include/ck/utility/is_detected.hpp
new file mode 100644
index 0000000000..31a4bf24d6
--- /dev/null
+++ b/include/ck/utility/is_detected.hpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+
+namespace detail {
+template <class Default, class AlwaysVoid, template <class...> class Op, class... Args>
+struct detector
+{
+    using value_t = std::false_type;
+    using type    = Default;
+};
+
+template <class Default, template <class...> class Op, class... Args>
+struct detector<Default, std::void_t<Op<Args...>>, Op, Args...>
+{
+    using value_t = std::true_type;
+    using type    = Op<Args...>;
+};
+} // namespace detail
+
+struct nonesuch
+{
+    ~nonesuch()               = delete;
+    nonesuch(nonesuch const&) = delete;
+    void operator=(nonesuch const&) = delete;
+};
+
+template <template <class...> class Op, class... Args>
+using is_detected = typename detail::detector<nonesuch, void, Op, Args...>::value_t;
+
+} // namespace ck
diff --git a/include/ck/utility/tuple.hpp b/include/ck/utility/tuple.hpp
index b616b3123f..3ff0af89db 100644
--- a/include/ck/utility/tuple.hpp
+++ b/include/ck/utility/tuple.hpp
@@ -177,6 +177,8 @@ struct Tuple : detail::TupleImpl<typename arithmetic_sequence_gen<0, sizeof...(X
     }
 
     __host__ __device__ static constexpr bool IsStaticBuffer() { return true; }
+
+    __host__ __device__ static constexpr bool IsTuple() { return true; }
 };
 
 template <>

From e2243a4d1e579c46a57f18afae536a162ab0f4bf Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Wed, 27 Sep 2023 17:19:06 +0200
Subject: [PATCH 145/194] Add column to image kernel (#930)

* Add column to image kernel

* Minor fixes for dtypes and client examples

* Disable tests for disabled dtypes

* Disable add instances functions for disabled data types

* Minor stylistic fixes

* Revert "Disable add instances functions for disabled data types"

This reverts commit 728b86956378dcd9415fd0f2557833a068fe1c10.

* Instances reduction

* Add comments in device_column_to_image_impl

* Update changelog and Copyrights

* Improve changelog
---
 CHANGELOG.md                                  |  13 +-
 .../20_image_to_column/CMakeLists.txt         |   2 -
 .../22_im2col_col2im/CMakeLists.txt           |   5 +
 .../22_im2col_col2im/column_to_image.cpp      | 173 +++++
 .../image_to_column.cpp                       |  26 +-
 .../CMakeLists.txt                            |   6 +-
 .../52_im2col_col2im/column_to_image_f32.cpp  | 165 +++++
 .../common.hpp                                |   4 +-
 .../image_to_column_f32.cpp                   |  23 +-
 .../gpu/device/conv_tensor_rearrange_op.hpp   |  33 +
 ...n.hpp => device_conv_tensor_rearrange.hpp} |  27 +-
 .../impl/device_column_to_image_impl.hpp      | 621 ++++++++++++++++++
 .../impl/device_image_to_column_impl.hpp      | 144 ++--
 ...lumn.hpp => gridwise_tensor_rearrange.hpp} |  75 ++-
 .../transform_conv_fwd_to_gemm.hpp            | 345 +---------
 include/ck/utility/dynamic_buffer.hpp         |  34 +-
 .../cpu/reference_column_to_image.hpp         | 363 ++++++++++
 .../cpu/reference_image_to_column.hpp         |  12 +-
 .../gpu/conv_tensor_rearrange.hpp             | 282 ++++++++
 .../device_column_to_image_instance.hpp       | 106 +++
 .../device_image_to_column_instance.hpp       |  32 +-
 .../gpu/image_to_column.hpp                   | 135 ----
 .../gpu/column_to_image/CMakeLists.txt        |   5 +
 ...evice_column_to_image_nhwc_1d_instance.cpp |  61 ++
 ...evice_column_to_image_nhwc_2d_instance.cpp |  62 ++
 ...evice_column_to_image_nhwc_3d_instance.cpp |  62 ++
 ...evice_image_to_column_nhwc_1d_instance.cpp |  40 +-
 ...evice_image_to_column_nhwc_2d_instance.cpp |  33 +-
 ...evice_image_to_column_nhwc_3d_instance.cpp |  41 +-
 profiler/README.md                            |   8 +-
 ...=> profile_conv_tensor_rearrange_impl.hpp} | 140 +++-
 profiler/src/CMakeLists.txt                   |   3 +-
 .../src/profile_conv_tensor_rearrange.cpp     | 251 +++++++
 profiler/src/profile_image_to_column.cpp      | 169 -----
 test/CMakeLists.txt                           |   2 +-
 test/conv_tensor_rearrange/CMakeLists.txt     |   4 +
 .../test_conv_tensor_rearrange.cpp            | 153 +++++
 .../test_conv_tensor_rearrange_interface.cpp  | 260 ++++++++
 test/image_to_column/CMakeLists.txt           |   4 -
 test/image_to_column/test_image_to_column.cpp | 121 ----
 .../test_image_to_column_interface.cpp        | 196 ------
 41 files changed, 3040 insertions(+), 1201 deletions(-)
 delete mode 100644 client_example/20_image_to_column/CMakeLists.txt
 create mode 100644 client_example/22_im2col_col2im/CMakeLists.txt
 create mode 100644 client_example/22_im2col_col2im/column_to_image.cpp
 rename client_example/{20_image_to_column => 22_im2col_col2im}/image_to_column.cpp (84%)
 rename example/{52_image_to_column => 52_im2col_col2im}/CMakeLists.txt (50%)
 create mode 100644 example/52_im2col_col2im/column_to_image_f32.cpp
 rename example/{52_image_to_column => 52_im2col_col2im}/common.hpp (94%)
 rename example/{52_image_to_column => 52_im2col_col2im}/image_to_column_f32.cpp (89%)
 create mode 100644 include/ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp
 rename include/ck/tensor_operation/gpu/device/{device_image_to_column.hpp => device_conv_tensor_rearrange.hpp} (74%)
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
 rename include/ck/tensor_operation/gpu/grid/{gridwise_image_to_column.hpp => gridwise_tensor_rearrange.hpp} (52%)
 create mode 100644 library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp
 rename library/include/ck/library/tensor_operation_instance/gpu/{image_to_column => conv_tensor_rearrange}/device_image_to_column_instance.hpp (70%)
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp
 rename profiler/include/profiler/{profile_image_to_column_impl.hpp => profile_conv_tensor_rearrange_impl.hpp} (56%)
 create mode 100644 profiler/src/profile_conv_tensor_rearrange.cpp
 delete mode 100644 profiler/src/profile_image_to_column.cpp
 create mode 100644 test/conv_tensor_rearrange/CMakeLists.txt
 create mode 100644 test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
 create mode 100644 test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
 delete mode 100644 test/image_to_column/CMakeLists.txt
 delete mode 100644 test/image_to_column/test_image_to_column.cpp
 delete mode 100644 test/image_to_column/test_image_to_column_interface.cpp

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 31f129b581..9d38a66382 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -2,6 +2,18 @@
 
 Full documentation for Composable Kernel is not yet available.
 
+## (Unreleased) CK for ROCm 6.0.0
+
+### Fixed
+
+### Optimizations
+
+### Added
+- Added image to column (#867) and column to image kernels (#930).
+
+### Changed
+
+
 ## CK 0.2.0 for ROCm 5.5.0
 
 ### Fixed
@@ -29,4 +41,3 @@ Full documentation for Composable Kernel is not yet available.
 - Added MaxPool backward (#750).
 
 ### Changed
-- Changed ...
diff --git a/client_example/20_image_to_column/CMakeLists.txt b/client_example/20_image_to_column/CMakeLists.txt
deleted file mode 100644
index 80edcd0416..0000000000
--- a/client_example/20_image_to_column/CMakeLists.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-add_executable(client_image_to_column image_to_column.cpp)
-target_link_libraries(client_image_to_column PRIVATE composable_kernel::device_operations)
diff --git a/client_example/22_im2col_col2im/CMakeLists.txt b/client_example/22_im2col_col2im/CMakeLists.txt
new file mode 100644
index 0000000000..47ac42fe87
--- /dev/null
+++ b/client_example/22_im2col_col2im/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_executable(client_image_to_column image_to_column.cpp)
+target_link_libraries(client_image_to_column PRIVATE composable_kernel::device_operations)
+
+add_executable(client_column_to_image column_to_image.cpp)
+target_link_libraries(client_column_to_image PRIVATE composable_kernel::device_operations)
diff --git a/client_example/22_im2col_col2im/column_to_image.cpp b/client_example/22_im2col_col2im/column_to_image.cpp
new file mode 100644
index 0000000000..43338ce408
--- /dev/null
+++ b/client_example/22_im2col_col2im/column_to_image.cpp
@@ -0,0 +1,173 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using OutDataType = ck::half_t;
+
+using ImageLayout = ck::tensor_layout::convolution::GNHWC;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t N             = 32; // batch size
+static constexpr ck::index_t C             = 32; // input channel (per group)
+static constexpr ck::index_t Y             = 3;  // filter H
+static constexpr ck::index_t X             = 3;  // filter W
+static constexpr ck::index_t Hi            = 28; // input H
+static constexpr ck::index_t Wi            = 28; // input W
+static constexpr ck::index_t Ho            = 28; // output H
+static constexpr ck::index_t Wo            = 28; // output W
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+
+    std::array<ck::index_t, 2> in_spatial_lengths{Hi, Wi};
+    std::array<ck::index_t, 2> wei_spatial_lengths{Y, X};
+    std::array<ck::index_t, 2> out_spatial_lengths{Ho, Wo};
+
+    // We have NHWGC in memory space (G is dummy)
+    // However, CK's API only accept length and stride with order of GNCHW
+    // Hence, we need to adjust the order of stride
+    std::array<ck::index_t, 5> image_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 2> gemm_strides{Y * X * C, 1};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Ho * Wo * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Hi * Wi * G * C);
+
+    using namespace ck::conv_tensor_rearrange_op;
+
+    using DeviceOp = ck::tensor_operation::device::DeviceConvTensorRearrange<NumDimSpatial,
+                                                                             ImageLayout,
+                                                                             InDataType,
+                                                                             OutDataType,
+                                                                             ColumnToImage>;
+
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        wei_spatial_lengths,
+                                                        image_strides,
+                                                        gemm_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads);
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
+                                    sizeof(OutDataType) * N * Ho * Wo * Y * X * C;
+
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
+                      << op_name << std::endl;
+
+            if(avg_time < best_avg_time)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_gb_per_sec
+              << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        wei_spatial_lengths,
+                                                        image_strides,
+                                                        gemm_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads);
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+}
diff --git a/client_example/20_image_to_column/image_to_column.cpp b/client_example/22_im2col_col2im/image_to_column.cpp
similarity index 84%
rename from client_example/20_image_to_column/image_to_column.cpp
rename to client_example/22_im2col_col2im/image_to_column.cpp
index ace4c1a681..a1447abf64 100644
--- a/client_example/20_image_to_column/image_to_column.cpp
+++ b/client_example/22_im2col_col2im/image_to_column.cpp
@@ -9,13 +9,14 @@
 #include <vector>
 
 #include "ck/ck.hpp"
-#include "ck/library/tensor_operation_instance/gpu/image_to_column.hpp"
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 
 using InDataType  = ck::half_t;
 using OutDataType = ck::half_t;
 
-using InLayout = ck::tensor_layout::convolution::GNHWC;
+using ImageLayout = ck::tensor_layout::convolution::GNHWC;
 
 static constexpr ck::index_t NumDimSpatial = 2;
 static constexpr ck::index_t G             = 1;
@@ -54,8 +55,8 @@ int main()
     // We have NHWGC in memory space (G is dummy)
     // However, CK's API only accept length and stride with order of GNCHW
     // Hence, we need to adjust the order of stride
-    std::array<ck::index_t, 5> in_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
-    std::array<ck::index_t, 2> out_strides{Y * X * C, 1};
+    std::array<ck::index_t, 5> image_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
+    std::array<ck::index_t, 2> gemm_strides{Y * X * C, 1};
 
     std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
     std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
@@ -65,8 +66,13 @@ int main()
     SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
     SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * Y * X * C);
 
-    using DeviceOp = ck::tensor_operation::device::
-        DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>;
+    using namespace ck::conv_tensor_rearrange_op;
+
+    using DeviceOp = ck::tensor_operation::device::DeviceConvTensorRearrange<NumDimSpatial,
+                                                                             ImageLayout,
+                                                                             InDataType,
+                                                                             OutDataType,
+                                                                             ImageToColumn>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -92,8 +98,8 @@ int main()
                                                         in_spatial_lengths,
                                                         out_spatial_lengths,
                                                         wei_spatial_lengths,
-                                                        in_strides,
-                                                        out_strides,
+                                                        image_strides,
+                                                        gemm_strides,
                                                         filter_strides,
                                                         filter_dilations,
                                                         input_left_pads,
@@ -148,8 +154,8 @@ int main()
                                                         in_spatial_lengths,
                                                         out_spatial_lengths,
                                                         wei_spatial_lengths,
-                                                        in_strides,
-                                                        out_strides,
+                                                        image_strides,
+                                                        gemm_strides,
                                                         filter_strides,
                                                         filter_dilations,
                                                         input_left_pads,
diff --git a/example/52_image_to_column/CMakeLists.txt b/example/52_im2col_col2im/CMakeLists.txt
similarity index 50%
rename from example/52_image_to_column/CMakeLists.txt
rename to example/52_im2col_col2im/CMakeLists.txt
index 226e1fc5ae..a2dec9e805 100644
--- a/example/52_image_to_column/CMakeLists.txt
+++ b/example/52_im2col_col2im/CMakeLists.txt
@@ -2,9 +2,11 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(example_image_to_column)
+   add_custom_target(example_im2col_col2im)
    add_example_executable(example_image_to_column_f32 image_to_column_f32.cpp)
-   add_dependencies(example_image_to_column example_image_to_column_f32)
+   add_dependencies(example_im2col_col2im example_image_to_column_f32)
+   add_example_executable(example_column_to_image_f32 column_to_image_f32.cpp)
+   add_dependencies(example_im2col_col2im example_column_to_image_f32)
    set(target 1)
  endif()
 endforeach()
diff --git a/example/52_im2col_col2im/column_to_image_f32.cpp b/example/52_im2col_col2im/column_to_image_f32.cpp
new file mode 100644
index 0000000000..52144e6885
--- /dev/null
+++ b/example/52_im2col_col2im/column_to_image_f32.cpp
@@ -0,0 +1,165 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+using InDataType  = FP32; // ck::bhalf_t;//FP32;
+using OutDataType = FP32; // ck::bhalf_t;//FP32;
+
+using ImLayout        = ck::tensor_layout::convolution::GNHWC;
+using ColumnToImageOp = ck::conv_tensor_rearrange_op::ColumnToImage;
+
+// clang-format off
+using DeviceColToImgInstance = ck::tensor_operation::device::DeviceColumnToImageImpl
+        //#####################|        Num| ImLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+                              < NDimSpatial, ImLayout, InDataType, OutDataType,   256,   128,   128, S<16, 16>,     1>;
+// clang-format on
+
+bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
+{
+
+    const auto N = conv_params.N_;
+    const auto C = conv_params.C_;
+
+    const ck::index_t NDoHoWo =
+        N * ck::accumulate_n<ck::index_t>(
+                conv_params.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+    const ck::index_t CZYX =
+        C * ck::accumulate_n<ck::index_t>(
+                conv_params.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+    const auto in_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+    const auto out_desc =
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params);
+
+    std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
+    std::array<ck::index_t, 2> gemm_m_k_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+
+    copy(conv_params.input_spatial_lengths_, input_spatial_lengths);
+    copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
+    copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
+    copy(in_desc.GetStrides(), gemm_m_k_strides);
+    copy(out_desc.GetStrides(), image_g_n_c_wis_strides);
+    copy(conv_params.conv_filter_strides_, conv_filter_strides);
+    copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_params.input_left_pads_, input_left_pads);
+    copy(conv_params.input_right_pads_, input_right_pads);
+
+    Tensor<InDataType> in(in_desc);
+    Tensor<OutDataType> out_device(out_desc);
+    Tensor<OutDataType> out_host(out_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "out: " << out_device.mDesc << std::endl;
+
+    switch(config.init_method)
+    {
+    case 0: break;
+    case 1: in.GenerateTensorValue(GeneratorTensor_2<InDataType>{1, 2}); break;
+    default: in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in.mData.data());
+
+    // reset input to zero
+    out_device_buf.SetZero();
+
+    static_assert(std::is_default_constructible_v<DeviceColToImgInstance>);
+
+    // do conv
+    auto col2img  = DeviceColToImgInstance{};
+    auto invoker  = col2img.MakeInvoker();
+    auto argument = col2img.MakeArgument(in_device_buf.GetDeviceBuffer(),
+                                         out_device_buf.GetDeviceBuffer(),
+                                         N,
+                                         C,
+                                         input_spatial_lengths,
+                                         filter_spatial_lengths,
+                                         output_spatial_lengths,
+                                         image_g_n_c_wis_strides,
+                                         gemm_m_k_strides,
+                                         conv_filter_strides,
+                                         conv_filter_dilations,
+                                         input_left_pads,
+                                         input_right_pads);
+
+    if(!col2img.IsSupportedArgument(argument))
+    {
+        std::cerr << "wrong! device_col2img with the specified compilation parameters does "
+                     "not support this col2img problem"
+                  << std::endl;
+
+        return false;
+    }
+
+    float ave_time        = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+    std::size_t num_btype = NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
+    float gb_per_sec      = num_btype / 1.E6 / ave_time;
+    std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
+
+    if(config.do_verification)
+    {
+        auto ref_column_to_image = ck::tensor_operation::host::
+            ReferenceColumnToImage<NDimSpatial, ImLayout, InDataType, OutDataType>();
+
+        auto ref_invoker = ref_column_to_image.MakeInvoker();
+
+        auto ref_argument = ref_column_to_image.MakeArgument(in,
+                                                             out_host,
+                                                             conv_params.filter_spatial_lengths_,
+                                                             conv_params.conv_filter_strides_,
+                                                             conv_params.conv_filter_dilations_,
+                                                             conv_params.input_left_pads_,
+                                                             conv_params.input_right_pads_);
+
+        if(!ref_column_to_image.IsSupportedArgument(&ref_argument))
+        {
+            std::cerr << "wrong! ref_col2img with the specified compilation parameters does "
+                         "not support this col2img problem"
+                      << std::endl;
+            return false;
+        }
+
+        ref_invoker.Run(ref_argument);
+        out_device_buf.FromDevice(out_device.mData.data());
+        return ck::utils::check_err(out_device.mData, out_host.mData);
+    }
+
+    return true;
+}
+
+int RunColumnToImageExample(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_params = DefaultConvParams;
+
+    if(!parse_cmd_args(argc, argv, config, conv_params))
+    {
+        return EXIT_FAILURE;
+    }
+
+    if(conv_params.num_dim_spatial_ != NDimSpatial)
+    {
+        std::cerr << "unsupported # of spatial dimensions" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    return !RunColumnToImage(config, conv_params);
+}
+
+int main(int argc, char* argv[]) { return RunColumnToImageExample(argc, argv); }
diff --git a/example/52_image_to_column/common.hpp b/example/52_im2col_col2im/common.hpp
similarity index 94%
rename from example/52_image_to_column/common.hpp
rename to example/52_im2col_col2im/common.hpp
index 8510fa1e6d..61d30c4cb4 100644
--- a/example/52_image_to_column/common.hpp
+++ b/example/52_im2col_col2im/common.hpp
@@ -10,6 +10,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 
 #include "ck/library/utility/algorithm.hpp"
@@ -20,6 +21,7 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp"
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
@@ -32,7 +34,7 @@ struct ExecutionConfig final
 {
     bool do_verification = true;
     int init_method      = 1;
-    bool time_kernel     = true;
+    bool time_kernel     = false;
 };
 
 #define DefaultConvParams                                                            \
diff --git a/example/52_image_to_column/image_to_column_f32.cpp b/example/52_im2col_col2im/image_to_column_f32.cpp
similarity index 89%
rename from example/52_image_to_column/image_to_column_f32.cpp
rename to example/52_im2col_col2im/image_to_column_f32.cpp
index c8a7e5f221..6d883460cc 100644
--- a/example/52_image_to_column/image_to_column_f32.cpp
+++ b/example/52_im2col_col2im/image_to_column_f32.cpp
@@ -6,15 +6,16 @@
 using InDataType  = FP32;
 using OutDataType = FP32;
 
-using InLayout = ck::tensor_layout::convolution::GNHWC;
+using ImLayout        = ck::tensor_layout::convolution::GNHWC;
+using ImageToColumnOp = ck::conv_tensor_rearrange_op::ImageToColumn;
 
 // clang-format off
 using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumnImpl
-        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Num| ImLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-                              < NDimSpatial, InLayout, InDataType, OutDataType,   256,   128,   128, S<16, 16>,     1>;
+                              < NDimSpatial, ImLayout, InDataType, OutDataType,   256,   128,   128, S<16, 16>,      1>;
 // clang-format on
 
 bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
@@ -31,14 +32,14 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
                 conv_params.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
 
     const auto in_desc =
-        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(conv_params);
+        ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params);
     const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
 
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
-    std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
-    std::array<ck::index_t, 2> output_m_k_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
+    std::array<ck::index_t, 2> gemm_m_k_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -49,8 +50,8 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
     copy(conv_params.input_spatial_lengths_, input_spatial_lengths);
     copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
     copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
-    copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
-    copy(out_desc.GetStrides(), output_m_k_strides);
+    copy(in_desc.GetStrides(), image_g_n_c_wis_strides);
+    copy(out_desc.GetStrides(), gemm_m_k_strides);
     copy(conv_params.conv_filter_strides_, conv_filter_strides);
     copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
     copy(conv_params.input_left_pads_, input_left_pads);
@@ -90,8 +91,8 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
                                          input_spatial_lengths,
                                          filter_spatial_lengths,
                                          output_spatial_lengths,
-                                         input_g_n_c_wis_strides,
-                                         output_m_k_strides,
+                                         image_g_n_c_wis_strides,
+                                         gemm_m_k_strides,
                                          conv_filter_strides,
                                          conv_filter_dilations,
                                          input_left_pads,
@@ -114,7 +115,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
     if(config.do_verification)
     {
         auto ref_image_to_column = ck::tensor_operation::host::
-            ReferenceImageToColumn<NDimSpatial, InLayout, InDataType, OutDataType>();
+            ReferenceImageToColumn<NDimSpatial, ImLayout, InDataType, OutDataType>();
 
         auto ref_invoker = ref_image_to_column.MakeInvoker();
 
diff --git a/include/ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp b/include/ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp
new file mode 100644
index 0000000000..dc08a2c88b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp
@@ -0,0 +1,33 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+namespace ck {
+namespace conv_tensor_rearrange_op {
+
+struct BaseConvTensorRearrangeOp
+{
+};
+
+struct ImageToColumn : public BaseConvTensorRearrangeOp
+{
+    static constexpr const char* name = "Image to Column";
+};
+
+struct ColumnToImage : public BaseConvTensorRearrangeOp
+{
+    static constexpr const char* name = "Column to Image";
+};
+
+template <typename Op,
+          typename std::enable_if<std::is_base_of<BaseConvTensorRearrangeOp, Op>::value,
+                                  bool>::type = false>
+std::ostream& operator<<(std::ostream& os, const BaseConvTensorRearrangeOp&)
+{
+    os << Op::name;
+    return os;
+}
+
+} // namespace conv_tensor_rearrange_op
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_image_to_column.hpp b/include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp
similarity index 74%
rename from include/ck/tensor_operation/gpu/device/device_image_to_column.hpp
rename to include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp
index 631d5189dd..898cfe0f2c 100644
--- a/include/ck/tensor_operation/gpu/device/device_image_to_column.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp
@@ -12,21 +12,26 @@ namespace tensor_operation {
 namespace device {
 
 /**
- * \brief Image to column.
+ * \brief Convolution Tensor Rearrange.
  *
- * This Device operator converts image ([G, N, Di, Hi, Wi, C]) to the gemm
- * problem([N * Do * Ho * Wo, Z *  Y * X * C]). G must be equal to 1.
+ * This Device operator supports conversion image ([G, N, Di, Hi, Wi, C]) to
+ * the gemm problem([N * Do * Ho * Wo, Z *  Y * X * C]) (Image to Column) and
+ * conversion gemm form to the image (Column to Image).
+ *
+ * Note that G must be equal to 1.
  *
  * \tparam NDimSpatial Number of spatial dimensions.
- * \tparam InputLayout Input Layout.
+ * \tparam ImageLayout Input Layout.
  * \tparam InputDataType Input Data Type.
  * \tparam OutputDataType Output Data Type.
+ * \tparam ConvTensorRearrangeOp Operation type: ImageToColumn, ColumnToImage.
  */
 template <index_t NDimSpatial,
-          typename InputLayout,
+          typename ImageLayout,
           typename InputDataType,
-          typename OutputDataType>
-struct DeviceImageToColumn : public BaseOperator
+          typename OutputDataType,
+          typename ConvTensorRearrangeOp>
+struct DeviceConvTensorRearrange : public BaseOperator
 {
 
     /**
@@ -39,8 +44,8 @@ struct DeviceImageToColumn : public BaseOperator
      * \param input_spatial_lengths Input spatial lengths.
      * \param filter_spatial_lengths Filter spatial lengths.
      * \param output_spatial_lengths Output spatial lengths.
-     * \param input_g_n_c_wis_strides Input strides in order [G, N, C, D, H, W].
-     * \param output_m_k_strides Output strides.
+     * \param image_g_n_c_wis_strides Image strides in order [G, N, C, D, H, W].
+     * \param gemm_m_k_strides Gemm form strides.
      * \param conv_filter_strides Convolution filter strides.
      * \param conv_filter_dilations Convolution filter dilations.
      * \param input_left_pads Convolution left pads.
@@ -55,8 +60,8 @@ struct DeviceImageToColumn : public BaseOperator
                         const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
-                        const std::array<index_t, 2>& output_m_k_strides,
+                        const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                        const std::array<index_t, 2>& gemm_m_k_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<index_t, NDimSpatial>& input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
new file mode 100644
index 0000000000..f8b4a01681
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
@@ -0,0 +1,621 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
+
+#include "ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp"
+
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+#include "ck/host_utility/io.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// Image to column for input layout NDHWC:
+//   input : image converted to the gemm problem [N * Do * Ho * Wo, Z * Y * X * C]
+//   output : image [N, Di, Hi, Wi, C]
+template <index_t NDimSpatial,
+          typename ImageLayout,
+          typename InputDataType,
+          typename OutputDataType,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t KPerBlock,
+          typename ThreadClusterLengths,
+          index_t ScalarPerVector,
+          typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
+struct DeviceColumnToImageImpl
+    : public DeviceConvTensorRearrange<NDimSpatial,
+                                       ImageLayout,
+                                       InputDataType,
+                                       OutputDataType,
+                                       conv_tensor_rearrange_op::ColumnToImage>
+{
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+
+    static constexpr auto ZIdx = Number<I0>{};
+    static constexpr auto YIdx = NDimSpatial == 1 ? I0 : Number<NDimSpatial - I2>{};
+    static constexpr auto XIdx = Number<NDimSpatial - I1>{};
+
+    static constexpr auto spatial_offset = Number<3>{};
+
+    static constexpr auto conv_to_gemm_transformer =
+        TransformConvFwdToGemm<NDimSpatial, ConvolutionForwardSpecialization::Default>{};
+    static constexpr auto matrix_padder =
+        MatrixPadder<GemmSpecialization::MKPadding, index_t, index_t, index_t>{
+            MPerBlock, 0 /* NPerBlock*/, KPerBlock};
+
+    // Calculate number of independent filters for given conv params
+    static index_t GetNumberOfIndependentFilters(const index_t input_spatial_len,
+                                                 const index_t left_pad,
+                                                 const index_t right_pad,
+                                                 const index_t filter_len,
+                                                 const index_t filter_stride,
+                                                 const index_t filter_dilation,
+                                                 const index_t image_offset)
+    {
+        const index_t x_eff = (filter_len - 1) * filter_dilation + 1;
+        const index_t next_filter_padded =
+            math::integer_divide_ceil(x_eff, filter_stride) * filter_stride;
+        // If filter_stride >= x_eff then each filter is independent
+        const index_t independent_filter_stride =
+            filter_stride >= x_eff ? filter_stride : next_filter_padded;
+        const index_t w_eff = input_spatial_len - image_offset + left_pad + right_pad - x_eff;
+        // There are no independent filters
+        if(w_eff < 0)
+            return 0;
+        const index_t independent_kernels_num = w_eff / independent_filter_stride + 1;
+        return independent_kernels_num;
+    }
+
+    // Make column form descriptor
+    static auto
+    MakeInputDescriptor_M_K(const ck::index_t N,
+                            const ck::index_t C,
+                            const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                            const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                            const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                            const std::array<index_t, 2>& gemm_m_k_strides,
+                            const std::array<index_t, NDimSpatial>& independent_filters,
+                            const std::array<index_t, NDimSpatial>& effs)
+    {
+        const index_t DoHoWo = ck::accumulate_n<index_t>(
+            output_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
+        const index_t CZYX =
+            C * ck::accumulate_n<index_t>(
+                    filter_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
+
+        const index_t NStride = DoHoWo * gemm_m_k_strides[I0] * gemm_m_k_strides[I1];
+        // Calculate the appropriate stride for each set of independent filters
+        // in each dimension
+        const index_t WStride =
+            math::integer_divide_ceil(effs[XIdx], conv_filter_strides[XIdx]) * gemm_m_k_strides[I0];
+        const index_t HStride = math::integer_divide_ceil(effs[YIdx], conv_filter_strides[YIdx]) *
+                                output_spatial_lengths[XIdx] * gemm_m_k_strides[I0];
+        const index_t DStride = math::integer_divide_ceil(effs[ZIdx], conv_filter_strides[ZIdx]) *
+                                output_spatial_lengths[YIdx] * output_spatial_lengths[XIdx] *
+                                gemm_m_k_strides[I0];
+        // Create descriptor for independent filters in each dimension and
+        // then merge them into column form
+        if constexpr(NDimSpatial == 1)
+        {
+            const auto desc_gemm_form =
+                make_naive_tensor_descriptor(make_tuple(N, independent_filters[XIdx], CZYX),
+                                             make_tuple(NStride, WStride, gemm_m_k_strides[I1]));
+            const auto desc_gemm_form_merged_filters = transform_tensor_descriptor(
+                desc_gemm_form,
+                make_tuple(make_merge_transform(make_tuple(N, independent_filters[XIdx])),
+                           make_pass_through_transform(CZYX)),
+                make_tuple(Sequence<0, 1>{}, Sequence<2>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto desc_m_k = matrix_padder.PadADescriptor_M_K(desc_gemm_form_merged_filters);
+            return desc_m_k;
+        }
+        else if constexpr(NDimSpatial == 2)
+        {
+            const auto desc_gemm_form = make_naive_tensor_descriptor(
+                make_tuple(N, independent_filters[YIdx], independent_filters[XIdx], CZYX),
+                make_tuple(NStride, HStride, WStride, gemm_m_k_strides[I1]));
+            const auto desc_gemm_form_merged_filters = transform_tensor_descriptor(
+                desc_gemm_form,
+                make_tuple(make_merge_transform(
+                               make_tuple(N, independent_filters[YIdx], independent_filters[XIdx])),
+                           make_pass_through_transform(CZYX)),
+                make_tuple(Sequence<0, 1, 2>{}, Sequence<3>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto desc_m_k = matrix_padder.PadADescriptor_M_K(desc_gemm_form_merged_filters);
+            return desc_m_k;
+        }
+        else if constexpr(NDimSpatial == 3)
+        {
+            const auto desc_gemm_form = make_naive_tensor_descriptor(
+                make_tuple(N,
+                           independent_filters[ZIdx],
+                           independent_filters[YIdx],
+                           independent_filters[XIdx],
+                           CZYX),
+                make_tuple(NStride, DStride, HStride, WStride, gemm_m_k_strides[I1]));
+            const auto desc_gemm_form_merged_filters = transform_tensor_descriptor(
+                desc_gemm_form,
+                make_tuple(make_merge_transform(make_tuple(N,
+                                                           independent_filters[ZIdx],
+                                                           independent_filters[YIdx],
+                                                           independent_filters[XIdx])),
+                           make_pass_through_transform(CZYX)),
+                make_tuple(Sequence<0, 1, 2, 3>{}, Sequence<4>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+            const auto desc_m_k = matrix_padder.PadADescriptor_M_K(desc_gemm_form_merged_filters);
+            return desc_m_k;
+        }
+    }
+
+    // Use MakeADescriptor_M_K from grouped convolution forward
+    static auto
+    MakeOutDescriptor_M_K(const ck::index_t N,
+                          const ck::index_t C,
+                          const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                          const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                          const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                          const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                          const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                          const std::array<index_t, NDimSpatial>& input_left_pads,
+                          const std::array<index_t, NDimSpatial>& input_right_pads,
+                          const std::array<index_t, NDimSpatial>& image_offsets,
+                          const std::array<index_t, NDimSpatial>& independent_filters,
+                          const std::array<index_t, NDimSpatial>& effs)
+    {
+        std::array<index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{1};
+        std::array<index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{1};
+        std::array<index_t, NDimSpatial + 3> c_g_n_k_wos_lengths{1};
+
+        auto copy = [](const auto& x, auto& y, index_t dst_offset) {
+            std::copy(x.begin(), x.end(), y.begin() + dst_offset);
+        };
+
+        copy(input_spatial_lengths, a_g_n_c_wis_lengths, spatial_offset);
+        copy(filter_spatial_lengths, b_g_k_c_xs_lengths, spatial_offset);
+        // Calculate descriptor only for independent filters
+        copy(independent_filters, c_g_n_k_wos_lengths, spatial_offset);
+
+        // fill only significant values (C and N)
+        a_g_n_c_wis_lengths[I1] = N;
+        a_g_n_c_wis_lengths[I2] = C;
+        b_g_k_c_xs_lengths[I2]  = C;
+        c_g_n_k_wos_lengths[I1] = N;
+
+        // Modify pads to apply offsets
+        std::array<index_t, NDimSpatial> input_left_pads_with_offset;
+        for(index_t i = 0; i < NDimSpatial; i++)
+        {
+            input_left_pads_with_offset[i] = math::max(0, input_left_pads[i] - image_offsets[i]);
+        }
+        // Modify input spatial lengths to apply offsets
+        for(index_t i = 0; i < NDimSpatial; i++)
+        {
+            a_g_n_c_wis_lengths[i + spatial_offset] -=
+                math::max(0, image_offsets[i] - input_left_pads[i]);
+        }
+
+        // Strides to next independent filters
+        std::array<index_t, NDimSpatial> independent_filter_strides;
+        for(index_t i = 0; i < NDimSpatial; i++)
+        {
+            index_t independent_filter_stride =
+                math::integer_divide_ceil(effs[i], conv_filter_strides[i]) * conv_filter_strides[i];
+            // If conv stride is greater than whole filter size, use conv stride
+            independent_filter_strides[i] = conv_filter_strides[i] >= effs[i]
+                                                ? conv_filter_strides[i]
+                                                : independent_filter_stride;
+        }
+
+        // Calculate image form descriptor for the modified convolution problem
+        const auto in_gemmmraw_gemmkraw_desc =
+            conv_to_gemm_transformer.template MakeADescriptor_M_K<ImageLayout>(
+                a_g_n_c_wis_lengths,
+                image_g_n_c_wis_strides,
+                b_g_k_c_xs_lengths,
+                {}, // not needed for A Descriptor
+                c_g_n_k_wos_lengths,
+                {}, // not needed for A Descriptor
+                // conv_filter_strides,
+                independent_filter_strides,
+                conv_filter_dilations,
+                input_left_pads_with_offset,
+                input_right_pads);
+
+        const auto in_gemmm_gemmk_desc =
+            matrix_padder.PadADescriptor_M_K(in_gemmmraw_gemmkraw_desc);
+        return in_gemmm_gemmk_desc;
+    }
+
+    using InputGridDesc =
+        remove_cvref_t<decltype(MakeInputDescriptor_M_K(1, 1, {}, {}, {}, {}, {}, {}))>;
+    using OutputGridDesc = remove_cvref_t<decltype(MakeOutDescriptor_M_K(
+        1, 1, {}, {}, {}, {}, {}, {}, {}, {}, {}, {}))>;
+
+    using Block2ETileMap = remove_cvref_t<
+        decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, InputGridDesc>(
+            InputGridDesc{}))>;
+
+    using GridwiseTensorRearrangeKernel = GridwiseTensorRearrange<InputGridDesc,
+                                                                  InputDataType,
+                                                                  OutputGridDesc,
+                                                                  OutputDataType,
+                                                                  BlockSize,
+                                                                  MPerBlock,
+                                                                  KPerBlock,
+                                                                  ThreadClusterLengths,
+                                                                  ScalarPerVector,
+                                                                  InMemoryDataOperationEnum::Add,
+                                                                  Block2ETileMap>;
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_in, // input image
+                 void* p_out,      // output image
+                 const ck::index_t N,
+                 const ck::index_t C,
+                 const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                 const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                 const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                 const std::array<index_t, 2>& gemm_m_k_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads)
+            : C_(C),
+              X_(filter_spatial_lengths[NDimSpatial - I1]),
+              p_in_{static_cast<const InputDataType*>(p_in)},
+              p_out_{static_cast<OutputDataType*>(p_out)},
+              image_g_n_c_wis_strides_{image_g_n_c_wis_strides},
+              conv_filter_strides_{conv_filter_strides},
+              conv_filter_dilations_{conv_filter_dilations},
+              input_left_pads_{input_left_pads},
+              input_right_pads_{input_right_pads}
+        {
+            const index_t x_eff =
+                (filter_spatial_lengths[XIdx] - 1) * conv_filter_dilations[XIdx] + 1;
+            const index_t y_eff =
+                NDimSpatial < 2
+                    ? I1
+                    : (filter_spatial_lengths[YIdx] - 1) * conv_filter_dilations[YIdx] + 1;
+            const index_t z_eff =
+                NDimSpatial < 3
+                    ? I1
+                    : (filter_spatial_lengths[ZIdx] - 1) * conv_filter_dilations[ZIdx] + 1;
+
+            // Iterate over sets of independent filters
+            for(int z_img_offset = 0; z_img_offset < z_eff;
+                z_img_offset += conv_filter_strides[ZIdx])
+            {
+                for(int y_img_offset = 0; y_img_offset < y_eff;
+                    y_img_offset += conv_filter_strides[YIdx])
+                {
+                    for(int x_img_offset = 0; x_img_offset < x_eff;
+                        x_img_offset += conv_filter_strides[XIdx])
+                    {
+
+                        std::array<index_t, NDimSpatial> image_offsets;
+                        std::array<index_t, NDimSpatial> effs;
+                        // Calculate the starting offset for a given set of
+                        // independent filters
+                        if constexpr(NDimSpatial == 1)
+                        {
+                            image_offsets = {x_img_offset};
+                            effs          = {x_eff};
+                        }
+                        if constexpr(NDimSpatial == 2)
+                        {
+                            image_offsets = {y_img_offset, x_img_offset};
+                            effs          = {y_eff, x_eff};
+                        }
+                        else if constexpr(NDimSpatial == 3)
+                        {
+                            image_offsets = {z_img_offset, y_img_offset, x_img_offset};
+                            effs          = {z_eff, y_eff, x_eff};
+                        }
+
+                        std::array<index_t, NDimSpatial> independent_filters;
+                        for(index_t i = 0; i < NDimSpatial; i++)
+                        {
+                            independent_filters[i] =
+                                GetNumberOfIndependentFilters(input_spatial_lengths[i],
+                                                              input_left_pads[i],
+                                                              input_right_pads[i],
+                                                              filter_spatial_lengths[i],
+                                                              conv_filter_strides[i],
+                                                              conv_filter_dilations[i],
+                                                              image_offsets[i]);
+                        }
+                        const index_t independent_filters_acum = ck::accumulate_n<index_t>(
+                            independent_filters.begin(), NDimSpatial, 1, std::multiplies<>());
+                        if(independent_filters_acum <= 0)
+                            continue;
+
+                        const auto in_grid_desc_m_k =
+                            MakeInputDescriptor_M_K(N,
+                                                    C,
+                                                    filter_spatial_lengths,
+                                                    output_spatial_lengths,
+                                                    conv_filter_strides,
+                                                    gemm_m_k_strides,
+                                                    independent_filters,
+                                                    effs);
+                        const auto out_grid_desc_m_k =
+                            MakeOutDescriptor_M_K(N,
+                                                  C,
+                                                  input_spatial_lengths,
+                                                  filter_spatial_lengths,
+                                                  image_g_n_c_wis_strides,
+                                                  conv_filter_strides,
+                                                  conv_filter_dilations,
+                                                  input_left_pads,
+                                                  input_right_pads,
+                                                  image_offsets,
+                                                  independent_filters,
+                                                  effs);
+                        in_grid_desc_m_k_container_.push_back(in_grid_desc_m_k);
+                        out_grid_desc_m_k_container_.push_back(out_grid_desc_m_k);
+
+                        const index_t x_idx = x_img_offset / conv_filter_strides[XIdx];
+                        const index_t y_idx = y_img_offset / conv_filter_strides[YIdx];
+                        const index_t z_idx = z_img_offset / conv_filter_strides[ZIdx];
+
+                        const index_t x_offset_with_pad =
+                            math::max(0, x_img_offset - input_left_pads[XIdx]);
+                        const index_t y_offset_with_pad =
+                            math::max(0, y_img_offset - input_left_pads[YIdx]);
+                        const index_t z_offset_with_pad =
+                            math::max(0, z_img_offset - input_left_pads[ZIdx]);
+
+                        // Memory offsets to next set of independent filters,
+                        // move to independent filters in each dimension
+                        const index_t in_offset =
+                            x_idx * gemm_m_k_strides[0] +
+                            y_idx * gemm_m_k_strides[0] * output_spatial_lengths[XIdx] +
+                            z_idx * gemm_m_k_strides[0] * output_spatial_lengths[YIdx] *
+                                output_spatial_lengths[XIdx];
+                        // Move to independent filters in appropriate dimensions
+                        const index_t out_offset =
+                            x_offset_with_pad * image_g_n_c_wis_strides[spatial_offset + XIdx] +
+                            y_offset_with_pad * image_g_n_c_wis_strides[spatial_offset + YIdx] +
+                            z_offset_with_pad * image_g_n_c_wis_strides[spatial_offset + ZIdx];
+
+                        const InputDataType* p_in_with_offset =
+                            static_cast<const InputDataType*>(p_in) + in_offset;
+                        OutputDataType* p_out_with_offset =
+                            static_cast<OutputDataType*>(p_out) + out_offset;
+                        p_in_container_.push_back(p_in_with_offset);
+                        p_out_container_.push_back(p_out_with_offset);
+                    }
+                }
+            }
+        }
+
+        void Print() const
+        {
+            for(std::size_t i = 0; i < in_grid_desc_m_k_container_.size(); i++)
+            {
+                std::cout << in_grid_desc_m_k_container_[i] << std::endl;
+                std::cout << out_grid_desc_m_k_container_[i] << std::endl;
+            }
+        }
+
+        const ck::index_t C_;
+        const ck::index_t X_;
+
+        const InputDataType* p_in_;
+        OutputDataType* p_out_;
+
+        const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides_;
+        const std::array<index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<index_t, NDimSpatial>& conv_filter_dilations_;
+        const std::array<index_t, NDimSpatial>& input_left_pads_;
+        const std::array<index_t, NDimSpatial>& input_right_pads_;
+
+        std::vector<InputGridDesc> in_grid_desc_m_k_container_;
+        std::vector<OutputGridDesc> out_grid_desc_m_k_container_;
+
+        std::vector<const InputDataType*> p_in_container_;
+        std::vector<OutputDataType*> p_out_container_;
+    };
+
+    struct Invoker : public BaseInvoker
+    {
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                arg.Print();
+            }
+
+            float elapsed_time = 0.f;
+            const auto kernel  = kernel_tensor_rearrange<InputGridDesc,
+                                                        InputDataType,
+                                                        OutputGridDesc,
+                                                        OutputDataType,
+                                                        Block2ETileMap,
+                                                        GridwiseTensorRearrangeKernel>;
+
+            // Execute each set of independent filters
+            for(std::size_t i = 0; i < arg.in_grid_desc_m_k_container_.size(); i++)
+            {
+                const auto block_2_tile_map =
+                    BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, InputGridDesc>(
+                        arg.out_grid_desc_m_k_container_[i]);
+                const index_t grid_size =
+                    block_2_tile_map.CalculateGridSize(arg.in_grid_desc_m_k_container_[i]);
+                elapsed_time += launch_and_time_kernel(stream_config,
+                                                       kernel,
+                                                       dim3(grid_size),
+                                                       dim3(BlockSize),
+                                                       0,
+                                                       arg.in_grid_desc_m_k_container_[i],
+                                                       arg.p_in_container_[i],
+                                                       arg.out_grid_desc_m_k_container_[i],
+                                                       arg.p_out_container_[i],
+                                                       block_2_tile_map);
+            }
+            return elapsed_time;
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    bool IsSupportedArgument(const Argument& arg)
+    {
+        using namespace tensor_layout::convolution;
+        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
+                       std::is_same_v<ImageLayout, GNDHWC>))
+        {
+            return false;
+        }
+
+        const auto w_pad_left  = arg.input_left_pads_[NDimSpatial - I1];
+        const auto w_pad_right = arg.input_right_pads_[NDimSpatial - I1];
+        const auto dilation_x  = arg.conv_filter_dilations_[NDimSpatial - I1];
+        const auto stride_x    = arg.conv_filter_strides_[NDimSpatial - I1];
+        bool is_w_packed       = arg.image_g_n_c_wis_strides_[NDimSpatial + I2] == arg.C_;
+        bool is_c_packed       = arg.image_g_n_c_wis_strides_[I2] == 1;
+
+        // check vector acces with c not packed
+        if(!is_c_packed && ScalarPerVector != 1)
+            return false;
+        // check vector access of filter window row (only C if C is not packed)
+        if(!is_w_packed && arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of filter window row (X * C)
+        if(arg.X_ * arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of pads (w_pad_left/w_pad_right * C)
+        if(w_pad_left * arg.C_ % ScalarPerVector != 0 ||
+           w_pad_right * arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of with stride and pad
+        if((w_pad_left != 0 || w_pad_right != 0) && stride_x > 1 && arg.C_ % ScalarPerVector != 0)
+            return false;
+        // check vector access of with dilation
+        if(dilation_x > 1 && arg.C_ % ScalarPerVector != 0)
+            return false;
+
+        bool valid = true;
+        for(std::size_t i = 0; i < arg.in_grid_desc_m_k_container_.size(); i++)
+        {
+            valid &= GridwiseTensorRearrangeKernel::CheckValidity(
+                arg.in_grid_desc_m_k_container_[i], arg.out_grid_desc_m_k_container_[i]);
+        }
+        return valid;
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const void* p_in, // input image
+                             void* p_out,      // output image
+                             const ck::index_t N,
+                             const ck::index_t C,
+                             const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                             const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                             const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                             const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                             const std::array<index_t, 2>& gemm_m_k_strides,
+                             const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                             const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                             const std::array<index_t, NDimSpatial>& input_left_pads,
+                             const std::array<index_t, NDimSpatial>& input_right_pads)
+    {
+        return Argument{static_cast<const InputDataType*>(p_in),
+                        static_cast<OutputDataType*>(p_out),
+                        N,
+                        C,
+                        input_spatial_lengths,
+                        filter_spatial_lengths,
+                        output_spatial_lengths,
+                        image_g_n_c_wis_strides,
+                        gemm_m_k_strides,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in, // input image
+                        void* p_out,      // output image
+                        const ck::index_t N,
+                        const ck::index_t C,
+                        const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+                        const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+                        const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+                        const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                        const std::array<index_t, 2>& gemm_m_k_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<index_t, NDimSpatial>& input_left_pads,
+                        const std::array<index_t, NDimSpatial>& input_right_pads) override
+    {
+        return std::make_unique<Argument>(static_cast<const InputDataType*>(p_in),
+                                          static_cast<OutputDataType*>(p_out),
+                                          N,
+                                          C,
+                                          input_spatial_lengths,
+                                          filter_spatial_lengths,
+                                          output_spatial_lengths,
+                                          image_g_n_c_wis_strides,
+                                          gemm_m_k_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceColumnToImage"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << KPerBlock << ", "
+            << ScalarPerVector
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
index 19f126e66f..8b4ae7875c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
@@ -5,64 +5,41 @@
 
 #include "ck/tensor_description/tensor_descriptor.hpp"
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp"
 #include "ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_forward_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
 #include "ck/host_utility/io.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 
-template <typename InputGridDesc,
-          typename InputDataType,
-          typename OutputGridDesc,
-          typename OutputDataType,
-          typename Block2ETileMap,
-          typename GridwiseImageToColumnKernel>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_image_to_column(const InputGridDesc in_grid_desc,
-                               const InputDataType* __restrict__ p_in_global,
-                               const OutputGridDesc out_grid_desc,
-                               OutputDataType* __restrict__ p_out_global,
-                               const Block2ETileMap block_2_tile_map)
-{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
-    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
-    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
-    GridwiseImageToColumnKernel::Run(
-        in_grid_desc, p_in_global, out_grid_desc, p_out_global, block_2_tile_map);
-#else
-    ignore = in_grid_desc;
-    ignore = p_in_global;
-    ignore = out_grid_desc;
-    ignore = p_out_global;
-    ignore = block_2_tile_map;
-#endif
-}
-
 // Image to column for input layout NDHWC:
-//   input : input image [N, Di, Hi, Wi, C],
-//   output : output image [N * Do * Ho * Wo, Z * Y * X * C]
+//   input : input image [N, Di, Hi, Wi, C]
+//   output : gemm form [N * Do * Ho * Wo, Z * Y * X * C]
 template <index_t NDimSpatial,
-          typename InputLayout,
+          typename ImageLayout,
           typename InputDataType,
           typename OutputDataType,
           index_t BlockSize,
           index_t MPerBlock,
           index_t KPerBlock,
           typename ThreadClusterLengths,
-          index_t ScalarPerVector>
+          index_t ScalarPerVector,
+          typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
 struct DeviceImageToColumnImpl
-    : public DeviceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>
+    : public DeviceConvTensorRearrange<NDimSpatial,
+                                       ImageLayout,
+                                       InputDataType,
+                                       OutputDataType,
+                                       conv_tensor_rearrange_op::ImageToColumn>
 {
 
     static constexpr auto I0 = Number<0>{};
@@ -83,7 +60,7 @@ struct DeviceImageToColumnImpl
                             const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                             const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                             const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                            const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
+                            const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
                             const std::array<index_t, NDimSpatial>& conv_filter_strides,
                             const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                             const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -110,9 +87,9 @@ struct DeviceImageToColumnImpl
         c_g_n_k_wos_lengths[I1] = N;
 
         const auto in_gemmmraw_gemmkraw_desc =
-            conv_to_gemm_transformer.template MakeADescriptor_M_K<InputLayout>(
+            conv_to_gemm_transformer.template MakeADescriptor_M_K<ImageLayout>(
                 a_g_n_c_wis_lengths,
-                input_g_n_c_wis_strides,
+                image_g_n_c_wis_strides,
                 b_g_k_c_xs_lengths,
                 {}, // not needed for A Descriptor
                 c_g_n_k_wos_lengths,
@@ -132,7 +109,7 @@ struct DeviceImageToColumnImpl
                           const ck::index_t C,
                           const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                           const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                          const std::array<index_t, 2>& output_m_k_strides)
+                          const std::array<index_t, 2>& gemm_m_k_strides)
     {
         const index_t NDoHoWo =
             N * ck::accumulate_n<index_t>(
@@ -141,7 +118,7 @@ struct DeviceImageToColumnImpl
             C * ck::accumulate_n<index_t>(
                     filter_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
         const auto desc_mraw_kraw = make_naive_tensor_descriptor(
-            make_tuple(NDoHoWo, CZYX), make_tuple(output_m_k_strides[I0], output_m_k_strides[I1]));
+            make_tuple(NDoHoWo, CZYX), make_tuple(gemm_m_k_strides[I0], gemm_m_k_strides[I1]));
 
         const auto desc_m_k = matrix_padder.PadADescriptor_M_K(desc_mraw_kraw);
         return desc_m_k;
@@ -155,28 +132,29 @@ struct DeviceImageToColumnImpl
         decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
             OutputGridDesc{}))>;
 
-    using GridwiseImageToColumnKernel = GridwiseImageToColumn<InputGridDesc,
-                                                              InputDataType,
-                                                              OutputGridDesc,
-                                                              OutputDataType,
-                                                              BlockSize,
-                                                              MPerBlock,
-                                                              KPerBlock,
-                                                              ThreadClusterLengths,
-                                                              ScalarPerVector,
-                                                              Block2ETileMap>;
+    using GridwiseTensorRearrangeKernel = GridwiseTensorRearrange<InputGridDesc,
+                                                                  InputDataType,
+                                                                  OutputGridDesc,
+                                                                  OutputDataType,
+                                                                  BlockSize,
+                                                                  MPerBlock,
+                                                                  KPerBlock,
+                                                                  ThreadClusterLengths,
+                                                                  ScalarPerVector,
+                                                                  InMemoryDataOperationEnum::Set,
+                                                                  Block2ETileMap>;
 
     struct Argument : public BaseArgument
     {
         Argument(const void* p_in, // input image
-                 void* p_out,      // output image
+                 void* p_out,      // gemm form
                  const ck::index_t N,
                  const ck::index_t C,
                  const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                  const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                  const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                 const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
-                 const std::array<index_t, 2>& output_m_k_strides,
+                 const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                 const std::array<index_t, 2>& gemm_m_k_strides,
                  const std::array<index_t, NDimSpatial>& conv_filter_strides,
                  const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                  const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -185,7 +163,7 @@ struct DeviceImageToColumnImpl
               X_(filter_spatial_lengths[NDimSpatial - I1]),
               p_in_{static_cast<const InputDataType*>(p_in)},
               p_out_{static_cast<OutputDataType*>(p_out)},
-              input_g_n_c_wis_strides_{input_g_n_c_wis_strides},
+              image_g_n_c_wis_strides_{image_g_n_c_wis_strides},
               conv_filter_strides_{conv_filter_strides},
               conv_filter_dilations_{conv_filter_dilations},
               input_left_pads_{input_left_pads},
@@ -197,7 +175,7 @@ struct DeviceImageToColumnImpl
                                                         input_spatial_lengths,
                                                         filter_spatial_lengths,
                                                         output_spatial_lengths,
-                                                        input_g_n_c_wis_strides,
+                                                        image_g_n_c_wis_strides,
 
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
@@ -205,7 +183,7 @@ struct DeviceImageToColumnImpl
                                                         input_right_pads);
 
             out_grid_desc_m_k_ = MakeOutDescriptor_M_K(
-                N, C, filter_spatial_lengths, output_spatial_lengths, output_m_k_strides);
+                N, C, filter_spatial_lengths, output_spatial_lengths, gemm_m_k_strides);
         }
 
         void Print() const
@@ -220,7 +198,7 @@ struct DeviceImageToColumnImpl
         const InputDataType* p_in_;
         OutputDataType* p_out_;
 
-        const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides_;
+        const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides_;
         const std::array<index_t, NDimSpatial>& conv_filter_strides_;
         const std::array<index_t, NDimSpatial>& conv_filter_dilations_;
         const std::array<index_t, NDimSpatial>& input_left_pads_;
@@ -243,12 +221,12 @@ struct DeviceImageToColumnImpl
                 BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
                     arg.out_grid_desc_m_k_);
             const index_t grid_size = block_2_tile_map.CalculateGridSize(arg.out_grid_desc_m_k_);
-            const auto kernel       = kernel_image_to_column<InputGridDesc,
-                                                       InputDataType,
-                                                       OutputGridDesc,
-                                                       OutputDataType,
-                                                       Block2ETileMap,
-                                                       GridwiseImageToColumnKernel>;
+            const auto kernel       = kernel_tensor_rearrange<InputGridDesc,
+                                                        InputDataType,
+                                                        OutputGridDesc,
+                                                        OutputDataType,
+                                                        Block2ETileMap,
+                                                        GridwiseTensorRearrangeKernel>;
 
             float elapsed_time = launch_and_time_kernel(stream_config,
                                                         kernel,
@@ -273,12 +251,8 @@ struct DeviceImageToColumnImpl
     bool IsSupportedArgument(const Argument& arg)
     {
         using namespace tensor_layout::convolution;
-        if(!(std::is_same_v<InputLayout, GNWC> || std::is_same_v<InputLayout, GNHWC> ||
-             std::is_same_v<InputLayout, GNDHWC>))
-        {
-            return false;
-        }
-        if(!(NDimSpatial >= 1 && NDimSpatial <= 3))
+        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
+                       std::is_same_v<ImageLayout, GNDHWC>))
         {
             return false;
         }
@@ -287,8 +261,8 @@ struct DeviceImageToColumnImpl
         const auto w_pad_right = arg.input_right_pads_[NDimSpatial - I1];
         const auto dilation_x  = arg.conv_filter_dilations_[NDimSpatial - I1];
         const auto stride_x    = arg.conv_filter_strides_[NDimSpatial - I1];
-        bool is_w_packed       = arg.input_g_n_c_wis_strides_[NDimSpatial + I2] == arg.C_;
-        bool is_c_packed       = arg.input_g_n_c_wis_strides_[I2] == 1;
+        bool is_w_packed       = arg.image_g_n_c_wis_strides_[NDimSpatial + I2] == arg.C_;
+        bool is_c_packed       = arg.image_g_n_c_wis_strides_[I2] == 1;
 
         // check vector acces with c not packed
         if(!is_c_packed && ScalarPerVector != 1)
@@ -310,8 +284,8 @@ struct DeviceImageToColumnImpl
         if(dilation_x > 1 && arg.C_ % ScalarPerVector != 0)
             return false;
 
-        return GridwiseImageToColumnKernel::CheckValidity(arg.in_grid_desc_m_k_,
-                                                          arg.out_grid_desc_m_k_);
+        return GridwiseTensorRearrangeKernel::CheckValidity(arg.in_grid_desc_m_k_,
+                                                            arg.out_grid_desc_m_k_);
     }
 
     bool IsSupportedArgument(const BaseArgument* p_arg) override
@@ -320,14 +294,14 @@ struct DeviceImageToColumnImpl
     }
 
     static auto MakeArgument(const void* p_in, // input image
-                             void* p_out,      // output image
+                             void* p_out,      // gemm form
                              const ck::index_t N,
                              const ck::index_t C,
                              const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                              const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                              const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                             const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
-                             const std::array<index_t, 2>& output_m_k_strides,
+                             const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                             const std::array<index_t, 2>& gemm_m_k_strides,
                              const std::array<index_t, NDimSpatial>& conv_filter_strides,
                              const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                              const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -340,8 +314,8 @@ struct DeviceImageToColumnImpl
                         input_spatial_lengths,
                         filter_spatial_lengths,
                         output_spatial_lengths,
-                        input_g_n_c_wis_strides,
-                        output_m_k_strides,
+                        image_g_n_c_wis_strides,
+                        gemm_m_k_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -352,14 +326,14 @@ struct DeviceImageToColumnImpl
 
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in, // input image
-                        void* p_out,      // output image
+                        void* p_out,      // gemm form
                         const ck::index_t N,
                         const ck::index_t C,
                         const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& input_g_n_c_wis_strides,
-                        const std::array<index_t, 2>& output_m_k_strides,
+                        const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
+                        const std::array<index_t, 2>& gemm_m_k_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -372,8 +346,8 @@ struct DeviceImageToColumnImpl
                                           input_spatial_lengths,
                                           filter_spatial_lengths,
                                           output_spatial_lengths,
-                                          input_g_n_c_wis_strides,
-                                          output_m_k_strides,
+                                          image_g_n_c_wis_strides,
+                                          gemm_m_k_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
similarity index 52%
rename from include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp
rename to include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
index 93625a324e..e4e47b4fae 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_image_to_column.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
@@ -16,6 +16,36 @@
 
 namespace ck {
 
+template <typename InputGridDesc,
+          typename InputDataType,
+          typename OutputGridDesc,
+          typename OutputDataType,
+          typename Block2ETileMap,
+          typename GridwiseTensorRearrangeKernel>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_tensor_rearrange(const InputGridDesc in_grid_desc,
+                                const InputDataType* __restrict__ p_in_global,
+                                const OutputGridDesc out_grid_desc,
+                                OutputDataType* __restrict__ p_out_global,
+                                const Block2ETileMap block_2_tile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
+    defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
+    defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
+    GridwiseTensorRearrangeKernel::Run(
+        in_grid_desc, p_in_global, out_grid_desc, p_out_global, block_2_tile_map);
+#else
+    ignore = in_grid_desc;
+    ignore = p_in_global;
+    ignore = out_grid_desc;
+    ignore = p_out_global;
+    ignore = block_2_tile_map;
+#endif
+}
+
 template <typename InputGridDesc,
           typename InputDataType,
           typename OutputGridDesc,
@@ -25,8 +55,9 @@ template <typename InputGridDesc,
           index_t KPerBlock,
           typename ThreadClusterLengths,
           index_t ScalarPerVector,
+          InMemoryDataOperationEnum DstInMemOp,
           typename Block2ETileMap>
-struct GridwiseImageToColumn
+struct GridwiseTensorRearrange
 {
 
     static constexpr auto I0 = Number<0>{};
@@ -55,27 +86,27 @@ struct GridwiseImageToColumn
         auto out_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_out_global, out_grid_desc.GetElementSpaceSize());
 
-        auto copy_global_to_global = ThreadGroupTensorSliceTransfer_v7<
-            ThisThreadBlock,
-            Tuple<InputDataType>,
-            Tuple<OutputDataType>,
-            decltype(tie(in_grid_desc)),
-            decltype(tie(out_grid_desc)),
-            tensor_operation::element_wise::PassThrough,
-            Sequence<static_cast<index_t>(InMemoryDataOperationEnum::Set)>,
-            Sequence<MPerBlock, KPerBlock>,
-            ThreadClusterLengths,
-            Sequence<0, 1>,
-            Sequence<0, 1>,
-            I1,
-            ScalarPerVector,
-            Sequence<true>,
-            Sequence<true>>{
-            in_grid_desc,
-            make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
-            out_grid_desc,
-            make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
-            tensor_operation::element_wise::PassThrough{}};
+        auto copy_global_to_global =
+            ThreadGroupTensorSliceTransfer_v7<ThisThreadBlock,
+                                              Tuple<InputDataType>,
+                                              Tuple<OutputDataType>,
+                                              decltype(tie(in_grid_desc)),
+                                              decltype(tie(out_grid_desc)),
+                                              tensor_operation::element_wise::PassThrough,
+                                              Sequence<static_cast<index_t>(DstInMemOp)>,
+                                              Sequence<MPerBlock, KPerBlock>,
+                                              ThreadClusterLengths,
+                                              Sequence<0, 1>,
+                                              Sequence<0, 1>,
+                                              I1,
+                                              ScalarPerVector,
+                                              Sequence<true>,
+                                              Sequence<true>>{
+                in_grid_desc,
+                make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
+                out_grid_desc,
+                make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
+                tensor_operation::element_wise::PassThrough{}};
 
         copy_global_to_global.Run(
             tie(in_grid_desc), tie(in_global_buf), tie(out_grid_desc), tie(out_global_buf));
diff --git a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
index cee3d2825b..6f546f1d6d 100644
--- a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
@@ -20,348 +20,13 @@ struct TransformConvFwdToGemm
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
 
-    template <typename ALayout,
-              typename std::enable_if<NDimSpatial == 1 &&
-                                          is_same_v<ALayout, tensor_layout::convolution::GNWC>,
-                                      bool>::type = false>
-    static auto
-    MakeADescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* a_g_n_c_wis_strides */,
-                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* b_g_k_c_xs_strides */,
-                        const std::array<index_t, NDimSpatial + 3>& c_g_n_k_wos_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* c_g_n_k_wos_strides */,
-                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
-                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
-                        const std::array<index_t, NDimSpatial>& input_left_pads,
-                        const std::array<index_t, NDimSpatial>& input_right_pads)
-    {
-        const index_t N = a_g_n_c_wis_lengths[1];
-        const index_t C = a_g_n_c_wis_lengths[2];
-
-        const index_t Wi = a_g_n_c_wis_lengths[3];
-
-        const index_t Wo = c_g_n_k_wos_lengths[3];
-
-        const index_t ConvStrideW = conv_filter_strides[0];
-
-        if constexpr(ConvForwardSpecialization ==
-                     device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
-        {
-            const index_t NWo =
-                N * ck::accumulate_n<index_t>(
-                        c_g_n_k_wos_lengths.begin() + 3, NDimSpatial, 1, std::multiplies<>());
-
-            const auto in_gemmm_gemmk_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(NWo, C));
-
-            return in_gemmm_gemmk_desc;
-        }
-        else if constexpr(ConvForwardSpecialization ==
-                          device::ConvolutionForwardSpecialization::Filter1x1Pad0)
-        {
-            const auto in_n_wi_c_desc = make_naive_tensor_descriptor_packed(make_tuple(N, Wi, C));
-
-            const auto in_n_wo_c_desc = transform_tensor_descriptor(
-                in_n_wi_c_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_embed_transform(make_tuple(Wo), make_tuple(ConvStrideW)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-
-            const auto in_gemmm_gemmk_desc = transform_tensor_descriptor(
-                in_n_wo_c_desc,
-                make_tuple(make_merge_transform(make_tuple(N, Wo)), make_pass_through_transform(C)),
-                make_tuple(Sequence<0, 1>{}, Sequence<2>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            return in_gemmm_gemmk_desc;
-        }
-        else
-        {
-            const index_t X             = b_g_k_c_xs_lengths[3];
-            const index_t ConvDilationW = conv_filter_dilations[0];
-            const index_t InLeftPadW    = input_left_pads[0];
-            const index_t InRightPadW   = input_right_pads[0];
-
-            const auto in_n_wi_c_desc = make_naive_tensor_descriptor_packed(make_tuple(N, Wi, C));
-
-            const auto in_n_wip_c_desc = transform_tensor_descriptor(
-                in_n_wi_c_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
-
-            const auto in_n_x_wo_c_desc = transform_tensor_descriptor(
-                in_n_wip_c_desc,
-                make_tuple(
-                    make_pass_through_transform(N),
-                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                    make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3>{}));
-
-            const auto in_gemmm_gemmk_desc =
-                transform_tensor_descriptor(in_n_x_wo_c_desc,
-                                            make_tuple(make_merge_transform(make_tuple(N, Wo)),
-                                                       make_merge_transform(make_tuple(X, C))),
-                                            make_tuple(Sequence<0, 2>{}, Sequence<1, 3>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            return in_gemmm_gemmk_desc;
-        }
-    }
-
-    template <typename ALayout,
-              typename std::enable_if<NDimSpatial == 2 &&
-                                          is_same_v<ALayout, tensor_layout::convolution::GNHWC>,
-                                      bool>::type = false>
-    static auto
-    MakeADescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* a_g_n_c_wis_strides */,
-                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* b_g_k_c_xs_strides */,
-                        const std::array<index_t, NDimSpatial + 3>& c_g_n_k_wos_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* c_g_n_k_wos_strides */,
-                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
-                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
-                        const std::array<index_t, NDimSpatial>& input_left_pads,
-                        const std::array<index_t, NDimSpatial>& input_right_pads)
-    {
-        const index_t N = a_g_n_c_wis_lengths[1];
-        const index_t C = a_g_n_c_wis_lengths[2];
-
-        const index_t Hi = a_g_n_c_wis_lengths[3];
-        const index_t Wi = a_g_n_c_wis_lengths[4];
-
-        const index_t Ho = c_g_n_k_wos_lengths[3];
-        const index_t Wo = c_g_n_k_wos_lengths[4];
-
-        const index_t ConvStrideH = conv_filter_strides[0];
-        const index_t ConvStrideW = conv_filter_strides[1];
-
-        if constexpr(ConvForwardSpecialization ==
-                     device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
-        {
-            const index_t NHoWo =
-                N * ck::accumulate_n<index_t>(
-                        c_g_n_k_wos_lengths.begin() + 3, NDimSpatial, 1, std::multiplies<>());
-
-            const auto in_gemmm_gemmk_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(NHoWo, C));
-
-            return in_gemmm_gemmk_desc;
-        }
-        else if constexpr(ConvForwardSpecialization ==
-                          device::ConvolutionForwardSpecialization::Filter1x1Pad0)
-        {
-            const auto in_n_hi_wi_c_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
-
-            const auto in_n_ho_wo_c_desc = transform_tensor_descriptor(
-                in_n_hi_wi_c_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_embed_transform(make_tuple(Ho), make_tuple(ConvStrideH)),
-                           make_embed_transform(make_tuple(Wo), make_tuple(ConvStrideW)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-            const auto in_gemmm_gemmk_desc =
-                transform_tensor_descriptor(in_n_ho_wo_c_desc,
-                                            make_tuple(make_merge_transform(make_tuple(N, Ho, Wo)),
-                                                       make_pass_through_transform(C)),
-                                            make_tuple(Sequence<0, 1, 2>{}, Sequence<3>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            return in_gemmm_gemmk_desc;
-        }
-        else
-        {
-            const index_t Y = b_g_k_c_xs_lengths[3];
-            const index_t X = b_g_k_c_xs_lengths[4];
-
-            const index_t ConvDilationH = conv_filter_dilations[0];
-            const index_t ConvDilationW = conv_filter_dilations[1];
-
-            const index_t InLeftPadH = input_left_pads[0];
-            const index_t InLeftPadW = input_left_pads[1];
-
-            const index_t InRightPadH = input_right_pads[0];
-            const index_t InRightPadW = input_right_pads[1];
-
-            const auto in_n_hi_wi_c_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Hi, Wi, C));
-
-            const auto in_n_hip_wip_c_desc = transform_tensor_descriptor(
-                in_n_hi_wi_c_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}));
-
-            const auto in_n_y_ho_x_wo_c_desc = transform_tensor_descriptor(
-                in_n_hip_wip_c_desc,
-                make_tuple(
-                    make_pass_through_transform(N),
-                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                    make_pass_through_transform(C)),
-                make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}),
-                make_tuple(Sequence<0>{}, Sequence<1, 2>{}, Sequence<3, 4>{}, Sequence<5>{}));
-
-            const auto in_gemmm_gemmk_desc =
-                transform_tensor_descriptor(in_n_y_ho_x_wo_c_desc,
-                                            make_tuple(make_merge_transform(make_tuple(N, Ho, Wo)),
-                                                       make_merge_transform(make_tuple(Y, X, C))),
-                                            make_tuple(Sequence<0, 2, 4>{}, Sequence<1, 3, 5>{}),
-                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            return in_gemmm_gemmk_desc;
-        }
-    }
-
-    template <typename ALayout,
-              typename std::enable_if<NDimSpatial == 3 &&
-                                          is_same_v<ALayout, tensor_layout::convolution::GNDHWC>,
-                                      bool>::type = false>
-    static auto
-    MakeADescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* a_g_n_c_wis_strides */,
-                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* b_g_k_c_xs_strides */,
-                        const std::array<index_t, NDimSpatial + 3>& c_g_n_k_wos_lengths,
-                        const std::array<index_t, NDimSpatial + 3>& /* c_g_n_k_wos_strides */,
-                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
-                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
-                        const std::array<index_t, NDimSpatial>& input_left_pads,
-                        const std::array<index_t, NDimSpatial>& input_right_pads)
-    {
-        const index_t N = a_g_n_c_wis_lengths[1];
-        const index_t C = a_g_n_c_wis_lengths[2];
-
-        const index_t Di = a_g_n_c_wis_lengths[3];
-        const index_t Hi = a_g_n_c_wis_lengths[4];
-        const index_t Wi = a_g_n_c_wis_lengths[5];
-
-        const index_t Do = c_g_n_k_wos_lengths[3];
-        const index_t Ho = c_g_n_k_wos_lengths[4];
-        const index_t Wo = c_g_n_k_wos_lengths[5];
-
-        const index_t ConvStrideD = conv_filter_strides[0];
-        const index_t ConvStrideH = conv_filter_strides[1];
-        const index_t ConvStrideW = conv_filter_strides[2];
-
-        if constexpr(ConvForwardSpecialization ==
-                     device::ConvolutionForwardSpecialization::Filter1x1Stride1Pad0)
-        {
-            const index_t NDoHoWo =
-                N * ck::accumulate_n<index_t>(
-                        c_g_n_k_wos_lengths.begin() + 3, NDimSpatial, 1, std::multiplies<>());
-
-            const auto in_gemmm_gemmk_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(NDoHoWo, C));
-
-            return in_gemmm_gemmk_desc;
-        }
-        else if constexpr(ConvForwardSpecialization ==
-                          device::ConvolutionForwardSpecialization::Filter1x1Pad0)
-        {
-            const auto in_n_di_hi_wi_c_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
-
-            const auto in_n_do_ho_wo_c_desc = transform_tensor_descriptor(
-                in_n_di_hi_wi_c_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_embed_transform(make_tuple(Do), make_tuple(ConvStrideD)),
-                           make_embed_transform(make_tuple(Ho), make_tuple(ConvStrideH)),
-                           make_embed_transform(make_tuple(Wo), make_tuple(ConvStrideW)),
-                           make_pass_through_transform(C)),
-                make_tuple(
-                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
-                make_tuple(
-                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
-
-            const auto in_gemmm_gemmk_desc = transform_tensor_descriptor(
-                in_n_do_ho_wo_c_desc,
-                make_tuple(make_merge_transform(make_tuple(N, Do, Ho, Wo)),
-                           make_pass_through_transform(C)),
-                make_tuple(Sequence<0, 1, 2, 3>{}, Sequence<4>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            return in_gemmm_gemmk_desc;
-        }
-        else
-        {
-            const index_t Z = b_g_k_c_xs_lengths[3];
-            const index_t Y = b_g_k_c_xs_lengths[4];
-            const index_t X = b_g_k_c_xs_lengths[5];
-
-            const index_t ConvDilationD = conv_filter_dilations[0];
-            const index_t ConvDilationH = conv_filter_dilations[1];
-            const index_t ConvDilationW = conv_filter_dilations[2];
-
-            const index_t InLeftPadD = input_left_pads[0];
-            const index_t InLeftPadH = input_left_pads[1];
-            const index_t InLeftPadW = input_left_pads[2];
-
-            const index_t InRightPadD = input_right_pads[0];
-            const index_t InRightPadH = input_right_pads[1];
-            const index_t InRightPadW = input_right_pads[2];
-
-            const auto in_n_di_hi_wi_c_desc =
-                make_naive_tensor_descriptor_packed(make_tuple(N, Di, Hi, Wi, C));
-
-            const auto in_n_hip_wip_c_desc = transform_tensor_descriptor(
-                in_n_di_hi_wi_c_desc,
-                make_tuple(make_pass_through_transform(N),
-                           make_pad_transform(Di, InLeftPadD, InRightPadD),
-                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
-                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
-                           make_pass_through_transform(C)),
-                make_tuple(
-                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
-                make_tuple(
-                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
-
-            const auto in_n_z_do_y_ho_x_wo_c_desc = transform_tensor_descriptor(
-                in_n_hip_wip_c_desc,
-                make_tuple(
-                    make_pass_through_transform(N),
-                    make_embed_transform(make_tuple(Z, Do), make_tuple(ConvDilationD, ConvStrideD)),
-                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
-                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
-                    make_pass_through_transform(C)),
-                make_tuple(
-                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
-                make_tuple(Sequence<0>{},
-                           Sequence<1, 2>{},
-                           Sequence<3, 4>{},
-                           Sequence<5, 6>{},
-                           Sequence<7>{}));
-
-            const auto in_gemmm_gemmk_desc = transform_tensor_descriptor(
-                in_n_z_do_y_ho_x_wo_c_desc,
-                make_tuple(make_merge_transform(make_tuple(N, Do, Ho, Wo)),
-                           make_merge_transform(make_tuple(Z, Y, X, C))),
-                make_tuple(Sequence<0, 2, 4, 6>{}, Sequence<1, 3, 5, 7>{}),
-                make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-            return in_gemmm_gemmk_desc;
-        }
-    }
-
     // TODO: implement ck::tensor_layout::convolution that describe packed/strided dimemsion as
     // properties
     template <typename ALayout,
               typename std::enable_if<NDimSpatial == 1 &&
                                           (is_same_v<ALayout, tensor_layout::convolution::G_NW_C> ||
-                                           is_same_v<ALayout, tensor_layout::convolution::NWGC>),
+                                           is_same_v<ALayout, tensor_layout::convolution::NWGC> ||
+                                           is_same_v<ALayout, tensor_layout::convolution::GNWC>),
                                       bool>::type = false>
     static auto
     MakeADescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
@@ -473,7 +138,8 @@ struct TransformConvFwdToGemm
     template <typename ALayout,
               typename std::enable_if<
                   NDimSpatial == 2 && (is_same_v<ALayout, tensor_layout::convolution::G_NHW_C> ||
-                                       is_same_v<ALayout, tensor_layout::convolution::NHWGC>),
+                                       is_same_v<ALayout, tensor_layout::convolution::NHWGC> ||
+                                       is_same_v<ALayout, tensor_layout::convolution::GNHWC>),
                   bool>::type = false>
     static auto
     MakeADescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
@@ -601,7 +267,8 @@ struct TransformConvFwdToGemm
     template <typename ALayout,
               typename std::enable_if<
                   NDimSpatial == 3 && (is_same_v<ALayout, tensor_layout::convolution::G_NDHW_C> ||
-                                       is_same_v<ALayout, tensor_layout::convolution::NDHWGC>),
+                                       is_same_v<ALayout, tensor_layout::convolution::NDHWGC> ||
+                                       is_same_v<ALayout, tensor_layout::convolution::GNDHWC>),
                   bool>::type = false>
     static auto
     MakeADescriptor_M_K(const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths,
diff --git a/include/ck/utility/dynamic_buffer.hpp b/include/ck/utility/dynamic_buffer.hpp
index 02d61f34ed..85756cd0d1 100644
--- a/include/ck/utility/dynamic_buffer.hpp
+++ b/include/ck/utility/dynamic_buffer.hpp
@@ -140,10 +140,36 @@ struct DynamicBuffer
         }
         else if constexpr(Op == InMemoryDataOperationEnum::Add)
         {
-            auto tmp = this->template Get<X>(i, is_valid_element);
-            this->template Set<X>(i, is_valid_element, x + tmp);
-            // tmp += x;
-            // this->template Set<X>(i, is_valid_element, tmp);
+            auto tmp       = this->template Get<X>(i, is_valid_element);
+            using scalar_t = typename scalar_type<remove_cvref_t<T>>::type;
+            // handle bfloat addition
+            if constexpr(is_same_v<scalar_t, bhalf_t>)
+            {
+                if constexpr(is_scalar_type<X>::value)
+                {
+                    // Scalar type
+                    auto result =
+                        type_convert<X>(type_convert<float>(x) + type_convert<float>(tmp));
+                    this->template Set<X>(i, is_valid_element, result);
+                }
+                else
+                {
+                    // Vector type
+                    constexpr auto vector_size = scalar_type<remove_cvref_t<X>>::vector_size;
+                    const vector_type<scalar_t, vector_size> a_vector{tmp};
+                    const vector_type<scalar_t, vector_size> b_vector{x};
+                    static_for<0, vector_size, 1>{}([&](auto idx) {
+                        auto result = type_convert<scalar_t>(
+                            type_convert<float>(a_vector.template AsType<scalar_t>()[idx]) +
+                            type_convert<float>(b_vector.template AsType<scalar_t>()[idx]));
+                        this->template Set<scalar_t>(i + idx, is_valid_element, result);
+                    });
+                }
+            }
+            else
+            {
+                this->template Set<X>(i, is_valid_element, x + tmp);
+            }
         }
     }
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp
new file mode 100644
index 0000000000..7ccfc6eb77
--- /dev/null
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp
@@ -0,0 +1,363 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <type_traits>
+#include <sstream>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace host {
+
+/**
+ * \brief Reference implementation for column to image.
+ *
+ * Input tensor descriptor has [N * Do * Ho * Wo, Z * Y * X * C] data layout.
+ * Memory layout is the same.
+ * Output tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
+ * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
+ *
+ * \tparam NDimSpatial Number of spatial dimensions.
+ * \tparam ImageLayout Image Layout.
+ * \tparam InDataType Input Data Type.
+ * \tparam OutDataType Output Data Type.
+ */
+template <ck::index_t NDimSpatial,
+          typename ImageLayout,
+          typename InDataType,
+          typename OutDataType,
+          typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
+struct ReferenceColumnToImage : public device::BaseOperator
+{
+    // Argument
+    struct Argument : public device::BaseArgument
+    {
+        public:
+        Argument(const Tensor<InDataType>& input,
+                 Tensor<OutDataType>& output,
+                 std::vector<ck::index_t> filter_spatial_lengths,
+                 std::vector<ck::index_t> conv_filter_strides,
+                 std::vector<ck::index_t> conv_filter_dilations,
+                 std::vector<ck::index_t> input_left_pads,
+                 std::vector<ck::index_t> input_right_pads)
+            : input_{input},
+              output_{output},
+              conv_strides_{conv_filter_strides},
+              conv_dilations_{conv_filter_dilations},
+              in_left_pads_{input_left_pads},
+              in_right_pads_{input_right_pads},
+              filter_spatial_lengths_{filter_spatial_lengths}
+        {
+            initOutputSpatialLengths();
+        }
+
+        const Tensor<InDataType>& input_;
+        Tensor<OutDataType>& output_;
+
+        std::vector<index_t> conv_strides_;
+        std::vector<index_t> conv_dilations_;
+        std::vector<index_t> in_left_pads_;
+        std::vector<index_t> in_right_pads_;
+
+        std::vector<index_t> filter_spatial_lengths_;
+        std::vector<index_t> output_spatial_lengths_;
+
+        private:
+        void initOutputSpatialLengths()
+        {
+            constexpr auto input_offset_to_spatial = 3;
+
+            for(ck::index_t i = 0; i < NDimSpatial; ++i)
+            {
+                // XEff = (X - 1) * conv_dilation_w + 1;
+                // Wo = (Wi + in_left_pad_w + in_right_pad_w - XEff) / conv_stride_w + 1;
+                const ck::index_t x_eff = (filter_spatial_lengths_[i] - 1) * conv_dilations_[i] + 1;
+
+                output_spatial_lengths_.push_back(
+                    (output_.GetLengths()[i + input_offset_to_spatial] + in_left_pads_[i] +
+                     in_right_pads_[i] - x_eff) /
+                        conv_strides_[i] +
+                    1);
+            }
+        }
+    };
+
+    struct Invoker : public device::BaseInvoker
+    {
+        using Argument = ReferenceColumnToImage::Argument;
+
+        float Run(const Argument& arg)
+        {
+            if(!(arg.output_.GetNumOfDimension() == NDimSpatial + 3 &&
+                 arg.input_.GetNumOfDimension() == 2))
+            {
+                throw std::runtime_error("wrong! inconsistent dimension");
+            }
+
+            const index_t N = arg.output_.GetLengths()[1];
+            const index_t C = arg.output_.GetLengths()[2];
+
+            if constexpr(NDimSpatial == 1)
+            {
+                const index_t Wo = arg.output_spatial_lengths_[0];
+                auto func        = [&](auto n) {
+                    for(index_t wo = 0; wo < Wo; ++wo)
+                    {
+                        index_t row    = n * Wo + wo;
+                        index_t column = 0;
+
+                        for(index_t x = 0; x < arg.filter_spatial_lengths_[0]; ++x)
+                        {
+                            auto wi = static_cast<ck::long_index_t>(wo * arg.conv_strides_[0]) +
+                                      static_cast<ck::long_index_t>(x * arg.conv_dilations_[0]) -
+                                      static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+
+                            for(index_t c = 0; c < C; ++c)
+                            {
+                                if(wi >= 0 &&
+                                   ck::type_convert<std::size_t>(wi) < arg.output_.GetLengths()[3])
+                                {
+                                    float v_in  = ck::type_convert<float>(arg.input_(row, column));
+                                    float v_out = ck::type_convert<float>(arg.output_(0, n, c, wi));
+                                    arg.output_(0, n, c, wi) =
+                                        ck::type_convert<OutDataType>(v_in + v_out);
+                                }
+                                column++;
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+            else if constexpr(NDimSpatial == 2)
+            {
+                const index_t Ho = arg.output_spatial_lengths_[0];
+                const index_t Wo = arg.output_spatial_lengths_[1];
+
+                auto func = [&](auto n) {
+                    for(index_t ho = 0; ho < Ho; ++ho)
+                    {
+                        for(index_t wo = 0; wo < Wo; ++wo)
+                        {
+                            index_t row    = n * Ho * Wo + ho * Wo + wo;
+                            index_t column = 0;
+
+                            for(index_t y = 0; y < arg.filter_spatial_lengths_[0]; ++y)
+                            {
+                                auto hi =
+                                    static_cast<ck::long_index_t>(ho * arg.conv_strides_[0]) +
+                                    static_cast<ck::long_index_t>(y * arg.conv_dilations_[0]) -
+                                    static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+
+                                for(index_t x = 0; x < arg.filter_spatial_lengths_[1]; ++x)
+                                {
+                                    auto wi =
+                                        static_cast<ck::long_index_t>(wo * arg.conv_strides_[1]) +
+                                        static_cast<ck::long_index_t>(x * arg.conv_dilations_[1]) -
+                                        static_cast<ck::long_index_t>(arg.in_left_pads_[1]);
+
+                                    for(index_t c = 0; c < C; ++c)
+                                    {
+
+                                        if(hi >= 0 &&
+                                           ck::type_convert<std::size_t>(hi) <
+                                               arg.output_.GetLengths()[3] &&
+                                           wi >= 0 &&
+                                           ck::type_convert<std::size_t>(wi) <
+                                               arg.output_.GetLengths()[4])
+                                        {
+                                            float v_in =
+                                                ck::type_convert<float>(arg.input_(row, column));
+                                            float v_out = ck::type_convert<float>(
+                                                arg.output_(0, n, c, hi, wi));
+                                            arg.output_(0, n, c, hi, wi) =
+                                                ck::type_convert<OutDataType>(v_in + v_out);
+                                        }
+                                        column++;
+                                    }
+                                }
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+            else if constexpr(NDimSpatial == 3)
+            {
+                const index_t Do = arg.output_spatial_lengths_[0];
+                const index_t Ho = arg.output_spatial_lengths_[1];
+                const index_t Wo = arg.output_spatial_lengths_[2];
+
+                auto func = [&](auto n) {
+                    for(index_t d_o = 0; d_o < Do; ++d_o)
+                    {
+                        for(index_t ho = 0; ho < Ho; ++ho)
+                        {
+                            for(index_t wo = 0; wo < Wo; ++wo)
+                            {
+                                index_t row    = n * Do * Ho * Wo + d_o * Ho * Wo + ho * Wo + wo;
+                                index_t column = 0;
+
+                                for(index_t z = 0; z < arg.filter_spatial_lengths_[0]; ++z)
+                                {
+                                    auto di =
+                                        static_cast<ck::long_index_t>(d_o * arg.conv_strides_[0]) +
+                                        static_cast<ck::long_index_t>(z * arg.conv_dilations_[0]) -
+                                        static_cast<ck::long_index_t>(arg.in_left_pads_[0]);
+                                    for(index_t y = 0; y < arg.filter_spatial_lengths_[1]; ++y)
+                                    {
+                                        auto hi =
+                                            static_cast<ck::long_index_t>(ho *
+                                                                          arg.conv_strides_[1]) +
+                                            static_cast<ck::long_index_t>(y *
+                                                                          arg.conv_dilations_[1]) -
+                                            static_cast<ck::long_index_t>(arg.in_left_pads_[1]);
+                                        for(index_t x = 0; x < arg.filter_spatial_lengths_[2]; ++x)
+                                        {
+                                            auto wi =
+                                                static_cast<ck::long_index_t>(
+                                                    wo * arg.conv_strides_[2]) +
+                                                static_cast<ck::long_index_t>(
+                                                    x * arg.conv_dilations_[2]) -
+                                                static_cast<ck::long_index_t>(arg.in_left_pads_[2]);
+                                            for(index_t c = 0; c < C; ++c)
+                                            {
+                                                if(di >= 0 &&
+                                                   ck::type_convert<std::size_t>(di) <
+                                                       arg.output_.GetLengths()[3] &&
+                                                   hi >= 0 &&
+                                                   ck::type_convert<std::size_t>(hi) <
+                                                       arg.output_.GetLengths()[4] &&
+                                                   wi >= 0 &&
+                                                   ck::type_convert<std::size_t>(wi) <
+                                                       arg.output_.GetLengths()[5])
+                                                {
+                                                    float v_in = ck::type_convert<float>(
+                                                        arg.input_(row, column));
+                                                    float v_out = ck::type_convert<float>(
+                                                        arg.output_(0, n, c, di, hi, wi));
+                                                    arg.output_(0, n, c, di, hi, wi) =
+                                                        ck::type_convert<OutDataType>(v_in + v_out);
+                                                }
+                                                column++;
+                                            }
+                                        }
+                                    }
+                                }
+                            }
+                        }
+                    }
+                };
+
+                make_ParallelTensorFunctor(func, N)(std::thread::hardware_concurrency());
+
+                return 0;
+            }
+        }
+
+        float Run(const device::BaseArgument* p_arg,
+                  const StreamConfig& /*stream_config*/ = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg));
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        using namespace tensor_layout::convolution;
+
+        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
+                       std::is_same_v<ImageLayout, GNDHWC>))
+        {
+            return false;
+        }
+        if constexpr(!(NDimSpatial >= 1 && NDimSpatial <= 3))
+        {
+            return false;
+        }
+        return true;
+    }
+
+    bool IsSupportedArgument(const Argument& arg)
+    {
+        const ck::index_t G = arg.output_.GetLengths()[0];
+        const ck::index_t N = arg.output_.GetLengths()[1];
+        const ck::index_t C = arg.output_.GetLengths()[2];
+
+        const index_t NDoHoWo =
+            N * ck::accumulate_n<index_t>(
+                    arg.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+        const index_t CZYX =
+            C * ck::accumulate_n<index_t>(
+                    arg.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+        if(!(arg.input_.GetLengths()[0] == static_cast<std::size_t>(NDoHoWo) &&
+             arg.input_.GetLengths()[1] == static_cast<std::size_t>(CZYX)))
+        {
+            return false;
+        }
+
+        if(G != 1)
+        {
+            return false;
+        }
+        return true;
+    }
+
+    bool IsSupportedArgument(const device::BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(const Tensor<InDataType>& input,
+                             Tensor<OutDataType>& output,
+                             std::vector<ck::index_t> filter_spatial_lengths,
+                             std::vector<ck::index_t> conv_filter_strides,
+                             std::vector<ck::index_t> conv_filter_dilations,
+                             std::vector<ck::index_t> input_left_pads,
+                             std::vector<ck::index_t> input_right_pads)
+    {
+        return Argument{input,
+                        output,
+                        filter_spatial_lengths,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    virtual std::unique_ptr<device::BaseInvoker> MakeInvokerPointer()
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "ReferenceColumnToImage"
+            << std::endl;
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace host
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
index 3f50ab88b3..9e12d07844 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
@@ -18,16 +18,18 @@ namespace host {
 /**
  * \brief Reference implementation for image to column.
  *
- * Tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
+ * Input tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
  * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
+ * Output tensor descriptor has [N * Do * Ho * Wo, Z * Y * X * C] data layout.
+ * Memory layout is the same.
  *
  * \tparam NDimSpatial Number of spatial dimensions.
- * \tparam InputLayout Input Layout.
+ * \tparam ImageLayout Image Layout.
  * \tparam InDataType Input Data Type.
  * \tparam OutDataType Output Data Type.
  */
 template <ck::index_t NDimSpatial,
-          typename InputLayout,
+          typename ImageLayout,
           typename InDataType,
           typename OutDataType,
           typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
@@ -240,8 +242,8 @@ struct ReferenceImageToColumn : public device::BaseOperator
     {
         using namespace tensor_layout::convolution;
 
-        if constexpr(!(std::is_same_v<InputLayout, GNWC> || std::is_same_v<InputLayout, GNHWC> ||
-                       std::is_same_v<InputLayout, GNDHWC>))
+        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
+                       std::is_same_v<ImageLayout, GNDHWC>))
         {
             return false;
         }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp
new file mode 100644
index 0000000000..57c0f88567
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp
@@ -0,0 +1,282 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <vector>
+#include <memory>
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+
+#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+// Image to Column
+// nhwc, 1d
+void add_device_image_to_column_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// nhwc, 2d
+void add_device_image_to_column_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// nhwc, 3d
+void add_device_image_to_column_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+
+// Column to Image
+// nhwc, 1d
+void add_device_column_to_image_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// nhwc, 2d
+void add_device_column_to_image_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// nhwc, 3d
+void add_device_column_to_image_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+
+template <ck::index_t NumDimSpatial,
+          typename ImageLayout,
+          typename InDataType,
+          typename OutDataType,
+          typename ConvTensorRearrangeOp>
+struct DeviceOperationInstanceFactory<
+    ck::tensor_operation::device::DeviceConvTensorRearrange<NumDimSpatial,
+                                                            ImageLayout,
+                                                            InDataType,
+                                                            OutDataType,
+                                                            ConvTensorRearrangeOp>>
+{
+    using DeviceOp = DeviceConvTensorRearrange<NumDimSpatial,
+                                               ImageLayout,
+                                               InDataType,
+                                               OutDataType,
+                                               ConvTensorRearrangeOp>;
+
+    static auto GetInstances()
+    {
+        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
+
+        if constexpr(is_same_v<ConvTensorRearrangeOp, ImageToColumn>)
+        {
+            if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, GNWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nwc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nwc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nwc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nwc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_ndhwc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_ndhwc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_ndhwc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_ndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+        }
+        else if constexpr(is_same_v<ConvTensorRearrangeOp, ColumnToImage>)
+        {
+            if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, GNWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nwc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nwc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nwc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nwc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nhwc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nhwc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nhwc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nhwc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_ndhwc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_ndhwc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_ndhwc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_ndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+        }
+
+        return op_ptrs;
+    }
+};
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp
new file mode 100644
index 0000000000..681f466677
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp
@@ -0,0 +1,106 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::tensor_layout::convolution;
+using namespace ck::conv_tensor_rearrange_op;
+
+using BF16 = ck::bhalf_t;
+using F16  = ck::half_t;
+using F32  = float;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_bf16_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    16,    16,   S<8, 8>,     1>,
+
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_f16_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,    64,    16,    16,   S<8, 8>,     1>,
+        
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     8>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_f32_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,    64,    16,    16,   S<8, 8>,     1>,
+        
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     4>
+    // clang-format on
+    >;
+
+template <ck::index_t NDimSpatial, typename InLayout>
+using device_column_to_image_i8_instances = std::tuple<
+    // clang-format off
+        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //#####################|           |         |           |            |      |      |      |          |       |
+        // generic instance
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    16,    16,   S<8, 8>,     1>,
+
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    32,    32,   S<8, 8>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    64,    64,   S<8, 8>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    32,    64,  S<8, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,   128,  S<8, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     4>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     8>,
+        DeviceColumnToImageImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   256,   256, S<16, 16>,     16>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp
similarity index 70%
rename from library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp
rename to library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp
index a2603218b2..74a2155a04 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp
@@ -13,6 +13,7 @@ namespace device {
 namespace instance {
 
 using namespace ck::tensor_layout::convolution;
+using namespace ck::conv_tensor_rearrange_op;
 
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
@@ -28,17 +29,12 @@ using device_image_to_column_bf16_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    16,    16,   S<8, 8>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    32,    32,   S<8, 8>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,    64,    64,    64,   S<8, 8>,     8>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    16,    16,  S<8, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,    64,  S<8, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    32,    64,  S<8, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   128,    64,   128,  S<8, 16>,     8>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    16,    16, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,    64,    64, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,       BF16,        BF16,   256,   128,   128, S<16, 16>,     8>
@@ -52,17 +48,13 @@ using device_image_to_column_f16_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance        
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    16,    16,   S<8, 8>,     1>,
+
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    32,    32,   S<8, 8>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,    64,    64,    64,   S<8, 8>,     8>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    16,    16,  S<8, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,    64,  S<8, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    32,    64,  S<8, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   128,    64,   128,  S<8, 16>,     8>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    16,    16, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,    64,    64, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F16,         F16,   256,   128,   128, S<16, 16>,     8>
@@ -76,15 +68,11 @@ using device_image_to_column_f32_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance      
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,    16,    16,   S<8, 8>,     1>,
+
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,    64,    32,    32,   S<8, 8>,     4>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    16,    16,  S<8, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    64,    64,  S<8, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   128,    32,    64,  S<8, 16>,     4>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    16,    16, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,    64,    64, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,        F32,         F32,   256,   128,   128, S<16, 16>,     4>
     // clang-format on
@@ -97,17 +85,13 @@ using device_image_to_column_i8_instances = std::tuple<
         //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
         //#####################|    Spatial|         |           |            |      |      |      |   Lengths| Vector|
         //#####################|           |         |           |            |      |      |      |          |       |
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,     8,     8,   S<8, 8>,     1>,
+        // generic instance
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    16,    16,   S<8, 8>,     1>,
+
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    32,    32,   S<8, 8>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,    64,    64,    64,   S<8, 8>,     8>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    16,    16,  S<8, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,    64,  S<8, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    32,    64,  S<8, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   128,    64,   128,  S<8, 16>,     8>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    16,    16, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     1>,
-        DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     1>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,    64,    64, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     4>,
         DeviceImageToColumnImpl<NDimSpatial, InLayout,     int8_t,      int8_t,   256,   128,   128, S<16, 16>,     8>,
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp b/library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp
deleted file mode 100644
index 6c4526ba4e..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/image_to_column.hpp
+++ /dev/null
@@ -1,135 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <vector>
-#include <memory>
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-
-#include "ck/library/tensor_operation_instance/device_operation_instance_factory.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// nhwc, 1d
-void add_device_image_to_column_nhwc_1d_bf16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances);
-
-void add_device_image_to_column_nhwc_1d_f16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances);
-
-void add_device_image_to_column_nhwc_1d_f32_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances);
-
-void add_device_image_to_column_nhwc_1d_i8_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances);
-// nhwc, 2d
-void add_device_image_to_column_nhwc_2d_bf16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances);
-
-void add_device_image_to_column_nhwc_2d_f16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances);
-
-void add_device_image_to_column_nhwc_2d_f32_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances);
-
-void add_device_image_to_column_nhwc_2d_i8_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances);
-// nhwc, 3d
-void add_device_image_to_column_nhwc_3d_bf16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances);
-
-void add_device_image_to_column_nhwc_3d_f16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances);
-
-void add_device_image_to_column_nhwc_3d_f32_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances);
-
-void add_device_image_to_column_nhwc_3d_i8_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances);
-
-template <ck::index_t NumDimSpatial, typename InLayout, typename InDataType, typename OutDataType>
-struct DeviceOperationInstanceFactory<
-    ck::tensor_operation::device::
-        DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>>
-{
-    using DeviceOp = DeviceImageToColumn<NumDimSpatial, InLayout, InDataType, OutDataType>;
-
-    static auto GetInstances()
-    {
-        std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
-
-        if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, GNWC>)
-        {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
-            {
-                add_device_image_to_column_nhwc_1d_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
-            {
-                add_device_image_to_column_nhwc_1d_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
-            {
-                add_device_image_to_column_nhwc_1d_bf16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
-            {
-                add_device_image_to_column_nhwc_1d_i8_instances(op_ptrs);
-            }
-        }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC>)
-        {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
-            {
-                add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
-            {
-                add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
-            {
-                add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
-            {
-                add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
-            }
-        }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC>)
-        {
-            if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
-            {
-                add_device_image_to_column_nhwc_3d_f32_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
-            {
-                add_device_image_to_column_nhwc_3d_f16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
-            {
-                add_device_image_to_column_nhwc_3d_bf16_instances(op_ptrs);
-            }
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
-            {
-                add_device_image_to_column_nhwc_3d_i8_instances(op_ptrs);
-            }
-        }
-
-        return op_ptrs;
-    }
-};
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt
new file mode 100644
index 0000000000..5d1a554524
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt
@@ -0,0 +1,5 @@
+add_instance_library(device_column_to_image_instance
+   device_column_to_image_nhwc_1d_instance.cpp
+   device_column_to_image_nhwc_2d_instance.cpp
+   device_column_to_image_nhwc_3d_instance.cpp
+)
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp
new file mode 100644
index 0000000000..8ba4d29775
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp
new file mode 100644
index 0000000000..4de665a63f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_nhwc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp
new file mode 100644
index 0000000000..9762b46c43
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
index c8463623c3..3a629f2346 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -9,28 +9,50 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_image_to_column_nhwc_1d_bf16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, BF16, BF16>>>& instances)
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_image_to_column_nwc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_BF16
     add_device_operation_instances(instances, device_image_to_column_bf16_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
-void add_device_image_to_column_nhwc_1d_f16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F16, F16>>>& instances)
+void add_device_image_to_column_nwc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_FP16
     add_device_operation_instances(instances, device_image_to_column_f16_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
-void add_device_image_to_column_nhwc_1d_f32_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, F32, F32>>>& instances)
+void add_device_image_to_column_nwc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_FP32
     add_device_operation_instances(instances, device_image_to_column_f32_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
-void add_device_image_to_column_nhwc_1d_i8_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<1, GNWC, int8_t, int8_t>>>& instances)
+void add_device_image_to_column_nwc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_INT8
     add_device_operation_instances(instances, device_image_to_column_i8_instances<1, GNWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
index 652c7fac2a..7115e75667 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -9,28 +9,51 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using namespace ck::conv_tensor_rearrange_op;
+
 void add_device_image_to_column_nhwc_2d_bf16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, BF16, BF16>>>& instances)
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_BF16
     add_device_operation_instances(instances, device_image_to_column_bf16_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
 void add_device_image_to_column_nhwc_2d_f16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F16, F16>>>& instances)
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_FP16
     add_device_operation_instances(instances, device_image_to_column_f16_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
 void add_device_image_to_column_nhwc_2d_f32_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, F32, F32>>>& instances)
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_FP32
     add_device_operation_instances(instances, device_image_to_column_f32_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
 void add_device_image_to_column_nhwc_2d_i8_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<2, GNHWC, int8_t, int8_t>>>& instances)
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_INT8
     add_device_operation_instances(instances, device_image_to_column_i8_instances<2, GNHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
index 07774504d7..8290dae928 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
@@ -1,7 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include "ck/library/tensor_operation_instance/gpu/image_to_column/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -9,28 +9,51 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-void add_device_image_to_column_nhwc_3d_bf16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, BF16, BF16>>>& instances)
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_image_to_column_ndhwc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_BF16
     add_device_operation_instances(instances, device_image_to_column_bf16_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
-void add_device_image_to_column_nhwc_3d_f16_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F16, F16>>>& instances)
+void add_device_image_to_column_ndhwc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_FP16
     add_device_operation_instances(instances, device_image_to_column_f16_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
-void add_device_image_to_column_nhwc_3d_f32_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, F32, F32>>>& instances)
+void add_device_image_to_column_ndhwc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_FP32
     add_device_operation_instances(instances, device_image_to_column_f32_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
-void add_device_image_to_column_nhwc_3d_i8_instances(
-    std::vector<std::unique_ptr<DeviceImageToColumn<3, GNDHWC, int8_t, int8_t>>>& instances)
+void add_device_image_to_column_ndhwc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ImageToColumn>>>&
+        instances)
 {
+#ifdef CK_ENABLE_INT8
     add_device_operation_instances(instances, device_image_to_column_i8_instances<3, GNDHWC>{});
+#else
+    ignore = instances;
+#endif
 }
 
 } // namespace instance
diff --git a/profiler/README.md b/profiler/README.md
index d03bfa7fc4..d081fc33e9 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -185,7 +185,7 @@ GB/s: 69.2301
 ```
 Note: This kernel use atomic add, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
 
-## Profile image to column kernels
+## Profile image to column/column to image kernels
 ```bash
 # arg1: tensor operation (" OP_NAME ": " OP_DESC ")
 # arg2: data type (0: Input fp32, Weight fp32, Output fp32
@@ -197,6 +197,7 @@ Note: This kernel use atomic add, this will cause output buffer to be accumulate
 # arg5: initialization (0: no init, 1: integer value, 2: decimal value)
 # arg6: print tensor value (0: no; 1: yes)
 # arg7: time kernel (0: no, 1: yes)
+# arg8: operation type (0: ImageToColumn, 1: ColumnToImage)
 # Following arguments (depending on number of spatial dims):
 #  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
 #  G, N, K, C, 
@@ -207,8 +208,8 @@ Note: This kernel use atomic add, this will cause output buffer to be accumulate
 #  <left padding>, (ie LeftPy, LeftPx for 2D)
 #  <right padding>, (ie RightPy, RightPx for 2D)
 
- ################             op   datatype  layout  verify  init  log  time  Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx
-./bin/ckProfiler image_to_column          0       0       1     1    0     1      2  1 256   1 512  3  3   28  28   1   1   1   1        0       0       0        0
+ ################                   op   datatype  layout  verify  init  log  time opType Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx
+./bin/ckProfiler conv_tensor_rearrange          0       0       0     1    0     1      0     2  1 256   1 512  3  3   28  28   1   1   1   1        0       0       0        0
 
  ```
 
@@ -222,3 +223,4 @@ name: DeviceImageToColumn<128, 32, 64, 4>
 avg_time: 3.12326
 GB/s: 2042.59
 ```
+Note: Column to image kernel adds to the output memory, this will cause output buffer to be accumulated multiple times, causing verification failure. To work around it, do not use CK's own timer and do verification at the same time.
diff --git a/profiler/include/profiler/profile_image_to_column_impl.hpp b/profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp
similarity index 56%
rename from profiler/include/profiler/profile_image_to_column_impl.hpp
rename to profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp
index cc929e9220..dc3d748bf6 100644
--- a/profiler/include/profiler/profile_image_to_column_impl.hpp
+++ b/profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp
@@ -9,9 +9,11 @@
 #include <limits>
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
 #include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
-#include "ck/library/tensor_operation_instance/gpu/image_to_column.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp"
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp"
 #include "ck/library/utility/check_err.hpp"
 #include "ck/library/utility/device_memory.hpp"
 #include "ck/library/utility/host_tensor.hpp"
@@ -19,22 +21,88 @@
 #include "ck/library/utility/convolution_parameter.hpp"
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp"
 
 namespace ck {
 namespace profiler {
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
+using namespace conv_tensor_rearrange_op;
+
+template <typename InputDataType, typename ConvTensorRearrangeOp>
+Tensor<InputDataType> create_input(const HostTensorDescriptor& image_desc,
+                                   const HostTensorDescriptor& gemm_desc)
+{
+    if constexpr(std::is_same_v<ConvTensorRearrangeOp, ImageToColumn>)
+    {
+        Tensor<InputDataType> input(image_desc);
+        return input;
+    }
+    else if constexpr(std::is_same_v<ConvTensorRearrangeOp, ColumnToImage>)
+    {
+        Tensor<InputDataType> input(gemm_desc);
+        return input;
+    }
+    else
+    {
+        throw std::runtime_error("Unsupported op!");
+    }
+}
+
+template <typename OutputDataType, typename ConvTensorRearrangeOp>
+Tensor<OutputDataType> create_output(const HostTensorDescriptor& image_desc,
+                                     const HostTensorDescriptor& gemm_desc)
+{
+    if constexpr(std::is_same_v<ConvTensorRearrangeOp, ImageToColumn>)
+    {
+        Tensor<OutputDataType> output(gemm_desc);
+        return output;
+    }
+    else if constexpr(std::is_same_v<ConvTensorRearrangeOp, ColumnToImage>)
+    {
+        Tensor<OutputDataType> output(image_desc);
+        return output;
+    }
+    else
+    {
+        throw std::runtime_error("Unsupported op!");
+    }
+}
 
 template <index_t NDimSpatial,
           typename InputLayout,
           typename InputDataType,
-          typename OutputDataType>
-bool profile_image_to_column_impl(int do_verification,
-                                  int init_method,
-                                  bool do_log,
-                                  bool time_kernel,
-                                  const ck::utils::conv::ConvParam& conv_param)
+          typename OutputDataType,
+          typename ConvTensorRearrangeOp>
+static auto make_ref_op()
+{
+    if constexpr(std::is_same_v<ConvTensorRearrangeOp, ImageToColumn>)
+    {
+        return ck::tensor_operation::host::
+            ReferenceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>{};
+    }
+    else if constexpr(std::is_same_v<ConvTensorRearrangeOp, ColumnToImage>)
+    {
+        return ck::tensor_operation::host::
+            ReferenceColumnToImage<NDimSpatial, InputLayout, InputDataType, OutputDataType>{};
+    }
+    else
+    {
+        throw std::runtime_error("Unsupported op!");
+    }
+}
+
+template <index_t NDimSpatial,
+          typename InputLayout,
+          typename InputDataType,
+          typename OutputDataType,
+          typename ConvTensorRearrangeOp>
+bool profile_conv_tensor_rearrange_impl(int do_verification,
+                                        int init_method,
+                                        bool do_log,
+                                        bool time_kernel,
+                                        const ck::utils::conv::ConvParam& conv_param)
 {
     const ck::index_t NDoHoWo =
         conv_param.N_ *
@@ -45,16 +113,16 @@ bool profile_image_to_column_impl(int do_verification,
         ck::accumulate_n<ck::index_t>(
             conv_param.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
 
-    const auto in_desc =
+    const auto image_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InputLayout>(
             conv_param);
-    const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+    const auto gemm_desc = HostTensorDescriptor({NDoHoWo, CZYX});
 
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
-    std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
-    std::array<ck::index_t, 2> output_m_k_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
+    std::array<ck::index_t, 2> gemm_m_k_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -65,16 +133,19 @@ bool profile_image_to_column_impl(int do_verification,
     copy(conv_param.input_spatial_lengths_, input_spatial_lengths);
     copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
     copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
-    copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
-    copy(out_desc.GetStrides(), output_m_k_strides);
+    copy(image_desc.GetStrides(), image_g_n_c_wis_strides);
+    copy(gemm_desc.GetStrides(), gemm_m_k_strides);
     copy(conv_param.conv_filter_strides_, conv_filter_strides);
     copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
     copy(conv_param.input_left_pads_, input_left_pads);
     copy(conv_param.input_right_pads_, input_right_pads);
 
-    Tensor<InputDataType> input(in_desc);
-    Tensor<OutputDataType> host_output(out_desc);
-    Tensor<OutputDataType> device_output(out_desc);
+    Tensor<InputDataType> input =
+        create_input<InputDataType, ConvTensorRearrangeOp>(image_desc, gemm_desc);
+    Tensor<OutputDataType> device_output =
+        create_output<OutputDataType, ConvTensorRearrangeOp>(image_desc, gemm_desc);
+    Tensor<OutputDataType> host_output =
+        create_output<OutputDataType, ConvTensorRearrangeOp>(image_desc, gemm_desc);
 
     std::cout << "input: " << input.mDesc << std::endl;
     std::cout << "output: " << host_output.mDesc << std::endl;
@@ -94,17 +165,21 @@ bool profile_image_to_column_impl(int do_verification,
     // run reference op
     if(do_verification)
     {
-        auto ref_image_to_column = ck::tensor_operation::host::
-            ReferenceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>{};
+        auto ref_conv_tensor_rearrange = make_ref_op<NDimSpatial,
+                                                     InputLayout,
+                                                     InputDataType,
+                                                     OutputDataType,
+                                                     ConvTensorRearrangeOp>();
 
-        auto ref_invoker  = ref_image_to_column.MakeInvoker();
-        auto ref_argument = ref_image_to_column.MakeArgument(input,
-                                                             host_output,
-                                                             conv_param.filter_spatial_lengths_,
-                                                             conv_param.conv_filter_strides_,
-                                                             conv_param.conv_filter_dilations_,
-                                                             conv_param.input_left_pads_,
-                                                             conv_param.input_right_pads_);
+        auto ref_invoker = ref_conv_tensor_rearrange.MakeInvoker();
+        auto ref_argument =
+            ref_conv_tensor_rearrange.MakeArgument(input,
+                                                   host_output,
+                                                   conv_param.filter_spatial_lengths_,
+                                                   conv_param.conv_filter_strides_,
+                                                   conv_param.conv_filter_dilations_,
+                                                   conv_param.input_left_pads_,
+                                                   conv_param.input_right_pads_);
 
         // init host output to zero
         host_output.SetZero();
@@ -112,8 +187,11 @@ bool profile_image_to_column_impl(int do_verification,
         ref_invoker.Run(ref_argument);
     }
 
-    using DeviceOp = ck::tensor_operation::device::
-        DeviceImageToColumn<NDimSpatial, InputLayout, InputDataType, OutputDataType>;
+    using DeviceOp = ck::tensor_operation::device::DeviceConvTensorRearrange<NDimSpatial,
+                                                                             InputLayout,
+                                                                             InputDataType,
+                                                                             OutputDataType,
+                                                                             ConvTensorRearrangeOp>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -139,8 +217,8 @@ bool profile_image_to_column_impl(int do_verification,
             input_spatial_lengths,
             filter_spatial_lengths,
             output_spatial_lengths,
-            input_g_n_c_wis_strides,
-            output_m_k_strides,
+            image_g_n_c_wis_strides,
+            gemm_m_k_strides,
             conv_filter_strides,
             conv_filter_dilations,
             input_left_pads,
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 7da7613f26..8d9a47cd6b 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -28,7 +28,7 @@ set(PROFILER_SOURCES
     profile_contraction_bilinear.cpp
     profile_contraction_scale.cpp
     profile_grouped_conv_bwd_data.cpp
-    profile_image_to_column.cpp
+    profile_conv_tensor_rearrange.cpp
 )
 if(DL_KERNELS)
   list(APPEND PROFILER_SOURCES profile_batched_gemm_multi_d.cpp)
@@ -84,6 +84,7 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instanc
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_image_to_column_instance)
+target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_column_to_image_instance)
 if(DL_KERNELS)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 endif()
diff --git a/profiler/src/profile_conv_tensor_rearrange.cpp b/profiler/src/profile_conv_tensor_rearrange.cpp
new file mode 100644
index 0000000000..bad5ce40a5
--- /dev/null
+++ b/profiler/src/profile_conv_tensor_rearrange.cpp
@@ -0,0 +1,251 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "profiler/profile_conv_tensor_rearrange_impl.hpp"
+#include "profiler_operation_registry.hpp"
+
+namespace {
+
+enum struct RearrangeOp
+{
+    ImageToColumn, // 0
+    ColumnToImage, // 1
+};
+
+enum struct ConvLayout
+{
+    NHWC, // 0
+};
+
+enum struct DataType
+{
+    F32_F32,   // 0
+    F16_F16,   // 1
+    BF16_BF16, // 2
+    INT8_INT8, // 3
+};
+
+#define OP_NAME "conv_tensor_rearrange"
+#define OP_DESC "Conv Tensor Rearrange"
+
+static void print_helper_msg()
+{
+    std::cout
+        // clang-format off
+        << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
+        << "arg2: data type (0: Input fp32, Weight fp32, Output fp32\n"
+        << "                 1: Input fp16, Weight fp16, Output fp16\n"
+        << "                 2: Input bf16, Weight bf16, Output bf16\n"
+        << "                 3: Input int8, Weight int8, Output int8)\n"
+        << "arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])\n"
+        << "arg4: verification (0: no, 1: yes)\n"
+        << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
+        << "arg6: print tensor value (0: no; 1: yes)\n"
+        << "arg7: time kernel (0: no, 1: yes)\n"
+        << "arg8: operation type (0: ImageToColumn, 1: ColumnToImage)\n"
+        << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+    // clang-format on
+}
+
+} // namespace
+
+int profile_conv_tensor_rearrange(int argc, char* argv[])
+{
+    // 9 for control, 1 for num_dim_spatial
+    if(argc < 10)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto data_type       = static_cast<DataType>(std::stoi(argv[2]));
+    const auto layout          = static_cast<ConvLayout>(std::stoi(argv[3]));
+    const bool do_verification = std::stoi(argv[4]);
+    const int init_method      = std::stoi(argv[5]);
+    const bool do_log          = std::stoi(argv[6]);
+    const bool time_kernel     = std::stoi(argv[7]);
+    const auto rearrange_op    = static_cast<RearrangeOp>(std::stoi(argv[8]));
+    const int num_dim_spatial  = std::stoi(argv[9]);
+
+    // 9 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
+    if(argc != 9 + 1 + 4 + 6 * num_dim_spatial)
+    {
+        print_helper_msg();
+        return 1;
+    }
+
+    const auto params = ck::utils::conv::parse_conv_param(num_dim_spatial, 10, argv);
+
+    using F32  = float;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+    using INT8 = int8_t;
+
+    using namespace ck::tensor_layout::convolution;
+    using namespace ck::conv_tensor_rearrange_op;
+
+    constexpr auto I1 = ck::Number<1>{};
+    constexpr auto I2 = ck::Number<2>{};
+    constexpr auto I3 = ck::Number<3>{};
+
+    auto profile = [&](auto num_dim_spatial_tmp,
+                       auto in_layout,
+                       auto in_type,
+                       auto out_type,
+                       auto rearrange_op_type) {
+        constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
+
+        using InLayout = decltype(in_layout);
+
+        using InDataType  = decltype(in_type);
+        using OutDataType = decltype(out_type);
+
+        using Op = decltype(rearrange_op_type);
+
+        bool pass = ck::profiler::
+            profile_conv_tensor_rearrange_impl<NDimSpatial, InLayout, InDataType, OutDataType, Op>(
+                do_verification, init_method, do_log, time_kernel, params);
+
+        return pass ? 0 : 1;
+    };
+
+    // Image To Column
+    if(rearrange_op == RearrangeOp::ImageToColumn)
+    {
+        // NHWC
+        if(layout == ConvLayout::NHWC)
+        {
+            if(num_dim_spatial == 1)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I1, GNWC{}, F32{}, F32{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I1, GNWC{}, F16{}, F16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I1, GNWC{}, BF16{}, BF16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I1, GNWC{}, INT8{}, INT8{}, ImageToColumn{});
+                }
+            }
+            else if(num_dim_spatial == 2)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I2, GNHWC{}, F32{}, F32{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I2, GNHWC{}, F16{}, F16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I2, GNHWC{}, BF16{}, BF16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I2, GNHWC{}, INT8{}, INT8{}, ImageToColumn{});
+                }
+            }
+            else if(num_dim_spatial == 3)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I3, GNDHWC{}, F32{}, F32{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I3, GNDHWC{}, F16{}, F16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I3, GNDHWC{}, BF16{}, BF16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I3, GNDHWC{}, INT8{}, INT8{}, ImageToColumn{});
+                }
+            }
+        }
+    }
+    else if(rearrange_op == RearrangeOp::ColumnToImage)
+    {
+        // NHWC
+        if(layout == ConvLayout::NHWC)
+        {
+            if(num_dim_spatial == 1)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I1, GNWC{}, F32{}, F32{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I1, GNWC{}, F16{}, F16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I1, GNWC{}, BF16{}, BF16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I1, GNWC{}, INT8{}, INT8{}, ColumnToImage{});
+                }
+            }
+            else if(num_dim_spatial == 2)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I2, GNHWC{}, F32{}, F32{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I2, GNHWC{}, F16{}, F16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I2, GNHWC{}, BF16{}, BF16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I2, GNHWC{}, INT8{}, INT8{}, ColumnToImage{});
+                }
+            }
+            else if(num_dim_spatial == 3)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I3, GNDHWC{}, F32{}, F32{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I3, GNDHWC{}, F16{}, F16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I3, GNDHWC{}, BF16{}, BF16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I3, GNDHWC{}, INT8{}, INT8{}, ColumnToImage{});
+                }
+            }
+        }
+    }
+
+    std::cout << "this data_type & layout is not implemented" << std::endl;
+    return 1;
+}
+
+REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_conv_tensor_rearrange);
diff --git a/profiler/src/profile_image_to_column.cpp b/profiler/src/profile_image_to_column.cpp
deleted file mode 100644
index bf4312a6cf..0000000000
--- a/profiler/src/profile_image_to_column.cpp
+++ /dev/null
@@ -1,169 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
-#include "profiler/profile_image_to_column_impl.hpp"
-#include "profiler_operation_registry.hpp"
-
-namespace {
-
-enum struct ConvLayout
-{
-    NHWC, // 0
-};
-
-enum struct DataType
-{
-    F32_F32,   // 0
-    F16_F16,   // 1
-    BF16_BF16, // 2
-    INT8_INT8, // 3
-};
-
-#define OP_NAME "image_to_column"
-#define OP_DESC "Image To Column"
-
-static void print_helper_msg()
-{
-    std::cout
-        // clang-format off
-        << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
-        << "arg2: data type (0: Input fp32, Weight fp32, Output fp32\n"
-        << "                 1: Input fp16, Weight fp16, Output fp16\n"
-        << "                 2: Input bf16, Weight bf16, Output bf16\n"
-        << "                 3: Input int8, Weight int8, Output int8)\n"
-        << "arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])\n"
-        << "arg4: verification (0: no, 1: yes)\n"
-        << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
-        << "arg6: print tensor value (0: no; 1: yes)\n"
-        << "arg7: time kernel (0: no, 1: yes)\n"
-        << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
-    // clang-format on
-}
-
-} // namespace
-
-int profile_image_to_column(int argc, char* argv[])
-{
-    // 8 for control, 1 for num_dim_spatial
-    if(argc < 9)
-    {
-        print_helper_msg();
-        return 1;
-    }
-
-    const auto data_type       = static_cast<DataType>(std::stoi(argv[2]));
-    const auto layout          = static_cast<ConvLayout>(std::stoi(argv[3]));
-    const bool do_verification = std::stoi(argv[4]);
-    const int init_method      = std::stoi(argv[5]);
-    const bool do_log          = std::stoi(argv[6]);
-    const bool time_kernel     = std::stoi(argv[7]);
-    const int num_dim_spatial  = std::stoi(argv[8]);
-
-    // 8 for control, 1 for num_dim_spatial, 4 for G/N/K/C, and 6 * num_dim_spatial
-    if(argc != 8 + 1 + 4 + 6 * num_dim_spatial)
-    {
-        print_helper_msg();
-        return 1;
-    }
-
-    const auto params = ck::utils::conv::parse_conv_param(num_dim_spatial, 9, argv);
-
-    using F32  = float;
-    using F16  = ck::half_t;
-    using BF16 = ck::bhalf_t;
-    using INT8 = int8_t;
-
-    using namespace ck::tensor_layout::convolution;
-
-    constexpr auto I1 = ck::Number<1>{};
-    constexpr auto I2 = ck::Number<2>{};
-    constexpr auto I3 = ck::Number<3>{};
-
-    auto profile = [&](auto num_dim_spatial_tmp, auto in_layout, auto in_type, auto out_type) {
-        constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
-
-        using InLayout = decltype(in_layout);
-
-        using InDataType  = decltype(in_type);
-        using OutDataType = decltype(out_type);
-
-        bool pass = ck::profiler::
-            profile_image_to_column_impl<NDimSpatial, InLayout, InDataType, OutDataType>(
-                do_verification, init_method, do_log, time_kernel, params);
-
-        return pass ? 0 : 1;
-    };
-
-    // NHWC
-    if(layout == ConvLayout::NHWC)
-    {
-        if(num_dim_spatial == 1)
-        {
-            if(data_type == DataType::F32_F32)
-            {
-                return profile(I1, GNWC{}, F32{}, F32{});
-            }
-            else if(data_type == DataType::F16_F16)
-            {
-                return profile(I1, GNWC{}, F16{}, F16{});
-            }
-            else if(data_type == DataType::BF16_BF16)
-            {
-                return profile(I1, GNWC{}, BF16{}, BF16{});
-            }
-            else if(data_type == DataType::INT8_INT8)
-            {
-                return profile(I1, GNWC{}, INT8{}, INT8{});
-            }
-        }
-        else if(num_dim_spatial == 2)
-        {
-            if(data_type == DataType::F32_F32)
-            {
-                return profile(I2, GNHWC{}, F32{}, F32{});
-            }
-            else if(data_type == DataType::F16_F16)
-            {
-                return profile(I2, GNHWC{}, F16{}, F16{});
-            }
-            else if(data_type == DataType::BF16_BF16)
-            {
-                return profile(I2, GNHWC{}, BF16{}, BF16{});
-            }
-            else if(data_type == DataType::INT8_INT8)
-            {
-                return profile(I2, GNHWC{}, INT8{}, INT8{});
-            }
-        }
-        else if(num_dim_spatial == 3)
-        {
-            if(data_type == DataType::F32_F32)
-            {
-                return profile(I3, GNDHWC{}, F32{}, F32{});
-            }
-            else if(data_type == DataType::F16_F16)
-            {
-                return profile(I3, GNDHWC{}, F16{}, F16{});
-            }
-            else if(data_type == DataType::BF16_BF16)
-            {
-                return profile(I3, GNDHWC{}, BF16{}, BF16{});
-            }
-            else if(data_type == DataType::INT8_INT8)
-            {
-                return profile(I3, GNDHWC{}, INT8{}, INT8{});
-            }
-        }
-    }
-
-    std::cout << "this data_type & layout is not implemented" << std::endl;
-
-    return 1;
-}
-
-REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_image_to_column);
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 368965aa8a..07dd675afc 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -155,7 +155,7 @@ add_subdirectory(contraction)
 add_subdirectory(pool)
 add_subdirectory(batched_gemm_multi_d)
 add_subdirectory(grouped_convnd_bwd_data)
-add_subdirectory(image_to_column)
+add_subdirectory(conv_tensor_rearrange)
 if(GPU_TARGETS MATCHES "gfx11")
     add_subdirectory(wmma_op)
 endif()
diff --git a/test/conv_tensor_rearrange/CMakeLists.txt b/test/conv_tensor_rearrange/CMakeLists.txt
new file mode 100644
index 0000000000..f6ad263242
--- /dev/null
+++ b/test/conv_tensor_rearrange/CMakeLists.txt
@@ -0,0 +1,4 @@
+add_gtest_executable(test_conv_tensor_rearrange test_conv_tensor_rearrange.cpp)
+target_link_libraries(test_conv_tensor_rearrange PRIVATE utility device_image_to_column_instance device_column_to_image_instance)
+add_gtest_executable(test_conv_tensor_rearrange_interface test_conv_tensor_rearrange_interface.cpp)
+target_link_libraries(test_conv_tensor_rearrange_interface PRIVATE utility)
diff --git a/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp b/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
new file mode 100644
index 0000000000..7065b03e0d
--- /dev/null
+++ b/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
@@ -0,0 +1,153 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "profiler/profile_conv_tensor_rearrange_impl.hpp"
+
+template <typename Tuple>
+class TestConvTensorRearrange : public ::testing::Test
+{
+    protected:
+    using ImLayout              = std::tuple_element_t<0, Tuple>;
+    using ConvTensorRearrangeOp = std::tuple_element_t<1, Tuple>;
+
+    std::vector<ck::utils::conv::ConvParam> conv_params;
+
+    template <ck::index_t NDimSpatial, typename InDataType, typename OutDataType>
+    void Run()
+    {
+        EXPECT_FALSE(conv_params.empty());
+        bool pass = true;
+        for(auto& param : conv_params)
+        {
+            pass = pass && ck::profiler::profile_conv_tensor_rearrange_impl<NDimSpatial,
+                                                                            ImLayout,
+                                                                            InDataType,
+                                                                            OutDataType,
+                                                                            ConvTensorRearrangeOp>(
+                               true,  // do_verification
+                               1,     // init_method: integer value
+                               false, // do_log
+                               false, // time_kernel
+                               param);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+using namespace ck::tensor_layout::convolution;
+using namespace ck::conv_tensor_rearrange_op;
+
+using KernelTypes1d =
+    ::testing::Types<std::tuple<GNWC, ImageToColumn>, std::tuple<GNWC, ColumnToImage>>;
+
+using KernelTypes2d =
+    ::testing::Types<std::tuple<GNHWC, ImageToColumn>, std::tuple<GNHWC, ColumnToImage>>;
+
+using KernelTypes3d =
+    ::testing::Types<std::tuple<GNDHWC, ImageToColumn>, std::tuple<GNDHWC, ColumnToImage>>;
+
+template <typename Tuple>
+class TestConvTensorRearrange1d : public TestConvTensorRearrange<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestConvTensorRearrange2d : public TestConvTensorRearrange<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestConvTensorRearrange3d : public TestConvTensorRearrange<Tuple>
+{
+};
+
+TYPED_TEST_SUITE(TestConvTensorRearrange1d, KernelTypes1d);
+TYPED_TEST_SUITE(TestConvTensorRearrange2d, KernelTypes2d);
+TYPED_TEST_SUITE(TestConvTensorRearrange3d, KernelTypes3d);
+
+TYPED_TEST(TestConvTensorRearrange1d, Test1D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back({1, 1, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {7}, {3}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
+    // ScalarPerVector should be 1
+    this->conv_params.push_back({1, 1, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
+    // stride != 1
+    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
+    // dilation != 1
+    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
+#ifdef CK_ENABLE_FP32
+    this->template Run<1, float, float>();
+#endif
+#ifdef CK_ENABLE_BF16
+    this->template Run<1, ck::bhalf_t, ck::bhalf_t>();
+#endif
+#ifdef CK_ENABLE_FP16
+    this->template Run<1, ck::half_t, ck::half_t>();
+#endif
+#ifdef CK_ENABLE_INT8
+    this->template Run<1, int8_t, int8_t>();
+#endif
+}
+
+TYPED_TEST(TestConvTensorRearrange2d, Test2D)
+{
+    this->conv_params.clear();
+
+    this->conv_params.push_back(
+        {2, 1, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back(
+        {2, 1, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+    this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {3, 3}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
+    this->conv_params.push_back(
+        {2, 1, 64, 1, 64, {3, 3}, {28, 28}, {2, 2}, {2, 2}, {1, 1}, {1, 1}});
+#ifdef CK_ENABLE_FP32
+    this->template Run<2, float, float>();
+#endif
+#ifdef CK_ENABLE_BF16
+    this->template Run<2, ck::bhalf_t, ck::bhalf_t>();
+#endif
+#ifdef CK_ENABLE_FP16
+    this->template Run<2, ck::half_t, ck::half_t>();
+#endif
+#ifdef CK_ENABLE_INT8
+    this->template Run<2, int8_t, int8_t>();
+#endif
+}
+
+TYPED_TEST(TestConvTensorRearrange3d, Test3D)
+{
+    this->conv_params.clear();
+    this->conv_params.push_back(
+        {3, 1, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {3, 3, 3}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 1, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+    this->conv_params.push_back(
+        {3, 1, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+    this->conv_params.push_back(
+        {3, 1, 64, 1, 64, {3, 3, 3}, {14, 14, 14}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}});
+#ifdef CK_ENABLE_FP32
+    this->template Run<3, float, float>();
+#endif
+#ifdef CK_ENABLE_BF16
+    this->template Run<3, ck::bhalf_t, ck::bhalf_t>();
+#endif
+#ifdef CK_ENABLE_FP16
+    this->template Run<3, ck::half_t, ck::half_t>();
+#endif
+#ifdef CK_ENABLE_INT8
+    this->template Run<3, int8_t, int8_t>();
+#endif
+}
diff --git a/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp b/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
new file mode 100644
index 0000000000..57fcdc4269
--- /dev/null
+++ b/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
@@ -0,0 +1,260 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp"
+#include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using DataType = float;
+using ImLayout = ck::tensor_layout::convolution::GNWC;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::conv_tensor_rearrange_op;
+
+template <ck::index_t ScalarPerVector, bool IsCPacked>
+class TestConvTensorRearrangeInterface : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 1;
+
+    // clang-format off
+    using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumnImpl
+        //        Num| ImLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //           |         |           |            |      |      |      |          |       |
+        < NDimSpatial, ImLayout,   DataType,    DataType,   256,   128,   128, S<16, 16>,ScalarPerVector>;
+    using DeviceColToimgInstance = ck::tensor_operation::device::DeviceColumnToImageImpl
+        //        Num| ImLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread| Scalar|
+        //        Dim|         |           |            |  Size| Block| Block|   Cluster|    Per|
+        //    Spatial|         |           |            |      |      |      |   Lengths| Vector|
+        //           |         |           |            |      |      |      |          |       |
+        < NDimSpatial, ImLayout,   DataType,    DataType,   256,   128,   128, S<16, 16>,ScalarPerVector>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    template <typename ConvTensorRearrangeOp>
+    bool Run()
+    {
+
+        const auto N = conv_param.N_;
+        const auto C = conv_param.C_;
+        const auto FakeC =
+            conv_param.C_ / 2; // Fake C to simulate the behavior that C is not packed
+
+        const ck::index_t NDoHoWo =
+            N *
+            ck::accumulate_n<ck::index_t>(
+                conv_param.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+        const ck::index_t CZYX =
+            C *
+            ck::accumulate_n<ck::index_t>(
+                conv_param.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
+
+        const auto image_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(
+                conv_param);
+        const auto gemm_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+
+        std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
+        std::array<ck::index_t, 2> output_m_k_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
+
+        copy(conv_param.input_spatial_lengths_, input_spatial_lengths);
+        copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
+        copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
+        copy(image_desc.GetStrides(), input_g_n_c_wis_strides);
+        copy(gemm_desc.GetStrides(), output_m_k_strides);
+        copy(conv_param.conv_filter_strides_, conv_filter_strides);
+        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+        copy(conv_param.input_left_pads_, input_left_pads);
+        copy(conv_param.input_right_pads_, input_right_pads);
+
+        if constexpr(std::is_same_v<ConvTensorRearrangeOp, ImageToColumn>)
+        {
+            auto img2col  = DeviceImgToColInstance{};
+            auto argument = img2col.MakeArgument(nullptr,
+                                                 nullptr,
+                                                 N,
+                                                 IsCPacked ? C : FakeC,
+                                                 input_spatial_lengths,
+                                                 filter_spatial_lengths,
+                                                 output_spatial_lengths,
+                                                 input_g_n_c_wis_strides,
+                                                 output_m_k_strides,
+                                                 conv_filter_strides,
+                                                 conv_filter_dilations,
+                                                 input_left_pads,
+                                                 input_right_pads);
+
+            return img2col.IsSupportedArgument(argument);
+        }
+        else if constexpr(std::is_same_v<ConvTensorRearrangeOp, ColumnToImage>)
+        {
+            auto col2img  = DeviceColToimgInstance{};
+            auto argument = col2img.MakeArgument(nullptr,
+                                                 nullptr,
+                                                 N,
+                                                 IsCPacked ? C : FakeC,
+                                                 input_spatial_lengths,
+                                                 filter_spatial_lengths,
+                                                 output_spatial_lengths,
+                                                 input_g_n_c_wis_strides,
+                                                 output_m_k_strides,
+                                                 conv_filter_strides,
+                                                 conv_filter_dilations,
+                                                 input_left_pads,
+                                                 input_right_pads);
+
+            return col2img.IsSupportedArgument(argument);
+        }
+    }
+};
+
+class TestConvTensorRearrangeInterface1ScalarPerVector
+    : public TestConvTensorRearrangeInterface<1, true>
+{
+};
+
+class TestConvTensorRearrangeInterface4ScalarPerVector
+    : public TestConvTensorRearrangeInterface<4, true>
+{
+};
+
+class TestConvTensorRearrangeInterface4ScalarPerVectorFakeC
+    : public TestConvTensorRearrangeInterface<4, false>
+{
+};
+
+TEST_F(TestConvTensorRearrangeInterface1ScalarPerVector, X1ScalarPerVector)
+{
+    // vector load C * X % ScalarPerVector
+    this->conv_param  = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
+    bool is_supported = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+    // vector load C * left_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+    // vector load C * right_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+    // vector load C % ScalarPerVector, right_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+    // vector load C % ScalarPerVector, left_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+    // vector load C % ScalarPerVector, dilation
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+    // C = 4
+    this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+}
+
+TEST_F(TestConvTensorRearrangeInterface4ScalarPerVector, X4ScalarPerVector)
+{
+    // vector load C * X % ScalarPerVector
+    this->conv_param  = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
+    bool is_supported = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // vector load C * left_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // vector load C * right_pad_x % ScalarPerVector
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // vector load C % ScalarPerVector, right_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // vector load C % ScalarPerVector, left_pad and stride
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // vector load C % ScalarPerVector, dilation
+    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // C = 4
+    this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+}
+
+TEST_F(TestConvTensorRearrangeInterface4ScalarPerVectorFakeC, X4ScalarPerVectorFakeC)
+{
+    // C = 3
+    this->conv_param  = {1, 1, 1, 1, 3, {4}, {3}, {1}, {1}, {0}, {0}};
+    bool is_supported = this->template Run<ImageToColumn>();
+    EXPECT_FALSE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_FALSE(is_supported);
+    // C = 4
+    this->conv_param = {1, 1, 1, 1, 8, {4}, {3}, {1}, {1}, {0}, {0}};
+    is_supported     = this->template Run<ImageToColumn>();
+    EXPECT_TRUE(is_supported);
+    is_supported = this->template Run<ColumnToImage>();
+    EXPECT_TRUE(is_supported);
+}
diff --git a/test/image_to_column/CMakeLists.txt b/test/image_to_column/CMakeLists.txt
deleted file mode 100644
index 0feb827b55..0000000000
--- a/test/image_to_column/CMakeLists.txt
+++ /dev/null
@@ -1,4 +0,0 @@
-add_gtest_executable(test_image_to_column test_image_to_column.cpp)
-target_link_libraries(test_image_to_column PRIVATE utility device_image_to_column_instance)
-add_gtest_executable(test_image_to_column_interface test_image_to_column_interface.cpp)
-target_link_libraries(test_image_to_column_interface PRIVATE utility)
diff --git a/test/image_to_column/test_image_to_column.cpp b/test/image_to_column/test_image_to_column.cpp
deleted file mode 100644
index 0b17cac2d0..0000000000
--- a/test/image_to_column/test_image_to_column.cpp
+++ /dev/null
@@ -1,121 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-#include <iostream>
-#include <initializer_list>
-#include <tuple>
-#include <vector>
-
-#include <gtest/gtest.h>
-
-#include "profiler/profile_image_to_column_impl.hpp"
-
-template <typename Tuple>
-class TestImageToColumn : public ::testing::Test
-{
-    protected:
-    using InDataType  = std::tuple_element_t<0, Tuple>;
-    using OutDataType = std::tuple_element_t<1, Tuple>;
-    using InLayout    = std::tuple_element_t<2, Tuple>;
-
-    std::vector<ck::utils::conv::ConvParam> conv_params;
-
-    template <ck::index_t NDimSpatial>
-    void Run()
-    {
-        EXPECT_FALSE(conv_params.empty());
-        bool pass = true;
-        for(auto& param : conv_params)
-        {
-            pass = pass && ck::profiler::profile_image_to_column_impl<NDimSpatial,
-                                                                      InLayout,
-                                                                      InDataType,
-                                                                      OutDataType>(
-                               true,  // do_verification
-                               1,     // init_method: integer value
-                               false, // do_log
-                               false, // time_kernel
-                               param);
-        }
-        EXPECT_TRUE(pass);
-    }
-};
-
-using namespace ck::tensor_layout::convolution;
-
-using KernelTypes1d = ::testing::Types<std::tuple<float, float, GNWC>,
-                                       std::tuple<ck::bhalf_t, ck::bhalf_t, GNWC>,
-                                       std::tuple<ck::half_t, ck::half_t, GNWC>,
-                                       std::tuple<int8_t, int8_t, GNWC>>;
-
-using KernelTypes2d = ::testing::Types<std::tuple<float, float, GNHWC>,
-                                       std::tuple<ck::bhalf_t, ck::bhalf_t, GNHWC>,
-                                       std::tuple<ck::half_t, ck::half_t, GNHWC>,
-                                       std::tuple<int8_t, int8_t, GNHWC>>;
-
-using KernelTypes3d = ::testing::Types<std::tuple<float, float, GNDHWC>,
-                                       std::tuple<ck::bhalf_t, ck::bhalf_t, GNDHWC>,
-                                       std::tuple<ck::half_t, ck::half_t, GNDHWC>,
-                                       std::tuple<int8_t, int8_t, GNDHWC>>;
-
-template <typename Tuple>
-class TestImageToColumn1d : public TestImageToColumn<Tuple>
-{
-};
-
-template <typename Tuple>
-class TestImageToColumn2d : public TestImageToColumn<Tuple>
-{
-};
-
-template <typename Tuple>
-class TestImageToColumn3d : public TestImageToColumn<Tuple>
-{
-};
-
-TYPED_TEST_SUITE(TestImageToColumn1d, KernelTypes1d);
-TYPED_TEST_SUITE(TestImageToColumn2d, KernelTypes2d);
-TYPED_TEST_SUITE(TestImageToColumn3d, KernelTypes3d);
-
-TYPED_TEST(TestImageToColumn1d, Test1D)
-{
-    this->conv_params.clear();
-
-    this->conv_params.push_back({1, 1, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
-    this->conv_params.push_back({1, 1, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
-    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {7}, {2}, {1}, {0}, {0}});
-    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
-    // ScalarPerVector should be 1
-    this->conv_params.push_back({1, 1, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
-    // stride != 1
-    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
-    // dilation != 1
-    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
-    this->template Run<1>();
-}
-
-TYPED_TEST(TestImageToColumn2d, Test2D)
-{
-    this->conv_params.clear();
-
-    this->conv_params.push_back(
-        {2, 1, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    this->conv_params.push_back(
-        {2, 1, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
-    this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}});
-    this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
-    this->template Run<2>();
-}
-
-TYPED_TEST(TestImageToColumn3d, Test3D)
-{
-    this->conv_params.clear();
-    this->conv_params.push_back(
-        {3, 1, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    this->conv_params.push_back(
-        {3, 1, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
-    this->conv_params.push_back(
-        {3, 1, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
-    this->template Run<3>();
-}
diff --git a/test/image_to_column/test_image_to_column_interface.cpp b/test/image_to_column/test_image_to_column_interface.cpp
deleted file mode 100644
index ea8b9632e1..0000000000
--- a/test/image_to_column/test_image_to_column_interface.cpp
+++ /dev/null
@@ -1,196 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <cstdlib>
-#include <iostream>
-#include <initializer_list>
-#include <tuple>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp"
-
-#include "ck/library/utility/convolution_parameter.hpp"
-#include "ck/library/utility/algorithm.hpp"
-#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
-
-#include <gtest/gtest.h>
-
-using DataType = float;
-using InLayout = ck::tensor_layout::convolution::GNWC;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-template <ck::index_t ScalarPerVector, bool IsCPacked>
-class TestImageToColumnInterface : public ::testing::Test
-{
-    protected:
-    static constexpr ck::index_t NDimSpatial = 1;
-
-    // clang-format off
-    using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumnImpl
-        //#####################|        Num| InLayout| InDataType| OutDataType| Block|  MPer|  KPer|    Thread|         Scalar|
-        //#####################|        Dim|         |           |            |  Size| Block| Block|   Cluster|            Per|
-        //#####################|    Spatial|         |           |            |      |      |      |   Lengths|         Vector|
-        //#####################|           |         |           |            |      |      |      |          |               |
-                              < NDimSpatial, InLayout,   DataType,    DataType,   256,   128,   128, S<16, 16>,ScalarPerVector>;
-    // clang-format on
-
-    ck::utils::conv::ConvParam conv_param;
-
-    bool Run()
-    {
-
-        const auto N = conv_param.N_;
-        const auto C = conv_param.C_;
-        const auto FakeC =
-            conv_param.C_ / 2; // Fake C to simulate the behavior that C is not packed
-
-        const ck::index_t NDoHoWo =
-            N *
-            ck::accumulate_n<ck::index_t>(
-                conv_param.output_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
-        const ck::index_t CZYX =
-            C *
-            ck::accumulate_n<ck::index_t>(
-                conv_param.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
-
-        const auto in_desc =
-            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
-                conv_param);
-        const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
-
-        std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
-        std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
-        std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
-        std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
-        std::array<ck::index_t, 2> output_m_k_strides{};
-        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
-        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
-        std::array<ck::index_t, NDimSpatial> input_left_pads{};
-        std::array<ck::index_t, NDimSpatial> input_right_pads{};
-
-        auto copy = [](const auto& x, auto& y) { std::copy(x.begin(), x.end(), y.begin()); };
-
-        copy(conv_param.input_spatial_lengths_, input_spatial_lengths);
-        copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
-        copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
-        copy(in_desc.GetStrides(), input_g_n_c_wis_strides);
-        copy(out_desc.GetStrides(), output_m_k_strides);
-        copy(conv_param.conv_filter_strides_, conv_filter_strides);
-        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
-        copy(conv_param.input_left_pads_, input_left_pads);
-        copy(conv_param.input_right_pads_, input_right_pads);
-
-        auto img2col  = DeviceImgToColInstance{};
-        auto argument = img2col.MakeArgument(nullptr,
-                                             nullptr,
-                                             N,
-                                             IsCPacked ? C : FakeC,
-                                             input_spatial_lengths,
-                                             filter_spatial_lengths,
-                                             output_spatial_lengths,
-                                             input_g_n_c_wis_strides,
-                                             output_m_k_strides,
-                                             conv_filter_strides,
-                                             conv_filter_dilations,
-                                             input_left_pads,
-                                             input_right_pads);
-
-        return img2col.IsSupportedArgument(argument);
-    }
-};
-
-class TestImageToColumnInterface1ScalarPerVector : public TestImageToColumnInterface<1, true>
-{
-};
-
-class TestImageToColumnInterface4ScalarPerVector : public TestImageToColumnInterface<4, true>
-{
-};
-
-class TestImageToColumnInterface4ScalarPerVectorFakeC : public TestImageToColumnInterface<4, false>
-{
-};
-
-TEST_F(TestImageToColumnInterface1ScalarPerVector, X1ScalarPerVector)
-{
-    // vector load C * X % ScalarPerVector
-    this->conv_param  = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
-    bool is_supported = this->Run();
-    EXPECT_TRUE(is_supported);
-    // vector load C * left_pad_x % ScalarPerVector
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-    // vector load C * right_pad_x % ScalarPerVector
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-
-    // vector load C % ScalarPerVector, right_pad and stride
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-    // vector load C % ScalarPerVector, left_pad and stride
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-    // vector load C % ScalarPerVector, dilation
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-
-    // C = 4
-    this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-}
-
-TEST_F(TestImageToColumnInterface4ScalarPerVector, X4ScalarPerVector)
-{
-    // vector load C * X % ScalarPerVector
-    this->conv_param  = {1, 1, 1, 1, 1, {3}, {3}, {1}, {1}, {0}, {0}};
-    bool is_supported = this->Run();
-    EXPECT_FALSE(is_supported);
-    // vector load C * left_pad_x % ScalarPerVector
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {3}, {0}};
-    is_supported     = this->Run();
-    EXPECT_FALSE(is_supported);
-    // vector load C * right_pad_x % ScalarPerVector
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {1}, {0}, {3}};
-    is_supported     = this->Run();
-    EXPECT_FALSE(is_supported);
-
-    // vector load C % ScalarPerVector, right_pad and stride
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {0}, {3}};
-    is_supported     = this->Run();
-    EXPECT_FALSE(is_supported);
-    // vector load C % ScalarPerVector, left_pad and stride
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {2}, {1}, {3}, {0}};
-    is_supported     = this->Run();
-    EXPECT_FALSE(is_supported);
-    // vector load C % ScalarPerVector, dilation
-    this->conv_param = {1, 1, 1, 1, 1, {4}, {3}, {1}, {2}, {0}, {0}};
-    is_supported     = this->Run();
-    EXPECT_FALSE(is_supported);
-
-    // C = 4
-    this->conv_param = {1, 1, 1, 1, 4, {3}, {3}, {1}, {1}, {3}, {3}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-}
-
-TEST_F(TestImageToColumnInterface4ScalarPerVectorFakeC, X4ScalarPerVectorFakeC)
-{
-    // C = 3
-    this->conv_param  = {1, 1, 1, 1, 3, {4}, {3}, {1}, {1}, {0}, {0}};
-    bool is_supported = this->Run();
-    EXPECT_FALSE(is_supported);
-    // C = 4
-    this->conv_param = {1, 1, 1, 1, 8, {4}, {3}, {1}, {1}, {0}, {0}};
-    is_supported     = this->Run();
-    EXPECT_TRUE(is_supported);
-}

From f4af5aed8bb02ed54bdd4702f958b01c10d0d59b Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Wed, 27 Sep 2023 22:02:42 +0200
Subject: [PATCH 146/194] Handle type conversions to a const datatype (#944)

* Handle type conversions to a const datatype

* Review: Handle X being const data type as well

* Review: Remove typo
---
 include/ck/utility/type_convert.hpp   | 19 +++++-
 test/data_type/CMakeLists.txt         |  2 +
 test/data_type/type_convert_const.cpp | 93 +++++++++++++++++++++++++++
 3 files changed, 112 insertions(+), 2 deletions(-)
 create mode 100644 test/data_type/type_convert_const.cpp

diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 70619ee0a5..935926070d 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -9,8 +9,10 @@
 
 namespace ck {
 
-// Convert X to Y
-template <typename Y, typename X>
+// Convert X to Y, both X and Y are non-const data types.
+template <typename Y,
+          typename X,
+          std::enable_if_t<!(std::is_const_v<Y> || std::is_const_v<X>), bool> = false>
 __host__ __device__ constexpr Y type_convert(X x)
 {
     static_assert(!std::is_reference_v<Y> && !std::is_reference_v<X>);
@@ -18,6 +20,19 @@ __host__ __device__ constexpr Y type_convert(X x)
     return static_cast<Y>(x);
 }
 
+// Convert X to Y, either X or Y is a const data type.
+template <typename Y,
+          typename X,
+          std::enable_if_t<std::is_const_v<Y> || std::is_const_v<X>, bool> = false>
+__host__ __device__ constexpr Y type_convert(X x)
+{
+    static_assert(!std::is_reference_v<Y> && !std::is_reference_v<X>);
+
+    using NonConstY = std::remove_const_t<Y>;
+    using NonConstX = std::remove_const_t<X>;
+    return static_cast<Y>(type_convert<NonConstY, NonConstX>(x));
+}
+
 // convert bfp16 to fp32
 template <>
 inline __host__ __device__ constexpr float type_convert<float, bhalf_t>(bhalf_t x)
diff --git a/test/data_type/CMakeLists.txt b/test/data_type/CMakeLists.txt
index b1606d2a75..2409ca05c2 100644
--- a/test/data_type/CMakeLists.txt
+++ b/test/data_type/CMakeLists.txt
@@ -13,3 +13,5 @@ add_gtest_executable(test_bf8 bf8.cpp)
 if(result EQUAL 0)
   target_link_libraries(test_bf8 PRIVATE utility)
 endif()
+
+add_gtest_executable(test_type_convert_const type_convert_const.cpp)
diff --git a/test/data_type/type_convert_const.cpp b/test/data_type/type_convert_const.cpp
new file mode 100644
index 0000000000..8b9c34861a
--- /dev/null
+++ b/test/data_type/type_convert_const.cpp
@@ -0,0 +1,93 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "gtest/gtest.h"
+#include "ck/utility/data_type.hpp"
+#include "ck/utility/type_convert.hpp"
+
+using ck::bhalf_t;
+using ck::type_convert;
+
+TEST(TypeConvertConst, ConvertToConst)
+{
+    constexpr float bf16_epsilon = 0.0078125;
+    constexpr float rel_tol      = 2 * bf16_epsilon;
+
+    const std::vector<float> cases = {0.0, -123.f, 3.981323f, 0.2429f};
+
+    for(float x : cases)
+    {
+        const float abs_tol = std::abs(rel_tol * x);
+        {
+            bhalf_t y = type_convert<bhalf_t>(x);
+            // Test non-const bhalf to const float.
+            const float y_float = type_convert<const float>(y);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+        {
+            // Test non-const float to const bhalf.
+            const bhalf_t y = type_convert<const bhalf_t>(x);
+            // Remove the constness manually to not rely on const casts anymore since the
+            // possible issue could hide after two casts.
+            bhalf_t& y_nonconst = const_cast<bhalf_t&>(y);
+            float y_float       = type_convert<float>(y_nonconst);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+    }
+}
+
+TEST(TypeConvertConst, ConvertFromConst)
+{
+    constexpr float bf16_epsilon = 0.0078125;
+    constexpr float rel_tol      = 2 * bf16_epsilon;
+
+    const std::vector<float> cases = {0.0, -123.f, 3.981323f, 0.2429f};
+
+    for(const float x : cases)
+    {
+        const float abs_tol = std::abs(rel_tol * x);
+        {
+            // Test const float to const bhalf_t.
+            const bhalf_t y = type_convert<const bhalf_t>(x);
+            // Remove the constness manually to not rely on const casts anymore since the
+            // possible issue could hide after two casts.
+            bhalf_t& y_nonconst = const_cast<bhalf_t&>(y);
+            float y_float       = type_convert<float>(y_nonconst);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+        {
+            // Test const float to non-const bhalf.
+            bhalf_t y     = type_convert<bhalf_t>(x);
+            float y_float = type_convert<float>(y);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+        {
+            const bhalf_t y = type_convert<const bhalf_t>(x);
+            // Test const bhalf to non-const float.
+            float y_float = type_convert<float>(y);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+        // Tests with full type specializations for X.
+        {
+            // Test const float to const bhalf_t.
+            const bhalf_t y = type_convert<const bhalf_t, const float>(x);
+            // Remove the constness manually to not rely on const casts anymore since the
+            // possible issue could hide after two casts.
+            bhalf_t& y_nonconst = const_cast<bhalf_t&>(y);
+            float y_float       = type_convert<float>(y_nonconst);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+        {
+            // Test const float to non-const bhalf.
+            bhalf_t y     = type_convert<bhalf_t, const float>(x);
+            float y_float = type_convert<float>(y);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+        {
+            const bhalf_t y = type_convert<const bhalf_t, const float>(x);
+            // Test const bhalf to non-const float.
+            float y_float = type_convert<float, const bhalf_t>(y);
+            ASSERT_NEAR(y_float, x, abs_tol);
+        }
+    }
+}

From bc1108bb3ee4e18c0417609c215e1779b37f8d39 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 27 Sep 2023 15:19:33 -0700
Subject: [PATCH 147/194] Fix gemm_splitk test, add hip_check_error after
 kernel calls in kernel_launch. (#951)

* Added error check after kernel launch (#919)

Co-authored-by: Xiaodong Wang <xdwang@meta.com>
Co-authored-by: Xiaodong Wang <xw285@cornell.edu>

* remove M=0 test cases for test_gemm_splitk

---------

Co-authored-by: Xiaodong Wang <xdwang@meta.com>
Co-authored-by: Xiaodong Wang <xw285@cornell.edu>
---
 include/ck/host_utility/kernel_launch.hpp       | 8 ++++++++
 test/gemm_split_k/test_gemm_splitk_ut_cases.inc | 8 ++++----
 2 files changed, 12 insertions(+), 4 deletions(-)

diff --git a/include/ck/host_utility/kernel_launch.hpp b/include/ck/host_utility/kernel_launch.hpp
index 3d27103dcb..df311620e1 100644
--- a/include/ck/host_utility/kernel_launch.hpp
+++ b/include/ck/host_utility/kernel_launch.hpp
@@ -34,6 +34,7 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
 #endif
         // warm up
         kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+        hip_check_error(hipGetLastError());
 
         const int nrepeat = 10;
 #if DEBUG_LOG
@@ -50,6 +51,7 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
         for(int i = 0; i < nrepeat; ++i)
         {
             kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+            hip_check_error(hipGetLastError());
         }
 
         hip_check_error(hipEventRecord(stop, stream_config.stream_id_));
@@ -64,11 +66,13 @@ float launch_and_time_kernel(const StreamConfig& stream_config,
     else
     {
         kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+        hip_check_error(hipGetLastError());
 
         return 0;
     }
 #else
     kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+    hip_check_error(hipGetLastError());
 
     return 0;
 #endif
@@ -101,6 +105,7 @@ float launch_and_time_kernel_with_preprocess(const StreamConfig& stream_config,
         // warm up
         preprocess();
         kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+        hip_check_error(hipGetLastError());
 
         const int nrepeat = 10;
 #if DEBUG_LOG
@@ -118,6 +123,7 @@ float launch_and_time_kernel_with_preprocess(const StreamConfig& stream_config,
         {
             preprocess();
             kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+            hip_check_error(hipGetLastError());
         }
 
         hip_check_error(hipEventRecord(stop, stream_config.stream_id_));
@@ -133,11 +139,13 @@ float launch_and_time_kernel_with_preprocess(const StreamConfig& stream_config,
     {
         preprocess();
         kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+        hip_check_error(hipGetLastError());
 
         return 0;
     }
 #else
     kernel<<<grid_dim, block_dim, lds_byte, stream_config.stream_id_>>>(args...);
+    hip_check_error(hipGetLastError());
 
     return 0;
 #endif
diff --git a/test/gemm_split_k/test_gemm_splitk_ut_cases.inc b/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
index 54b9c6c9e3..d583a3e377 100644
--- a/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
+++ b/test/gemm_split_k/test_gemm_splitk_ut_cases.inc
@@ -2,7 +2,7 @@
 
 TYPED_TEST(TestGemmSplitK_MK_KN, SmallM)
 {
-    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
     constexpr int N = 512;
     constexpr int K = 320;
 
@@ -16,7 +16,7 @@ TYPED_TEST(TestGemmSplitK_MK_KN, SmallM)
 
 TYPED_TEST(TestGemmSplitK_MK_NK, SmallM)
 {
-    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
     constexpr int N = 512;
     constexpr int K = 320;
 
@@ -30,7 +30,7 @@ TYPED_TEST(TestGemmSplitK_MK_NK, SmallM)
 
 TYPED_TEST(TestGemmSplitK_KM_KN, SmallM)
 {
-    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
     constexpr int N = 512;
     constexpr int K = 320;
 
@@ -43,7 +43,7 @@ TYPED_TEST(TestGemmSplitK_KM_KN, SmallM)
 
 TYPED_TEST(TestGemmSplitK_KM_NK, SmallM)
 {
-    std::vector<int> Ms{0, 1, 2, 3, 4, 5, 6};
+    std::vector<int> Ms{1, 2, 3, 4, 5, 6};
     constexpr int N = 512;
     constexpr int K = 320;
 

From 271ef645ac8335967791ab9e24c2a41f26bcf404 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Thu, 28 Sep 2023 18:18:32 +0200
Subject: [PATCH 148/194] Add grouped convolution changes to changelog (#952)

* Add grouped convolution changes to changelog

* Fix 0.2.0 ck release rocm version

* Suggested CHANGELOG.md edits

* Update CHANGELOG.md

* Update CHANGELOG.md

* Update CHANGELOG.md

* Update CHANGELOG.md

* Update CHANGELOG.md

* Update CHANGELOG.md

---------

Co-authored-by: Lisa <lisajdelaney@gmail.com>
---
 CHANGELOG.md | 66 ++++++++++++++++++++++++++++++----------------------
 1 file changed, 38 insertions(+), 28 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 9d38a66382..10d92d8e3e 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,43 +1,53 @@
-# Change Log for Composable Kernel
+# Changelog for Composable Kernel
 
 Full documentation for Composable Kernel is not yet available.
 
 ## (Unreleased) CK for ROCm 6.0.0
 
-### Fixed
+### Fixes
+ - Fixed a hazard associated with inline v_dot (#808)
+ - Fixed two bugs in grouped convolution backward data without K padding (#848 #876)
 
 ### Optimizations
+None
 
-### Added
-- Added image to column (#867) and column to image kernels (#930).
+### Additions
+- Added an image to a column kernel (#867)
+- Added a column to an image kernel (#930)
+- Support for 3D grouped convolution forward on RDNA 3 GPUs (#935)
+- Grouped convolution support for small K and C (#822 #879 #897)
+- Support for NHWGC (2D and 3D) grouped convolution backward weight (#769 #804)
+- Support for bf16/f32/f16 and NHWGC (2D and 3d) grouped convolution backward data (#757 #799)
+- Support for Batched Gemm DL (#732)
 
-### Changed
+### Changes
+ - Changed the grouped convolution API to maintain consistency with other convolution kernels (#817)
 
+## CK 0.2.0 for ROCm 5.7.0
 
-## CK 0.2.0 for ROCm 5.5.0
-
-### Fixed
-- Fixed a bug in 6-dimensional kernels (#555).
-- Fixed grouped ConvBwdWeight test case failure (#524).
+### Fixes
+- Fixed a bug in 6-dimensional kernels (#555)
+- Fixed a test case failure with grouped convolution backward weight (#524)
 
 ### Optimizations
-- Improve proformance of normalization kernel
+- Improved the performance of the normalization kernel
 
-### Added
-- Added new cmake flag "DL_KERNELS" must be set to "ON" in order to build the gemm_dl and batched_gemm_multi_d_dl instances.
-- Added new cmake flag "DTYPES" which could be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build instance of select data types.
-- Added new cmake flag "INSTANCES_ONLY" which will only build CK library and instances without the tests, examples, or profiler.
-- Added new feature: if GPU_TARGETS is not set on cmake command line, CK will be built for all targets supported by compiler.
-- Added support on MI300A/MI300X.
-- Added support on NAVI3x.
-- Added user tutorial (#563).
-- Added more instances for irregular GEMM sizes (#560).
-- Added inter-wave consumer-producer programming model for GEMM kernels (#310).
-- Added multi-D GEMM client APIs (#534).
-- Added multi-embeddings support (#542).
-- Added Navi3x blockwise GEMM and real GEMM support (#541).
-- Added Navi grouped ConvBwdWeight support (#505).
-- Added MaxPool, AvgPool forward (#815).
-- Added MaxPool backward (#750).
+### Additions
+- New CMake flags:
+  - "DL_KERNELS"-- Must be set to "ON" in order to build the gemm_dl and batched_gemm_multi_d_dl instances
+  - "DTYPES" -- Can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build an instance of the specified data types
+  - "INSTANCES_ONLY" -- Only builds CK library and instances without tests, examples, or profiler
+- New feature: if GPU_TARGETS is not set in the CMake command line, CK will be built for all targets supported by the compiler
+- Support for MI300A/MI300X
+- Support for AMD RDNA 3
+- New user tutorial (#563)
+- Additional instances for irregular GEMM sizes (#560)
+- New inter-wave consumer-producer programming model for GEMM kernels (#310)
+- GEMM with support multiple elementwise fusions (multi-D) (#534)
+- Multi-embeddings support (#542)
+- AMD RDNA 3 blockwise GEMM and real GEMM support (#541)
+- AMD RDNA grouped convolution backward weight support (#505)
+- MaxPool and AvgPool forward (#815); MaxPool backward (#750)
 
-### Changed
+### Changes
+None

From cb538740021d5a41d89d9c21ec2eb28b2088d046 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Thu, 28 Sep 2023 23:10:18 +0200
Subject: [PATCH 149/194] Add grouped conv bwd data wmma (#950)

* Add grouped conv bwd data wmma

* Fix copyrights

* Add instances with smaller NPerBlock

* Update interface test

* Minor stylistic fixes

* Minor stylistic fixes
---
 .../CMakeLists.txt                            |  24 +-
 .../common.hpp                                |   1 -
 ...uped_conv_bwd_data_bias_relu_xdl_fp16.cpp} |   1 +
 .../grouped_conv_bwd_data_wmma_fp16.cpp       |  34 +
 ...cpp => grouped_conv_bwd_data_xdl_fp16.cpp} |   1 +
 ...conv_bwd_data_multiple_d_wmma_cshuffle.hpp | 879 ++++++++++++++++++
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  50 +-
 ..._conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp |  46 +-
 ...uped_conv_fwd_multiple_d_wmma_cshuffle.hpp |  50 +-
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |  46 +-
 .../device/impl/device_grouped_conv_utils.hpp |  59 ++
 ...ce_grouped_conv_bwd_data_wmma_instance.hpp | 118 +++
 .../gpu/grouped_convolution_backward_data.hpp | 283 ++++++
 .../grouped_conv2d_bwd_data/CMakeLists.txt    |  22 +-
 ...gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp |  40 +
 ...ta_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp |  40 +
 ..._gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp |  40 +
 ...ata_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp |  40 +
 ...nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp |  40 +
 ...ta_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp |  40 +
 ..._nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp |  40 +
 ...ata_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp |  40 +
 ...ta_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |   0
 ...ata_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   0
 ...ata_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   0
 ...ta_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |   0
 ...ata_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |   0
 ...ata_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |   0
 .../grouped_conv3d_bwd_data/CMakeLists.txt    |  21 +-
 ...hwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp |  40 +
 ...wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |  40 +
 ...dhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp |  40 +
 ..._wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp |  40 +
 ...wgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp |  40 +
 ...wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |  40 +
 ...hwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp |  40 +
 ..._wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp |  40 +
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |   0
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |   0
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |   0
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |   0
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |   0
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |   0
 test/grouped_convnd_bwd_data/CMakeLists.txt   |  25 +-
 .../test_grouped_convnd_bwd_data.cpp          |   8 +-
 ...grouped_convnd_bwd_data_interface_wmma.cpp | 178 ++++
 ...grouped_convnd_bwd_data_interface_xdl.cpp} |   0
 47 files changed, 2272 insertions(+), 214 deletions(-)
 rename example/38_grouped_conv_bwd_data_multiple_d/{grouped_conv_bwd_data_bias_relu_fp16.cpp => grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp} (97%)
 create mode 100644 example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_wmma_fp16.cpp
 rename example/38_grouped_conv_bwd_data_multiple_d/{grouped_conv_bwd_data_fp16.cpp => grouped_conv_bwd_data_xdl_fp16.cpp} (97%)
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/{ => xdl}/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/{ => xdl}/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/{ => xdl}/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/{ => xdl}/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/{ => xdl}/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/{ => xdl}/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp (100%)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/{ => xdl}/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/{ => xdl}/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/{ => xdl}/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/{ => xdl}/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/{ => xdl}/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/{ => xdl}/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp (100%)
 create mode 100644 test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface_wmma.cpp
 rename test/grouped_convnd_bwd_data/{test_grouped_convnd_bwd_data_interface.cpp => test_grouped_convnd_bwd_data_interface_xdl.cpp} (100%)

diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index 1675e3663a..96d1a6c3c0 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -1,15 +1,27 @@
-list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
+ if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_data)
-   add_example_executable(example_grouped_conv_bwd_data_fp16 grouped_conv_bwd_data_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_data_xdl_fp16 grouped_conv_bwd_data_xdl_fp16.cpp)
    if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_fp16)
+      add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_xdl_fp16)
    endif()
-   add_example_executable(example_grouped_conv_bwd_data_bias_relu_fp16 grouped_conv_bwd_data_bias_relu_fp16.cpp)
+   add_example_executable(example_grouped_conv_bwd_data_bias_relu_xdl_fp16 grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp)
    if(result EQUAL 0)
-     add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_fp16)
+     add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_xdl_fp16)
+   endif()
+   set(target 1)
+ endif()
+endforeach()
+
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
+   add_custom_target(example_grouped_conv_bwd_data)
+   add_example_executable(example_grouped_conv_bwd_data_wmma_fp16 grouped_conv_bwd_data_wmma_fp16.cpp)
+   if(result EQUAL 0)
+      add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_wmma_fp16)
    endif()
    set(target 1)
  endif()
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/common.hpp b/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
index ca824b1075..ebb1c606cb 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/common.hpp
@@ -10,7 +10,6 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp
similarity index 97%
rename from example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
rename to example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp
index a3533bb4cc..4c28e25e01 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_fp16.cpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
 #include "common.hpp"
 
 using OutDataType      = FP16;
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_wmma_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_wmma_fp16.cpp
new file mode 100644
index 0000000000..5baa521501
--- /dev/null
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_wmma_fp16.cpp
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp"
+#include "common.hpp"
+
+using OutDataType      = FP16;
+using WeiDataType      = FP16;
+using AccDataType      = FP32;
+using CShuffleDataType = FP16;
+using DsDataType       = ck::Tuple<>;
+using InDataType       = FP16;
+
+using OutLayout = ck::tensor_layout::convolution::GNHWK;
+using WeiLayout = ck::tensor_layout::convolution::GKYXC;
+using DsLayout  = ck::Tuple<>;
+using InLayout  = ck::tensor_layout::convolution::GNHWC;
+
+using OutElementOp = PassThrough;
+using WeiElementOp = PassThrough;
+using InElementOp  = PassThrough;
+
+// clang-format off
+using DeviceConvInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle
+        //|    NumDim|        A|         B|       Ds|       E|        AData|        BData|    AccData|          CShuffle|     DsData|       EData|           A|           B|          CDE|       ConvForward| Block|  MPer|  NPer| K0Per| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //|   Spatial|   Layout|    Layout|   Layout|  Layout|         Type|         Type|       Type|          DataType|       Type|        Type| Elementwise| Elementwise|  Elementwise|    Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //|          |         |          |         |        |             |             |           |                  |           |            |   Operation|   Operation|    Operation|                  |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //|          |         |          |         |        |             |             |           |                  |           |            |            |            |             |                  |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+           <        2,OutLayout, WeiLayout, DsLayout, InLayout, OutDataType,  WeiDataType, AccDataType, CShuffleDataType, DsDataType,  InDataType, OutElementOp, WeiElementOp, InElementOp, ConvBwdDataDefault, 128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>;
+// clang-format on
+
+#include "run_grouped_conv_bwd_data_example.inc"
+
+int main(int argc, char* argv[]) { return run_grouped_conv_bwd_data_example(argc, argv); }
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16.cpp
similarity index 97%
rename from example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
rename to example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16.cpp
index fb688b6f3f..b1554412b1 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_fp16.cpp
+++ b/example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16.cpp
@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
 #include "common.hpp"
 
 using OutDataType      = FP16;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp
new file mode 100644
index 0000000000..35581de72b
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp
@@ -0,0 +1,879 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+#include "ck/host_utility/io.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// Conv backward data multiple D:
+//   input : output image A: [G, N, K, Ho, Wo]
+//   input : weight B: [G, K, C, Y, X],
+//   input : D0, D1, ... : [G, N, K, Ho, Wo]
+//   output : input image E: [G, N, C, Hi, Wi]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+template <index_t NDimSpatial,
+          typename ALayout,   // output image
+          typename BLayout,   // weight
+          typename DsLayout,  // bias
+          typename ELayout,   // input image
+          typename ADataType, // output image
+          typename BDataType, // weight
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,       // bias
+          typename EDataType,        // input image
+          typename AElementwiseOp,   // output image
+          typename BElementwiseOp,   // weight
+          typename CDEElementwiseOp, // C, bias, and input image
+          ConvolutionBackwardDataSpecialization ConvBackwardDataSpecialization,
+          ck::index_t BlockSize,
+          ck::index_t MPerBlock,
+          ck::index_t NPerBlock,
+          ck::index_t K0PerBlock,
+          ck::index_t K1,
+          ck::index_t MPerWMMA,
+          ck::index_t NPerWMMA,
+          ck::index_t MRepeat,
+          ck::index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          bool ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          bool BBlockLdsExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+          index_t NumGemmKPrefetchStage   = 1,
+          LoopScheduler LoopSched         = make_default_loop_scheduler(),
+          ck::PipelineVersion PipelineVer = ck::PipelineVersion::v1>
+struct DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle
+    : public DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
+                                               ALayout,    // output image
+                                               BLayout,    // weight
+                                               DsLayout,   // bias
+                                               ELayout,    // input image
+                                               ADataType,  // output image
+                                               BDataType,  // weight
+                                               DsDataType, // bias
+                                               EDataType,  // input image
+                                               AElementwiseOp,
+                                               BElementwiseOp,
+                                               CDEElementwiseOp>
+{
+    // TODO: Extend support for more spatial dimensions.
+    static_assert(NDimSpatial == 2 || NDimSpatial == 3,
+                  "wrong! only implemented for 2D and 3D now");
+
+    using DeviceOp = DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle;
+
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    // TODO: Add support for different A and B data types.
+    using ABDataType = ADataType;
+
+    static constexpr auto I0           = Number<0>{};
+    static constexpr auto I1           = Number<1>{};
+    static constexpr auto I2           = Number<2>{};
+    static constexpr auto I3           = Number<3>{};
+    static constexpr index_t KPerBlock = K0PerBlock * K1;
+
+    static constexpr auto transform_conv_to_gemm =
+        TransformConvBwdDataToGemm_v1<NDimSpatial,
+                                      ConvBackwardDataSpecialization,
+                                      K1,
+                                      K1,
+                                      MPerBlock,
+                                      NPerBlock,
+                                      KPerBlock,
+                                      true /* DoPadGemmM */,
+                                      true /* DoPadGemmN */>{};
+
+    static auto GetDummyABDsEGridDescriptor()
+    {
+        const std::array<index_t, NDimSpatial + 3> dummy_tensor_lengths = {1};
+        const std::array<index_t, NDimSpatial + 3> dummy_tensor_strides = {1};
+        const std::array<index_t, NDimSpatial> dummy_spatial_lengths    = {1};
+
+        const auto a_grid_desc_ak0_m_ak1 =
+            transform_conv_to_gemm.template MakeADescriptor_AK0_M_AK1<ALayout>(
+                dummy_tensor_lengths,
+                dummy_tensor_strides,
+                dummy_tensor_lengths,
+                dummy_tensor_strides,
+                dummy_tensor_lengths,
+                dummy_tensor_strides,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths);
+
+        const auto b_grid_desc_bk0_n_bk1 =
+            transform_conv_to_gemm.template MakeBDescriptor_BK0_N_BK1<BLayout>(
+                dummy_tensor_lengths,
+                dummy_tensor_strides,
+                dummy_tensor_lengths,
+                dummy_tensor_strides,
+                dummy_tensor_lengths,
+                dummy_tensor_strides,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths,
+                dummy_spatial_lengths);
+
+        const auto ds_grid_desc_m_n = generate_tuple(
+            [&](auto i) {
+                using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                return transform_conv_to_gemm.template MakeCDescriptor_M_N<DLayout>(
+                    dummy_tensor_lengths,
+                    dummy_tensor_strides,
+                    dummy_tensor_lengths,
+                    dummy_tensor_strides,
+                    dummy_tensor_lengths,
+                    dummy_tensor_strides,
+                    dummy_spatial_lengths,
+                    dummy_spatial_lengths,
+                    dummy_spatial_lengths,
+                    dummy_spatial_lengths,
+                    dummy_spatial_lengths);
+            },
+            Number<NumDTensor>{});
+
+        const auto e_grid_desc_m_n =
+            transform_conv_to_gemm.template MakeCDescriptor_M_N<ELayout>(dummy_tensor_lengths,
+                                                                         dummy_tensor_strides,
+                                                                         dummy_tensor_lengths,
+                                                                         dummy_tensor_strides,
+                                                                         dummy_tensor_lengths,
+                                                                         dummy_tensor_strides,
+                                                                         dummy_spatial_lengths,
+                                                                         dummy_spatial_lengths,
+                                                                         dummy_spatial_lengths,
+                                                                         dummy_spatial_lengths,
+                                                                         dummy_spatial_lengths);
+
+        return make_tuple(
+            a_grid_desc_ak0_m_ak1, b_grid_desc_bk0_n_bk1, ds_grid_desc_m_n, e_grid_desc_m_n);
+    }
+
+    // desc
+    using ABDsEGridDesc = decltype(GetDummyABDsEGridDescriptor());
+
+    using AGridDesc_AK0_M_AK1 = remove_cvref_t<tuple_element_t<0, ABDsEGridDesc>>;
+    using BGridDesc_BK0_N_BK1 = remove_cvref_t<tuple_element_t<1, ABDsEGridDesc>>;
+    using DsGridDesc_M_N      = remove_cvref_t<tuple_element_t<2, ABDsEGridDesc>>;
+    using EGridDesc_M_N       = remove_cvref_t<tuple_element_t<3, ABDsEGridDesc>>;
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle<
+        // DataType Family
+        ADataType,
+        BDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        // InMemory Data Descriptor
+        AGridDesc_AK0_M_AK1,
+        BGridDesc_BK0_N_BK1,
+        DsGridDesc_M_N,
+        EGridDesc_M_N,
+        // ElementwiseOp Family
+        AElementwiseOp,
+        BElementwiseOp,
+        CDEElementwiseOp,
+        InMemoryDataOperationEnum::Set,
+        // Tiling Family
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerWMMA,
+        NPerWMMA,
+        K1,
+        MRepeat,
+        NRepeat,
+        // ThreadCluster Family
+        BlockSize,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CDEShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEShuffleBlockTransferScalarPerVector_NPerBlock,
+        NumGemmKPrefetchStage,
+        LoopSched,
+        PipelineVer>;
+
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}));
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}));
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(const void* p_a,                                 // output image
+                 const void* p_b,                                 // weight
+                 const std::array<const void*, NumDTensor>& p_ds, // bias
+                 void* p_e,                                       // input image
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_k_wos_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_k_wos_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+                     ds_g_n_c_wis_lengths,
+                 const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+                     ds_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_c_wis_lengths,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads,
+                 const AElementwiseOp& a_element_op,
+                 const BElementwiseOp& b_element_op,
+                 const CDEElementwiseOp& cde_element_op)
+            : p_a_grid_{static_cast<const ADataType*>(p_a)},
+              p_b_grid_{static_cast<const BDataType*>(p_b)},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e)},
+              num_group_{a_g_n_k_wos_lengths[0]},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op},
+              a_g_n_k_wos_lengths_{a_g_n_k_wos_lengths},
+              a_g_n_k_wos_strides_{a_g_n_k_wos_strides},
+              b_g_k_c_xs_lengths_{b_g_k_c_xs_lengths},
+              b_g_k_c_xs_strides_{b_g_k_c_xs_strides},
+              ds_g_n_c_wis_lengths_{ds_g_n_c_wis_lengths},
+              ds_g_n_c_wis_strides_{ds_g_n_c_wis_strides},
+              e_g_n_c_wis_lengths_{e_g_n_c_wis_lengths},
+              e_g_n_c_wis_strides_{e_g_n_c_wis_strides},
+              conv_filter_strides_{conv_filter_strides},
+              conv_filter_dilations_{conv_filter_dilations},
+              input_left_pads_{input_left_pads},
+              input_right_pads_{input_right_pads}
+        {
+            // populate Ds pointer
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds[i]);
+            });
+
+            // A/B/Ds/E Batch Stride
+            compute_ptr_offset_of_batch_.BatchStrideA_ = a_g_n_k_wos_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = b_g_k_c_xs_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideE_ = e_g_n_c_wis_strides[0];
+
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                compute_ptr_offset_of_batch_.BatchStrideDs_(i) = ds_g_n_c_wis_strides[i][0];
+            });
+
+            static constexpr auto NonSpatialDimsNum = Number<3>{};
+
+            static constexpr auto DIdx = Number<NonSpatialDimsNum>{};
+            static constexpr auto HIdx =
+                NDimSpatial == 2 ? Number<NonSpatialDimsNum>{} : Number<NonSpatialDimsNum + 1>{};
+            static constexpr auto WIdx = NDimSpatial == 2 ? Number<NonSpatialDimsNum + 1>{}
+                                                          : Number<NonSpatialDimsNum + 2>{};
+
+            static constexpr auto ZIdx = Number<NonSpatialDimsNum>{};
+            static constexpr auto YIdx =
+                NDimSpatial == 2 ? Number<NonSpatialDimsNum>{} : Number<NonSpatialDimsNum + 1>{};
+            static constexpr auto XIdx = NDimSpatial == 2 ? Number<NonSpatialDimsNum + 1>{}
+                                                          : Number<NonSpatialDimsNum + 2>{};
+
+            // problem definition
+            const index_t Z = b_g_k_c_xs_lengths[ZIdx];
+            const index_t Y = b_g_k_c_xs_lengths[YIdx];
+            const index_t X = b_g_k_c_xs_lengths[XIdx];
+
+            const index_t ConvStrideD = conv_filter_strides[DIdx - NonSpatialDimsNum];
+            const index_t ConvStrideH = conv_filter_strides[HIdx - NonSpatialDimsNum];
+            const index_t ConvStrideW = conv_filter_strides[WIdx - NonSpatialDimsNum];
+
+            const index_t ConvDilationD = conv_filter_dilations[DIdx - NonSpatialDimsNum];
+            const index_t ConvDilationH = conv_filter_dilations[HIdx - NonSpatialDimsNum];
+            const index_t ConvDilationW = conv_filter_dilations[WIdx - NonSpatialDimsNum];
+
+            const auto GcdStrideDilationD = math::gcd(ConvStrideD, ConvDilationD);
+            const auto GcdStrideDilationH = math::gcd(ConvStrideH, ConvDilationH);
+            const auto GcdStrideDilationW = math::gcd(ConvStrideW, ConvDilationW);
+
+            const auto ZTilde = NDimSpatial == 3 ? ConvStrideD / GcdStrideDilationD : 1;
+            const auto YTilde = ConvStrideH / GcdStrideDilationH;
+            const auto XTilde = ConvStrideW / GcdStrideDilationW;
+
+            for(index_t i_ztilde = 0; i_ztilde < ZTilde; ++i_ztilde)
+            {
+
+                for(index_t i_ytilde = 0; i_ytilde < YTilde; ++i_ytilde)
+                {
+                    for(index_t i_xtilde = 0; i_xtilde < XTilde; ++i_xtilde)
+                    {
+                        // check slice is valid
+                        const auto ZDotSlice =
+                            NDimSpatial == 3 ? math::integer_divide_ceil(Z - i_ztilde, ZTilde) : 1;
+                        const auto YDotSlice = math::integer_divide_ceil(Y - i_ytilde, YTilde);
+                        const auto XDotSlice = math::integer_divide_ceil(X - i_xtilde, XTilde);
+
+                        if(YDotSlice * XDotSlice * ZDotSlice <= 0)
+                        {
+                            continue;
+                        }
+
+                        std::array<index_t, NDimSpatial> tildes;
+                        if constexpr(NDimSpatial == 2)
+                        {
+                            tildes = {i_ytilde, i_xtilde};
+                        }
+                        else if constexpr(NDimSpatial == 3)
+                        {
+                            tildes = {i_ztilde, i_ytilde, i_xtilde};
+                        }
+                        else
+                        {
+                            throw std::runtime_error("wrong! only implemented for 2D and 3D now");
+                        }
+
+                        const auto a_grid_desc_ak0_m_ak1 =
+                            transform_conv_to_gemm.template MakeADescriptor_AK0_M_AK1<ALayout>(
+                                a_g_n_k_wos_lengths,
+                                a_g_n_k_wos_strides,
+                                b_g_k_c_xs_lengths,
+                                b_g_k_c_xs_strides,
+                                e_g_n_c_wis_lengths,
+                                e_g_n_c_wis_strides,
+                                conv_filter_strides,
+                                conv_filter_dilations,
+                                input_left_pads,
+                                input_right_pads,
+                                tildes);
+
+                        const auto b_grid_desc_bk0_n_bk1 =
+                            transform_conv_to_gemm.template MakeBDescriptor_BK0_N_BK1<BLayout>(
+                                a_g_n_k_wos_lengths,
+                                a_g_n_k_wos_strides,
+                                b_g_k_c_xs_lengths,
+                                b_g_k_c_xs_strides,
+                                e_g_n_c_wis_lengths,
+                                e_g_n_c_wis_strides,
+                                conv_filter_strides,
+                                conv_filter_dilations,
+                                input_left_pads,
+                                input_right_pads,
+                                tildes);
+
+                        DsGridDesc_M_N ds_grid_desc_m_n;
+
+                        // populate Ds desc
+                        static_for<0, NumDTensor, 1>{}([&](auto i) {
+                            using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+                            ds_grid_desc_m_n(i) =
+                                transform_conv_to_gemm.template MakeCDescriptor_M_N<DLayout>(
+                                    a_g_n_k_wos_lengths,
+                                    a_g_n_k_wos_strides,
+                                    b_g_k_c_xs_lengths,
+                                    b_g_k_c_xs_strides,
+                                    ds_g_n_c_wis_lengths[i],
+                                    ds_g_n_c_wis_strides[i],
+                                    conv_filter_strides,
+                                    conv_filter_dilations,
+                                    input_left_pads,
+                                    input_right_pads,
+                                    tildes);
+                        });
+
+                        const auto e_grid_desc_m_n =
+                            transform_conv_to_gemm.template MakeCDescriptor_M_N<ELayout>(
+                                a_g_n_k_wos_lengths,
+                                a_g_n_k_wos_strides,
+                                b_g_k_c_xs_lengths,
+                                b_g_k_c_xs_strides,
+                                e_g_n_c_wis_lengths,
+                                e_g_n_c_wis_strides,
+                                conv_filter_strides,
+                                conv_filter_dilations,
+                                input_left_pads,
+                                input_right_pads,
+                                tildes);
+
+                        // for check validity
+                        ds_grid_desc_m_n_container_.push_back(ds_grid_desc_m_n);
+                        e_grid_desc_m_n_container_.push_back(e_grid_desc_m_n);
+
+                        // desc for blockwise copy
+                        a_grid_desc_ak0_m_ak1_container_.push_back(a_grid_desc_ak0_m_ak1);
+                        b_grid_desc_bk0_n_bk1_container_.push_back(b_grid_desc_bk0_n_bk1);
+
+                        // block-to-e-tile-map
+                        auto block_2_ctile_map = GridwiseGemm::MakeDefaultBlock2CTileMap(
+                            e_grid_desc_m_n, 1 /* M01 */, 1 /* N01 */);
+
+                        block_2_ctile_map_container_.push_back(block_2_ctile_map);
+
+                        ds_grid_desc_mblock_mperblock_nblock_nperblock_container_.push_back(
+                            GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                                ds_grid_desc_m_n));
+                        e_grid_desc_mblock_mperblock_nblock_nperblock_container_.push_back(
+                            GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                                e_grid_desc_m_n));
+                    }
+                }
+            }
+        }
+
+        void Print() const
+        {
+            for(std::size_t i = 0; i < a_grid_desc_ak0_m_ak1_container_.size(); i++)
+            {
+                std::cout << "a_grid_desc_ak0_m_ak1_container_"
+                          << a_grid_desc_ak0_m_ak1_container_[i] << std::endl;
+
+                std::cout << "b_grid_desc_bk0_n_bk1_container_"
+                          << b_grid_desc_bk0_n_bk1_container_[i] << std::endl;
+
+                static_for<0, NumDTensor, 1>{}([&](auto j) {
+                    std::cout << "ds_grid_desc_mblock_mperblock_nblock_nperblock_container_"
+                              << ds_grid_desc_mblock_mperblock_nblock_nperblock_container_[i][j]
+                              << std::endl;
+                });
+
+                std::cout << "e_grid_desc_mblock_mperblock_nblock_nperblock_container_"
+                          << e_grid_desc_mblock_mperblock_nblock_nperblock_container_[i]
+                          << std::endl;
+            }
+        }
+
+        // pointers
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // tensor descriptor for problem definition
+        index_t num_group_;
+        std::vector<DsGridDesc_M_N> ds_grid_desc_m_n_container_;
+        std::vector<EGridDesc_M_N> e_grid_desc_m_n_container_;
+
+        // tensor descriptor for block-wise copy
+        std::vector<AGridDesc_AK0_M_AK1> a_grid_desc_ak0_m_ak1_container_;
+        std::vector<BGridDesc_BK0_N_BK1> b_grid_desc_bk0_n_bk1_container_;
+        std::vector<DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_container_;
+        std::vector<EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>
+            e_grid_desc_mblock_mperblock_nblock_nperblock_container_;
+
+        // block-to-e-tile map
+        std::vector<typename GridwiseGemm::DefaultBlock2CTileMap> block_2_ctile_map_container_;
+
+        // for computing batch offset
+        ComputePtrOffsetOfStridedBatch<NumDTensor> compute_ptr_offset_of_batch_;
+
+        // element-wise op
+        AElementwiseOp a_element_op_;
+        BElementwiseOp b_element_op_;
+        CDEElementwiseOp cde_element_op_;
+
+        // for checking IsSupportedArgument()
+        std::array<index_t, NDimSpatial + 3> a_g_n_k_wos_lengths_;
+        std::array<index_t, NDimSpatial + 3> a_g_n_k_wos_strides_;
+        std::array<index_t, NDimSpatial + 3> b_g_k_c_xs_lengths_;
+        std::array<index_t, NDimSpatial + 3> b_g_k_c_xs_strides_;
+        std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor> ds_g_n_c_wis_lengths_;
+        std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor> ds_g_n_c_wis_strides_;
+        std::array<index_t, NDimSpatial + 3> e_g_n_c_wis_lengths_;
+        std::array<index_t, NDimSpatial + 3> e_g_n_c_wis_strides_;
+        std::array<index_t, NDimSpatial> conv_filter_strides_;
+        std::array<index_t, NDimSpatial> conv_filter_dilations_;
+        std::array<index_t, NDimSpatial> input_left_pads_;
+        std::array<index_t, NDimSpatial> input_right_pads_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                arg.Print();
+            }
+
+            float ave_time = 0;
+
+            for(std::size_t i = 0; i < arg.a_grid_desc_ak0_m_ak1_container_.size(); i++)
+            {
+                const index_t grid_size = arg.block_2_ctile_map_container_[i].CalculateGridSize(
+                                              arg.e_grid_desc_m_n_container_[i]) *
+                                          arg.num_group_;
+
+                const auto GemmK = arg.a_grid_desc_ak0_m_ak1_container_[i].GetLength(I0) *
+                                   arg.a_grid_desc_ak0_m_ak1_container_[i].GetLength(I2);
+
+                auto launch_kernel = [&](auto has_main_k_block_loop) {
+                    constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                    const auto kernel = kernel_grouped_conv_fwd_multiple_d_wmma_cshuffle<
+                        GridwiseGemm,
+                        ADataType,
+                        BDataType,
+                        typename GridwiseGemm::DsGridPointer,
+                        EDataType,
+                        AElementwiseOp,
+                        BElementwiseOp,
+                        CDEElementwiseOp,
+                        DeviceOp::AGridDesc_AK0_M_AK1,
+                        DeviceOp::BGridDesc_BK0_N_BK1,
+                        typename GridwiseGemm::DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                        typename GridwiseGemm::EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+                        remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
+                        ComputePtrOffsetOfStridedBatch<NumDTensor>,
+                        has_main_loop>;
+
+                    return launch_and_time_kernel(
+                        stream_config,
+                        kernel,
+                        dim3(grid_size),
+                        dim3(BlockSize),
+                        0,
+                        arg.p_a_grid_,
+                        arg.p_b_grid_,
+                        arg.p_ds_grid_,
+                        arg.p_e_grid_,
+                        arg.a_element_op_,
+                        arg.b_element_op_,
+                        arg.cde_element_op_,
+                        arg.a_g_n_k_wos_lengths_[0], // Group count
+                        arg.a_grid_desc_ak0_m_ak1_container_[i],
+                        arg.b_grid_desc_bk0_n_bk1_container_[i],
+                        arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_container_[i],
+                        arg.e_grid_desc_mblock_mperblock_nblock_nperblock_container_[i],
+                        arg.block_2_ctile_map_container_[i],
+                        arg.compute_ptr_offset_of_batch_);
+                };
+
+                if(GridwiseGemm::CalculateHasMainKBlockLoop(GemmK))
+                {
+                    ave_time += launch_kernel(integral_constant<bool, true>{});
+                }
+                else
+                {
+                    ave_time += launch_kernel(integral_constant<bool, false>{});
+                }
+            }
+
+            return ave_time;
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        // check device
+        if(get_device_name() == "gfx1100" || get_device_name() == "gfx1101" ||
+           ck::get_device_name() == "gfx1102")
+        {
+            if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        const index_t ConvK = arg.b_g_k_c_xs_lengths_[1];
+        const index_t ConvC = arg.b_g_k_c_xs_lengths_[2];
+
+        // Specialization
+        if constexpr(ConvBackwardDataSpecialization ==
+                     ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0)
+        {
+            // check if it's a 1x1 convolution with stride=1 and no padding
+            for(int i = 0; i < NDimSpatial; i++)
+            {
+                if(!(arg.b_g_k_c_xs_lengths_[3 + i] == 1 && arg.conv_filter_strides_[i] == 1 &&
+                     arg.input_left_pads_[i] == 0 && arg.input_right_pads_[i] == 0))
+                {
+                    return false;
+                }
+            }
+        }
+
+        // vector load for A matrix from global memory to LDS
+        if constexpr(is_same_v<ALayout, tensor_layout::convolution::GNHWK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::GNDHWK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::NHWGK> ||
+                     is_same_v<ALayout, tensor_layout::convolution::NDHWGK>)
+        {
+            if(!(ABlockTransferSrcVectorDim == 2 && ConvK % ABlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // vector load for B matrix from global memory to LDS
+        if constexpr(is_same_v<BLayout, tensor_layout::convolution::GKYXC> ||
+                     is_same_v<BLayout, tensor_layout::convolution::GKZYXC>)
+        {
+            if(!(BBlockTransferSrcVectorDim == 1 && ConvC % BBlockTransferSrcScalarPerVector == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // vector store for Ds
+        bool ds_valid = true;
+
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            using DLayout = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+
+            if constexpr(is_same_v<DLayout, tensor_layout::convolution::GNHWC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::GNDHWC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::NHWGC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::NDHWGC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::G_NHW_C> ||
+                         is_same_v<DLayout, tensor_layout::convolution::GC> ||
+                         is_same_v<DLayout, tensor_layout::convolution::G_C>)
+            {
+                // vector load D matrix from global memory
+                if(!(ConvC % CDEShuffleBlockTransferScalarPerVector_NPerBlock == 0))
+                {
+                    ds_valid = false;
+                }
+            }
+            else
+            {
+                ds_valid = false;
+            }
+        });
+
+        if(!ds_valid)
+        {
+            return false;
+        }
+
+        // vector store for E
+        if constexpr(is_same_v<ELayout, tensor_layout::convolution::GNHWC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::GNDHWC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::NHWGC> ||
+                     is_same_v<ELayout, tensor_layout::convolution::NDHWGC>)
+        {
+            // vector store C matrix into global memory
+            if(!(ConvC % CDEShuffleBlockTransferScalarPerVector_NPerBlock == 0))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // Gridwise GEMM size
+        for(std::size_t i = 0; i < arg.a_grid_desc_ak0_m_ak1_container_.size(); i++)
+        {
+            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_ak0_m_ak1_container_[i],
+                                            arg.b_grid_desc_bk0_n_bk1_container_[i],
+                                            arg.ds_grid_desc_m_n_container_[i],
+                                            arg.e_grid_desc_m_n_container_[i],
+                                            arg.block_2_ctile_map_container_[i]))
+            {
+                return false;
+            }
+        }
+
+        return true;
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto
+    MakeArgument(const void* p_a,                                                 // output image
+                 const void* p_b,                                                 // weight
+                 const std::array<const void*, NumDTensor>& p_ds,                 // bias
+                 void* p_e,                                                       // input image
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_k_wos_lengths, // output image
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_k_wos_strides, // output image
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,  // weight
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,  // weight
+                 const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+                     ds_g_n_c_wis_lengths, // bias
+                 const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+                     ds_g_n_c_wis_strides,                                        // bias
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_c_wis_lengths, // input image
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_c_wis_strides, // input image
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads,
+                 const AElementwiseOp& a_element_op,
+                 const BElementwiseOp& b_element_op,
+                 const CDEElementwiseOp& cde_element_op)
+    {
+        return Argument{p_a,
+                        p_b,
+                        p_ds,
+                        p_e,
+                        a_g_n_k_wos_lengths,
+                        a_g_n_k_wos_strides,
+                        b_g_k_c_xs_lengths,
+                        b_g_k_c_xs_strides,
+                        ds_g_n_c_wis_lengths,
+                        ds_g_n_c_wis_strides,
+                        e_g_n_c_wis_lengths,
+                        e_g_n_c_wis_strides,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    std::unique_ptr<BaseArgument> MakeArgumentPointer(
+        const void* p_a,                                                 // output image
+        const void* p_b,                                                 // weight
+        const std::array<const void*, NumDTensor>& p_ds,                 // bias
+        void* p_e,                                                       // input image
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_k_wos_lengths, // output image
+        const std::array<index_t, NDimSpatial + 3>& a_g_n_k_wos_strides, // output image
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths,  // weight
+        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,  // weight
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+            ds_g_n_c_wis_lengths, // bias
+        const std::array<std::array<index_t, NDimSpatial + 3>, NumDTensor>&
+            ds_g_n_c_wis_strides,                                        // bias
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_c_wis_lengths, // input image
+        const std::array<index_t, NDimSpatial + 3>& e_g_n_c_wis_strides, // input image
+        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<index_t, NDimSpatial>& input_left_pads,
+        const std::array<index_t, NDimSpatial>& input_right_pads,
+        const AElementwiseOp& a_element_op,
+        const BElementwiseOp& b_element_op,
+        const CDEElementwiseOp& cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_a,
+                                          p_b,
+                                          p_ds,
+                                          p_e,
+                                          a_g_n_k_wos_lengths,
+                                          a_g_n_k_wos_strides,
+                                          b_g_k_c_xs_lengths,
+                                          b_g_k_c_xs_strides,
+                                          ds_g_n_c_wis_lengths,
+                                          ds_g_n_c_wis_strides,
+                                          e_g_n_c_wis_lengths,
+                                          e_g_n_c_wis_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << getConvBackwardDataSpecializationString(ConvBackwardDataSpecialization) << ", "
+            << K1 << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index c828269acf..37d7b0b1fc 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -14,6 +14,7 @@
 #include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
 #include "ck/tensor_operation/operator_transform/transform_conv_bwd_data_to_gemm_v1.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/io.hpp"
@@ -24,51 +25,6 @@ namespace device {
 
 namespace {
 
-template <index_t NumDTensor>
-struct ComputePtrOffsetOfStridedBatch
-{
-    ComputePtrOffsetOfStridedBatch() = default;
-
-    ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
-                                   index_t BatchStrideB,
-                                   Array<ck::index_t, NumDTensor> BatchStrideDs,
-                                   index_t BatchStrideE)
-        : BatchStrideA_(BatchStrideA),
-          BatchStrideB_(BatchStrideB),
-          BatchStrideDs_(BatchStrideDs),
-          BatchStrideE_(BatchStrideE)
-    {
-    }
-
-    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideA_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideB_);
-    }
-
-    __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
-    {
-        Array<long_index_t, NumDTensor> ds_offset;
-        static_for<0, NumDTensor, 1>{}(
-            [&](auto i) { ds_offset(i) = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]); });
-        return ds_offset;
-    }
-
-    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideE_);
-    }
-
-    index_t BatchStrideA_;
-    index_t BatchStrideB_;
-    Array<ck::index_t, NumDTensor> BatchStrideDs_;
-    index_t BatchStrideE_;
-};
-
 /*
  * \brief Wrapper function of GridwiseGemm::Run to realize BatchedGEMM.
  *
@@ -257,7 +213,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
                                                BElementwiseOp,
                                                CDEElementwiseOp>
 {
-    // FIXME
+    // TODO: Extend support for more spatial dimensions.
     static_assert(NDimSpatial == 2 || NDimSpatial == 3,
                   "wrong! only implemented for 2D and 3D now");
 
@@ -265,7 +221,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
     static constexpr index_t NumDTensor = DsDataType::Size();
 
-    // TODO make A/B datatype different
+    // TODO: Add support for different A and B data types.
     using ABDataType = ADataType;
 
     static constexpr auto I0 = Number<0>{};
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 8b22bd2090..8cfdd04e55 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -19,6 +19,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/io.hpp"
@@ -29,51 +30,6 @@ namespace device {
 
 namespace {
 
-template <index_t NumDTensor>
-struct ComputePtrOffsetOfStridedBatch
-{
-    ComputePtrOffsetOfStridedBatch() = default;
-
-    ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
-                                   index_t BatchStrideB,
-                                   Array<ck::index_t, NumDTensor> BatchStrideDs,
-                                   index_t BatchStrideE)
-        : BatchStrideA_(BatchStrideA),
-          BatchStrideB_(BatchStrideB),
-          BatchStrideDs_(BatchStrideDs),
-          BatchStrideE_(BatchStrideE)
-    {
-    }
-
-    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideA_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideB_);
-    }
-
-    __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
-    {
-        Array<long_index_t, NumDTensor> ds_offset;
-        static_for<0, NumDTensor, 1>{}(
-            [&](auto i) { ds_offset(i) = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]); });
-        return ds_offset;
-    }
-
-    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideE_);
-    }
-
-    index_t BatchStrideA_;
-    index_t BatchStrideB_;
-    Array<ck::index_t, NumDTensor> BatchStrideDs_;
-    index_t BatchStrideE_;
-};
-
 /*
  * \brief Wrapper function of GridwiseGemm::Run to realize BatchedGEMM.
  *
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
index 1d73a72387..e4e793eff8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -19,6 +19,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/io.hpp"
@@ -27,55 +28,6 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-namespace {
-
-template <index_t NumDTensor>
-struct ComputePtrOffsetOfStridedBatch
-{
-    ComputePtrOffsetOfStridedBatch() = default;
-
-    ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
-                                   index_t BatchStrideB,
-                                   Array<ck::index_t, NumDTensor> BatchStrideDs,
-                                   index_t BatchStrideE)
-        : BatchStrideA_(BatchStrideA),
-          BatchStrideB_(BatchStrideB),
-          BatchStrideDs_(BatchStrideDs),
-          BatchStrideE_(BatchStrideE)
-    {
-    }
-
-    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideA_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideB_);
-    }
-
-    __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
-    {
-        Array<long_index_t, NumDTensor> ds_offset;
-        static_for<0, NumDTensor, 1>{}(
-            [&](auto i) { ds_offset(i) = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]); });
-        return ds_offset;
-    }
-
-    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideE_);
-    }
-
-    index_t BatchStrideA_;
-    index_t BatchStrideB_;
-    Array<ck::index_t, NumDTensor> BatchStrideDs_;
-    index_t BatchStrideE_;
-};
-
-} // namespace
-
 //
 // @brief      Device Convolution operation.
 //
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index bcef5c3b62..b04594fceb 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -19,6 +19,7 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/io.hpp"
@@ -29,51 +30,6 @@ namespace device {
 
 namespace {
 
-template <index_t NumDTensor>
-struct ComputePtrOffsetOfStridedBatch
-{
-    ComputePtrOffsetOfStridedBatch() = default;
-
-    ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
-                                   index_t BatchStrideB,
-                                   Array<ck::index_t, NumDTensor> BatchStrideDs,
-                                   index_t BatchStrideE)
-        : BatchStrideA_(BatchStrideA),
-          BatchStrideB_(BatchStrideB),
-          BatchStrideDs_(BatchStrideDs),
-          BatchStrideE_(BatchStrideE)
-    {
-    }
-
-    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideA_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideB_);
-    }
-
-    __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
-    {
-        Array<long_index_t, NumDTensor> ds_offset;
-        static_for<0, NumDTensor, 1>{}(
-            [&](auto i) { ds_offset(i) = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]); });
-        return ds_offset;
-    }
-
-    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideE_);
-    }
-
-    index_t BatchStrideA_;
-    index_t BatchStrideB_;
-    Array<ck::index_t, NumDTensor> BatchStrideDs_;
-    index_t BatchStrideE_;
-};
-
 /*
  * \brief Wrapper function of GridwiseGemm::Run to realize BatchedGEMM.
  *
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
new file mode 100644
index 0000000000..948232ed6f
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
@@ -0,0 +1,59 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/utility/common_header.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <index_t NumDTensor>
+struct ComputePtrOffsetOfStridedBatch
+{
+    ComputePtrOffsetOfStridedBatch() = default;
+
+    ComputePtrOffsetOfStridedBatch(index_t BatchStrideA,
+                                   index_t BatchStrideB,
+                                   Array<ck::index_t, NumDTensor> BatchStrideDs,
+                                   index_t BatchStrideE)
+        : BatchStrideA_(BatchStrideA),
+          BatchStrideB_(BatchStrideB),
+          BatchStrideDs_(BatchStrideDs),
+          BatchStrideE_(BatchStrideE)
+    {
+    }
+
+    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideA_);
+    }
+
+    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideB_);
+    }
+
+    __host__ __device__ constexpr auto GetDsPtrOffset(index_t g_idx) const
+    {
+        Array<long_index_t, NumDTensor> ds_offset;
+        static_for<0, NumDTensor, 1>{}(
+            [&](auto i) { ds_offset(i) = g_idx * static_cast<long_index_t>(BatchStrideDs_[i]); });
+        return ds_offset;
+    }
+
+    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideE_);
+    }
+
+    index_t BatchStrideA_;
+    index_t BatchStrideB_;
+    Array<ck::index_t, NumDTensor> BatchStrideDs_;
+    index_t BatchStrideE_;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp
new file mode 100644
index 0000000000..5db8226e11
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp
@@ -0,0 +1,118 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+using I8  = int8_t;
+using I32 = int32_t;
+
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdDataDefault =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Default;
+
+static constexpr auto ConvBwdData1x1S1P0 =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_wmma_f16_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|    AccData|  CShuffle|      DsData| EData|           A|           B|          CDE|    ConvForward| Block|  MPer|  NPer| K0Per| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|   Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|       Type|  DataType|        Type|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|          |        |        |         |        |      |      |           |          |            |      |   Operation|   Operation|    Operation|               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|          |        |        |         |        |      |      |           |          |            |      |            |            |             |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,   128,   256,     8,  8,    16,   16,       4,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,    64,   256,     8,  8,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,   128,   256,     8,  8,    16,   16,       4,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,   128,    64,     8,  8,    16,   16,       4,       1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,   128,     8,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,   128,     8,  8,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   128,    64,     8,  8,    16,   16,       4,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   128,   128,     8,  8,    16,   16,       4,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    32,   256,     8,  8,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        // blocksize=64
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,    64,     8,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    64,    64,     8,  8,    16,   16,       2,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,    64,     8,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   128,     8,  8,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    64,     8,  8,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    64,    32,     8,  8,    16,   16,       4,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    32,    32,     8,  8,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout,  F16,   F16,     F32,       F16, Empty_Tuple,   F16,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    32,     8,  8,    16,   16,       1,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          typename DsDatatype,
+          typename CDEElementOp,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_wmma_i8_instances = std::tuple<
+    // clang-format off
+        //########################################|    NumDim|       A|       B|       Ds|       E| AData| BData|    AccData|  CShuffle|      DsData| EData|           A|           B|          CDE|    ConvForward| Block|  MPer|  NPer| K0Per| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //########################################|   Spatial|  Layout|  Layout|   Layout|  Layout|  Type|  Type|       Type|  DataType|        Type|  Type| Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //########################################|          |        |        |         |        |      |      |           |          |            |      |   Operation|   Operation|    Operation|               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //########################################|          |        |        |         |        |      |      |           |          |            |      |            |            |             |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,    64,     4,  16,    16,   16,       2,       2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,               1,              16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            1,              16,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 256,    64,   256,     8,  16,    16,   16,       2,       4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        // blocksize=128
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,   256,     8,  16,    16,   16,       2,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    64,   128,     8,  16,    16,   16,       2,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   128,   256,     8,  16,    16,   16,       4,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,    32,   256,     8,  16,    16,   16,       1,       8,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec, 128,   256,   128,     8,  16,    16,   16,       8,       4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 4>,               8>,      
+        // blocksize=64
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   128,     8,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    64,   128,     8,  16,    16,   16,       2,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   128,     8,  16,    16,   16,       1,       8,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  64,    32,   64,      8,  16,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    64,     8,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    64,    64,     8,  16,    16,   16,       4,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    32,    32,     8,  16,    16,   16,       2,       2,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle<NDSpatial, ALayout, BLayout, DsLayout, ELayout, I8,    I8,        I32,        I8, Empty_Tuple,    I8,  PassThrough, PassThrough, CDEElementOp,       ConvSpec,  32,    16,    64,     8,  16,    16,   16,       1,       4,     S<2, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              16,              16,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              16,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index 2b49e17187..b259c640da 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -30,6 +30,34 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f32_instances(
@@ -61,6 +89,35 @@ void add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+#endif
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
@@ -75,6 +132,35 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f32_instances(
@@ -106,6 +192,35 @@ void add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+#endif
 // conv3d backward data
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
@@ -121,6 +236,34 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f32_instances(
@@ -152,6 +295,35 @@ void add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+#endif
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
@@ -166,6 +338,34 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                                                                   PassThrough,
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f32_instances(
@@ -197,6 +397,35 @@ void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances);
+#endif
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
@@ -247,6 +476,10 @@ struct DeviceOperationInstanceFactory<
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_f16_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -263,6 +496,15 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv2d_bwd_data_xdl_gnhwk_gkyxc_gnhwc_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_INT8
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                                  is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
             else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
@@ -273,6 +515,10 @@ struct DeviceOperationInstanceFactory<
                              is_same_v<OutDataType, F16>)
                 {
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_f16_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -289,6 +535,15 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv2d_bwd_data_xdl_nhwgk_gkyxc_nhwgc_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_INT8
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                                  is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_instances(op_ptrs);
+                    add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
         }
@@ -304,6 +559,10 @@ struct DeviceOperationInstanceFactory<
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_f16_instances(
                         op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -321,6 +580,16 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv3d_bwd_data_xdl_gndhwk_gkzyxc_gndhwc_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_INT8
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                                  is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
@@ -332,6 +601,10 @@ struct DeviceOperationInstanceFactory<
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                         op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_FP32
@@ -349,6 +622,16 @@ struct DeviceOperationInstanceFactory<
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_INT8
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                                  is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
         }
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
index 85ec0f55aa..d8bd0de692 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/CMakeLists.txt
@@ -1,8 +1,16 @@
 add_instance_library(device_grouped_conv2d_bwd_data_instance
-   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
-   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
-   device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
-   device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
-)
+   xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+   xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+   xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
+   xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+
+   wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
+   wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..ac2ba91b63
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<2,
+                                                        GNHWK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        GNHWC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
new file mode 100644
index 0000000000..39af70f623
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_f16_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<2,
+                                                        GNHWK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        GNHWC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..1de9f7a95c
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<2,
+                                                       GNHWK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       GNHWC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
new file mode 100644
index 0000000000..8eb6558a2a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_gnhwc_gkyxc_gnhwk_i8_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_gnhwk_gkyxc_gnhwc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  GNHWK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  GNHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<2,
+                                                       GNHWK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       GNHWC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..f46cdf7f18
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<2,
+                                                        NHWGK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        NHWGC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..2f4659524f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_f16_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<2,
+                                                        NHWGK,
+                                                        GKYXC,
+                                                        Empty_Tuple,
+                                                        NHWGC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..789b80b8ac
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<2,
+                                                       NHWGK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       NHWGC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
new file mode 100644
index 0000000000..71394762b4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/wmma/device_grouped_conv2d_bwd_data_wmma_nhwgc_gkyxc_nhwgk_i8_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv2d_bwd_data_wmma_nhwgk_gkyxc_nhwgc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<2,
+                                                                  NHWGK,
+                                                                  GKYXC,
+                                                                  Empty_Tuple,
+                                                                  NHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<2,
+                                                       NHWGK,
+                                                       GKYXC,
+                                                       Empty_Tuple,
+                                                       NHWGC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_data/xdl/device_grouped_conv2d_bwd_data_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
index 8383d622b2..f44e97ed75 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
@@ -1,8 +1,17 @@
 add_instance_library(device_grouped_conv3d_bwd_data_instance
-   device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
-   device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
-   device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
-   device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
-   device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
-   device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+   xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+   xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+   xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+   xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+   xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+
+   wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
+   wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..ba2f6bdc0a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<3,
+                                                        GNDHWK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        GNDHWC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
new file mode 100644
index 0000000000..26403bf0e5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<3,
+                                                        GNDHWK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        GNDHWC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..9e453a36c5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<3,
+                                                       GNDHWK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       GNDHWC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
new file mode 100644
index 0000000000..6f7d4f79d1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_gndhwk_gkzyxc_gndhwc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  GNDHWK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  GNDHWC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<3,
+                                                       GNDHWK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       GNDHWC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..4475a0e456
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<3,
+                                                        NDHWGK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        NDHWGC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..f3941636f8
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_f16_instances<3,
+                                                        NDHWGK,
+                                                        GKZYXC,
+                                                        Empty_Tuple,
+                                                        NDHWGC,
+                                                        Empty_Tuple,
+                                                        PassThrough,
+                                                        ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..b8479c9c9d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<3,
+                                                       NDHWGK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       NDHWGC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdData1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
new file mode 100644
index 0000000000..d8a74e744b
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  int8_t,
+                                                                  int8_t,
+                                                                  Empty_Tuple,
+                                                                  int8_t,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_wmma_i8_instances<3,
+                                                       NDHWGK,
+                                                       GKZYXC,
+                                                       Empty_Tuple,
+                                                       NDHWGC,
+                                                       Empty_Tuple,
+                                                       PassThrough,
+                                                       ConvBwdDataDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
diff --git a/test/grouped_convnd_bwd_data/CMakeLists.txt b/test/grouped_convnd_bwd_data/CMakeLists.txt
index 436ab2af76..9773e5a9c6 100644
--- a/test/grouped_convnd_bwd_data/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_data/CMakeLists.txt
@@ -1,6 +1,19 @@
-if(GPU_TARGETS MATCHES "gfx908" OR GPU_TARGETS MATCHES "gfx90a" OR GPU_TARGETS MATCHES "gfx940")
-    add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
-    target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
-    add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface.cpp)
-    target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
-endif()
\ No newline at end of file
+list(APPEND gpu_list_xdl gfx908 gfx90a gfx940)
+list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+    if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
+        add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
+        target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
+        add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface_xdl.cpp)
+        target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+        set(target 1)
+    endif()
+    if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
+        add_gtest_executable(test_grouped_convnd_bwd_data test_grouped_convnd_bwd_data.cpp)
+        target_link_libraries(test_grouped_convnd_bwd_data PRIVATE utility device_grouped_conv2d_bwd_data_instance device_grouped_conv3d_bwd_data_instance)
+        add_gtest_executable(test_grouped_convnd_bwd_data_interface test_grouped_convnd_bwd_data_interface_wmma.cpp)
+        target_link_libraries(test_grouped_convnd_bwd_data_interface PRIVATE utility device_grouped_conv2d_bwd_data_instance)
+        set(target 1)
+    endif()
+endforeach()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
index bdad37fe14..96506b876d 100644
--- a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data.cpp
@@ -51,16 +51,20 @@ using namespace ck::tensor_layout::convolution;
 using KernelTypes2d = ::testing::Types<std::tuple<float, GNHWK, GKYXC, GNHWC>,
                                        std::tuple<ck::half_t, GNHWK, GKYXC, GNHWC>,
                                        std::tuple<ck::bhalf_t, GNHWK, GKYXC, GNHWC>,
+                                       std::tuple<int8_t, GNHWK, GKYXC, GNHWC>,
                                        std::tuple<float, NHWGK, GKYXC, NHWGC>,
                                        std::tuple<ck::half_t, NHWGK, GKYXC, NHWGC>,
-                                       std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>>;
+                                       std::tuple<ck::bhalf_t, NHWGK, GKYXC, NHWGC>,
+                                       std::tuple<int8_t, NHWGK, GKYXC, NHWGC>>;
 
 using KernelTypes3d = ::testing::Types<std::tuple<float, GNDHWK, GKZYXC, GNDHWC>,
                                        std::tuple<ck::half_t, GNDHWK, GKZYXC, GNDHWC>,
                                        std::tuple<ck::bhalf_t, GNDHWK, GKZYXC, GNDHWC>,
+                                       std::tuple<int8_t, GNDHWK, GKZYXC, GNDHWC>,
                                        std::tuple<float, NDHWGK, GKZYXC, NDHWGC>,
                                        std::tuple<ck::half_t, NDHWGK, GKZYXC, NDHWGC>,
-                                       std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>>;
+                                       std::tuple<ck::bhalf_t, NDHWGK, GKZYXC, NDHWGC>,
+                                       std::tuple<int8_t, NDHWGK, GKZYXC, NDHWGC>>;
 
 template <typename Tuple>
 class TestGroupedConvndBwdData2d : public TestGroupedConvndBwdData<Tuple>
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface_wmma.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface_wmma.cpp
new file mode 100644
index 0000000000..c0429c6d09
--- /dev/null
+++ b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface_wmma.cpp
@@ -0,0 +1,178 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_data_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using DataType    = ck::half_t;
+using AccDataType = float;
+using Pass        = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvBackwardDataSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardDataSpecialization;
+
+static constexpr auto ConvBwdDataDefault   = ConvBackwardDataSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 = ConvBackwardDataSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvBackwardDataSpecialization ConvSpec>
+class TestGroupedConvndBwdData : public ::testing::Test
+{
+    protected:
+    static constexpr ck::index_t NDimSpatial = 2;
+
+    using OutLayout = std::tuple_element_t<0, Tuple>;
+    using WeiLayout = std::tuple_element_t<1, Tuple>;
+    using InLayout  = std::tuple_element_t<2, Tuple>;
+
+    // clang-format off
+    using GroupedConvBwdDataDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle
+            //|    NumDim|        A|         B|          Ds|       E|        AData|        BData|    AccData|          CShuffle|     DsData|       EData|           A|           B|          CDE|       ConvForward| Block|  MPer|  NPer| K0Per| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+            //|   Spatial|   Layout|    Layout|      Layout|  Layout|         Type|         Type|       Type|          DataType|       Type|        Type| Elementwise| Elementwise|  Elementwise|    Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+            //|          |         |          |            |        |             |             |           |                  |           |            |   Operation|   Operation|    Operation|                  |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+            //|          |         |          |            |        |             |             |           |                  |           |            |            |            |             |                  |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+            < NDimSpatial,OutLayout, WeiLayout, ck::Tuple<>, InLayout,       DataType,  DataType, AccDataType,          DataType,  ck::Tuple<>,   DataType,        Pass,        Pass,        Pass,         ConvSpec, 64,    32,    64,     8,  8,    16,   16,       1,       4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    template <ck::index_t NDimSpatial>
+    bool Run()
+    {
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial + 3> out_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> out_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> wei_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> in_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> in_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+        copy(out_g_n_k_wos_desc.GetLengths(), out_lengths);
+        copy(out_g_n_k_wos_desc.GetStrides(), out_strides);
+        copy(wei_g_k_c_xs_desc.GetLengths(), wei_lengths);
+        copy(wei_g_k_c_xs_desc.GetStrides(), wei_strides);
+        copy(in_g_n_c_wis_desc.GetLengths(), in_lengths);
+        copy(in_g_n_c_wis_desc.GetStrides(), in_strides);
+        copy(conv_param.conv_filter_strides_, conv_filter_strides);
+        copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+        copy(conv_param.input_left_pads_, input_left_pads);
+        copy(conv_param.input_right_pads_, input_right_pads);
+
+        auto conv = GroupedConvBwdDataDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          std::array<const void*, 0>{},
+                                          nullptr,
+                                          out_lengths,
+                                          out_strides,
+                                          wei_lengths,
+                                          wei_strides,
+                                          {},
+                                          {},
+                                          in_lengths,
+                                          in_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          Pass{},
+                                          Pass{},
+                                          Pass{});
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using GNHWC = ck::tensor_layout::convolution::GNHWC;
+using NHWGC = ck::tensor_layout::convolution::NHWGC;
+
+using GKYXC = ck::tensor_layout::convolution::GKYXC;
+
+using GNHWK = ck::tensor_layout::convolution::GNHWK;
+using NHWGK = ck::tensor_layout::convolution::NHWGK;
+
+using KernelTypes =
+    ::testing::Types<std::tuple<GNHWK, GKYXC, GNHWC>, std::tuple<NHWGK, GKYXC, NHWGC>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataDefault : public TestGroupedConvndBwdData<Tuple, ConvBwdDataDefault>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdDataFilter1x1
+    : public TestGroupedConvndBwdData<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataDefault, KernelTypes);
+TYPED_TEST_SUITE(TestGroupedConvndBwdDataFilter1x1, KernelTypes);
+
+TYPED_TEST(TestGroupedConvndBwdDataFilter1x1, SpecializationCheck)
+{
+    // Check filter 3,3 instead of 1,1
+    this->conv_param  = {2, 2, 4, 192, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2,2 instead of 1,1
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {2, 2, 4, 192, 192, {1, 1}, {28, 28}, {1, 1}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdDataDefault, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param  = {2, 2, 128, 129, 256, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    bool is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {2, 2, 128, 128, 257, {1, 1}, {7, 7}, {2, 2}, {1, 1}, {0, 0}, {0, 0}};
+    is_supported     = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}
diff --git a/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp b/test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface_xdl.cpp
similarity index 100%
rename from test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface.cpp
rename to test/grouped_convnd_bwd_data/test_grouped_convnd_bwd_data_interface_xdl.cpp

From f07485060e23b871872640a5b1d39fe05bd53a5d Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Fri, 29 Sep 2023 17:54:31 +0200
Subject: [PATCH 150/194] Add support for mixed precision in contraction scale
 and bilinear (#936)

* Extract common functionality to separate files

* Reference contraction: Remove incorrect consts from type_converts

* Reference contraction: Add missing type_convert for dst value

* Reference contraction: Fix incorrect order of B matrix dimensions

* Add support for mixed precision in contraction scale and bilinear

* Move using statements from instances to a common file

* Move using statements from examples to a common file

* Fix the order of B matrix dimensions across examples and profiler

* Fix the computation of error threshold

* Make ComputeDataType an optional argument

* Include possible DataType -> ComputeDataType casting error in the threshold

* Remove commented code
---
 example/26_contraction/CMakeLists.txt         |  48 ++
 example/26_contraction/common_instances.hpp   | 183 ++++++++
 ...raction_bilinear_xdl_bf16_compute_fp32.cpp |  86 ++++
 ...raction_bilinear_xdl_fp16_compute_fp32.cpp |  86 ++++
 .../contraction_bilinear_xdl_fp32.cpp         | 319 +++----------
 ...raction_bilinear_xdl_fp32_compute_bf16.cpp |  86 ++++
 ...raction_bilinear_xdl_fp32_compute_fp16.cpp |  86 ++++
 .../contraction_bilinear_xdl_fp64.cpp         | 319 +++----------
 ...raction_bilinear_xdl_fp64_compute_fp32.cpp |  86 ++++
 ...ontraction_scale_xdl_bf16_compute_fp32.cpp |  85 ++++
 ...ontraction_scale_xdl_fp16_compute_fp32.cpp |  85 ++++
 .../contraction_scale_xdl_fp32.cpp            | 303 +++----------
 ...ontraction_scale_xdl_fp32_compute_bf16.cpp |  85 ++++
 ...ontraction_scale_xdl_fp32_compute_fp16.cpp |  85 ++++
 .../contraction_scale_xdl_fp64.cpp            | 303 +++----------
 ...ontraction_scale_xdl_fp64_compute_fp32.cpp |  85 ++++
 .../run_contraction_bilinear_example.inc      | 234 ++++++++++
 .../run_contraction_scale_example.inc         | 217 +++++++++
 .../device/device_contraction_multiple_d.hpp  |   3 +-
 ...ce_contraction_multiple_d_xdl_cshuffle.hpp |   6 +-
 .../element/binary_element_wise_operation.hpp |  19 +
 .../element/unary_element_wise_operation.hpp  |  26 ++
 .../cpu/reference_contraction.hpp             |  16 +-
 .../device_operation_instance_factory.hpp     |   2 +
 .../device_contraction_instance.hpp           | 292 ++++++++++++
 .../gpu/contraction_bilinear.hpp              | 424 ++++++++++++++++--
 .../gpu/contraction_scale.hpp                 | 420 +++++++++++++++--
 .../gpu/contraction_bilinear/CMakeLists.txt   |  46 +-
 ...16_bf16_bf16_compute_f32_kknn_instance.cpp |  57 +++
 ...16_bf16_bf16_compute_f32_knnn_instance.cpp |  57 +++
 ...16_bf16_bf16_compute_f32_mknn_instance.cpp |  57 +++
 ...16_bf16_bf16_compute_f32_mnnn_instance.cpp |  57 +++
 ..._f16_f16_f16_compute_f32_kknn_instance.cpp |  57 +++
 ..._f16_f16_f16_compute_f32_knnn_instance.cpp |  57 +++
 ..._f16_f16_f16_compute_f32_mknn_instance.cpp |  57 +++
 ..._f16_f16_f16_compute_f32_mnnn_instance.cpp |  57 +++
 ...f32_f32_f32_compute_bf16_kknn_instance.cpp |  57 +++
 ...f32_f32_f32_compute_bf16_knnn_instance.cpp |  57 +++
 ...f32_f32_f32_compute_bf16_mknn_instance.cpp |  57 +++
 ...f32_f32_f32_compute_bf16_mnnn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_f16_kknn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_f16_knnn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_f16_mknn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_f16_mnnn_instance.cpp |  57 +++
 ..._shuffle_f32_f32_f32_f32_kknn_instance.cpp |  52 +--
 ..._shuffle_f32_f32_f32_f32_knnn_instance.cpp |  55 +--
 ..._shuffle_f32_f32_f32_f32_mknn_instance.cpp |  55 +--
 ..._shuffle_f32_f32_f32_f32_mnnn_instance.cpp |  55 +--
 ..._f64_f64_f64_compute_f32_kknn_instance.cpp |  57 +++
 ..._f64_f64_f64_compute_f32_knnn_instance.cpp |  57 +++
 ..._f64_f64_f64_compute_f32_mknn_instance.cpp |  57 +++
 ..._f64_f64_f64_compute_f32_mnnn_instance.cpp |  57 +++
 ..._shuffle_f64_f64_f64_f64_kknn_instance.cpp |  49 +-
 ..._shuffle_f64_f64_f64_f64_knnn_instance.cpp |  49 +-
 ..._shuffle_f64_f64_f64_f64_mknn_instance.cpp |  49 +-
 ..._shuffle_f64_f64_f64_f64_mnnn_instance.cpp |  49 +-
 .../gpu/contraction_scale/CMakeLists.txt      |  32 +-
 ...f16_bf16_bf16_compute_f32_kkn_instance.cpp |  57 +++
 ...f16_bf16_bf16_compute_f32_knn_instance.cpp |  57 +++
 ...f16_bf16_bf16_compute_f32_mkn_instance.cpp |  57 +++
 ...f16_bf16_bf16_compute_f32_mnn_instance.cpp |  57 +++
 ...e_f16_f16_f16_compute_f32_kkn_instance.cpp |  57 +++
 ...e_f16_f16_f16_compute_f32_knn_instance.cpp |  57 +++
 ...e_f16_f16_f16_compute_f32_mkn_instance.cpp |  57 +++
 ...e_f16_f16_f16_compute_f32_mnn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_bf16_kkn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_bf16_knn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_bf16_mkn_instance.cpp |  57 +++
 ..._f32_f32_f32_compute_bf16_mnn_instance.cpp |  57 +++
 ...e_f32_f32_f32_compute_f16_kkn_instance.cpp |  57 +++
 ...e_f32_f32_f32_compute_f16_knn_instance.cpp |  57 +++
 ...e_f32_f32_f32_compute_f16_mkn_instance.cpp |  57 +++
 ...e_f32_f32_f32_compute_f16_mnn_instance.cpp |  57 +++
 ...xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp |  56 +--
 ...xdl_c_shuffle_f32_f32_f32_knn_instance.cpp |  59 +--
 ...xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp |  59 +--
 ...xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp |  59 +--
 ...e_f64_f64_f64_compute_f32_kkn_instance.cpp |  57 +++
 ...e_f64_f64_f64_compute_f32_knn_instance.cpp |  57 +++
 ...e_f64_f64_f64_compute_f32_mkn_instance.cpp |  57 +++
 ...e_f64_f64_f64_compute_f32_mnn_instance.cpp |  57 +++
 ...xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp |  49 +-
 ...xdl_c_shuffle_f64_f64_f64_knn_instance.cpp |  49 +-
 ...xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp |  49 +-
 ...xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp |  49 +-
 profiler/README.md                            |  24 +-
 .../profiler/profile_contraction_impl.hpp     |  62 ++-
 .../profiler/profile_contraction_utils.hpp    |  14 +-
 profiler/src/profile_contraction_bilinear.cpp | 219 +++++----
 profiler/src/profile_contraction_scale.cpp    | 215 +++++----
 test/contraction/test_contraction.cpp         | 147 +++---
 .../test_contraction_interface.cpp            |  10 +-
 92 files changed, 5918 insertions(+), 1962 deletions(-)
 create mode 100644 example/26_contraction/common_instances.hpp
 create mode 100644 example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp
 create mode 100644 example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp
 create mode 100644 example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp
 create mode 100644 example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp
 create mode 100644 example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp
 create mode 100644 example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp
 create mode 100644 example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp
 create mode 100644 example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp
 create mode 100644 example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp
 create mode 100644 example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp
 create mode 100644 example/26_contraction/run_contraction_bilinear_example.inc
 create mode 100644 example/26_contraction/run_contraction_scale_example.inc
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp

diff --git a/example/26_contraction/CMakeLists.txt b/example/26_contraction/CMakeLists.txt
index 167a9f147d..1a0489ce9c 100644
--- a/example/26_contraction/CMakeLists.txt
+++ b/example/26_contraction/CMakeLists.txt
@@ -1,4 +1,52 @@
+add_custom_target(example_contraction)
+add_custom_target(example_contraction_scale)
+add_custom_target(example_contraction_bilinear)
+
+# FP32
 add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32)
+
 add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32)
+
+add_example_executable(example_contraction_bilinear_xdl_fp32_compute_bf16 contraction_bilinear_xdl_fp32_compute_bf16.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_bf16)
+
+add_example_executable(example_contraction_scale_xdl_fp32_compute_bf16 contraction_scale_xdl_fp32_compute_bf16.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_bf16)
+
+add_example_executable(example_contraction_bilinear_xdl_fp32_compute_fp16 contraction_bilinear_xdl_fp32_compute_fp16.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_fp16)
+
+add_example_executable(example_contraction_scale_xdl_fp32_compute_fp16 contraction_scale_xdl_fp32_compute_fp16.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_fp16)
+
+# FP64
 add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64)
+
 add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64)
+
+add_example_executable(example_contraction_bilinear_xdl_fp64_compute_fp32 contraction_bilinear_xdl_fp64_compute_fp32.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64_compute_fp32)
+
+add_example_executable(example_contraction_scale_xdl_fp64_compute_fp32 contraction_scale_xdl_fp64_compute_fp32.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64_compute_fp32)
+
+# FP16
+add_example_executable(example_contraction_bilinear_xdl_fp16_compute_fp32 contraction_bilinear_xdl_fp16_compute_fp32.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp16_compute_fp32)
+
+add_example_executable(example_contraction_scale_xdl_fp16_compute_fp32 contraction_scale_xdl_fp16_compute_fp32.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp16_compute_fp32)
+
+# BF16
+add_example_executable(example_contraction_bilinear_xdl_bf16_compute_fp32 contraction_bilinear_xdl_bf16_compute_fp32.cpp)
+add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_bf16_compute_fp32)
+
+add_example_executable(example_contraction_scale_xdl_bf16_compute_fp32 contraction_scale_xdl_bf16_compute_fp32.cpp)
+add_dependencies(example_contraction_scale example_contraction_scale_xdl_bf16_compute_fp32)
+
+add_dependencies(example_contraction example_contraction_scale)
+add_dependencies(example_contraction example_contraction_bilinear)
diff --git a/example/26_contraction/common_instances.hpp b/example/26_contraction/common_instances.hpp
new file mode 100644
index 0000000000..eb2fa5ce73
--- /dev/null
+++ b/example/26_contraction/common_instances.hpp
@@ -0,0 +1,183 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
+using F64  = double;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// Generic instances for fp32, fp16 and bf16 data types.
+// clang-format off
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceKK_Generic = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
+
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceKN_Generic = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
+
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceMK_Generic = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
+
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceMN_Generic = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
+
+// Fp64 instances.
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceKK_FP64 = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceKN_FP64 = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceMK_FP64 = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+template <ck::index_t NumDimM,
+          ck::index_t NumDimN,
+          ck::index_t NumDimK,
+          typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementOp,
+          typename BElementOp,
+          typename CDEElementOp>
+using DeviceOpInstanceMN_FP64 = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+// clang-format on
diff --git a/example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp
new file mode 100644
index 0000000000..77ec5669a7
--- /dev/null
+++ b/example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = BF16;
+using BDataType        = BF16;
+using AccDataType      = F32;
+using CShuffleDataType = BF16;
+using DDataType        = BF16;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = BF16;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp
new file mode 100644
index 0000000000..25c4d9eac4
--- /dev/null
+++ b/example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F16;
+using DDataType        = F16;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F16;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
index 78522160c8..d440e7394e 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
@@ -1,29 +1,10 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/numeric.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F32 = float;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+#include "common_instances.hpp"
 
 using ADataType        = F32;
 using BDataType        = F32;
@@ -32,6 +13,7 @@ using CShuffleDataType = F32;
 using DDataType        = F32;
 using DsDataType       = ck::Tuple<DDataType>;
 using EDataType        = F32;
+using ComputeDataType  = F32;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -41,253 +23,64 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
 
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
 
-// clang-format off
-using DeviceOpInstanceKKNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
 
-using DeviceOpInstanceKNNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
 
-using DeviceOpInstanceMKNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
-
-using DeviceOpInstanceMNNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
-// clang-format on
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
+#include "run_contraction_bilinear_example.inc"
 
-    // A[M0, M1, K0, K1]
-    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
-    // D[M0, M1, N0, N1]
-    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
-
-    float alpha = 1.f;
-    float beta  = 1.f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 28)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        const ck::index_t M0 = std::stoi(argv[4]);
-        const ck::index_t M1 = std::stoi(argv[5]);
-
-        const ck::index_t N0 = std::stoi(argv[6]);
-        const ck::index_t N1 = std::stoi(argv[7]);
-
-        const ck::index_t K0 = std::stoi(argv[8]);
-        const ck::index_t K1 = std::stoi(argv[9]);
-
-        a_ms_ks_lengths = {M0, M1, K0, K1};
-        a_ms_ks_strides = {
-            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
-
-        b_ns_ks_lengths = {N0, N1, K0, K1};
-        b_ns_ks_strides = {
-            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
-
-        d_ms_ns_lengths = {M0, M1, N0, N1};
-        d_ms_ns_strides = {
-            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
-
-        e_ms_ns_lengths = {M0, M1, N0, N1};
-        e_ms_ns_strides = {
-            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
-
-        alpha = std::stof(argv[26]);
-        beta  = std::stof(argv[27]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 7: M0, M1, N0, N1, K0, K1\n");
-        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
-        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
-        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
-        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
-        printf("arg26 to 27: alpha, beta\n");
-        exit(0);
-    }
-
-    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-    d_device_buf.ToDevice(d_ms_ns.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{alpha, beta};
-
-    // device operation
-    auto op       = DeviceOpInstance{};
-    auto invoker  = op.MakeInvoker();
-    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
-                                    b_device_buf.GetDeviceBuffer(),
-                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
-                                    e_device_buf.GetDeviceBuffer(),
-                                    a_ms_ks_lengths,
-                                    a_ms_ks_strides,
-                                    b_ns_ks_lengths,
-                                    b_ns_ks_strides,
-                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
-                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
-                                    e_ms_ns_lengths,
-                                    e_ms_ns_strides,
-                                    a_element_op,
-                                    b_element_op,
-                                    cde_element_op);
-
-    if(!op.IsSupportedArgument(argument))
-    {
-        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    ck::index_t M =
-        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-    ck::index_t N = ck::accumulate_n<ck::index_t>(
-        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-    ck::index_t K = ck::accumulate_n<ck::index_t>(
-        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-    std::size_t flop      = std::size_t(2) * M * N * K;
-    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
-                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
-              << op.GetTypeString() << std::endl;
-
-    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      AElementOp,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1),
-                                       d_ms_ns(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp
new file mode 100644
index 0000000000..5b748949bd
--- /dev/null
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F32;
+using BDataType        = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F32;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F32;
+using ComputeDataType  = BF16;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp
new file mode 100644
index 0000000000..334cd615d1
--- /dev/null
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F32;
+using BDataType        = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F32;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F32;
+using ComputeDataType  = F16;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                        NumDimN,
+                                                        NumDimK,
+                                                        ADataType,
+                                                        BDataType,
+                                                        AccDataType,
+                                                        CShuffleDataType,
+                                                        DsDataType,
+                                                        EDataType,
+                                                        ComputeDataType,
+                                                        AElementOp,
+                                                        BElementOp,
+                                                        CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
index 6cceed5bc1..86ea7e0212 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
@@ -1,29 +1,10 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/numeric.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F64 = double;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+#include "common_instances.hpp"
 
 using ADataType        = F64;
 using BDataType        = F64;
@@ -32,6 +13,7 @@ using CShuffleDataType = F64;
 using DDataType        = F64;
 using DsDataType       = ck::Tuple<DDataType>;
 using EDataType        = F64;
+using ComputeDataType  = F64;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -41,253 +23,64 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
 
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
 
-// clang-format off
-using DeviceOpInstanceKKNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
 
-using DeviceOpInstanceKNNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
 
-using DeviceOpInstanceMKNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
-
-using DeviceOpInstanceMNNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
-// clang-format on
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
+#include "run_contraction_bilinear_example.inc"
 
-    // A[M0, M1, K0, K1]
-    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
-    // D[M0, M1, N0, N1]
-    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
-
-    float alpha = 1.f;
-    float beta  = 1.f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 28)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        const ck::index_t M0 = std::stoi(argv[4]);
-        const ck::index_t M1 = std::stoi(argv[5]);
-
-        const ck::index_t N0 = std::stoi(argv[6]);
-        const ck::index_t N1 = std::stoi(argv[7]);
-
-        const ck::index_t K0 = std::stoi(argv[8]);
-        const ck::index_t K1 = std::stoi(argv[9]);
-
-        a_ms_ks_lengths = {M0, M1, K0, K1};
-        a_ms_ks_strides = {
-            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
-
-        b_ns_ks_lengths = {N0, N1, K0, K1};
-        b_ns_ks_strides = {
-            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
-
-        d_ms_ns_lengths = {M0, M1, N0, N1};
-        d_ms_ns_strides = {
-            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
-
-        e_ms_ns_lengths = {M0, M1, N0, N1};
-        e_ms_ns_strides = {
-            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
-
-        alpha = std::stof(argv[26]);
-        beta  = std::stof(argv[27]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 7: M0, M1, N0, N1, K0, K1\n");
-        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
-        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
-        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
-        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
-        printf("arg26 to 27: alpha, beta\n");
-        exit(0);
-    }
-
-    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-    d_device_buf.ToDevice(d_ms_ns.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{alpha, beta};
-
-    // device operation
-    auto op       = DeviceOpInstance{};
-    auto invoker  = op.MakeInvoker();
-    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
-                                    b_device_buf.GetDeviceBuffer(),
-                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
-                                    e_device_buf.GetDeviceBuffer(),
-                                    a_ms_ks_lengths,
-                                    a_ms_ks_strides,
-                                    b_ns_ks_lengths,
-                                    b_ns_ks_strides,
-                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
-                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
-                                    e_ms_ns_lengths,
-                                    e_ms_ns_strides,
-                                    a_element_op,
-                                    b_element_op,
-                                    cde_element_op);
-
-    if(!op.IsSupportedArgument(argument))
-    {
-        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    ck::index_t M =
-        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-    ck::index_t N = ck::accumulate_n<ck::index_t>(
-        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-    ck::index_t K = ck::accumulate_n<ck::index_t>(
-        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-    std::size_t flop      = std::size_t(2) * M * N * K;
-    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
-                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
-              << op.GetTypeString() << std::endl;
-
-    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      AElementOp,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1),
-                                       d_ms_ns(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp
new file mode 100644
index 0000000000..b0a2a9a10f
--- /dev/null
+++ b/example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F32;
+using CShuffleDataType = F64;
+using DDataType        = F64;
+using DsDataType       = ck::Tuple<DDataType>;
+using EDataType        = F64;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
+
+using DeviceOpInstanceKKNN = DeviceOpInstanceKK_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstanceKNNN = DeviceOpInstanceKN_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstanceMKNN = DeviceOpInstanceMK_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstanceMNNN = DeviceOpInstanceMN_FP64<NumDimM,
+                                                     NumDimN,
+                                                     NumDimK,
+                                                     ADataType,
+                                                     BDataType,
+                                                     AccDataType,
+                                                     CShuffleDataType,
+                                                     DsDataType,
+                                                     EDataType,
+                                                     ComputeDataType,
+                                                     AElementOp,
+                                                     BElementOp,
+                                                     CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKNN;
+
+#include "run_contraction_bilinear_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp b/example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp
new file mode 100644
index 0000000000..32652fc6f3
--- /dev/null
+++ b/example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = BF16;
+using BDataType        = BF16;
+using AccDataType      = F32;
+using CShuffleDataType = BF16;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = BF16;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Scale;
+
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKN;
+
+#include "run_contraction_scale_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp
new file mode 100644
index 0000000000..a4797e2400
--- /dev/null
+++ b/example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F16;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F16;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Scale;
+
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKN;
+
+#include "run_contraction_scale_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp32.cpp
index 1574f5d18f..fe984b29d6 100644
--- a/example/26_contraction/contraction_scale_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp32.cpp
@@ -1,29 +1,10 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/numeric.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F32 = float;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+#include "common_instances.hpp"
 
 using ADataType        = F32;
 using BDataType        = F32;
@@ -31,6 +12,7 @@ using AccDataType      = F32;
 using CShuffleDataType = F32;
 using DsDataType       = ck::Tuple<>;
 using EDataType        = F32;
+using ComputeDataType  = F32;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -40,237 +22,64 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Scale;
 
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
 
-// clang-format off
-using DeviceOpInstanceKKN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
 
-using DeviceOpInstanceKNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
 
-using DeviceOpInstanceMKN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
-
-using DeviceOpInstanceMNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
-// clang-format on
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
+#include "run_contraction_scale_example.inc"
 
-    // A[M0, M1, K0, K1]
-    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
-
-    float scale = 1.f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 23)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        const ck::index_t M0 = std::stoi(argv[4]);
-        const ck::index_t M1 = std::stoi(argv[5]);
-
-        const ck::index_t N0 = std::stoi(argv[6]);
-        const ck::index_t N1 = std::stoi(argv[7]);
-
-        const ck::index_t K0 = std::stoi(argv[8]);
-        const ck::index_t K1 = std::stoi(argv[9]);
-
-        a_ms_ks_lengths = {M0, M1, K0, K1};
-        a_ms_ks_strides = {
-            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
-
-        b_ns_ks_lengths = {N0, N1, K0, K1};
-        b_ns_ks_strides = {
-            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
-
-        e_ms_ns_lengths = {M0, M1, N0, N1};
-        e_ms_ns_strides = {
-            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
-
-        scale = std::stof(argv[22]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
-        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
-        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
-        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
-        printf("arg22: scale\n");
-        exit(0);
-    }
-
-    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{scale};
-
-    // device operation
-    auto op       = DeviceOpInstance{};
-    auto invoker  = op.MakeInvoker();
-    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
-                                    b_device_buf.GetDeviceBuffer(),
-                                    std::array<const void*, 0>{},
-                                    e_device_buf.GetDeviceBuffer(),
-                                    a_ms_ks_lengths,
-                                    a_ms_ks_strides,
-                                    b_ns_ks_lengths,
-                                    b_ns_ks_strides,
-                                    std::array<std::vector<ck::index_t>, 0>{},
-                                    std::array<std::vector<ck::index_t>, 0>{},
-                                    e_ms_ns_lengths,
-                                    e_ms_ns_strides,
-                                    a_element_op,
-                                    b_element_op,
-                                    cde_element_op);
-
-    if(!op.IsSupportedArgument(argument))
-    {
-        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    ck::index_t M =
-        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-    ck::index_t N = ck::accumulate_n<ck::index_t>(
-        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-    ck::index_t K = ck::accumulate_n<ck::index_t>(
-        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-    std::size_t flop = std::size_t(2) * M * N * K;
-    std::size_t num_btype =
-        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
-              << op.GetTypeString() << std::endl;
-
-    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      AElementOp,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp b/example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp
new file mode 100644
index 0000000000..83c8fec179
--- /dev/null
+++ b/example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F32;
+using BDataType        = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F32;
+using ComputeDataType  = BF16;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Scale;
+
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKN;
+
+#include "run_contraction_scale_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp b/example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp
new file mode 100644
index 0000000000..27e42e8203
--- /dev/null
+++ b/example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F32;
+using BDataType        = F32;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F32;
+using ComputeDataType  = F16;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Scale;
+
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
+                                                       NumDimN,
+                                                       NumDimK,
+                                                       ADataType,
+                                                       BDataType,
+                                                       AccDataType,
+                                                       CShuffleDataType,
+                                                       DsDataType,
+                                                       EDataType,
+                                                       ComputeDataType,
+                                                       AElementOp,
+                                                       BElementOp,
+                                                       CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKN;
+
+#include "run_contraction_scale_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp64.cpp b/example/26_contraction/contraction_scale_xdl_fp64.cpp
index 3dacc70887..37374780b6 100644
--- a/example/26_contraction/contraction_scale_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp64.cpp
@@ -1,29 +1,10 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/numeric.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F64 = double;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+#include "common_instances.hpp"
 
 using ADataType        = F64;
 using BDataType        = F64;
@@ -31,6 +12,7 @@ using AccDataType      = F64;
 using CShuffleDataType = F64;
 using DsDataType       = ck::Tuple<>;
 using EDataType        = F64;
+using ComputeDataType  = F64;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -40,237 +22,64 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Scale;
 
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
 
-// clang-format off
-using DeviceOpInstanceKKN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
 
-using DeviceOpInstanceKNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
 
-using DeviceOpInstanceMKN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
-
-using DeviceOpInstanceMNN = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
-// clang-format on
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
+#include "run_contraction_scale_example.inc"
 
-    // A[M0, M1, K0, K1]
-    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
-
-    float scale = 1.f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 23)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        const ck::index_t M0 = std::stoi(argv[4]);
-        const ck::index_t M1 = std::stoi(argv[5]);
-
-        const ck::index_t N0 = std::stoi(argv[6]);
-        const ck::index_t N1 = std::stoi(argv[7]);
-
-        const ck::index_t K0 = std::stoi(argv[8]);
-        const ck::index_t K1 = std::stoi(argv[9]);
-
-        a_ms_ks_lengths = {M0, M1, K0, K1};
-        a_ms_ks_strides = {
-            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
-
-        b_ns_ks_lengths = {N0, N1, K0, K1};
-        b_ns_ks_strides = {
-            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
-
-        e_ms_ns_lengths = {M0, M1, N0, N1};
-        e_ms_ns_strides = {
-            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
-
-        scale = std::stof(argv[22]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
-        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
-        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
-        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
-        printf("arg22: scale\n");
-        exit(0);
-    }
-
-    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{scale};
-
-    // device operation
-    auto op       = DeviceOpInstance{};
-    auto invoker  = op.MakeInvoker();
-    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
-                                    b_device_buf.GetDeviceBuffer(),
-                                    std::array<const void*, 0>{},
-                                    e_device_buf.GetDeviceBuffer(),
-                                    a_ms_ks_lengths,
-                                    a_ms_ks_strides,
-                                    b_ns_ks_lengths,
-                                    b_ns_ks_strides,
-                                    std::array<std::vector<ck::index_t>, 0>{},
-                                    std::array<std::vector<ck::index_t>, 0>{},
-                                    e_ms_ns_lengths,
-                                    e_ms_ns_strides,
-                                    a_element_op,
-                                    b_element_op,
-                                    cde_element_op);
-
-    if(!op.IsSupportedArgument(argument))
-    {
-        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    ck::index_t M =
-        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-    ck::index_t N = ck::accumulate_n<ck::index_t>(
-        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-    ck::index_t K = ck::accumulate_n<ck::index_t>(
-        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-    std::size_t flop = std::size_t(2) * M * N * K;
-    std::size_t num_btype =
-        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
-              << op.GetTypeString() << std::endl;
-
-    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      AElementOp,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp
new file mode 100644
index 0000000000..5dbeb23154
--- /dev/null
+++ b/example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "common_instances.hpp"
+
+using ADataType        = F64;
+using BDataType        = F64;
+using AccDataType      = F32;
+using CShuffleDataType = F64;
+using DsDataType       = ck::Tuple<>;
+using EDataType        = F64;
+using ComputeDataType  = F32;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
+using CDEElementOp = ck::tensor_operation::element_wise::Scale;
+
+using DeviceOpInstanceKKN = DeviceOpInstanceKK_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
+
+using DeviceOpInstanceKNN = DeviceOpInstanceKN_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
+
+using DeviceOpInstanceMKN = DeviceOpInstanceMK_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
+
+using DeviceOpInstanceMNN = DeviceOpInstanceMN_FP64<NumDimM,
+                                                    NumDimN,
+                                                    NumDimK,
+                                                    ADataType,
+                                                    BDataType,
+                                                    AccDataType,
+                                                    CShuffleDataType,
+                                                    DsDataType,
+                                                    EDataType,
+                                                    ComputeDataType,
+                                                    AElementOp,
+                                                    BElementOp,
+                                                    CDEElementOp>;
+
+using DeviceOpInstance = DeviceOpInstanceKKN;
+
+#include "run_contraction_scale_example.inc"
+
+int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/run_contraction_bilinear_example.inc b/example/26_contraction/run_contraction_bilinear_example.inc
new file mode 100644
index 0000000000..7b9a01e0ae
--- /dev/null
+++ b/example/26_contraction/run_contraction_bilinear_example.inc
@@ -0,0 +1,234 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <iostream>
+#include <string>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+int run_contraction_bilinear_example(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float alpha = 1.f;
+    float beta  = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 28)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        d_ms_ns_lengths = {M0, M1, N0, N1};
+        d_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
+
+        alpha = std::stof(argv[26]);
+        beta  = std::stof(argv[27]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
+        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg26 to 27: alpha, beta\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop      = std::size_t(2) * M * N * K;
+    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      ComputeDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1),
+                                       d_ms_ns(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/26_contraction/run_contraction_scale_example.inc b/example/26_contraction/run_contraction_scale_example.inc
new file mode 100644
index 0000000000..65ca182da2
--- /dev/null
+++ b/example/26_contraction/run_contraction_scale_example.inc
@@ -0,0 +1,217 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <iostream>
+#include <string>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+int run_contraction_scale_example(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float scale = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 23)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        scale = std::stof(argv[22]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg22: scale\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{scale};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 0>{},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      ComputeDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
index 118ade8978..dffb51d0ba 100644
--- a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
@@ -33,7 +33,8 @@ template <index_t NumDimM,
           typename EDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CDEElementwiseOperation>
+          typename CDEElementwiseOperation,
+          typename ComputeDataType = ADataType>
 struct DeviceContractionMultipleD : public BaseOperator
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index 5ae06836fa..db119178b7 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -112,6 +112,7 @@ template <index_t NumDimM,
           typename CShuffleDataType,
           typename DsDataType,
           typename EDataType,
+          typename ComputeDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CDEElementwiseOperation,
@@ -156,7 +157,8 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
                                         EDataType,
                                         AElementwiseOperation,
                                         BElementwiseOperation,
-                                        CDEElementwiseOperation>
+                                        CDEElementwiseOperation,
+                                        ComputeDataType>
 {
     using DeviceOp = DeviceContractionMultipleD_Xdl_CShuffle;
 
@@ -310,8 +312,6 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({{}}, {{}}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N({}, {}));
 
-    using ComputeDataType = ADataType;
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
diff --git a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
index 9fe0931cba..1b9e37b74a 100644
--- a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
@@ -186,6 +186,25 @@ struct Bilinear
         y = type_convert<half_t>(alpha_ * x0 + beta_ * ck::type_convert<float>(x1));
     };
 
+    template <>
+    __host__ __device__ constexpr void
+    operator()<bhalf_t, bhalf_t, bhalf_t>(bhalf_t& y, const bhalf_t& x0, const bhalf_t& x1) const
+    {
+        const float x0_tmp = type_convert<float>(x0);
+        const float x1_tmp = type_convert<float>(x1);
+        const float y_tmp  = alpha_ * x0_tmp + beta_ * x1_tmp;
+        y                  = type_convert<bhalf_t>(y_tmp);
+    };
+
+    template <>
+    __host__ __device__ constexpr void
+    operator()<bhalf_t, float, bhalf_t>(bhalf_t& y, const float& x0, const bhalf_t& x1) const
+    {
+        const float x1_tmp = ck::type_convert<float>(x1);
+        const float y_tmp  = alpha_ * x0 + beta_ * x1_tmp;
+        y                  = y_tmp;
+    };
+
     template <>
     __host__ __device__ constexpr void operator()<std::int8_t, std::int32_t, std::int8_t>(
         std::int8_t& y, const std::int32_t& x0, const std::int8_t& x1) const
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 59cd54a122..b181b66009 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -33,6 +33,12 @@ struct PassThrough
         y = type_convert<float>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<double, float>(double& y, const float& x) const
+    {
+        y = type_convert<double>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<float, float>(float& y, const float& x) const
     {
@@ -69,6 +75,12 @@ struct PassThrough
         y = type_convert<bhalf_t>(x);
     }
 
+    template <>
+    __host__ __device__ void operator()<float, bhalf_t>(float& y, const bhalf_t& x) const
+    {
+        y = type_convert<float>(x);
+    }
+
     template <>
     __host__ __device__ void operator()<bhalf_t, half_t>(bhalf_t& y, const half_t& x) const
     {
@@ -198,6 +210,20 @@ struct Scale
     template <typename Y, typename X>
     __host__ __device__ void operator()(Y& y, const X& x) const;
 
+    template <>
+    __host__ __device__ void operator()<half_t, half_t>(half_t& y, const half_t& x) const
+    {
+        y = ck::type_convert<half_t>(scale_) * x;
+    };
+
+    template <>
+    __host__ __device__ void operator()<bhalf_t, bhalf_t>(bhalf_t& y, const bhalf_t& x) const
+    {
+        const float x_tmp = ck::type_convert<float>(x);
+        const float y_tmp = scale_ * x_tmp;
+        y                 = ck::type_convert<bhalf_t>(y_tmp);
+    };
+
     template <>
     __host__ __device__ void operator()<float, float>(float& y, const float& x) const
     {
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
index 92a8c82a6e..527dac6d3b 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
@@ -23,6 +23,7 @@ template <ck::index_t NumDimM,
           typename BDataType,
           typename CDataType,
           typename AccDataType,
+          typename ComputeDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
@@ -69,19 +70,24 @@ struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::Base
                 {
                     for(ck::index_t k1 = 0; k1 < K1; ++k1)
                     {
+                        // Simulate the possible casting when ComputeDataType is different than the
+                        // A/B data types
+                        ComputeDataType v_a_compute_input =
+                            ck::type_convert<ComputeDataType>(arg.a_ms_ks_(m0, m1, k0, k1));
+                        ComputeDataType v_b_compute_input =
+                            ck::type_convert<ComputeDataType>(arg.b_ns_ks_(n0, n1, k0, k1));
+
                         AccDataType v_a;
                         AccDataType v_b;
 
-                        arg.a_element_op_(
-                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
-                        arg.b_element_op_(
-                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
+                        arg.a_element_op_(v_a, ck::type_convert<AccDataType>(v_a_compute_input));
+                        arg.b_element_op_(v_b, ck::type_convert<AccDataType>(v_b_compute_input));
 
                         v_acc += v_a * v_b;
                     }
                 }
 
-                arg.c_ms_ns_(m0, m1, n0, n1) = v_acc;
+                arg.c_ms_ns_(m0, m1, n0, n1) = ck::type_convert<CDataType>(v_acc);
             };
 
             make_ParallelTensorFunctor(f_ms_ns,
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index ea11fd2e1a..f6fb3f9eee 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -29,6 +29,8 @@ using BF8 = ck::bf8_t;
 
 using Empty_Tuple = ck::Tuple<>;
 
+using BF16_Tuple = ck::Tuple<BF16>;
+
 using F16_Tuple     = ck::Tuple<F16>;
 using F16_F16_Tuple = ck::Tuple<F16, F16>;
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp
new file mode 100644
index 0000000000..96f4d4e30e
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp
@@ -0,0 +1,292 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+#pragma once
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
+using F64  = double;
+
+using F16_Tuple   = ck::Tuple<F16>;
+using BF16_Tuple  = ck::Tuple<BF16>;
+using F32_Tuple   = ck::Tuple<F32>;
+using F64_Tuple   = ck::Tuple<F64>;
+using Empty_Tuple = ck::Tuple<>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_kk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_kn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,      
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,      
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,      
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_mk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_mn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_kk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_kn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_mk_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+template <typename ADataType,
+          typename BDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename ComputeDataType,
+          typename AElementwiseOp,
+          typename BElementwiseOp,
+          typename CDEElementwiseOp>
+using device_contraction_f64_mn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index bd3af891ec..d06cab1196 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -17,7 +17,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #ifdef CK_ENABLE_FP32
-// float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -28,7 +27,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F32>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -40,7 +40,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F32>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -52,7 +53,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F32>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -64,10 +66,115 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
-#endif
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances);
+#endif // CK_ENABLE_FP32
+
 #ifdef CK_ENABLE_FP64
-// double
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -78,7 +185,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F64>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -90,7 +198,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F64>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -102,7 +211,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
+                                                           Bilinear,
+                                                           F64>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -114,8 +224,170 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances);
-#endif
+                                                           Bilinear,
+                                                           F64>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP64
+
+#ifdef CK_ENABLE_FP16
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
+#ifdef CK_ENABLE_BF16
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
 // Contraction + Bilinear
 template <index_t NumDimM,
           index_t NumDimN,
@@ -123,7 +395,8 @@ template <index_t NumDimM,
           typename ADataType,
           typename BDataType,
           typename DDataType,
-          typename EDataType>
+          typename EDataType,
+          typename ComputeDataType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContractionMultipleD<
     NumDimM,
     NumDimN,
@@ -134,7 +407,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     EDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::Bilinear>>
+    ck::tensor_operation::element_wise::Bilinear,
+    ComputeDataType>>
 {
     using DeviceOp = DeviceContractionMultipleD<NumDimM,
                                                 NumDimN,
@@ -145,45 +419,125 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                                                 EDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::Bilinear>;
+                                                ck::tensor_operation::element_wise::Bilinear,
+                                                ComputeDataType>;
 
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 #ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
-                     is_same_v<DDataType, float> && is_same_v<EDataType, float>)
+                     is_same_v<EDataType, float>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::half_t>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::bhalf_t>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_FP32
 #ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
-                     is_same_v<DDataType, double> && is_same_v<EDataType, double>)
+                     is_same_v<EDataType, double>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
-                    op_ptrs);
-                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, double>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_FP64
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
+                     is_same_v<EDataType, ck::half_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_FP16
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
+                     is_same_v<EDataType, ck::bhalf_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
+                        op_ptrs);
+                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_BF16
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index 4dd735d062..8e994d61ae 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -17,7 +17,6 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #ifdef CK_ENABLE_FP32
-// float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -28,7 +27,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F32>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -40,7 +40,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F32>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -52,7 +53,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F32>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -64,10 +66,115 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
-#endif
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances);
+#endif // CK_ENABLE_FP32
+
 #ifdef CK_ENABLE_FP64
-// double
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -78,7 +185,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F64>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -90,7 +198,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F64>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -102,7 +211,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
+                                                           Scale,
+                                                           F64>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -114,15 +224,178 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances);
-#endif
+                                                           Scale,
+                                                           F64>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP64
+
+#ifdef CK_ENABLE_FP16
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
+#ifdef CK_ENABLE_BF16
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances);
+#endif // CK_ENABLE_FP16
+
 // Contraction + Scale
 template <index_t NumDimM,
           index_t NumDimN,
           index_t NumDimK,
           typename ADataType,
           typename BDataType,
-          typename EDataType>
+          typename EDataType,
+          typename ComputeDataType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContractionMultipleD<
     NumDimM,
     NumDimN,
@@ -133,7 +406,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     EDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::Scale>>
+    ck::tensor_operation::element_wise::Scale,
+    ComputeDataType>>
 {
     using DeviceOp = DeviceContractionMultipleD<NumDimM,
                                                 NumDimN,
@@ -144,7 +418,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                                                 EDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::Scale>;
+                                                ck::tensor_operation::element_wise::Scale,
+                                                ComputeDataType>;
 
     static auto GetInstances()
     {
@@ -155,34 +430,113 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::half_t>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, ck::bhalf_t>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_FP32
 #ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<EDataType, double>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
-                    op_ptrs);
-                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
-                    op_ptrs);
+                if constexpr(is_same_v<ComputeDataType, double>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
+                        op_ptrs);
+                }
+                else if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
+                        op_ptrs);
+                }
             }
         }
-#endif
+#endif // CK_ENABLE_FP64
+#ifdef CK_ENABLE_FP16
+        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
+                     is_same_v<EDataType, ck::half_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_FP16
+#ifdef CK_ENABLE_BF16
+        if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
+                     is_same_v<EDataType, ck::bhalf_t>)
+        {
+            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
+            {
+                if constexpr(is_same_v<ComputeDataType, float>)
+                {
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
+                        op_ptrs);
+                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
+                        op_ptrs);
+                }
+            }
+        }
+#endif // CK_ENABLE_BF16
         return op_ptrs;
     }
 };
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
index ee9c419e1a..87a6bbba4e 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
@@ -1,14 +1,42 @@
 set(DEVICE_CONTRACTION_BILINEAR_INSTANCES)
-#float
-list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
-                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
-                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
-                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
 
-#double
+# FP32
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
+
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp)
+
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp)
+
+# FP64
 list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
-                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
-                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
-                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
+
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp)
+
+# FP16
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp)
+
+# BF16
+list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
+                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp)
 
 add_instance_library(device_contraction_bilinear_instance ${DEVICE_CONTRACTION_BILINEAR_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
new file mode 100644
index 0000000000..56dc1d2c71
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance =
+    device_contraction_kk_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   BF16_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
new file mode 100644
index 0000000000..7b8163814e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance =
+    device_contraction_kn_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   BF16_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
new file mode 100644
index 0000000000..ba202202f9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance =
+    device_contraction_mk_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   BF16_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp
new file mode 100644
index 0000000000..b6acab1c18
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance =
+    device_contraction_mn_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   BF16_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           BF16_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
new file mode 100644
index 0000000000..1bd3495b36
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance =
+    device_contraction_kk_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   F16_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
new file mode 100644
index 0000000000..6817546d3d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance =
+    device_contraction_kn_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   F16_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
new file mode 100644
index 0000000000..cedace2cb2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance =
+    device_contraction_mk_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   F16_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp
new file mode 100644
index 0000000000..b8383e5268
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance =
+    device_contraction_mn_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   F16_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           F16_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
new file mode 100644
index 0000000000..2e73bdb804
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance =
+    device_contraction_kk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
new file mode 100644
index 0000000000..ae69db03ac
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance =
+    device_contraction_kn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
new file mode 100644
index 0000000000..fc6b6003f7
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance =
+    device_contraction_mk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp
new file mode 100644
index 0000000000..b81984acc4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance =
+    device_contraction_mn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
new file mode 100644
index 0000000000..8da6a79f0a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance =
+    device_contraction_kk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
new file mode 100644
index 0000000000..b05c443e85
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance =
+    device_contraction_kn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
new file mode 100644
index 0000000000..c2cbb9f25f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance =
+    device_contraction_mk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp
new file mode 100644
index 0000000000..887eaad7eb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance =
+    device_contraction_mn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           F32_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
index 5587db77e0..9a3c71acb5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,40 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32       = float;
-using F32_Tuple = ck::Tuple<F32>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance =
+    device_contraction_kk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
index 26262855ea..5659d477d4 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,43 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32       = float;
-using F32_Tuple = ck::Tuple<F32>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance =
+    device_contraction_kn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -69,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
index befc0dcd10..7cbe1e34be 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,43 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32       = float;
-using F32_Tuple = ck::Tuple<F32>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance =
+    device_contraction_mk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -69,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
index e45b47cf94..e26dcf394b 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,43 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32       = float;
-using F32_Tuple = ck::Tuple<F32>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance =
+    device_contraction_mn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   F32_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -69,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
new file mode 100644
index 0000000000..74c5c9c653
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance =
+    device_contraction_f64_kk_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
new file mode 100644
index 0000000000..8a2705f722
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance =
+    device_contraction_f64_kn_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
new file mode 100644
index 0000000000..47924a843d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance =
+    device_contraction_f64_mk_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp
new file mode 100644
index 0000000000..d6c684e2b0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance =
+    device_contraction_f64_mn_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
+
+void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           F64_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Bilinear,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
index f437a227d5..75bd8e3f4a 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64       = double;
-using F64_Tuple = ck::Tuple<F64>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance =
+    device_contraction_f64_kk_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
index 13fdbeb35c..c515f37916 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64       = double;
-using F64_Tuple = ck::Tuple<F64>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance =
+    device_contraction_f64_kn_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
index 95ef8c4929..ff9fa61e0b 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64       = double;
-using F64_Tuple = ck::Tuple<F64>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance =
+    device_contraction_f64_mk_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
index 290f81d7c9..f274c45b58 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64       = double;
-using F64_Tuple = ck::Tuple<F64>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance =
+    device_contraction_f64_mn_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       F64_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Bilinear>;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear>>>& instances)
+                                                           Bilinear,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
index 673fae3373..a0918d9d3f 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
@@ -1,15 +1,43 @@
 set(DEVICE_CONTRACTION_SCALE_INSTANCES)
-#float
+
+# FP32
 list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp)
 
-#double
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp)
+
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp)
+
+# FP64
 list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp)
 
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp)
+
+# FP16
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp)
+
+# BF16
+list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
+                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp)
+
 add_instance_library(device_contraction_scale_instance ${DEVICE_CONTRACTION_SCALE_INSTANCES})
 
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
new file mode 100644
index 0000000000..ace4d4a330
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance =
+    device_contraction_kk_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   Empty_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
new file mode 100644
index 0000000000..d2a1d45f7a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance =
+    device_contraction_kn_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   Empty_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
new file mode 100644
index 0000000000..aa4e4df9d9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance =
+    device_contraction_mk_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   Empty_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp
new file mode 100644
index 0000000000..a59bce0a1f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance =
+    device_contraction_mn_instance<BF16,
+                                   BF16,
+                                   F32,
+                                   BF16,
+                                   Empty_Tuple,
+                                   BF16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           BF16,
+                                                           BF16,
+                                                           Empty_Tuple,
+                                                           BF16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
new file mode 100644
index 0000000000..876acd3f69
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance =
+    device_contraction_kk_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   Empty_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
new file mode 100644
index 0000000000..96e961d3bb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance =
+    device_contraction_kn_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   Empty_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
new file mode 100644
index 0000000000..69cae90889
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance =
+    device_contraction_mk_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   Empty_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp
new file mode 100644
index 0000000000..e6ac98273a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance =
+    device_contraction_mn_instance<F16,
+                                   F16,
+                                   F32,
+                                   F16,
+                                   Empty_Tuple,
+                                   F16,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F16,
+                                                           F16,
+                                                           Empty_Tuple,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
new file mode 100644
index 0000000000..da1df4c43e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance =
+    device_contraction_kk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
new file mode 100644
index 0000000000..21c4e87fb1
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance =
+    device_contraction_kn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
new file mode 100644
index 0000000000..9d22f8feff
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance =
+    device_contraction_mk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp
new file mode 100644
index 0000000000..8e861f74c9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance =
+    device_contraction_mn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   BF16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           BF16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
new file mode 100644
index 0000000000..390b5526e3
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance =
+    device_contraction_kk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
new file mode 100644
index 0000000000..385a13cc25
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance =
+    device_contraction_kn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
new file mode 100644
index 0000000000..1afd432bac
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance =
+    device_contraction_mk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp
new file mode 100644
index 0000000000..bfb4d4630f
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance =
+    device_contraction_mn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F16,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F32,
+                                                           F32,
+                                                           Empty_Tuple,
+                                                           F32,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F16>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
index 16fd1cb407..51bc64eaad 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,40 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32         = float;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
-// k/k/n are the fast changing dimension for A/B/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>
-    // clang-format on
-    >;
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance =
+    device_contraction_kk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
index ff37bf7cce..4cd28d52dd 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,43 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32         = float;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
-// k/n/n are the fast changing dimension for A/B/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance =
+    device_contraction_kn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -69,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
index 8a1f6f9334..82bba62715 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,43 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32         = float;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
-// m/k/n are the fast changing dimension for A/B/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance =
+    device_contraction_mk_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -69,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
index d333f59726..06bf4f34b5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,43 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F32         = float;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
-// m/n/n are the fast changing dimension for A/B/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance =
+    device_contraction_mn_instance<F32,
+                                   F32,
+                                   F32,
+                                   F32,
+                                   Empty_Tuple,
+                                   F32,
+                                   F32,
+                                   PassThrough,
+                                   PassThrough,
+                                   Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -69,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F32>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
new file mode 100644
index 0000000000..c047153bd9
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance =
+    device_contraction_f64_kk_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
new file mode 100644
index 0000000000..ee33c56223
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// k/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance =
+    device_contraction_f64_kn_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
new file mode 100644
index 0000000000..b62d4bb2aa
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/k/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance =
+    device_contraction_f64_mk_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp
new file mode 100644
index 0000000000..346f91beb0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
+// setting Don't use this hack unless absolutely necessary!
+// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
+#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
+// m/n/n/n are the fast changing dimension for A/B/D/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance =
+    device_contraction_f64_mn_instance<F64,
+                                       F64,
+                                       F32,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F32,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
+
+void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
+    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
+                                                           2,
+                                                           2,
+                                                           F64,
+                                                           F64,
+                                                           Empty_Tuple,
+                                                           F64,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           Scale,
+                                                           F32>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
index 4c87b51a9e..30cc3b9652 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64         = double;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
-    // clang-format on
-    >;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance =
+    device_contraction_f64_kk_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
index fd3f57c6b6..0a36ef7ffd 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64         = double;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance =
+    device_contraction_f64_kn_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
index 1e53f0b2f5..617332c2c7 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64         = double;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance =
+    device_contraction_f64_mk_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
index d02d146a9b..1df7f31201 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
@@ -9,11 +9,9 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
+#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -21,37 +19,19 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-using F64         = double;
-using Empty_Tuple = ck::Tuple<>;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance =
+    device_contraction_f64_mn_instance<F64,
+                                       F64,
+                                       F64,
+                                       F64,
+                                       Empty_Tuple,
+                                       F64,
+                                       F64,
+                                       PassThrough,
+                                       PassThrough,
+                                       Scale>;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -63,7 +43,8 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale>>>& instances)
+                                                           Scale,
+                                                           F64>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance{});
diff --git a/profiler/README.md b/profiler/README.md
index d081fc33e9..b7b5be420b 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -50,21 +50,23 @@ Best Perf: 1.42509 ms, 102.988 TFlops, 234.086 GB/s
 ## Profile contraction kernels
 ```bash
 #arg1: tensor operation (contraction_bilinear=CONTRACTION+Bilinear)
-#arg2: data type (0: fp32; 1: f64)\n"
-#arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#arg2: data type (0: fp32; 1: f64; 2: f16; 3: bf16)
+#arg3: compute data type (0: fp32; 1: f64; 2: f16; 3: bf16)
+#arg4: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
 #                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
 #                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
 #                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1])
-#arg4: verification (0: no; 1: yes)
-#arg5: initialization (0: no init; 1: integer value; 2: decimal value)
-#arg6: print tensor value (0: no; 1: yes)
-#arg7: time kernel (0: no, 1: yes)
-#arg8 and arg9: alpha and beta
-#arg10 to 15: M0, M1, N0, N1, K0, K1
-#arg16 to 31: Strides for A, B, D and E (skip for default)
+#arg5: verification (0: no; 1: yes)
+#arg6: initialization (0: no init; 1: integer value; 2: decimal value)
+#arg7: print tensor value (0: no; 1: yes)
+#arg8: time kernel (0: no, 1: yes)
+#arg9: alpha
+#arg10: beta
+#arg11 to 16: M0, M1, N0, N1, K0, K1
+#arg17 to 32: Strides for A, B, D and E (skip for default)
 
-################                   op  datatype  layout  verify  init  log  time  alpha  beta  M0  M1  N0  N1  K0  K1
-./bin/ckProfiler contraction_bilinear         0       1       0     0    0     1    1.0   1.0 128 128 128 128 128 128
+################                   op  datatype  compute_datatype  layout  verify  init  log  time  alpha  beta  M0  M1  N0  N1  K0  K1
+./bin/ckProfiler contraction_bilinear         0                 0       1       0     0    0     1    1.0   1.0 128 128 128 128 128 128
 ```
 
 Result (MI100)
diff --git a/profiler/include/profiler/profile_contraction_impl.hpp b/profiler/include/profiler/profile_contraction_impl.hpp
index 660cc3f9e5..f6e4b3f395 100644
--- a/profiler/include/profiler/profile_contraction_impl.hpp
+++ b/profiler/include/profiler/profile_contraction_impl.hpp
@@ -31,10 +31,14 @@ namespace profiler {
 using Bilinear = ck::tensor_operation::element_wise::Bilinear;
 using Scale    = ck::tensor_operation::element_wise::Scale;
 
+using F32 = float;
+using F64 = double;
+
 template <typename ALayout,
           typename BLayout,
           typename CDELayout,
           typename DataType,
+          typename ComputeDataType,
           typename DTupleDataType,
           typename CDElementOp>
 int profile_contraction_impl(ck::index_t do_verification,
@@ -45,10 +49,10 @@ int profile_contraction_impl(ck::index_t do_verification,
                              const std::vector<ck::index_t>& M,
                              const std::vector<ck::index_t>& N,
                              const std::vector<ck::index_t>& K,
-                             const std::vector<ck::index_t>& StridesA,
-                             const std::vector<ck::index_t>& StridesB,
-                             const std::vector<ck::index_t>& StridesE,
-                             const std::vector<ck::index_t>& StridesD)
+                             const std::vector<ck::index_t>& StridesA, // [M0, M1, K0, K1]
+                             const std::vector<ck::index_t>& StridesB, // [N0, N1, K0, K1]
+                             const std::vector<ck::index_t>& StridesE, // [M0, M1, N0, N1]
+                             const std::vector<ck::index_t>& StridesD) // [M0, M1, N0, N1]
 {
     bool pass = true;
 
@@ -63,13 +67,13 @@ int profile_contraction_impl(ck::index_t do_verification,
     };
 
     Tensor<DataType> a_m_k(f_host_tensor_descriptor(M, K, StridesA));
-    Tensor<DataType> b_k_n(f_host_tensor_descriptor(K, N, StridesB));
+    Tensor<DataType> b_n_k(f_host_tensor_descriptor(N, K, StridesB));
     Tensor<DataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
     Tensor<DataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StridesE));
     Tensor<DataType> d_m_n(f_host_tensor_descriptor(M, N, StridesD));
 
     std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "b_n_k: " << b_n_k.mDesc << std::endl;
     std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
     std::cout << "e_m_n: " << e_m_n_device_result.mDesc << std::endl;
 
@@ -78,12 +82,12 @@ int profile_contraction_impl(ck::index_t do_verification,
     case 0: break;
     case 1:
         a_m_k.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        b_n_k.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
         d_m_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
         break;
     default:
         a_m_k.GenerateTensorValue(GeneratorTensor_3<DataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
+        b_n_k.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
         d_m_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
     }
 
@@ -91,12 +95,12 @@ int profile_contraction_impl(ck::index_t do_verification,
     using BElementOp = ck::tensor_operation::element_wise::PassThrough;
 
     DeviceMem a_device_buf(sizeof(DataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(DataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(DataType) * b_n_k.mDesc.GetElementSpaceSize());
     DeviceMem e_device_buf(sizeof(DataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
     DeviceMem d_device_buf(sizeof(DataType) * d_m_n.mDesc.GetElementSpaceSize());
 
     a_device_buf.ToDevice(a_m_k.mData.data());
-    b_device_buf.ToDevice(b_k_n.mData.data());
+    b_device_buf.ToDevice(b_n_k.mData.data());
     e_device_buf.SetZero();
     d_device_buf.ToDevice(d_m_n.mData.data());
 
@@ -118,7 +122,8 @@ int profile_contraction_impl(ck::index_t do_verification,
                                                                               DataType,
                                                                               AElementOp,
                                                                               BElementOp,
-                                                                              CDElementOp>;
+                                                                              CDElementOp,
+                                                                              ComputeDataType>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -126,6 +131,9 @@ int profile_contraction_impl(ck::index_t do_verification,
 
     std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
 
+    using AccDataType =
+        typename std::conditional<std::is_same<ComputeDataType, F64>::value, F64, F32>::type;
+
     // Run reference op
     if(do_verification)
     {
@@ -136,7 +144,8 @@ int profile_contraction_impl(ck::index_t do_verification,
                                                                       DataType,
                                                                       DataType,
                                                                       DataType,
-                                                                      DataType,
+                                                                      AccDataType,
+                                                                      ComputeDataType,
                                                                       AElementOp,
                                                                       BElementOp>;
 
@@ -146,7 +155,7 @@ int profile_contraction_impl(ck::index_t do_verification,
         Tensor<DataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
 
         auto ref_argument =
-            ref_op.MakeArgument(a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op);
+            ref_op.MakeArgument(a_m_k, b_n_k, c_m_n_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
@@ -272,8 +281,29 @@ int profile_contraction_impl(ck::index_t do_verification,
             {
                 e_device_buf.FromDevice(e_m_n_device_result.mData.data());
 
-                float threshold =
-                    static_cast<DataType>(nelems_k) * std::numeric_limits<DataType>::epsilon();
+                // Both the kernel and the reference use `AccDataType`, so an absolute error of both
+                // of them is bounded by `nelems_k * std::numeric_limits<AccDataType>::epsilon()`.
+                // Comparing one to another can result in an absolute error as high as twice that
+                // value.
+                double threshold = 2 * nelems_k * std::numeric_limits<AccDataType>::epsilon();
+                // Handle the possible casting error of either AccDataType -> DataType or
+                // DataType -> ComputeDataType.
+                // TODO: Add a generic solution for calculating thresholds in CK.
+                if constexpr(ck::is_same_v<DataType, ck::bhalf_t> ||
+                             ck::is_same_v<ComputeDataType, ck::bhalf_t>)
+                {
+                    const double epsilon = std::pow(2, -7);
+                    // Maximum relative casting error when rounding to zero.
+                    threshold += epsilon * 2;
+                }
+                else if constexpr(ck::is_same_v<DataType, ck::half_t> ||
+                                  ck::is_same_v<ComputeDataType, ck::half_t>)
+                {
+                    const double epsilon = std::pow(2, -10);
+                    // Maximum relative casting error when rounding to zero.
+                    threshold += epsilon * 2;
+                }
+
                 pass = pass & ck::utils::check_err(e_m_n_device_result,
                                                    e_m_n_host_result,
                                                    "Error: incorrect results!",
@@ -283,7 +313,7 @@ int profile_contraction_impl(ck::index_t do_verification,
                 if(do_log)
                 {
                     LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
-                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "b: ", b_n_k.mData, ",") << std::endl;
                     LogRangeAsType<float>(std::cout << "c_host  : ", e_m_n_host_result.mData, ",")
                         << std::endl;
                     LogRangeAsType<float>(std::cout << "c_device: ", e_m_n_device_result.mData, ",")
diff --git a/profiler/include/profiler/profile_contraction_utils.hpp b/profiler/include/profiler/profile_contraction_utils.hpp
index 076bbd4559..05ec7daf9d 100644
--- a/profiler/include/profiler/profile_contraction_utils.hpp
+++ b/profiler/include/profiler/profile_contraction_utils.hpp
@@ -23,8 +23,18 @@ enum struct ContractionMatrixLayout
 
 enum struct ContractionDataType
 {
-    F32_F32_F32_F32, // 0
-    F64_F64_F64_F64, // 1
+    F32_F32_F32_F32,     // 0
+    F64_F64_F64_F64,     // 1
+    F16_F16_F16_F16,     // 2
+    BF16_BF16_BF16_BF16, // 3
+};
+
+enum struct ContractionComputeDataType
+{
+    F32 = 0,
+    F64,
+    F16,
+    BF16,
 };
 
 inline void collect_index_params(char* argv[],
diff --git a/profiler/src/profile_contraction_bilinear.cpp b/profiler/src/profile_contraction_bilinear.cpp
index 6ed1841204..8cb34ccc6c 100644
--- a/profiler/src/profile_contraction_bilinear.cpp
+++ b/profiler/src/profile_contraction_bilinear.cpp
@@ -17,8 +17,9 @@
 static void print_helper_msg()
 {
     std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
-              << "arg2: data type (0: fp32; 1: f64)\n"
-              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+              << "arg2: data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
+              << "arg3: compute data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
+              << "arg4: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
@@ -26,40 +27,42 @@ static void print_helper_msg()
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
-              << "arg4: verification (0: no; 1: yes)\n"
-              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
+              << "arg5: verification (0: no; 1: yes)\n"
+              << "arg6: initialization (0: no init; 1: integer value; 2: decimal "
               << "value)\n"
-              << "arg6: print tensor value (0: no; 1: yes)\n"
-              << "arg7: time kernel (0: no, 1: yes)\n"
-              << "arg8 and arg9: alpha and beta\n"
-              << "arg10 to 15: M0, M1, N0, N1, K0, K1\n"
-              << "arg16 to 31: Strides for A, B, D and E (skip for default)\n"
+              << "arg7: print tensor value (0: no; 1: yes)\n"
+              << "arg8: time kernel (0: no, 1: yes)\n"
+              << "arg9: alpha\n"
+              << "arg10: beta\n"
+              << "arg11 to 16: M0, M1, N0, N1, K0, K1\n"
+              << "arg17 to 32: Strides for A, B, D and E (skip for default)\n"
               << std::endl;
 }
 
 int profile_contraction_bilinear(int argc, char* argv[])
 {
-    const bool default_strides = argc == 16;
+    const bool default_strides = argc == 17;
 
-    if(argc != 32 && argc != 16)
+    if(argc != 33 && argc != 17)
     {
         print_helper_msg();
         exit(1);
     }
 
     const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
-    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
-    const bool do_verification    = std::stoi(argv[4]);
-    const ck::index_t init_method = std::stoi(argv[5]);
-    const bool do_log             = std::stoi(argv[6]);
-    const bool time_kernel        = std::stoi(argv[7]);
-    const float alpha             = std::stof(argv[8]);
-    const float beta              = std::stof(argv[9]);
+    const auto compute_data_type  = static_cast<ContractionComputeDataType>(std::stoi(argv[3]));
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[4]));
+    const bool do_verification    = std::stoi(argv[5]);
+    const ck::index_t init_method = std::stoi(argv[6]);
+    const bool do_log             = std::stoi(argv[7]);
+    const bool time_kernel        = std::stoi(argv[8]);
+    const float alpha             = std::stof(argv[9]);
+    const float beta              = std::stof(argv[10]);
 
     std::vector<ck::index_t> M;
     std::vector<ck::index_t> N;
     std::vector<ck::index_t> K;
-    const ck::index_t dims_arg_num = 10;
+    const ck::index_t dims_arg_num = 11;
     collect_index_params(argv, M, dims_arg_num, 2);
     collect_index_params(argv, N, dims_arg_num + 2, 2);
     collect_index_params(argv, K, dims_arg_num + 4, 2);
@@ -76,90 +79,130 @@ int profile_contraction_bilinear(int argc, char* argv[])
         collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
     }
 
-    using F32 = float;
-    using F64 = double;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+    using F32  = float;
+    using F64  = double;
 
-    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
-        using ALayout   = decltype(a_layout);
-        using BLayout   = decltype(b_layout);
-        using CDELayout = decltype(cde_layout);
+    auto profile =
+        [&](auto a_layout, auto b_layout, auto cde_layout, auto type, auto compute_type) {
+            using ALayout   = decltype(a_layout);
+            using BLayout   = decltype(b_layout);
+            using CDELayout = decltype(cde_layout);
 
-        using DataType = decltype(type);
+            using DataType        = decltype(type);
+            using ComputeDataType = decltype(compute_type);
 
-        if(default_strides)
+            if(default_strides)
+            {
+                assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+                assign_default_strides(b_layout, StridesB, {N[0], N[1], K[0], K[1]});
+                assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+                assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+            }
+            bool pass = ck::profiler::profile_contraction_impl<ALayout,
+                                                               BLayout,
+                                                               CDELayout,
+                                                               DataType,
+                                                               ComputeDataType,
+                                                               ck::Tuple<DataType>,
+                                                               Bilinear>(do_verification,
+                                                                         init_method,
+                                                                         do_log,
+                                                                         time_kernel,
+                                                                         Bilinear{alpha, beta},
+                                                                         M,
+                                                                         N,
+                                                                         K,
+                                                                         StridesA,
+                                                                         StridesB,
+                                                                         StridesE,
+                                                                         StridesD);
+
+            return pass;
+        };
+
+    auto run_profile_for_datatype = [&](auto type, auto compute_type) {
+        if(layout == ContractionMatrixLayout::MK_KN_MN_MN)
         {
-            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
-            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
-            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
-            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+            return profile(Row{}, Row{}, Row{}, type, compute_type);
         }
-        bool pass = ck::profiler::profile_contraction_impl<ALayout,
-                                                           BLayout,
-                                                           CDELayout,
-                                                           DataType,
-                                                           ck::Tuple<DataType>,
-                                                           Bilinear>(do_verification,
-                                                                     init_method,
-                                                                     do_log,
-                                                                     time_kernel,
-                                                                     Bilinear{alpha, beta},
-                                                                     M,
-                                                                     N,
-                                                                     K,
-                                                                     StridesA,
-                                                                     StridesB,
-                                                                     StridesE,
-                                                                     StridesD);
-
-        return pass;
+        else if(layout == ContractionMatrixLayout::MK_NK_MN_MN)
+        {
+            return profile(Row{}, Col{}, Row{}, type, compute_type);
+        }
+        else if(layout == ContractionMatrixLayout::KM_KN_MN_MN)
+        {
+            return profile(Col{}, Row{}, Row{}, type, compute_type);
+        }
+        else if(layout == ContractionMatrixLayout::KM_NK_MN_MN)
+        {
+            return profile(Col{}, Col{}, Row{}, type, compute_type);
+        }
+        return false;
     };
 
-    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-       layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    if(data_type == ContractionDataType::F32_F32_F32_F32)
     {
-        return profile(Row{}, Row{}, Row{}, F32{});
+        if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(F32{}, F32{});
+        }
+        else if(compute_data_type == ContractionComputeDataType::F16)
+        {
+            return run_profile_for_datatype(F32{}, F16{});
+        }
+        else if(compute_data_type == ContractionComputeDataType::BF16)
+        {
+            return run_profile_for_datatype(F32{}, BF16{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
-    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    else if(data_type == ContractionDataType::F64_F64_F64_F64)
     {
-        return profile(Row{}, Col{}, Row{}, F32{});
+        if(compute_data_type == ContractionComputeDataType::F64)
+        {
+            return run_profile_for_datatype(F64{}, F64{});
+        }
+        else if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(F64{}, F32{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
-    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    else if(data_type == ContractionDataType::F16_F16_F16_F16)
     {
-        return profile(Col{}, Row{}, Row{}, F32{});
+        if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(F16{}, F32{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
-    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    else if(data_type == ContractionDataType::BF16_BF16_BF16_BF16)
     {
-        return profile(Col{}, Col{}, Row{}, F32{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::MK_KN_MN_MN)
-    {
-        return profile(Row{}, Row{}, Row{}, F64{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::MK_NK_MN_MN)
-    {
-        return profile(Row{}, Col{}, Row{}, F64{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::KM_KN_MN_MN)
-    {
-        return profile(Col{}, Row{}, Row{}, F64{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::KM_NK_MN_MN)
-    {
-        return profile(Col{}, Col{}, Row{}, F64{});
-    }
-    else
-    {
-        std::cout << "this data_type & layout is not implemented" << std::endl;
-
-        return 1;
+        if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(BF16{}, F32{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
+    return 1;
 }
 
 REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_bilinear);
diff --git a/profiler/src/profile_contraction_scale.cpp b/profiler/src/profile_contraction_scale.cpp
index 6784b916f6..ca9c199988 100644
--- a/profiler/src/profile_contraction_scale.cpp
+++ b/profiler/src/profile_contraction_scale.cpp
@@ -17,8 +17,9 @@
 static void print_helper_msg()
 {
     std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
-              << "arg2: data type (0: fp32; 1: f64)\n"
-              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+              << "arg2: data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
+              << "arg3: compute data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
+              << "arg4: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
@@ -26,39 +27,40 @@ static void print_helper_msg()
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
-              << "arg4: verification (0: no; 1: yes)\n"
-              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
+              << "arg5: verification (0: no; 1: yes)\n"
+              << "arg6: initialization (0: no init; 1: integer value; 2: decimal "
               << "value)\n"
-              << "arg6: print tensor value (0: no; 1: yes)\n"
-              << "arg7: time kernel (0: no, 1: yes)\n"
-              << "arg8: alpha\n"
-              << "arg9 to 14: M0, M1, N0, N1, K0, K1\n"
-              << "arg15 to 30: Strides for A, B, D and E (skip for default)\n"
+              << "arg7: print tensor value (0: no; 1: yes)\n"
+              << "arg8: time kernel (0: no, 1: yes)\n"
+              << "arg9: alpha\n"
+              << "arg10 to 15: M0, M1, N0, N1, K0, K1\n"
+              << "arg16 to 31: Strides for A, B, D and E (skip for default)\n"
               << std::endl;
 }
 
 int profile_contraction_scale(int argc, char* argv[])
 {
-    const bool default_strides = argc == 15;
+    const bool default_strides = argc == 16;
 
-    if(argc != 31 && argc != 15)
+    if(argc != 32 && argc != 16)
     {
         print_helper_msg();
         exit(1);
     }
 
     const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
-    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
-    const bool do_verification    = std::stoi(argv[4]);
-    const ck::index_t init_method = std::stoi(argv[5]);
-    const bool do_log             = std::stoi(argv[6]);
-    const bool time_kernel        = std::stoi(argv[7]);
-    const float alpha             = std::stof(argv[8]);
+    const auto compute_data_type  = static_cast<ContractionComputeDataType>(std::stoi(argv[3]));
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[4]));
+    const bool do_verification    = std::stoi(argv[5]);
+    const ck::index_t init_method = std::stoi(argv[6]);
+    const bool do_log             = std::stoi(argv[7]);
+    const bool time_kernel        = std::stoi(argv[8]);
+    const float alpha             = std::stof(argv[9]);
 
     std::vector<ck::index_t> M;
     std::vector<ck::index_t> N;
     std::vector<ck::index_t> K;
-    const ck::index_t dims_arg_num = 9;
+    const ck::index_t dims_arg_num = 10;
     collect_index_params(argv, M, dims_arg_num, 2);
     collect_index_params(argv, N, dims_arg_num + 2, 2);
     collect_index_params(argv, K, dims_arg_num + 4, 2);
@@ -75,88 +77,131 @@ int profile_contraction_scale(int argc, char* argv[])
         collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
     }
 
-    using F32 = float;
-    using F64 = double;
+    using F16  = ck::half_t;
+    using BF16 = ck::bhalf_t;
+    using F32  = float;
+    using F64  = double;
 
-    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
-        using ALayout   = decltype(a_layout);
-        using BLayout   = decltype(b_layout);
-        using CDELayout = decltype(cde_layout);
+    auto profile =
+        [&](auto a_layout, auto b_layout, auto cde_layout, auto type, auto compute_type) {
+            using ALayout   = decltype(a_layout);
+            using BLayout   = decltype(b_layout);
+            using CDELayout = decltype(cde_layout);
 
-        using DataType = decltype(type);
+            using DataType        = decltype(type);
+            using ComputeDataType = decltype(compute_type);
 
-        if(default_strides)
+            if(default_strides)
+            {
+                assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+                assign_default_strides(b_layout, StridesB, {N[0], N[1], K[0], K[1]});
+                assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+                assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+            }
+
+            bool pass = ck::profiler::profile_contraction_impl<ALayout,
+                                                               BLayout,
+                                                               CDELayout,
+                                                               DataType,
+                                                               ComputeDataType,
+                                                               ck::Tuple<>,
+                                                               Scale>(do_verification,
+                                                                      init_method,
+                                                                      do_log,
+                                                                      time_kernel,
+                                                                      Scale{alpha},
+                                                                      M,
+                                                                      N,
+                                                                      K,
+                                                                      StridesA,
+                                                                      StridesB,
+                                                                      StridesE,
+                                                                      StridesD);
+
+            return pass;
+        };
+
+    auto run_profile_for_datatype = [&](auto type, auto compute_type) {
+        if(layout == ContractionMatrixLayout::MK_KN_MN_MN)
         {
-            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
-            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
-            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
-            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
+            return profile(Row{}, Row{}, Row{}, type, compute_type);
         }
-
-        bool pass = ck::profiler::
-            profile_contraction_impl<ALayout, BLayout, CDELayout, DataType, ck::Tuple<>, Scale>(
-                do_verification,
-                init_method,
-                do_log,
-                time_kernel,
-                Scale{alpha},
-                M,
-                N,
-                K,
-                StridesA,
-                StridesB,
-                StridesE,
-                StridesD);
-
-        return pass;
+        else if(layout == ContractionMatrixLayout::MK_NK_MN_MN)
+        {
+            return profile(Row{}, Col{}, Row{}, type, compute_type);
+        }
+        else if(layout == ContractionMatrixLayout::KM_KN_MN_MN)
+        {
+            return profile(Col{}, Row{}, Row{}, type, compute_type);
+        }
+        else if(layout == ContractionMatrixLayout::KM_NK_MN_MN)
+        {
+            return profile(Col{}, Col{}, Row{}, type, compute_type);
+        }
+        return false;
     };
 
-    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-       layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    if(data_type == ContractionDataType::F32_F32_F32_F32)
     {
-        return profile(Row{}, Row{}, Row{}, F32{});
+        if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(F32{}, F32{});
+        }
+        else if(compute_data_type == ContractionComputeDataType::F16)
+        {
+            return run_profile_for_datatype(F32{}, F16{});
+        }
+        else if(compute_data_type == ContractionComputeDataType::BF16)
+        {
+            return run_profile_for_datatype(F32{}, BF16{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
-    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    else if(data_type == ContractionDataType::F64_F64_F64_F64)
     {
-        return profile(Row{}, Col{}, Row{}, F32{});
+        if(compute_data_type == ContractionComputeDataType::F64)
+        {
+            return run_profile_for_datatype(F64{}, F64{});
+        }
+        else if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(F64{}, F32{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
-    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    else if(data_type == ContractionDataType::F16_F16_F16_F16)
     {
-        return profile(Col{}, Row{}, Row{}, F32{});
+        if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(F16{}, F32{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
-    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
-            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    else if(data_type == ContractionDataType::BF16_BF16_BF16_BF16)
     {
-        return profile(Col{}, Col{}, Row{}, F32{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::MK_KN_MN_MN)
-    {
-        return profile(Row{}, Row{}, Row{}, F64{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::MK_NK_MN_MN)
-    {
-        return profile(Row{}, Col{}, Row{}, F64{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::KM_KN_MN_MN)
-    {
-        return profile(Col{}, Row{}, Row{}, F64{});
-    }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
-            layout == ContractionMatrixLayout::KM_NK_MN_MN)
-    {
-        return profile(Col{}, Col{}, Row{}, F64{});
-    }
-    else
-    {
-        std::cout << "this data_type & layout is not implemented" << std::endl;
-
-        return 1;
+        if(compute_data_type == ContractionComputeDataType::F32)
+        {
+            return run_profile_for_datatype(BF16{}, F32{});
+        }
+        else
+        {
+            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
+            return 1;
+        }
     }
+    return 1;
 }
 
 REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_scale);
diff --git a/test/contraction/test_contraction.cpp b/test/contraction/test_contraction.cpp
index c86b849235..958d5be38c 100644
--- a/test/contraction/test_contraction.cpp
+++ b/test/contraction/test_contraction.cpp
@@ -10,9 +10,12 @@
 #include <gtest/gtest.h>
 
 #include "profiler/profile_contraction_impl.hpp"
+#include "profiler/profile_contraction_utils.hpp"
 
-using F32 = float;
-using F64 = double;
+using F16  = ck::half_t;
+using BF16 = ck::bhalf_t;
+using F32  = float;
+using F64  = double;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -20,49 +23,49 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
 using Bilinear = ck::tensor_operation::element_wise::Bilinear;
 using Scale    = ck::tensor_operation::element_wise::Scale;
 
-struct MemoryParams
+struct Dimensions
 {
     std::vector<ck::index_t> M;
     std::vector<ck::index_t> N;
     std::vector<ck::index_t> K;
-    std::vector<ck::index_t> StridesA;
-    std::vector<ck::index_t> StridesB;
-    std::vector<ck::index_t> StridesC;
-    std::vector<ck::index_t> StridesD;
 };
 
 template <typename Tuple>
 class TestContraction : public ::testing::Test
 {
     protected:
-    using ALayout        = std::tuple_element_t<0, Tuple>;
-    using BLayout        = std::tuple_element_t<1, Tuple>;
-    using CDLayout       = std::tuple_element_t<2, Tuple>;
-    using DataType       = std::tuple_element_t<3, Tuple>;
-    using DTupleDataType = std::tuple_element_t<4, Tuple>;
-    using CDElementOp    = std::tuple_element_t<5, Tuple>;
+    using ALayout         = std::tuple_element_t<0, Tuple>;
+    using BLayout         = std::tuple_element_t<1, Tuple>;
+    using CDLayout        = std::tuple_element_t<2, Tuple>;
+    using DataType        = std::tuple_element_t<3, Tuple>;
+    using DTupleDataType  = std::tuple_element_t<4, Tuple>;
+    using ComputeDataType = std::tuple_element_t<5, Tuple>;
+    using CDElementOp     = std::tuple_element_t<6, Tuple>;
 
-    std::vector<MemoryParams> list_of_memory_params = {{{32, 32},
-                                                        {32, 32},
-                                                        {32, 32},
-                                                        {32768, 1024, 32, 1},
-                                                        {32768, 1024, 32, 1},
-                                                        {32768, 1024, 32, 1},
-                                                        {32768, 1024, 32, 1}},
-                                                       {{16, 16},
-                                                        {32, 32},
-                                                        {16, 16},
-                                                        {4096, 256, 16, 1},
-                                                        {16, 1, 8192, 256},
-                                                        {16384, 1024, 32, 1},
-                                                        {16384, 1024, 32, 1}}};
+    std::vector<Dimensions> dimension_list = {{{32, 32}, {32, 32}, {32, 32}},
+                                              {{16, 16}, {32, 32}, {16, 16}}};
 
-    std::vector<ck::index_t> init_methods = {0, 1, 2};
+    std::vector<ck::index_t> init_methods = {1, 2};
     std::unique_ptr<CDElementOp> p_cd_element_op;
+
     void Run()
     {
-        for(auto& memory_params : list_of_memory_params)
+        for(auto& dimension_params : dimension_list)
         {
+            std::vector<ck::index_t> StridesA;
+            std::vector<ck::index_t> StridesB;
+            std::vector<ck::index_t> StridesC;
+            std::vector<ck::index_t> StridesD;
+
+            const auto& M = dimension_params.M;
+            const auto& N = dimension_params.N;
+            const auto& K = dimension_params.K;
+
+            assign_default_strides(ALayout{}, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(BLayout{}, StridesB, {N[0], N[1], K[0], K[1]});
+            assign_default_strides(CDLayout{}, StridesC, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(CDLayout{}, StridesD, {M[0], M[1], N[0], N[1]});
+
             for(const ck::index_t init_method : init_methods)
             {
                 bool pass =
@@ -70,19 +73,20 @@ class TestContraction : public ::testing::Test
                                                            BLayout,
                                                            CDLayout,
                                                            DataType,
+                                                           ComputeDataType,
                                                            DTupleDataType,
                                                            CDElementOp>(true /*do_verification*/,
                                                                         init_method,
                                                                         false /*do_logs*/,
                                                                         false /*time_kernel*/,
                                                                         *p_cd_element_op,
-                                                                        memory_params.M,
-                                                                        memory_params.N,
-                                                                        memory_params.K,
-                                                                        memory_params.StridesA,
-                                                                        memory_params.StridesB,
-                                                                        memory_params.StridesC,
-                                                                        memory_params.StridesD);
+                                                                        dimension_params.M,
+                                                                        dimension_params.N,
+                                                                        dimension_params.K,
+                                                                        StridesA,
+                                                                        StridesB,
+                                                                        StridesC,
+                                                                        StridesD);
                 EXPECT_TRUE(pass);
             }
         }
@@ -99,24 +103,18 @@ class TestContractionBilinear : public TestContraction<Tuple>
 {
 };
 
-using BilinearKernelTypes =
-    ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Row, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Col, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Col, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Row, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Row, Col, Row, F64, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Col, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
-                     std::tuple<Col, Col, Row, F64, ck::Tuple<F32>, Bilinear>>;
+#define ALL_LAYOUT_COMBINATIONS(dt, tuple_dt, compute_dt, op)    \
+    std::tuple<Row, Row, Row, dt, tuple_dt, compute_dt, op>,     \
+        std::tuple<Row, Col, Row, dt, tuple_dt, compute_dt, op>, \
+        std::tuple<Col, Row, Row, dt, tuple_dt, compute_dt, op>, \
+        std::tuple<Col, Col, Row, dt, tuple_dt, compute_dt, op>
 
-using ScaleKernelTypes = ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<>, Scale>,
-                                          std::tuple<Row, Col, Row, F32, ck::Tuple<>, Scale>,
-                                          std::tuple<Col, Row, Row, F32, ck::Tuple<>, Scale>,
-                                          std::tuple<Col, Col, Row, F32, ck::Tuple<>, Scale>,
-                                          std::tuple<Row, Row, Row, F64, ck::Tuple<>, Scale>,
-                                          std::tuple<Row, Col, Row, F64, ck::Tuple<>, Scale>,
-                                          std::tuple<Col, Row, Row, F64, ck::Tuple<>, Scale>,
-                                          std::tuple<Col, Col, Row, F64, ck::Tuple<>, Scale>>;
+using BilinearKernelTypes =
+    ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F32, Bilinear),
+                     ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F64, Bilinear)>;
+
+using ScaleKernelTypes = ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F32, Scale),
+                                          ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F64, Scale)>;
 
 TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes);
 TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
@@ -136,3 +134,46 @@ TYPED_TEST(TestContractionScale, scale)
     this->p_cd_element_op = std::make_unique<Scale>(0.5f);
     this->Run();
 }
+
+template <typename Tuple>
+class TestContractionScaleMixedPrecision : public TestContraction<Tuple>
+{
+};
+
+template <typename Tuple>
+class TestContractionBilinearMixedPrecision : public TestContraction<Tuple>
+{
+};
+
+using BilinearKernelTypesMixedPrecision =
+    ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F16, Bilinear),
+                     ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, BF16, Bilinear),
+                     ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F32, Bilinear),
+                     ALL_LAYOUT_COMBINATIONS(F16, ck::Tuple<F16>, F32, Bilinear),
+                     ALL_LAYOUT_COMBINATIONS(BF16, ck::Tuple<BF16>, F32, Bilinear)>;
+
+using ScaleKernelTypesMixedPrecision =
+    ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F16, Scale),
+                     ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, BF16, Scale),
+                     ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F32, Scale),
+                     ALL_LAYOUT_COMBINATIONS(F16, ck::Tuple<>, F32, Scale),
+                     ALL_LAYOUT_COMBINATIONS(BF16, ck::Tuple<>, F32, Scale)>;
+
+TYPED_TEST_SUITE(TestContractionBilinearMixedPrecision, BilinearKernelTypesMixedPrecision);
+TYPED_TEST_SUITE(TestContractionScaleMixedPrecision, ScaleKernelTypesMixedPrecision);
+
+TYPED_TEST(TestContractionBilinearMixedPrecision, bilinear)
+{
+    this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
+    this->Run();
+    this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
+    this->Run();
+}
+
+TYPED_TEST(TestContractionScaleMixedPrecision, scale)
+{
+    this->p_cd_element_op = std::make_unique<Scale>(1.f);
+    this->Run();
+    this->p_cd_element_op = std::make_unique<Scale>(0.5f);
+    this->Run();
+}
diff --git a/test/contraction/test_contraction_interface.cpp b/test/contraction/test_contraction_interface.cpp
index 12a307f5da..e0f547445e 100644
--- a/test/contraction/test_contraction_interface.cpp
+++ b/test/contraction/test_contraction_interface.cpp
@@ -34,11 +34,11 @@ class ContractionInstanceWrapper
     static constexpr ck::index_t NumDim = 2;
     // clang-format off
     using ContractionDeviceInstance = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|         DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer|             ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer|              BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|                  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |               |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |               |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
-        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,         Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|         DsData| EData| Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer|             ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer|              BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|                  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|    Data| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |               |      |    Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |               |      |        |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
+        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,     F32,        Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
     // clang-format on
 
     bool isSupported(std::vector<ck::index_t>& ADims,

From 6b5f6473719501cae0d333836245daacb6cd7248 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 29 Sep 2023 08:55:25 -0700
Subject: [PATCH 151/194] add gfx942 target to the daily ckprofiler package
 (#955)

---
 Jenkinsfile | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 0fdadf8a87..60a53c54fb 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -713,8 +713,8 @@ pipeline {
                     }
                     agent{ label rocmnode("gfx908 || gfx90a") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx908;gfx90a;gfx940;gfx941;gfx942" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """ 
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')

From 9d58c42103613ad09725e44b29316eba99ddc916 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Mon, 2 Oct 2023 09:18:36 -0500
Subject: [PATCH 152/194] Contraction multi abd (#957)

* add gridwise_multi_abd

* move element_op into RunRead

* merge element_wise op with data read

* add multiABD example

* allow packed elementwise_op

* changed example

* clean

* clean

* add is_detected

* fix

* minor fix

* add scaleAdd_vec4 example

* init commit for contraction_multi_ABD

* add examples

* add examples of multiA and broadcast

* update example

* fixed comments

* Update cmake-ck-dev.sh

* Update cmake-ck-dev.sh

* Add comments into the example

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/60_gemm_multi_ABD/CMakeLists.txt      |  10 +
 .../gemm_multi_ABD_xdl_fp16.cpp               | 362 ++++++++
 .../61_contraction_multi_ABD/CMakeLists.txt   |  10 +
 .../contraction_multi_ABD_xdl_fp16.cpp        | 328 +++++++
 .../device_contraction_multiple_abd.hpp       |  61 ++
 ..._contraction_multiple_abd_xdl_cshuffle.hpp | 846 ++++++++++++++++++
 ...ridwise_gemm_multiple_abd_xdl_cshuffle.hpp |   2 +-
 7 files changed, 1618 insertions(+), 1 deletion(-)
 create mode 100644 example/60_gemm_multi_ABD/CMakeLists.txt
 create mode 100644 example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
 create mode 100644 example/61_contraction_multi_ABD/CMakeLists.txt
 create mode 100644 example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp

diff --git a/example/60_gemm_multi_ABD/CMakeLists.txt b/example/60_gemm_multi_ABD/CMakeLists.txt
new file mode 100644
index 0000000000..0c6ce7ae87
--- /dev/null
+++ b/example/60_gemm_multi_ABD/CMakeLists.txt
@@ -0,0 +1,10 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+     add_example_executable(example_gemm_multi_ABD_xdl_fp16 gemm_multi_ABD_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
+endif()
diff --git a/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
new file mode 100644
index 0000000000..6bd7a889b8
--- /dev/null
+++ b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
@@ -0,0 +1,362 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F16;
+using EDataType        = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using DLayout = Row;
+using ELayout = Row;
+
+struct AddScale
+{
+    static constexpr auto I0 = ck::Number<0>{};
+    static constexpr auto I1 = ck::Number<1>{};
+    static constexpr auto I2 = ck::Number<2>{};
+    static constexpr auto I3 = ck::Number<3>{};
+
+    __host__ __device__ constexpr void
+    operator()(ck::half4_t& a, const ck::half4_t& a0, const ck::half4_t& a1) const
+    {
+        const auto a0_v_t = ck::vector_type<ck::half_t, 4>{a0};
+        const auto a1_v_t = ck::vector_type<ck::half_t, 4>{a1};
+
+        auto r_v_t = ck::vector_type<ck::half_t, 4>{};
+
+        r_v_t.AsType<ck::half_t>()(I0) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I0] + a1_v_t.AsType<ck::half_t>()[I0]);
+        r_v_t.AsType<ck::half_t>()(I1) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I1] + a1_v_t.AsType<ck::half_t>()[I1]);
+        r_v_t.AsType<ck::half_t>()(I2) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I2] + a1_v_t.AsType<ck::half_t>()[I2]);
+        r_v_t.AsType<ck::half_t>()(I3) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I3] + a1_v_t.AsType<ck::half_t>()[I3]);
+
+        a = r_v_t.AsType<ck::half4_t>()[I0];
+    }
+
+    __host__ __device__ constexpr void
+    operator()(ck::half_t& a, const ck::half_t& a0, const ck::half_t& a1) const
+    {
+        a = scale * (a0 + a1);
+    }
+
+    //this attribute will force copy_function applying element_wise with vector_type
+    static constexpr ck::index_t vec_len = 4;
+
+    float scale = 1.0;
+};
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
+        ck::half_t& e, const float& c, const ck::half_t& d) const
+    {
+        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
+    };
+
+    float alpha_;
+    float beta_;
+};
+
+using AElementOp   = AddScale;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleABD_Xdl_CShuffle<
+    ck::Tuple<ALayout, ALayout>,
+    ck::Tuple<BLayout>,
+    ck::Tuple<DLayout>,
+    ELayout,
+    ck::Tuple<ADataType, ADataType>,
+    ck::Tuple<BDataType>,
+    AccDataType,
+    CShuffleDataType,
+    ck::Tuple<DDataType>,
+    EDataType,
+    AElementOp,
+    BElementOp,
+    CDEElementOp,
+    GemmSpec,
+    1,
+    256,
+    256,
+    128,
+    32,
+    8,
+    8,
+    32,
+    32,
+    4,
+    2,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    1,
+    1,
+    S<1, 32, 1, 8>,
+    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideD = 4096;
+    ck::index_t StrideE = 4096;
+
+    float alpha = 1.0f;
+    float beta  = 1.0f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 6)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        alpha = std::stof(argv[4]);
+        beta  = std::stof(argv[5]);
+    }
+    else if(argc == 13)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        StrideA = std::stoi(argv[7]);
+        StrideB = std::stoi(argv[8]);
+        StrideD = std::stoi(argv[9]);
+        StrideE = std::stoi(argv[10]);
+
+        alpha = std::stof(argv[11]);
+        beta  = std::stof(argv[12]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
+               "beta\n");
+        exit(0);
+    }
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<ADataType> a1_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
+    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
+        break;
+    default:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem a0_device_buf(sizeof(ADataType) * a0_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem a1_device_buf(sizeof(ADataType) * a1_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_m_k.mData.data());
+    a1_device_buf.ToDevice(a1_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d_device_buf.ToDevice(d_m_n.mData.data());
+    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
+
+    auto a_element_op   = AElementOp{0.2};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
+                                                          a1_device_buf.GetDeviceBuffer()},
+                               std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
+                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                               e_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               std::array<ck::index_t, 2>{StrideA, StrideA},
+                               std::array<ck::index_t, 1>{StrideB},
+                               std::array<ck::index_t, 1>{StrideD},
+                               StrideE,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_m_n({M, N});
+
+        Tensor<ADataType> a_m_k({M, K});
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int k = 0; k < K; ++k)
+            {
+                a_element_op(a_m_k(m, k), a0_m_k(m, k), a1_m_k(m, k));
+            }
+        }
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                PassThrough,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/61_contraction_multi_ABD/CMakeLists.txt b/example/61_contraction_multi_ABD/CMakeLists.txt
new file mode 100644
index 0000000000..0420fdb9c3
--- /dev/null
+++ b/example/61_contraction_multi_ABD/CMakeLists.txt
@@ -0,0 +1,10 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+     add_example_executable(example_contraction_multi_ABD_xdl_fp16 contraction_multi_ABD_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
+endif()
diff --git a/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
new file mode 100644
index 0000000000..6634dacd25
--- /dev/null
+++ b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
@@ -0,0 +1,328 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/numeric.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using A0DataType       = F16;
+using A1DataType       = F32;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F16;
+using EDataType        = F16;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
+        ck::half_t& e, const float& c, const ck::half_t& d) const
+    {
+        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
+    };
+
+    float alpha_;
+    float beta_;
+};
+
+struct Multiply
+{
+    __host__ __device__ constexpr void
+    operator()(ck::half_t& a, const ck::half_t& a0, const float& a1) const
+    {
+        a = ck::type_convert<ck::half_t>(ck::type_convert<float>(a0) * a1);
+    }
+};
+
+using AElementOp   = Multiply;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceOpInstance = ck::tensor_operation::device::DeviceContractionMultipleABD_Xdl_CShuffle<
+    NumDimM,
+    NumDimN,
+    NumDimK,
+    ck::Tuple<A0DataType, A1DataType>,
+    ck::Tuple<BDataType>,
+    AccDataType,
+    CShuffleDataType,
+    ck::Tuple<DDataType>,
+    EDataType,
+    AElementOp,
+    BElementOp,
+    CDEElementOp,
+    GemmSpec,
+    1,
+    256,
+    256,
+    128,
+    32,
+    8,
+    8,
+    32,
+    32,
+    4,
+    2,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    1,
+    1,
+    S<1, 32, 1, 8>,
+    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    float alpha = 1.0f;
+    float beta  = 1.0f;
+
+    // A0[M0, M1, K0, K1]
+    std::vector<ck::index_t> a0_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a0_ms_ks_strides{128 * 32 * 64, 32 * 64, 64, 1};
+    // A1[M1, K1] -> A1[M0, M1, K0, K1]
+    std::vector<ck::index_t> a1_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a1_ms_ks_strides{0, 64, 0, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{64 * 32 * 64, 32 * 64, 64, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{128 * 32 * 64, 32 * 64, 64, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{128 * 32 * 64, 32 * 64, 64, 1};
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        exit(0);
+    }
+
+    Tensor<A0DataType> a0_ms_ks(a0_ms_ks_lengths, a0_ms_ks_strides);
+    Tensor<A1DataType> a1_ms_ks(a1_ms_ks_lengths, a1_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a0_ms_ks: " << a0_ms_ks.mDesc << std::endl;
+    std::cout << "a1_ms_ks: " << a1_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_ms_ks.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-5, 5});
+        a1_ms_ks.GenerateTensorValue(GeneratorTensor_2<A1DataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a0_ms_ks.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
+        a1_ms_ks.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a0_device_buf(sizeof(A0DataType) * a0_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem a1_device_buf(sizeof(A1DataType) * a1_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_ms_ks.mData.data());
+    a1_device_buf.ToDevice(a1_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument  = device_op.MakeArgument(
+        std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
+                                   a1_device_buf.GetDeviceBuffer()},
+        std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
+        std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+        e_device_buf.GetDeviceBuffer(),
+        std::array<std::vector<ck::index_t>, 2>{a0_ms_ks_lengths, a1_ms_ks_lengths},
+        std::array<std::vector<ck::index_t>, 2>{a0_ms_ks_strides, a1_ms_ks_strides},
+        std::array<std::vector<ck::index_t>, 1>{b_ns_ks_lengths},
+        std::array<std::vector<ck::index_t>, 1>{b_ns_ks_strides},
+        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+        e_ms_ns_lengths,
+        e_ms_ns_strides,
+        a_element_op,
+        b_element_op,
+        cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_contraction with the specified compilation parameters does "
+            "not support this problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    if(time_kernel)
+    {
+        ck::index_t M =
+            ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+        ck::index_t N = ck::accumulate_n<ck::index_t>(
+            e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+        ck::index_t K = ck::accumulate_n<ck::index_t>(
+            a0_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+        std::size_t flop = std::size_t(2) * M * N * K;
+        std::size_t num_btype =
+            sizeof(A0DataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
+
+        float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+        float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                  << " GB/s" << std::endl;
+    }
+
+    if(do_verification)
+    {
+
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        Tensor<A0DataType> a_ms_ks(a0_ms_ks_lengths, a0_ms_ks_strides);
+
+        for(size_t m0 = 0; m0 < a_ms_ks.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < a_ms_ks.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t k0 = 0; k0 < a_ms_ks.mDesc.GetLengths()[2]; ++k0)
+                {
+                    for(size_t k1 = 0; k1 < a_ms_ks.mDesc.GetLengths()[3]; ++k1)
+                    {
+                        a_element_op(a_ms_ks(m0, m1, k0, k1),
+                                     a0_ms_ks(m0, m1, k0, k1),
+                                     a1_ms_ks(m0, m1, k0, k1));
+                    }
+                }
+            }
+        }
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      A0DataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      PassThrough,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, PassThrough{}, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1),
+                                       d_ms_ns(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
new file mode 100644
index 0000000000..3116d40474
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A0[M0, M1, ... K0, K1, ...], ...
+//   input : B0[N0, N1, ... K0, K1, ...], ...
+//   input : D0[M0, M1, ... N0, N1, ...], D1[M0, M1, ... N0, N1, ...], ...
+//   output : E[M0, M1, ... N0, N1, ...]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          typename AsDataType,
+          typename BsDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceContractionMultipleABD : public BaseOperator
+{
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
+                        std::array<const void*, NumBTensor> p_bs,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
+                        const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
+                        const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
+                        const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
+                        const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
+                        const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
+                        const std::vector<index_t>& e_ms_ns_length,
+                        const std::vector<index_t>& e_ms_ns_stride,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
new file mode 100644
index 0000000000..a4c6767434
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
@@ -0,0 +1,846 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm,
+          typename AsPointer,
+          typename BsPointer,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AsGridDesc_AK0_M_AK1,
+          typename BsGridDesc_BK0_N_BK1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename Block2ETileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_contraction_multiple_abd_xdl_cshuffle(
+            AsPointer p_as_grid,
+            BsPointer p_bs_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1,
+            const BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const Block2ETileMap block_2_etile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_as_grid,
+                                                  p_bs_grid,
+                                                  p_ds_grid,
+                                                  p_e_grid,
+                                                  p_shared,
+                                                  a_element_op,
+                                                  b_element_op,
+                                                  cde_element_op,
+                                                  as_grid_desc_ak0_m_ak1,
+                                                  bs_grid_desc_bk0_n_bk1,
+                                                  ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  block_2_etile_map);
+#else
+    ignore = p_as_grid;
+    ignore = p_bs_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = as_grid_desc_ak0_m_ak1;
+    ignore = bs_grid_desc_bk0_n_bk1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = block_2_etile_map;
+#endif
+}
+
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          typename AsDataType,
+          typename BsDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1,
+          index_t BK1,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct DeviceContractionMultipleABD_Xdl_CShuffle
+    : public DeviceContractionMultipleABD<NumDimM,
+                                          NumDimN,
+                                          NumDimK,
+                                          AsDataType,
+                                          BsDataType,
+                                          DsDataType,
+                                          EDataType,
+                                          AElementwiseOperation,
+                                          BElementwiseOperation,
+                                          CDEElementwiseOperation>
+{
+    using DeviceOp = DeviceContractionMultipleABD_Xdl_CShuffle;
+
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+    using ComputeDataType = EDataType;
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemmMultipleABD_xdl_cshuffle<
+        AsDataType,
+        BsDataType,
+        ComputeDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        NumGemmKPrefetchStage,
+        BlockSize,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        AK1,
+        BK1,
+        MPerXDL,
+        NPerXDL,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEBlockTransferScalarPerVector_NPerBlock,
+        LoopSched,
+        PipelineVer>;
+
+    static constexpr auto matrix_padder =
+        ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+            MPerBlock, NPerBlock, KPerBlock};
+
+    static auto MakeAGridDescriptor_M_K(const std::vector<index_t>& a_ms_ks_lengths_,
+                                        const std::vector<index_t>& a_ms_ks_strides_)
+    {
+        assert(a_ms_ks_lengths_.size() == NumDimM + NumDimK &&
+               a_ms_ks_strides_.size() == NumDimM + NumDimK);
+
+        const auto to_tuple = [&](auto& vec, auto num) {
+            return generate_tuple([&](auto i) { return vec[i]; }, num);
+        };
+
+        const auto a_ms_ks_lengths = to_tuple(a_ms_ks_lengths_, Number<NumDimM + NumDimK>{});
+        const auto a_ms_ks_strides = to_tuple(a_ms_ks_strides_, Number<NumDimM + NumDimK>{});
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
+
+        // dimension Ids for K0, K1, ...
+        constexpr auto kDimIds =
+            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimK, 1>::type{};
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(a_ms_ks_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto kLengths = get_container_subset(a_ms_ks_lengths, kDimIds);
+
+        // naive tensor A[M0, M1, M2, ..., K0, K1, K2...]
+        const auto a_grid_desc_ms_ks =
+            make_naive_tensor_descriptor(a_ms_ks_lengths, a_ms_ks_strides);
+
+        // transformed tensor A[MRaw = M0 * M1 * M2 * ... , KRaw = K0 * K1 * K2 * ...]
+        const auto a_grid_desc_mraw_kraw = transform_tensor_descriptor(
+            a_grid_desc_ms_ks,
+            make_tuple(make_merge_transform(mLengths), make_merge_transform(kLengths)),
+            make_tuple(mDimIds, kDimIds),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+    }
+
+    __host__ __device__ static auto
+    MakeAsGridDescriptor_M_K(const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_lengths,
+                             const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeAGridDescriptor_M_K(as_ms_ks_lengths[i], as_ms_ks_strides[i]);
+            },
+            Number<NumATensor>{});
+    }
+
+    // Assume: B[N0, N1, N2, ..., K0, K1, K2, ...]
+    static auto MakeBGridDescriptor_N_K(const std::vector<index_t>& b_ns_ks_lengths_,
+                                        const std::vector<index_t>& b_ns_ks_strides_)
+    {
+        assert(b_ns_ks_lengths_.size() == NumDimN + NumDimK &&
+               b_ns_ks_strides_.size() == NumDimN + NumDimK);
+
+        const auto to_tuple = [&](auto& vec, auto num) {
+            return generate_tuple([&](auto i) { return vec[i]; }, num);
+        };
+
+        const auto b_ns_ks_lengths = to_tuple(b_ns_ks_lengths_, Number<NumDimN + NumDimK>{});
+        const auto b_ns_ks_strides = to_tuple(b_ns_ks_strides_, Number<NumDimN + NumDimK>{});
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds = typename arithmetic_sequence_gen<0, NumDimN, 1>::type{};
+
+        // dimension Ids for K0, K1, ...
+        constexpr auto kDimIds =
+            typename arithmetic_sequence_gen<NumDimN, NumDimN + NumDimK, 1>::type{};
+
+        // lengths for K0, K1, ...
+        const auto kLengths = get_container_subset(b_ns_ks_lengths, kDimIds);
+
+        // lengths for N0, N1, ...
+        const auto nLengths = get_container_subset(b_ns_ks_lengths, nDimIds);
+
+        // naive tensor B[N0, N1, N2, ..., K0, K1, K2, ...]
+        const auto b_grid_desc_ns_ks =
+            make_naive_tensor_descriptor(b_ns_ks_lengths, b_ns_ks_strides);
+
+        // transformed tensor B[NRaw = N0 * N1 * N2 * ..., KRaw = K0 * K1 * K2 * ...]
+        const auto b_grid_desc_nraw_kraw = transform_tensor_descriptor(
+            b_grid_desc_ns_ks,
+            make_tuple(make_merge_transform(nLengths), make_merge_transform(kLengths)),
+            make_tuple(nDimIds, kDimIds),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+    }
+
+    __host__ __device__ static auto
+    MakeBsGridDescriptor_N_K(const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_lengths,
+                             const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeBGridDescriptor_N_K(bs_ns_ks_lengths[i], bs_ns_ks_strides[i]);
+            },
+            Number<NumBTensor>{});
+    }
+
+    // assume E[M0, M1, M2, ..., N0, N1, N2...]
+    static auto MakeEGridDescriptor_M_N(const std::vector<index_t>& e_ms_ns_lengths_,
+                                        const std::vector<index_t>& e_ms_ns_strides_)
+    {
+        assert(e_ms_ns_lengths_.size() == NumDimM + NumDimN &&
+               e_ms_ns_strides_.size() == NumDimM + NumDimN);
+
+        const auto to_tuple = [&](auto& vec, auto num) {
+            return generate_tuple([&](auto i) { return vec[i]; }, num);
+        };
+
+        const auto e_ms_ns_lengths = to_tuple(e_ms_ns_lengths_, Number<NumDimM + NumDimN>{});
+        const auto e_ms_ns_strides = to_tuple(e_ms_ns_strides_, Number<NumDimM + NumDimN>{});
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds =
+            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimN, 1>::type{};
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(e_ms_ns_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto nLengths = get_container_subset(e_ms_ns_lengths, nDimIds);
+
+        // naive tensor E[M0, M1, M2, ..., N0, N1, N2...]
+        const auto e_grid_desc_ms_ns =
+            make_naive_tensor_descriptor(e_ms_ns_lengths, e_ms_ns_strides);
+
+        // transformed tensor E[MRaw = M0 * M1 * M2 * ... , NRaw = N0 * N1 * N2 * ...]
+        const auto e_grid_desc_mraw_nraw = transform_tensor_descriptor(
+            e_grid_desc_ms_ns,
+            make_tuple(make_merge_transform(mLengths), make_merge_transform(nLengths)),
+            make_tuple(mDimIds, nDimIds),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    static auto
+    MakeDsGridDescriptor_M_N(const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_lengths,
+                             const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_M_N(ds_ms_ns_lengths[i], ds_ms_ns_strides[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // desc for problem definition
+    using AsGridDesc_M_K = remove_cvref_t<decltype(MakeAsGridDescriptor_M_K({}, {}))>;
+    using BsGridDesc_N_K = remove_cvref_t<decltype(MakeBsGridDescriptor_N_K({}, {}))>;
+    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
+    using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N({}, {}))>;
+
+    // desc for blockwise copy
+    using AsGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(AsGridDesc_M_K{}))>;
+    using BsGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(BsGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
+
+    // block-to-e-tile map
+    using Block2ETileMap =
+        remove_cvref_t<decltype(GridwiseGemm::MakeBlock2ETileMap(EGridDesc_M_N{}))>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(std::array<const void*, NumATensor> p_as_grid,
+                 std::array<const void*, NumBTensor> p_bs_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
+                 const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
+                 const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
+                 const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
+                 const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
+                 const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
+                 const std::vector<index_t>& e_ms_ns_length,
+                 const std::vector<index_t>& e_ms_ns_stride,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_as_grid_{},
+              p_bs_grid_{},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              as_grid_desc_m_k_{},
+              bs_grid_desc_n_k_{},
+              ds_grid_desc_m_n_{},
+              e_grid_desc_m_n_{MakeEGridDescriptor_M_N(e_ms_ns_length, e_ms_ns_stride)},
+              as_grid_desc_ak0_m_ak1_{},
+              bs_grid_desc_bk0_n_bk1_{},
+              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              e_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              block_2_etile_map_{GridwiseGemm::MakeBlock2ETileMap(e_grid_desc_m_n_)},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+            // populate pointer, desc for As
+            static_for<0, NumATensor, 1>{}([&](auto i) {
+                // using ALayout   = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
+                using ADataType = remove_cvref_t<tuple_element_t<i.value, AsDataType>>;
+
+                // A pointer
+                p_as_grid_(i) = static_cast<const ADataType*>(p_as_grid[i]);
+
+                // A desc
+                as_grid_desc_m_k_(i) =
+                    MakeAGridDescriptor_M_K(a_ms_ks_lengths[i], a_ms_ks_strides[i]);
+            });
+
+            // populate pointer, desc for Bs
+            static_for<0, NumBTensor, 1>{}([&](auto i) {
+                // using BLayout   = remove_cvref_t<tuple_element_t<i.value, BsLayout>>;
+                using BDataType = remove_cvref_t<tuple_element_t<i.value, BsDataType>>;
+
+                // B pointer
+                p_bs_grid_(i) = static_cast<const BDataType*>(p_bs_grid[i]);
+
+                // B desc
+                bs_grid_desc_n_k_(i) =
+                    MakeBGridDescriptor_N_K(b_ns_ks_lengths[i], b_ns_ks_strides[i]);
+            });
+
+            // populate pointer, desc for Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                // using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    MakeEGridDescriptor_M_N(d_ms_ns_lengths[i], d_ms_ns_strides[i]);
+            });
+
+            // populate desc for Ds/E
+            if(GridwiseGemm::CheckValidity(as_grid_desc_m_k_,
+                                           bs_grid_desc_n_k_,
+                                           ds_grid_desc_m_n_,
+                                           e_grid_desc_m_n_,
+                                           block_2_etile_map_))
+            {
+                as_grid_desc_ak0_m_ak1_ =
+                    GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_);
+
+                bs_grid_desc_bk0_n_bk1_ =
+                    GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_);
+
+                ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        ds_grid_desc_m_n_);
+
+                e_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        e_grid_desc_m_n_);
+            }
+
+            // for sanity check of vector memory access
+            for(index_t i = 0; i < NumATensor; ++i)
+            {
+                a_mz_stride_[i] = a_ms_ks_strides[i][NumDimM - 1];
+                a_kz_stride_[i] = a_ms_ks_strides[i][NumDimM + NumDimK - 1];
+            }
+
+            for(index_t i = 0; i < NumBTensor; ++i)
+            {
+                b_nz_stride_[i] = b_ns_ks_strides[i][NumDimN - 1];
+                b_kz_stride_[i] = b_ns_ks_strides[i][NumDimN + NumDimK - 1];
+            }
+
+            for(index_t i = 0; i < NumDTensor; ++i)
+            {
+                ds_nz_stride_[i] = d_ms_ns_strides[i][NumDimM + NumDimN - 1];
+            }
+
+            e_nz_stride_ = e_ms_ns_stride[NumDimM + NumDimN - 1];
+        }
+
+        // pointers
+        typename GridwiseGemm::AsGridPointer p_as_grid_;
+        typename GridwiseGemm::BsGridPointer p_bs_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // tensor descriptors for problem definiton
+        AsGridDesc_M_K as_grid_desc_m_k_;
+        BsGridDesc_N_K bs_grid_desc_n_k_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        // tensor descriptors for block/thread-wise copy
+        AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1_;
+        BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1_;
+        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
+        EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock e_grid_desc_mblock_mperblock_nblock_nperblock_;
+
+        // block-to-e-tile map
+        Block2ETileMap block_2_etile_map_;
+
+        // element-wise op
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        // Strides for the last M/N/K dimensions of A/B/Ds/E
+        //   for sanity check of vector load/store
+        std::array<index_t, NumATensor> a_mz_stride_;
+        std::array<index_t, NumATensor> a_kz_stride_;
+
+        std::array<index_t, NumBTensor> b_nz_stride_;
+        std::array<index_t, NumBTensor> b_kz_stride_;
+
+        std::array<index_t, NumDTensor> ds_nz_stride_;
+        index_t e_nz_stride_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(!GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
+                                            arg.bs_grid_desc_n_k_,
+                                            arg.ds_grid_desc_m_n_,
+                                            arg.e_grid_desc_m_n_,
+                                            arg.block_2_etile_map_))
+            {
+                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_contraction_multiple_abd_xdl_cshuffle<
+                    GridwiseGemm,
+                    typename GridwiseGemm::AsGridPointer,
+                    typename GridwiseGemm::BsGridPointer,
+                    typename GridwiseGemm::DsGridPointer,
+                    EDataType,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    DeviceOp::AsGridDesc_AK0_M_AK1,
+                    DeviceOp::BsGridDesc_BK0_N_BK1,
+                    DeviceOp::DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    DeviceOp::EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    DeviceOp::Block2ETileMap,
+                    has_main_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_as_grid_,
+                                              arg.p_bs_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.as_grid_desc_ak0_m_ak1_,
+                                              arg.bs_grid_desc_bk0_n_bk1_,
+                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.block_2_etile_map_);
+            };
+
+            const auto K = arg.as_grid_desc_m_k_[I0].GetLength(I1);
+
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
+        // check vector load/store
+        {
+            bool all_valid = true;
+
+            static_for<0, NumATensor, 1>{}([&](auto i) {
+                // vector memory access of A: could be on M or AK1 dimension
+                if constexpr(ABlockTransferSrcVectorDim == 1)
+                {
+                    if(!(arg.a_mz_stride_[i] == 1 && arg.as_grid_desc_ak0_m_ak1_[i].GetLength(I1) %
+                                                             ABlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+                else
+                {
+                    if(!(arg.a_kz_stride_[i] == 1 && arg.as_grid_desc_ak0_m_ak1_[i].GetLength(I2) %
+                                                             ABlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+            });
+
+            // vector memory access of B: could be on N or BK1 dimension
+            static_for<0, NumBTensor, 1>{}([&](auto i) {
+                if constexpr(BBlockTransferSrcVectorDim == 1)
+                {
+                    if(!(arg.b_nz_stride_[i] == 1 && arg.bs_grid_desc_bk0_n_bk1_[i].GetLength(I1) %
+                                                             BBlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+                else
+                {
+                    if(!(arg.b_kz_stride_[i] == 1 && arg.bs_grid_desc_bk0_n_bk1_[i].GetLength(I2) %
+                                                             BBlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+            });
+
+            // check vector load of Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                if(!(arg.ds_nz_stride_[i] == 1 &&
+                     arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_[i].GetLength(I3) %
+                             CDEBlockTransferScalarPerVector_NPerBlock ==
+                         0))
+                {
+                    all_valid = false;
+                }
+            });
+
+            // vector memory access of E: always on NPerBlock dimension
+            if(!(arg.e_nz_stride_ == 1 &&
+                 arg.e_grid_desc_mblock_mperblock_nblock_nperblock_.GetLength(I3) %
+                         CDEBlockTransferScalarPerVector_NPerBlock ==
+                     0))
+            {
+                all_valid = false;
+            }
+
+            if(!all_valid)
+            {
+                return false;
+            }
+        }
+
+        return GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
+                                           arg.bs_grid_desc_n_k_,
+                                           arg.ds_grid_desc_m_n_,
+                                           arg.e_grid_desc_m_n_,
+                                           arg.block_2_etile_map_);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::array<const void*, NumATensor> p_as,
+                             std::array<const void*, NumBTensor> p_bs,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
+                             const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
+                             const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
+                             const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
+                             const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
+                             const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
+                             const std::vector<index_t>& e_ms_ns_length,
+                             const std::vector<index_t>& e_ms_ns_stride,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_as,
+                        p_bs,
+                        p_ds,
+                        p_e,
+                        a_ms_ks_lengths,
+                        a_ms_ks_strides,
+                        b_ns_ks_lengths,
+                        b_ns_ks_strides,
+                        d_ms_ns_lengths,
+                        d_ms_ns_strides,
+                        e_ms_ns_length,
+                        e_ms_ns_stride,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
+                        std::array<const void*, NumBTensor> p_bs,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_lengths,
+                        const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_strides,
+                        const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_lengths,
+                        const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_strides,
+                        const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_lengths,
+                        const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_strides,
+                        const std::vector<index_t>& e_ms_ns_length,
+                        const std::vector<index_t>& e_ms_ns_stride,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_as,
+                                          p_bs,
+                                          p_ds,
+                                          p_e,
+                                          as_ms_ks_lengths,
+                                          as_ms_ks_strides,
+                                          bs_ns_ks_lengths,
+                                          bs_ns_ks_strides,
+                                          ds_ms_ns_lengths,
+                                          ds_ms_ns_strides,
+                                          e_ms_ns_length,
+                                          e_ms_ns_stride,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceContractionMultipleABD_Xdl_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
+            << ">"
+            << " LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
index d98cbea166..e7e012c009 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
@@ -428,7 +428,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
             [&](auto i) {
                 using ALayout = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
 
-                return MakeEGridDescriptor_M_N<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
+                return MakeAGridDescriptor_M_K<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
             },
             Number<NumATensor>{});
     }

From 59dbb01fd11e869ff69705f959a0871475858282 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 2 Oct 2023 12:01:11 -0700
Subject: [PATCH 153/194] get rid of gfx900/906, set rocm5.7 as default (#958)

---
 CMakeLists.txt                                        | 4 ++--
 Dockerfile                                            | 2 +-
 example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index f7a9d37bba..b09da41a83 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -106,7 +106,7 @@ message("checking which targets are supported")
 #Setting GPU_TARGETS on command line will override this list
 if(NOT PROFILER_ONLY)
     rocm_check_target_ids(DEFAULT_GPU_TARGETS
-        TARGETS "gfx900;gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
+        TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
 else()
     add_definitions(-DPROFILER_ONLY)
     set(GPU_TARGETS "" CACHE STRING "" FORCE)
@@ -114,7 +114,7 @@ else()
         message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx90, gfx94, gfx10, or gfx11")
     endif()
     if(GPU_ARCH MATCHES "gfx90")
-        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx900;gfx906;gfx908;gfx90a")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx908;gfx90a")
     elseif(GPU_ARCH MATCHES "gfx94")
         rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx940;gfx941;gfx942")
     elseif(GPU_ARCH MATCHES "gfx10")
diff --git a/Dockerfile b/Dockerfile
index 4f227d0362..75ca0c7f9f 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -1,6 +1,6 @@
 FROM ubuntu:20.04
 ARG DEBIAN_FRONTEND=noninteractive
-ARG ROCMVERSION=5.6
+ARG ROCMVERSION=5.7
 ARG compiler_version=""
 ARG compiler_commit=""
 
diff --git a/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
index 6bd7a889b8..90558364a4 100644
--- a/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
+++ b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
@@ -74,7 +74,7 @@ struct AddScale
         a = scale * (a0 + a1);
     }
 
-    //this attribute will force copy_function applying element_wise with vector_type
+    // this attribute will force copy_function applying element_wise with vector_type
     static constexpr ck::index_t vec_len = 4;
 
     float scale = 1.0;

From bd09b5c5389c9f7f51459e5a38d3d7253b0c8dc0 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Mon, 2 Oct 2023 16:39:03 -0500
Subject: [PATCH 154/194] Add fp8 @ bf8 gemm support and example (#933)

* Add f8 bf8 gemm example

* Add element-wise ops

* Add intrinsics

* Update reference calculation

* Add an additional type option for xdlops gemm

* Fix build process

* Add bf8 to buffer addressing

* Update blockwise op, split typeA and typeB

* Update for compatibility

* Uppdate naming to f8->fp8

* Update naming

* Format
---
 example/01_gemm/CMakeLists.txt                |  17 +-
 ..._xdl_fp16_f8.cpp => gemm_xdl_fp16_fp8.cpp} |   0
 .../{gemm_xdl_f8.cpp => gemm_xdl_fp8.cpp}     |   0
 example/01_gemm/gemm_xdl_fp8_bf8.cpp          |  49 ++++++
 .../gpu/block/blockwise_gemm_xdlops.hpp       | 127 ++++++++-------
 .../device/impl/device_gemm_xdl_cshuffle.hpp  |   6 +-
 .../element/unary_element_wise_operation.hpp  |  32 ++++
 ...iple_d_welford_first_half_xdl_cshuffle.hpp |   1 +
 ...wise_batched_gemm_gemm_xdl_cshuffle_v1.hpp |   3 +-
 ...iple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp |   7 +-
 ...ultiple_d_softmax_gemm_xdl_cshuffle_v1.hpp |   3 +-
 ...ched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp |   3 +-
 ...e_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp |   1 +
 ...emm_multiple_d_multiple_r_xdl_cshuffle.hpp |   1 +
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |   1 +
 ...se_gemm_multiple_d_xdl_splitk_cshuffle.hpp |   1 +
 .../gridwise_gemm_reduce_xdl_cshuffle_v1.hpp  |   1 +
 ...e_gemm_split_k_multiple_d_xdl_cshuffle.hpp |   2 +
 .../grid/gridwise_gemm_xdl_cshuffle_v1.hpp    |  24 +--
 ...ridwise_gemm_xdl_layernorm_cshuffle_v1.hpp |   1 +
 ...ridwise_gemm_xdl_waveletmodel_cshuffle.hpp |   1 +
 .../grid/gridwise_gemm_xdlops_bwd_weight.hpp  |   1 +
 .../gpu/grid/gridwise_gemm_xdlops_streamk.hpp |   1 +
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |   2 +
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |   1 +
 .../gpu/grid/gridwise_gemm_xdlops_v3r1.hpp    |   1 +
 .../gpu/grid/gridwise_gemm_xdlops_v3r2.hpp    |   1 +
 .../gpu/grid/gridwise_gemm_xdlops_v3r3.hpp    |   1 +
 .../tensor_operation/gpu/warp/xdlops_gemm.hpp |  98 ++++++++++--
 include/ck/utility/amd_buffer_addressing.hpp  | 147 +++++++++++-------
 include/ck/utility/amd_xdlops.hpp             |  65 ++++++++
 .../cpu/reference_gemm.hpp                    |   7 +-
 .../library/utility/host_tensor_generator.hpp |   4 +-
 .../gpu/grouped_gemm_fixed_nk/CMakeLists.txt  |  24 ++-
 ...ixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp} |   0
 ...ixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp} |   0
 36 files changed, 475 insertions(+), 159 deletions(-)
 rename example/01_gemm/{gemm_xdl_fp16_f8.cpp => gemm_xdl_fp16_fp8.cpp} (100%)
 rename example/01_gemm/{gemm_xdl_f8.cpp => gemm_xdl_fp8.cpp} (100%)
 create mode 100644 example/01_gemm/gemm_xdl_fp8_bf8.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/{device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp => device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp} (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/{device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp => device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp} (100%)

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 2351d3f105..e0124e57f7 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -67,13 +67,20 @@ add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
 
 
 if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-  add_example_executable(example_gemm_xdl_f8 gemm_xdl_f8.cpp)
+  add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
   if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_f8)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
   endif()
 endif()
 
-add_example_executable(example_gemm_xdl_fp16_f8 gemm_xdl_fp16_f8.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_f8)
+if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+  add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
+  if(result EQUAL 0)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
+  endif()
+endif()
+
+add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp)
+if(result EQUAL 0)
+  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
 endif()
diff --git a/example/01_gemm/gemm_xdl_fp16_f8.cpp b/example/01_gemm/gemm_xdl_fp16_fp8.cpp
similarity index 100%
rename from example/01_gemm/gemm_xdl_fp16_f8.cpp
rename to example/01_gemm/gemm_xdl_fp16_fp8.cpp
diff --git a/example/01_gemm/gemm_xdl_f8.cpp b/example/01_gemm/gemm_xdl_fp8.cpp
similarity index 100%
rename from example/01_gemm/gemm_xdl_f8.cpp
rename to example/01_gemm/gemm_xdl_fp8.cpp
diff --git a/example/01_gemm/gemm_xdl_fp8_bf8.cpp b/example/01_gemm/gemm_xdl_fp8_bf8.cpp
new file mode 100644
index 0000000000..0d69b7a90f
--- /dev/null
+++ b/example/01_gemm/gemm_xdl_fp8_bf8.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
+
+using ADataType        = ck::f8_t;
+using BDataType        = ck::bf8_t;
+using CDataType        = ck::f8_t;
+using AccDataType      = float;
+using CShuffleDataType = ck::f8_t;
+
+using ALayout = Row;
+using BLayout = Col;
+using CLayout = Row;
+
+using AElementOp = PassThrough;
+using BElementOp = PassThrough;
+using CElementOp = PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto LoopSched   = ck::make_default_loop_scheduler();
+static constexpr auto PipelineVer = ck::PipelineVersion::v1;
+using ComputeTypeA                = ck::f8_t;
+using ComputeTypeB                = ck::bf8_t;
+
+// clang-format off
+using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
+// ######| ALayout| BLayout| CLayout|     AData|     BData|     CData|     AccData|         CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+// ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+// ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+// ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16,  LoopSched, PipelineVer, ComputeTypeA, ComputeTypeB>;
+// clang-format on
+
+using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                        BDataType,
+                                                                        CDataType,
+                                                                        AccDataType,
+                                                                        AElementOp,
+                                                                        BElementOp,
+                                                                        CElementOp,
+                                                                        ComputeTypeA,
+                                                                        ComputeTypeB>;
+
+#include "run_gemm_example.inc"
+
+int main(int argc, char* argv[]) { return !run_gemm_example(argc, argv); }
diff --git a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
index 1fee9c3225..904a96cc9f 100644
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
@@ -28,7 +28,8 @@ MakeGemmMmaTileDescriptor_MN0_MN1_MN2_K(const TileDesc_K0_MN_K1&)
 }
 
 template <index_t BlockSize,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatAcc,
           typename AK0MK1BlockDesc,
           typename BK0NK1BlockDesc,
@@ -58,7 +59,7 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
     static constexpr index_t A_K1 = AK0MK1BlockDesc{}.GetLength(I2);
     static constexpr index_t B_K1 = BK0NK1BlockDesc{}.GetLength(I2);
 
-    static constexpr auto xdlops_gemm = XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack>{};
+    static constexpr auto xdlops_gemm = XdlopsGemm<FloatA, MPerXDL, NPerXDL, KPack, FloatB>{};
 
     static constexpr index_t KPerThread = KPerBlock / xdlops_gemm.K0PerXdlops;
 
@@ -294,9 +295,9 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                         const BBlockBuffer& b_block_buf,
                         CThreadBuffer& c_thread_buf) const
     {
-        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAB>(
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatA>(
             a_thread_desc_.GetElementSpaceSize());
-        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAB>(
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
             b_thread_desc_.GetElementSpaceSize());
 
         static_for<0, MRepeat, 1>{}([&](auto m0) {
@@ -318,25 +319,27 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                                    b_thread_buf);
 
                 static_for<0, KPerThread, KPack>{}([&](auto k) {
-                    vector_type<FloatAB, KPack> a_thread_vec;
-                    vector_type<FloatAB, KPack> b_thread_vec;
+                    vector_type<FloatA, KPack> a_thread_vec;
+                    vector_type<FloatB, KPack> b_thread_vec;
 
                     static_for<0, KPack, 1>{}([&](auto i) {
-                        a_thread_vec.template AsType<FloatAB>()(i) = a_thread_buf
+                        a_thread_vec.template AsType<FloatA>()(i) = a_thread_buf
                             [Number<a_thread_desc_.CalculateOffset(make_tuple(0, 0, 0, k + i))>{}];
-                        b_thread_vec.template AsType<FloatAB>()(i) = b_thread_buf
+                        b_thread_vec.template AsType<FloatB>()(i) = b_thread_buf
                             [Number<b_thread_desc_.CalculateOffset(make_tuple(0, 0, 0, k + i))>{}];
                     });
 
-                    using mfma_input_type =
-                        typename vector_type<FloatAB, xdlops_gemm.K1PerXdlops>::type;
+                    using mfma_input_type_a =
+                        typename vector_type<FloatA, xdlops_gemm.K1PerXdlops>::type;
+                    using mfma_input_type_b =
+                        typename vector_type<FloatB, xdlops_gemm.K1PerXdlops>::type;
 
                     constexpr index_t c_offset =
                         c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
 
                     xdlops_gemm.template Run(
-                        a_thread_vec.template AsType<mfma_input_type>(),
-                        b_thread_vec.template AsType<mfma_input_type>(),
+                        a_thread_vec.template AsType<mfma_input_type_a>(),
+                        b_thread_vec.template AsType<mfma_input_type_b>(),
                         c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
                 });
             });
@@ -356,8 +359,8 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
     static constexpr auto c_thread_desc_ = make_naive_tensor_descriptor_packed(
         make_tuple(Number<MRepeat>{}, Number<NRepeat>{}, xdlops_gemm.GetRegSizePerXdlops()));
 
-    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatAB,
-                                                         FloatAB,
+    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatA,
+                                                         FloatA,
                                                          decltype(a_block_desc_m0_m1_m2_k),
                                                          decltype(a_thread_desc_),
                                                          Sequence<1, 1, 1, KPerThread>,
@@ -366,8 +369,8 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                                                          A_K1,
                                                          A_K1>;
 
-    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatAB,
-                                                         FloatAB,
+    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatB,
+                                                         FloatB,
                                                          decltype(b_block_desc_n0_n1_n2_k),
                                                          decltype(b_thread_desc_),
                                                          Sequence<1, 1, 1, KPerThread>,
@@ -385,7 +388,8 @@ struct BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
 // the latest ROCm release. For unsupported compilers, inter-wave loop scheduler falls back to the
 // default loop scheduler which is given by the macro CK_EXPERIMENTAL_INTER_WAVE_SCHEDULING=0
 template <index_t BlockSize,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatAcc,
           typename AK0MK1BlockDesc,
           typename BK0NK1BlockDesc,
@@ -397,7 +401,8 @@ template <index_t BlockSize,
           index_t NumMacClusters = CK_EXPERIMENTAL_INTER_WAVE_SCHEDULING_MAC_CLUSTERS>
 struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
     : public BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                 FloatAB,
+                                                                 FloatA,
+                                                                 FloatB,
                                                                  FloatAcc,
                                                                  AK0MK1BlockDesc,
                                                                  BK0NK1BlockDesc,
@@ -408,7 +413,8 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                                                                  KPack>
 {
     using Base = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                     FloatAB,
+                                                                     FloatA,
+                                                                     FloatB,
                                                                      FloatAcc,
                                                                      AK0MK1BlockDesc,
                                                                      BK0NK1BlockDesc,
@@ -440,9 +446,9 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                         const BBlockBuffer& b_block_buf,
                         CThreadBuffer& c_thread_buf) const
     {
-        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAB>(
+        auto a_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatA>(
             a_thread_desc_.GetElementSpaceSize());
-        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatAB>(
+        auto b_thread_buf = make_static_buffer<AddressSpaceEnum::Vgpr, FloatB>(
             b_thread_desc_.GetElementSpaceSize());
 
         static_for<0, KPerThread, KPerInnerLoop>{}([&](auto k) {
@@ -479,20 +485,22 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
             static_for<0, KPerInnerLoop, KPack>{}([&](auto k_) {
                 static_for<0, MRepeat, 1>{}([&](auto m0) {
                     static_for<0, NRepeat, 1>{}([&](auto n0) {
-                        vector_type<FloatAB, KPack> a_thread_vec;
-                        vector_type<FloatAB, KPack> b_thread_vec;
+                        vector_type<FloatA, KPack> a_thread_vec;
+                        vector_type<FloatB, KPack> b_thread_vec;
 
                         static_for<0, KPack, 1>{}([&](auto i) {
-                            a_thread_vec.template AsType<FloatAB>()(i) =
+                            a_thread_vec.template AsType<FloatA>()(i) =
                                 a_thread_buf[Number<a_thread_desc_.CalculateOffset(
                                     make_tuple(m0, 0, 0, k_ + i))>{}];
-                            b_thread_vec.template AsType<FloatAB>()(i) =
+                            b_thread_vec.template AsType<FloatB>()(i) =
                                 b_thread_buf[Number<b_thread_desc_.CalculateOffset(
                                     make_tuple(n0, 0, 0, k_ + i))>{}];
                         });
 
-                        using mfma_input_type =
-                            typename vector_type<FloatAB, xdlops_gemm.K1PerXdlops>::type;
+                        using mfma_input_type_a =
+                            typename vector_type<FloatA, xdlops_gemm.K1PerXdlops>::type;
+                        using mfma_input_type_b =
+                            typename vector_type<FloatB, xdlops_gemm.K1PerXdlops>::type;
 
                         constexpr index_t c_offset =
                             c_thread_desc_.CalculateOffset(make_tuple(m0, n0, 0));
@@ -514,8 +522,8 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                         // TODO: insert setprio in more precise manner since we
                         // could have more than >1 MFMA instructions in single call
                         xdlops_gemm.template Run(
-                            a_thread_vec.template AsType<mfma_input_type>(),
-                            b_thread_vec.template AsType<mfma_input_type>(),
+                            a_thread_vec.template AsType<mfma_input_type_a>(),
+                            b_thread_vec.template AsType<mfma_input_type_b>(),
                             c_thread_buf.GetVectorTypeReference(Number<c_offset>{}));
                         if constexpr(k_.value == 0 && m0.value == 0 && n0.value == 0)
                         {
@@ -541,8 +549,8 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
     static constexpr auto b_thread_desc_ = make_naive_tensor_descriptor_packed(
         make_tuple(Number<NRepeat>{}, I1, I1, Number<KPerInnerLoop>{}));
 
-    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatAB,
-                                                         FloatAB,
+    using AThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatA,
+                                                         FloatA,
                                                          decltype(a_block_desc_m0_m1_m2_k),
                                                          decltype(a_thread_desc_),
                                                          Sequence<1, 1, 1, KPerInnerLoop>,
@@ -551,8 +559,8 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
                                                          A_K1,
                                                          A_K1>;
 
-    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatAB,
-                                                         FloatAB,
+    using BThreadCopy = ThreadwiseTensorSliceTransfer_v4<FloatB,
+                                                         FloatB,
                                                          decltype(b_block_desc_n0_n1_n2_k),
                                                          decltype(b_thread_desc_),
                                                          Sequence<1, 1, 1, KPerInnerLoop>,
@@ -568,7 +576,8 @@ struct BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1
 };
 
 template <index_t BlockSize,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatAcc,
           typename AK0MK1BlockDesc,
           typename BK0NK1BlockDesc,
@@ -583,7 +592,8 @@ constexpr auto BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector()
     if constexpr(LoopSched == LoopScheduler::Default)
     {
         return BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                   FloatAB,
+                                                                   FloatA,
+                                                                   FloatB,
                                                                    FloatAcc,
                                                                    AK0MK1BlockDesc,
                                                                    BK0NK1BlockDesc,
@@ -596,7 +606,8 @@ constexpr auto BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector()
     else if constexpr(LoopSched == LoopScheduler::Interwave)
     {
         return BlockwiseGemmXdlopsInterwave_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                            FloatAB,
+                                                                            FloatA,
+                                                                            FloatB,
                                                                             FloatAcc,
                                                                             AK0MK1BlockDesc,
                                                                             BK0NK1BlockDesc,
@@ -618,26 +629,27 @@ constexpr auto BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector()
  * 3. configurable k index starting position and step size after each FMA/XDL instruction
  */
 
-template <index_t BlockSize,
-          typename FloatAB,
-          typename FloatAcc,
-          typename ATileDesc,
-          typename BTileDesc,
-          typename AMmaTileDesc,
-          typename BMmaTileDesc,
-          index_t MPerBlock,
-          index_t NPerBlock,
-          index_t KPerBlock,
-          index_t MPerXDL,
-          index_t NPerXDL,
-          index_t MRepeat,
-          index_t NRepeat,
-          index_t KPack,
-          bool TransposeC = false,
-          index_t AMmaKStride =
-              KPack* XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack, TransposeC>{}.K0PerXdlops,
-          index_t BMmaKStride =
-              KPack* XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack, TransposeC>{}.K0PerXdlops>
+template <
+    index_t BlockSize,
+    typename FloatAB,
+    typename FloatAcc,
+    typename ATileDesc,
+    typename BTileDesc,
+    typename AMmaTileDesc,
+    typename BMmaTileDesc,
+    index_t MPerBlock,
+    index_t NPerBlock,
+    index_t KPerBlock,
+    index_t MPerXDL,
+    index_t NPerXDL,
+    index_t MRepeat,
+    index_t NRepeat,
+    index_t KPack,
+    bool TransposeC = false,
+    index_t AMmaKStride =
+        KPack* XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack, FloatAB, TransposeC>{}.K0PerXdlops,
+    index_t BMmaKStride =
+        KPack* XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack, FloatAB, TransposeC>{}.K0PerXdlops>
 struct BlockwiseGemmXdlops_v2
 {
     static constexpr auto I0 = Number<0>{};
@@ -654,7 +666,8 @@ struct BlockwiseGemmXdlops_v2
     static constexpr index_t A_K1 = ATileDesc{}.GetLength(I2);
     static constexpr index_t B_K1 = BTileDesc{}.GetLength(I2);
 
-    static constexpr auto xdlops_gemm = XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack, TransposeC>{};
+    static constexpr auto xdlops_gemm =
+        XdlopsGemm<FloatAB, MPerXDL, NPerXDL, KPack, FloatAB, TransposeC>{};
 
     static constexpr index_t KPerThread = KPerBlock / xdlops_gemm.K0PerXdlops;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
index 86bc7b5bbb..693fd7048c 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp
@@ -66,7 +66,8 @@ template <typename ALayout,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched     = make_default_loop_scheduler(),
           PipelineVersion PipelineVer = PipelineVersion::v1,
-          typename ComputeType        = CDataType>
+          typename ComputeTypeA       = CDataType,
+          typename ComputeTypeB       = ComputeTypeA>
 struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
                                                    BLayout,
                                                    CLayout,
@@ -131,7 +132,8 @@ struct DeviceGemm_Xdl_CShuffle : public DeviceGemm<ALayout,
         CShuffleBlockTransferScalarPerVector_NPerBlock,
         LoopSched,
         PipelineVer,
-        ComputeType>;
+        ComputeTypeA,
+        ComputeTypeB>;
 
     using Argument = typename GridwiseGemm::Argument;
 
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index b181b66009..927574dfde 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -156,6 +156,38 @@ struct PassThrough
         y = type_convert<f8_t>(x);
     }
 #endif
+
+#if defined CK_ENABLE_BF8
+    template <>
+    __host__ __device__ void operator()<bf8_t, bf8_t>(bf8_t& y, const bf8_t& x) const
+    {
+        y = x;
+    }
+
+    template <>
+    __host__ __device__ void operator()<float, bf8_t>(float& y, const bf8_t& x) const
+    {
+        y = type_convert<float>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<bf8_t, float>(bf8_t& y, const float& x) const
+    {
+        y = type_convert<bf8_t>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<half_t, bf8_t>(half_t& y, const bf8_t& x) const
+    {
+        y = type_convert<half_t>(x);
+    }
+
+    template <>
+    __host__ __device__ void operator()<bf8_t, half_t>(bf8_t& y, const half_t& x) const
+    {
+        y = type_convert<bf8_t>(x);
+    }
+#endif
 };
 
 struct UnaryConvert
diff --git a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
index b25f136a37..206ea00b9d 100644
--- a/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gemm_layernorm/gridwise_gemm_multiple_d_welford_first_half_xdl_cshuffle.hpp
@@ -522,6 +522,7 @@ struct GridwiseGemmMultipleDWelfordFirstHalf_xdl_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ABDataType,
+            ABDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
index c2f47bd444..9469fa7bc7 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_gemm_xdl_cshuffle_v1.hpp
@@ -628,7 +628,8 @@ struct GridwiseBatchedGemmGemm_Xdl_CShuffle
             Gemm1KPack,
             false,      // TransposeC
             Gemm1KPack, // AMmaKStride
-            Gemm1KPack * XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, false>{}.K0PerXdlops>{
+            Gemm1KPack *
+                XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, FloatAB, false>{}.K0PerXdlops>{
             // BMmaKStride
             make_tuple(0, 0, 0, 0)}; // A_origin
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
index d2920570e4..a0924ae3b0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_gemm_multiple_d_xdl_cshuffle_v1.hpp
@@ -880,7 +880,12 @@ struct GridwiseBatchedGemmMultipleDGemmMultipleD_Xdl_CShuffle
             Gemm1KPack,
             false,      // TransposeC
             Gemm1KPack, // AMmaKStride
-            Gemm1KPack * XdlopsGemm<A0B0B1DataType, Gemm0MPerXdl, Gemm0NPerXdl, Gemm1KPack, false>{}
+            Gemm1KPack * XdlopsGemm<A0B0B1DataType,
+                                    Gemm0MPerXdl,
+                                    Gemm0NPerXdl,
+                                    Gemm1KPack,
+                                    A0B0B1DataType,
+                                    false>{}
                              .K0PerXdlops>{                         // BMmaKStride
                                            make_tuple(0, 0, 0, 0)}; // A_origin
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
index 18cfeebcf3..bc76d4cc4f 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_multiple_d_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -794,7 +794,8 @@ struct GridwiseBatchedGemmMultipleDSoftmaxGemm_Xdl_CShuffle
             Gemm1KPack,
             true,       // TransposeC
             Gemm1KPack, // AMmaKStride
-            Gemm1KPack * XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, false>{}.K0PerXdlops>{
+            Gemm1KPack *
+                XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, FloatAB, false>{}.K0PerXdlops>{
             // BMmaKStride
             make_tuple(0, 0, 0, 0)}; // A_origin
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
index f4b82badf1..afb2ad2e76 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp
@@ -649,7 +649,8 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
             Gemm1KPack,
             true,       // TransposeC
             Gemm1KPack, // AMmaKStride
-            Gemm1KPack * XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, false>{}.K0PerXdlops>{
+            Gemm1KPack *
+                XdlopsGemm<FloatAB, MPerXdl, NPerXdl, Gemm1KPack, FloatAB, false>{}.K0PerXdlops>{
             // BMmaKStride
             make_tuple(0, 0, 0, 0)}; // A_origin
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
index 3ced4b9ad6..9f5df8bdbf 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_bias_add_reduce_xdl_cshuffle_v1.hpp
@@ -504,6 +504,7 @@ struct GridwiseGemmBiasAddReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatAB,
+            FloatAB,
             FloatGemmAcc,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
index d710fc1894..5c9f40b51a 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_multiple_r_xdl_cshuffle.hpp
@@ -470,6 +470,7 @@ struct GridwiseGemmMultipleDMultipleR_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatAB,
+            FloatAB,
             FloatGemmAcc,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index 1d920fb44d..c7ec435477 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -568,6 +568,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ComputeDataType,
+            ComputeDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
index a3343bd3a0..4f37049b20 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_splitk_cshuffle.hpp
@@ -602,6 +602,7 @@ struct GridwiseGemmMultipleD_xdl_splitk_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ComputeType,
+            ComputeType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
index 99e410f688..d75b631e61 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_reduce_xdl_cshuffle_v1.hpp
@@ -457,6 +457,7 @@ struct GridwiseGemmReduce_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatAB,
+            FloatAB,
             FloatGemmAcc,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
index 18cf80041b..e7dc0d3eb0 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_split_k_multiple_d_xdl_cshuffle.hpp
@@ -588,6 +588,7 @@ struct GridwiseGemmSplitKMultipleD_xdl_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ABDataType,
+            ABDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -1012,6 +1013,7 @@ struct GridwiseGemmSplitKMultipleD_xdl_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ABDataType,
+            ABDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
index 9c09f3a539..d700314a26 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_cshuffle_v1.hpp
@@ -108,7 +108,8 @@ template <typename ALayout,
           index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched,
           PipelineVersion PipelineVer = PipelineVersion::v1,
-          typename ComputeType        = FloatC>
+          typename ComputeTypeA       = FloatC,
+          typename ComputeTypeB       = ComputeTypeA>
 struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
 {
     static constexpr auto I0 = Number<0>{};
@@ -547,8 +548,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         constexpr auto c_block_size =
             c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
 
-        return math::max((a_block_space_size_aligned * sizeof(ComputeType) +
-                          b_block_space_size_aligned * sizeof(ComputeType)),
+        return math::max((a_block_space_size_aligned * sizeof(ComputeTypeA) +
+                          b_block_space_size_aligned * sizeof(ComputeTypeB)),
                          c_block_size * sizeof(FloatCShuffle));
     }
 
@@ -750,7 +751,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 FloatA,
-                                                ComputeType,
+                                                ComputeTypeA,
                                                 decltype(a_grid_desc_ak0_m_ak1),
                                                 decltype(a_block_desc_ak0_m_ak1),
                                                 ABlockTransferSrcAccessOrder,
@@ -781,7 +782,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 FloatB,
-                                                ComputeType,
+                                                ComputeTypeB,
                                                 decltype(b_grid_desc_bk0_n_bk1),
                                                 decltype(b_block_desc_bk0_n_bk1),
                                                 BBlockTransferSrcAccessOrder,
@@ -809,13 +810,14 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-        constexpr index_t KPack =
-            math::max(math::lcm(AK1Number, BK1Number),
-                      MfmaSelector<ComputeType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+        constexpr index_t KPack = math::max(
+            math::lcm(AK1Number, BK1Number),
+            MfmaSelector<ComputeTypeA, MPerXdl, NPerXdl, ComputeTypeB>::selected_mfma.k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ComputeType,
+            ComputeTypeA,
+            ComputeTypeB,
             FloatGemmAcc,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -833,10 +835,10 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ComputeType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+            static_cast<ComputeTypeA*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ComputeType*>(p_shared) + a_block_space_size_aligned,
+            static_cast<ComputeTypeB*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1Number, 0, 0);
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
index 0404d88ab8..013120c540 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_layernorm_cshuffle_v1.hpp
@@ -495,6 +495,7 @@ struct GridwiseGemmLayernorm_k0mk1_k0nk1_mn_xdl_cshuffle_v1
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatAB,
+            FloatAB,
             FloatGemmAcc,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
index bbd01a238e..8675a9242a 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdl_waveletmodel_cshuffle.hpp
@@ -494,6 +494,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdl_waveletmodel_cshuffle
             auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<
                 TileMathThreadGroupSize,
                 ABDataType,
+                ABDataType,
                 FloatGemmAcc,
                 decltype(a_block_desc_ak0_m_ak1),
                 decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index 0920a17fc7..8897ce5a10 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -737,6 +737,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatABAdjusted,
                                                                 FloatABAdjusted,
                                                                 FloatAcc,
                                                                 decltype(a_k0_m_k1_block_desc),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
index 70abcac0b1..2b1814c03b 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_streamk.hpp
@@ -490,6 +490,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_streamk
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatAB,
                                                                 FloatAB,
                                                                 FloatAcc,
                                                                 decltype(a_block_desc_k0_m_k1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index e941f96553..54760b0556 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -424,6 +424,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
 
         using BlockwiseGemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatABAdjusted,
                                                                 FloatABAdjusted,
                                                                 FloatAcc,
                                                                 decltype(a_block_desc_k0_m_k1),
@@ -569,6 +570,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             FloatABAdjusted,
+            FloatABAdjusted,
             FloatAcc,
             decltype(a_block_desc_k0_m_k1),
             decltype(b_block_desc_k0_n_k1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 371281dfe2..f9744d7167 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -762,6 +762,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ComputeType,
+            ComputeType,
             FloatAcc,
             decltype(a_k0_m_k1_block_desc),
             decltype(b_k0_n_k1_block_desc),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
index 8d7bd9a8d1..b766b70a67 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r1.hpp
@@ -451,6 +451,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r1
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatAB,
                                                                 FloatAB,
                                                                 FloatAcc,
                                                                 decltype(a_block_desc_ak0_m_ak1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
index 79202cb5cf..fbe4dd409d 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r2.hpp
@@ -471,6 +471,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r2
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatAB,
                                                                 FloatAB,
                                                                 FloatAcc,
                                                                 decltype(a_block_desc_k0_m_k1),
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
index 0d461b4fbf..1dc8d31efe 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v3r3.hpp
@@ -489,6 +489,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v3r3
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
+                                                                FloatAB,
                                                                 FloatAB,
                                                                 FloatAcc,
                                                                 decltype(a_block_desc_k0_m_k1),
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index 9ee07b84ae..06b95aef21 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -31,7 +31,9 @@ enum struct MfmaInstr
     mfma_i32_16x16x32i8,
     mfma_f64_16x16x4f64,
     mfma_f32_32x32x16f8f8,
-    mfma_f32_16x16x32f8f8
+    mfma_f32_16x16x32f8f8,
+    mfma_f32_32x32x16f8bf8,
+    mfma_f32_16x16x32f8bf8
 };
 
 template <MfmaInstr instr>
@@ -502,10 +504,62 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8f8>
 };
 #endif
 
-template <typename base_type, index_t MPerXdlops, index_t NPerXdlops>
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8bf8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_32x32x16f8bf8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8bf8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_16x16x32f8bf8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+#endif
+
+template <typename base_type,
+          index_t MPerXdlops,
+          index_t NPerXdlops,
+          typename additional_type = base_type>
 struct MfmaSelector
 {
-    template <typename base_type_, index_t MPerXdlops_, index_t NPerXdlops_>
+    template <typename base_type_,
+              index_t MPerXdlops_,
+              index_t NPerXdlops_,
+              typename additional_type_ = base_type_>
     static constexpr auto GetMfma();
 
     template <>
@@ -656,7 +710,22 @@ struct MfmaSelector
     }
 #endif
 
-    static constexpr auto selected_mfma = mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops>()>{};
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+    template <>
+    static constexpr auto GetMfma<f8_t, 32, 32, bf8_t>()
+    {
+        return MfmaInstr::mfma_f32_32x32x16f8bf8;
+    }
+
+    template <>
+    static constexpr auto GetMfma<f8_t, 16, 16, bf8_t>()
+    {
+        return MfmaInstr::mfma_f32_16x16x32f8bf8;
+    }
+#endif
+
+    static constexpr auto selected_mfma =
+        mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops, additional_type>()>{};
 
     __host__ __device__ constexpr MfmaSelector()
     {
@@ -703,7 +772,8 @@ template <typename base_type,
           index_t MPerXdlops,
           index_t NPerXdlops,
           index_t KPack,
-          bool TransposeC = false>
+          typename additional_type = base_type,
+          bool TransposeC          = false>
 struct XdlopsGemm
 {
     static constexpr auto I0 = Number<0>{};
@@ -854,14 +924,18 @@ struct XdlopsGemm
     template <class FloatA, class FloatB, class FloatC>
     __device__ void Run(const FloatA& p_a_wave, const FloatB& p_b_wave, FloatC& p_c_thread) const
     {
-        static_assert(is_same<base_type, double>::value || is_same<base_type, float>::value ||
-                          is_same<base_type, half_t>::value || is_same<base_type, bhalf_t>::value ||
-                          is_same<base_type, int8_t>::value
+        static_assert(
+            is_same<base_type, double>::value || is_same<base_type, float>::value ||
+                is_same<base_type, half_t>::value || is_same<base_type, bhalf_t>::value ||
+                is_same<base_type, int8_t>::value
 #if defined CK_ENABLE_FP8
-                          || is_same<base_type, f8_t>::value
+                || is_same<base_type, f8_t>::value
 #endif
-                      ,
-                      "base base_type must be double, float, half, bfloat16, and int8_t!");
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+                || (is_same<base_type, f8_t>::value && is_same<additional_type, bf8_t>::value)
+#endif
+                ,
+            "base base_type must be double, float, half, bfloat16, int8_t, f8_t or bf8_t!");
 
         static_for<0, KPack / mfma_instr.k_per_blk, 1>{}([&](auto k) {
             if constexpr(!TransposeC)
@@ -957,7 +1031,7 @@ struct XdlopsGemm
         return TransposeC ? CIndex4D{blk_td, I0, blk_id, I0} : CIndex4D{I0, blk_id, I0, blk_td};
     }
 
-    static constexpr auto mfma = MfmaSelector<base_type, MPerXdlops, NPerXdlops>{};
+    static constexpr auto mfma = MfmaSelector<base_type, MPerXdlops, NPerXdlops, additional_type>{};
 
     static constexpr auto mfma_instr = mfma.selected_mfma;
 
diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index 694027100f..d8094d84d2 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -1127,37 +1127,53 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 
 #if CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK
     uint32_t src_addr_shift = src_thread_element_valid ? 0 : 0x80000000;
-#if defined CK_ENABLE_FP8
-    if constexpr(is_same<scalar_t, f8_t>::value)
-    {
-        auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
-            src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
-        return bit_cast<vector_t>(tmp);
-    }
-    else
-    {
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+    if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
 #endif
-        return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
-            src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
-#if defined CK_ENABLE_FP8
-    }
+#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
+        if constexpr(is_same<scalar_t, f8_t>::value)
+#endif
+#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+            if constexpr(is_same<scalar_t, bf8_t>::value)
+#endif
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+            {
+                auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
+                    src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+                return bit_cast<vector_t>(tmp);
+            }
+            else
+            {
+#endif
+                return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+                    src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+            }
 #endif
 #else
-#if defined CK_ENABLE_FP8
-    if constexpr(is_same<scalar_t, f8_t>::value)
-    {
-        auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
-            src_wave_buffer_resource, src_thread_addr_offset, 0);
-        return src_thread_element_valid ? bit_cast<vector_t>(tmp) : vector_t(0);
-    }
-    else
-    {
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+    if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
 #endif
-        vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
-            src_wave_buffer_resource, src_thread_addr_offset, 0);
-        return src_thread_element_valid ? tmp : vector_t(0);
-#if defined CK_ENABLE_FP8
-    }
+#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
+        if constexpr(is_same<scalar_t, f8_t>::value)
+#endif
+#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+            if constexpr(is_same<scalar_t, bf8_t>::value)
+#endif
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+            {
+                auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
+                    src_wave_buffer_resource, src_thread_addr_offset, 0);
+                return src_thread_element_valid ? bit_cast<vector_t>(tmp) : vector_t(0);
+            }
+            else
+            {
+#endif
+                vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+                    src_wave_buffer_resource, src_thread_addr_offset, 0);
+                return src_thread_element_valid ? tmp : vector_t(0);
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+            }
 #endif
 #endif
 }
@@ -1216,40 +1232,61 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
 
 #if CK_EXPERIMENTAL_USE_BUFFER_STORE_OOB_CHECK_OFFSET_TRICK
     uint32_t dst_addr_shift = dst_thread_element_valid ? 0 : 0x80000000;
-#if defined CK_ENABLE_FP8
-    if constexpr(is_same<scalar_t, f8_t>::value)
-    {
-        auto tmp =
-            bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(src_thread_data);
-        amd_buffer_store_impl<int8_t, vector_size, coherence>(
-            tmp, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
-    }
-    else
-    {
+
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+    if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
 #endif
-        amd_buffer_store_impl<scalar_t, vector_size, coherence>(
-            src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
-#if defined CK_ENABLE_FP8
-    }
+#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
+        if constexpr(is_same<scalar_t, f8_t>::value)
+#endif
+#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+            if constexpr(is_same<scalar_t, bf8_t>::value)
+#endif
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+            {
+                auto tmp = bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
+                    src_thread_data);
+                amd_buffer_store_impl<int8_t, vector_size, coherence>(
+                    tmp, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
+            }
+            else
+            {
+#endif
+                amd_buffer_store_impl<scalar_t, vector_size, coherence>(src_thread_data,
+                                                                        dst_wave_buffer_resource,
+                                                                        dst_addr_shift +
+                                                                            dst_thread_addr_offset,
+                                                                        0);
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+            }
 #endif
 #else
     if(dst_thread_element_valid)
     {
-#if defined CK_ENABLE_FP8
-        if constexpr(is_same<scalar_t, f8_t>::value)
-        {
-            auto tmp = bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
-                src_thread_data);
-            amd_buffer_store_impl<int8_t, vector_size, coherence>(
-                tmp, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
-        }
-        else
-        {
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+        if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
 #endif
-            amd_buffer_store_impl<scalar_t, vector_size, coherence>(
-                src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
-#if defined CK_ENABLE_FP8
-        }
+#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
+            if constexpr(is_same<scalar_t, f8_t>::value)
+#endif
+#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+                if constexpr(is_same<scalar_t, bf8_t>::value)
+#endif
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+                {
+                    auto tmp =
+                        bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
+                            src_thread_data);
+                    amd_buffer_store_impl<int8_t, vector_size, coherence>(
+                        tmp, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+                }
+                else
+                {
+#endif
+                    amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+                        src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
+#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+                }
 #endif
     }
 #endif
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index a80540515a..3768f92633 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -419,5 +419,70 @@ struct intrin_mfma_f32_16x16x32f8f8<16, 16>
     }
 };
 #endif
+
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_32x32x16f8bf8;
+
+template <>
+struct intrin_mfma_f32_32x32x16f8bf8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const f8x8_t& reg_a, const bf8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_f32_32x32x16_fp8_bf8(
+                bit_cast<long>(reg_a),
+                bit_cast<long>(reg_b),
+                reg_c.template AsType<float16_t>()[Number<0>{}],
+                0,
+                0,
+                0);
+#else
+        vector_type<f8_t, 8> reg_a_v(reg_a);
+        vector_type<bf8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<f8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<bf8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_32x32x2f32<32, 32>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_16x16x32f8bf8;
+
+template <>
+struct intrin_mfma_f32_16x16x32f8bf8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const f8x8_t& reg_a, const bf8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f32_16x16x32_fp8_bf8(
+            bit_cast<long>(reg_a),
+            bit_cast<long>(reg_b),
+            reg_c.template AsType<float4_t>()[Number<0>{}],
+            0,
+            0,
+            0);
+#else
+        vector_type<f8_t, 8> reg_a_v(reg_a);
+        vector_type<bf8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<f8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<bf8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_16x16x4f32<16, 16>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+#endif
 } // namespace ck
 #endif
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
index 95bd1e13d9..6e39dee71c 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_gemm.hpp
@@ -21,7 +21,8 @@ template <typename ADataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CElementwiseOperation,
-          typename ComputType = ADataType>
+          typename ComputeTypeA = ADataType,
+          typename ComputeTypeB = ComputeTypeA>
 struct ReferenceGemm : public device::BaseOperator
 {
     // Argument
@@ -65,8 +66,8 @@ struct ReferenceGemm : public device::BaseOperator
 
                 for(int k = 0; k < K; ++k)
                 {
-                    ComputType v_a;
-                    ComputType v_b;
+                    ComputeTypeA v_a;
+                    ComputeTypeB v_b;
 
                     // use PassThrough instead of ConvertBF16RTN for reference calculation
                     if constexpr(is_same_v<AElementwiseOperation,
diff --git a/library/include/ck/library/utility/host_tensor_generator.hpp b/library/include/ck/library/utility/host_tensor_generator.hpp
index 2fdb0b141d..07d3a35a18 100644
--- a/library/include/ck/library/utility/host_tensor_generator.hpp
+++ b/library/include/ck/library/utility/host_tensor_generator.hpp
@@ -95,7 +95,7 @@ struct GeneratorTensor_2<int8_t>
     }
 };
 
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+#if defined CK_ENABLE_FP8
 template <>
 struct GeneratorTensor_2<ck::f8_t>
 {
@@ -143,7 +143,7 @@ struct GeneratorTensor_3<ck::bhalf_t>
     }
 };
 
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
+#if defined CK_ENABLE_FP8
 template <>
 struct GeneratorTensor_3<ck::f8_t>
 {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
index 45f8130c57..51a42c3d8d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
@@ -1,10 +1,18 @@
-add_instance_library(device_grouped_gemm_fixed_nk_instance
-   device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
-   device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
+set(GROUPED_GEMM_FIXED_NK_INSTANCES)
 
-   device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
-   device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp)
+  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp)
+endif()
 
-   device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
-   device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp
-)
+if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp)
+  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp)
+endif()
+
+if((DTYPES MATCHES "int8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp)
+  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp)
+endif()
+
+add_instance_library(device_grouped_gemm_fixed_nk_instance ${GROUPED_GEMM_FIXED_NK_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_kn_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_f8_f16_mk_nk_mn_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp

From f477fca4361ed1ff7ff7913ba1eb75b7ffe3179d Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 3 Oct 2023 09:32:28 -0500
Subject: [PATCH 155/194] add generic instances (#947)

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp | 5 ++++-
 .../device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp | 5 ++++-
 .../device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp | 5 ++++-
 .../device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp | 5 ++++-
 4 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
index fed2cbbfb9..b6312a6f51 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instance.cpp
@@ -26,7 +26,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances = std::tuple<
@@ -35,6 +35,9 @@ using device_gemm_xdl_splitk_f16_f16_f16_km_kn_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        //Generic instance
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               2>,
+        //
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
index 44ac4c08cd..0b1dcad6f2 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instance.cpp
@@ -26,7 +26,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances = std::tuple<
@@ -35,6 +35,9 @@ using device_gemm_xdl_splitk_f16_f16_f16_km_nk_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        //Generic instance
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              1,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               2>,
+        //
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8>,
         DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8>,
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
index 9d15ccd362..1fcaad1ff7 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instance.cpp
@@ -26,7 +26,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // Compilation parameters for a[k, m] * b[k, n] = c[m, n]
 using device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances = std::tuple<
@@ -35,6 +35,9 @@ using device_gemm_xdl_splitk_f32_f32_f32_km_kn_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        //Generic instance
+        DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   128,     4,  4,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              4,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              4,      true,           1,           1,                   S<1, 32, 1, 8>,               1>,
+        //
         DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              4,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              4,      true,           1,           1,                   S<1, 32, 1, 8>,               4>,
         DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              4,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              4,      true,           1,           1,                   S<1, 32, 1, 8>,               4>,
         DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              4,      true,           1,           1,                   S<1, 16, 1, 8>,               4>,
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
index 4e9ad58742..2b53de60d5 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instance.cpp
@@ -26,7 +26,7 @@ using S = ck::Sequence<Is...>;
 
 using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
-static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
 // Compilation parameters for a[k, m] * b[n, k] = c[m, n]
 using device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances = std::tuple<
@@ -35,6 +35,9 @@ using device_gemm_xdl_splitk_f32_f32_f32_km_nk_mn_instances = std::tuple<
         //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|
         //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|
         //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |
+        //Generic instances
+        DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              4,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              1,              4,      true,           1,           1,                   S<1, 32, 1, 8>,               1>,
+        //
         DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              4,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              4,              4,      true,           1,           1,                   S<1, 32, 1, 8>,               4>,
         DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  4,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              4,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              4,              4,      true,           1,           1,                   S<1, 32, 1, 8>,               4>,
         DeviceGemmXdlSplitKCShuffle<  F32,   F32,   F32,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  4,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              4,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              4,              4,      true,           1,           1,                   S<1, 16, 1, 8>,               4>,

From aa46039f2d25874c799d85f5cd28bc636892deef Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 3 Oct 2023 09:32:44 -0500
Subject: [PATCH 156/194] Fixed contraction issues (#960)

* add missing ComputeType

* fixed

* Update cmake-ck-dev.sh

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp  | 1 +
 .../gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp         | 1 +
 2 files changed, 2 insertions(+)

diff --git a/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
index 6634dacd25..5a274d485e 100644
--- a/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
+++ b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
@@ -291,6 +291,7 @@ int main(int argc, char* argv[])
                                                                       BDataType,
                                                                       CShuffleDataType,
                                                                       AccDataType,
+                                                                      A0DataType,
                                                                       PassThrough,
                                                                       BElementOp>;
 
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
index e7e012c009..6321de73a9 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
@@ -657,6 +657,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ComputeDataType,
+            ComputeDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),

From 5311d1b32579e0427954d0ecba3a44280b75c165 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 3 Oct 2023 09:32:58 -0500
Subject: [PATCH 157/194] changed test for grouped_gemm to be random (#959)

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../grouped_gemm_fixed_nk_bias_fp16.cpp                |  9 ++++-----
 .../22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp     | 10 ++++------
 .../22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp      |  9 ++++-----
 .../22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp       |  9 ++++-----
 .../15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp |  8 +++-----
 .../15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp  |  8 +++-----
 .../15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp   |  8 +++-----
 7 files changed, 25 insertions(+), 36 deletions(-)

diff --git a/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp b/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp
index 3b6dd9a2a9..c758720e10 100644
--- a/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp
+++ b/client_example/21_grouped_gemm_bias/grouped_gemm_fixed_nk_bias_fp16.cpp
@@ -60,14 +60,13 @@ int main()
 
     int sum_of_m = 0;
 
-    Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-
-    int group_count = Ms.size();
+    const int group_count = 16;
 
     for(int i = 0; i < group_count; ++i)
     {
-        Ns.push_back(768);
-        Ks.push_back(4608);
+        Ms.push_back(256 + 256 * i);
+        Ns.push_back(128 + 128 * i);
+        Ks.push_back(128 + 64 * i);
 
         StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
         StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
diff --git a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp
index 3503ae8b24..b16fe90387 100644
--- a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp
+++ b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp16.cpp
@@ -57,15 +57,13 @@ int main()
 
     int sum_of_m = 0;
 
-    // Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-    Ms = {0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0};
-
-    int group_count = Ms.size();
+    const int group_count = 16;
 
     for(int i = 0; i < group_count; ++i)
     {
-        Ns.push_back(768);
-        Ks.push_back(4608);
+        Ms.push_back(256 + 256 * i);
+        Ns.push_back(128 + 128 * i);
+        Ks.push_back(128 + 64 * i);
 
         StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
         StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
diff --git a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp
index b288550b74..045fe47c4f 100644
--- a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp
+++ b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_fp8.cpp
@@ -58,14 +58,13 @@ int main()
 
     int sum_of_m = 0;
 
-    Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-
-    int group_count = Ms.size();
+    const int group_count = 16;
 
     for(int i = 0; i < group_count; ++i)
     {
-        Ns.push_back(768);
-        Ks.push_back(4608);
+        Ms.push_back(256 + 256 * i);
+        Ns.push_back(128 + 128 * i);
+        Ks.push_back(128 + 64 * i);
 
         StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
         StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
diff --git a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp
index c60daa3b36..8f82140f3f 100644
--- a/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp
+++ b/client_example/22_grouped_gemm/grouped_gemm_fixed_nk_i8.cpp
@@ -58,14 +58,13 @@ int main()
 
     int sum_of_m = 0;
 
-    Ms = {167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-
-    int group_count = Ms.size();
+    const int group_count = 16;
 
     for(int i = 0; i < group_count; ++i)
     {
-        Ns.push_back(768);
-        Ks.push_back(4608);
+        Ms.push_back(256 + 256 * i);
+        Ns.push_back(128 + 128 * i);
+        Ks.push_back(128 + 64 * i);
 
         StrideAs.push_back(std::is_same<Row, ALayout>::value ? Ks[i] : Ms[i]);
         StrideBs.push_back(std::is_same<Row, BLayout>::value ? Ns[i] : Ks[i]);
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp
index 89d4789c12..95b8526094 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp16.cpp
@@ -296,13 +296,11 @@ int main(int argc, char* argv[])
 
     problem_size.group_count = 16;
 
-    problem_size.Ms = {
-        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-
     for(int i = 0; i < problem_size.group_count; i++)
     {
-        problem_size.Ns.push_back(768);
-        problem_size.Ks.push_back(4608);
+        problem_size.Ms.push_back(256 + 256 * i);
+        problem_size.Ns.push_back(128 + 128 * i);
+        problem_size.Ks.push_back(128 + 64 * i);
 
         problem_size.stride_As.push_back(problem_size.Ks[i]);
         problem_size.stride_Bs.push_back(problem_size.Ks[i]);
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp
index 1c50dc051b..84abe1d1db 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_fixed_nk_fp8.cpp
@@ -297,13 +297,11 @@ int main(int argc, char* argv[])
 
     problem_size.group_count = 16;
 
-    problem_size.Ms = {
-        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-
     for(int i = 0; i < problem_size.group_count; i++)
     {
-        problem_size.Ns.push_back(768);
-        problem_size.Ks.push_back(4608);
+        problem_size.Ms.push_back(256 + 256 * i);
+        problem_size.Ns.push_back(128 + 128 * i);
+        problem_size.Ks.push_back(128 + 64 * i);
 
         problem_size.stride_As.push_back(problem_size.Ks[i]);
         problem_size.stride_Bs.push_back(problem_size.Ks[i]);
diff --git a/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
index 743ab96be6..9f8f6cb1e4 100644
--- a/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
+++ b/example/15_grouped_gemm/grouped_gemm_xdl_splitk_fp16.cpp
@@ -66,13 +66,11 @@ int main(int argc, char* argv[])
 
     problem_size.group_count = 16;
 
-    problem_size.Ms = {
-        167, 183, 177, 181, 153, 139, 156, 173, 163, 150, 204, 184, 168, 156, 168, 148};
-
     for(int i = 0; i < problem_size.group_count; i++)
     {
-        problem_size.Ns.push_back(768);
-        problem_size.Ks.push_back(4608);
+        problem_size.Ms.push_back(256 + 256 * i);
+        problem_size.Ns.push_back(128 + 128 * i);
+        problem_size.Ks.push_back(128 + 64 * i);
 
         problem_size.stride_As.push_back(problem_size.Ks[i]);
         problem_size.stride_Bs.push_back(problem_size.Ks[i]);

From e921e1f08dc04bc4bdf8a1efeb2c1623ff336a6d Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 3 Oct 2023 20:04:26 -0500
Subject: [PATCH 158/194] 3d grouped conv fwd with input/output fp16 and comp
 fp8 (#931)

* add f8 comp instance

* fixed

* fixed comments

* rename

* fixed dtype

* format

* fixed CI

* fixed ci

* add missing ComputeType

* fixed cit

* fixed

* Update cmake-ck-dev.sh

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 client_example/16_convnd_fwd/CMakeLists.txt   |  18 +-
 client_example/16_convnd_fwd/common.hpp       |   6 +-
 .../conv3d_fwd_fp16_comp_fp8.cpp              |  46 ++
 .../device_grouped_conv_fwd_multiple_d.hpp    |   3 +-
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |   8 +-
 .../device_grouped_conv_fwd_xdl_instance.hpp  |  40 ++
 .../gpu/grouped_convolution_forward.hpp       | 495 +++++++++++-------
 .../gpu/grouped_conv3d_fwd/CMakeLists.txt     |   2 +
 ...gc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp |  56 ++
 9 files changed, 472 insertions(+), 202 deletions(-)
 create mode 100644 client_example/16_convnd_fwd/conv3d_fwd_fp16_comp_fp8.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp

diff --git a/client_example/16_convnd_fwd/CMakeLists.txt b/client_example/16_convnd_fwd/CMakeLists.txt
index e2580a370c..249c2c030f 100644
--- a/client_example/16_convnd_fwd/CMakeLists.txt
+++ b/client_example/16_convnd_fwd/CMakeLists.txt
@@ -1,5 +1,15 @@
-add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp)
-add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
+if((DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+    add_executable(client_conv3d_fwd_fp16 conv3d_fwd_fp16.cpp)
+    target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_operations)
 
-target_link_libraries(client_conv3d_fwd_fp16 PRIVATE composable_kernel::device_operations)
-target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_operations)
+endif()
+
+if((DTYPES MATCHES "fp8") OR NOT DEFINED DTYPES)
+    add_executable(client_conv3d_fwd_fp16_comp_fp8 conv3d_fwd_fp16_comp_fp8.cpp)
+    target_link_libraries(client_conv3d_fwd_fp16_comp_fp8 PRIVATE composable_kernel::device_operations)
+endif()
+
+if((DTYPES MATCHES "fp32") OR NOT DEFINED DTYPES)
+    add_executable(client_conv3d_fwd_fp32 conv3d_fwd_fp32.cpp)
+    target_link_libraries(client_conv3d_fwd_fp32 PRIVATE composable_kernel::device_operations)
+endif()
diff --git a/client_example/16_convnd_fwd/common.hpp b/client_example/16_convnd_fwd/common.hpp
index 449c9466e8..50c03c49d8 100644
--- a/client_example/16_convnd_fwd/common.hpp
+++ b/client_example/16_convnd_fwd/common.hpp
@@ -94,7 +94,8 @@ template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
           typename OutLayout,
-          ck::index_t NumNonSpatialDim = 3>
+          ck::index_t NumNonSpatialDim = 3,
+          typename ComputeType         = InDataType>
 bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> in_lengths,
                           std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> wei_lengths,
                           std::array<ck::index_t, NumDimSpatial + NumNonSpatialDim> out_lengths)
@@ -184,7 +185,8 @@ bool run_grouped_conv_fwd(std::array<ck::index_t, NumDimSpatial + NumNonSpatialD
                                                                                  OutDataType,
                                                                                  PassThrough,
                                                                                  PassThrough,
-                                                                                 PassThrough>;
+                                                                                 PassThrough,
+                                                                                 ComputeType>;
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
         DeviceOp>::GetInstances();
diff --git a/client_example/16_convnd_fwd/conv3d_fwd_fp16_comp_fp8.cpp b/client_example/16_convnd_fwd/conv3d_fwd_fp16_comp_fp8.cpp
new file mode 100644
index 0000000000..1651ec2f39
--- /dev/null
+++ b/client_example/16_convnd_fwd/conv3d_fwd_fp16_comp_fp8.cpp
@@ -0,0 +1,46 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout  = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::NDHWGK;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t N             = 64;
+static constexpr ck::index_t K             = 128;
+static constexpr ck::index_t C             = 64;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 28;
+static constexpr ck::index_t Hi            = 28;
+static constexpr ck::index_t Wi            = 3;
+static constexpr ck::index_t Do            = 28;
+static constexpr ck::index_t Ho            = 28;
+static constexpr ck::index_t Wo            = 3;
+
+int main()
+{
+    return run_grouped_conv_fwd<NumDimSpatial,
+                                InDataType,
+                                WeiDataType,
+                                OutDataType,
+                                InLayout,
+                                WeiLayout,
+                                OutLayout,
+                                3,
+                                ck::f8_t>(
+               {N, Di, Hi, Wi, G, C}, {G, K, Z, Y, X, C}, {N, Do, Ho, Wo, G, K})
+               ? EXIT_SUCCESS
+               : EXIT_FAILURE;
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
index 1cc30fd9e6..2ca82dc6da 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d.hpp
@@ -29,7 +29,8 @@ template <index_t NDimSpatial,
           typename EDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CDEElementwiseOperation>
+          typename CDEElementwiseOperation,
+          typename ComputeType = ADataType>
 struct DeviceGroupedConvFwdMultipleD : public BaseOperator
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index b04594fceb..f4b8d66ecf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -211,7 +211,8 @@ template <index_t NDimSpatial,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CDEBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched = make_default_loop_scheduler()>
+          typename ComputeDataType = ADataType,
+          LoopScheduler LoopSched  = make_default_loop_scheduler()>
 struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
     : public DeviceGroupedConvFwdMultipleD<NDimSpatial,
                                            ALayout,
@@ -224,7 +225,8 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
                                            EDataType,
                                            AElementwiseOperation,
                                            BElementwiseOperation,
-                                           CDEElementwiseOperation>
+                                           CDEElementwiseOperation,
+                                           ComputeDataType>
 {
     using DeviceOp = DeviceGroupedConvFwdMultipleD_Xdl_CShuffle;
 
@@ -323,8 +325,6 @@ struct DeviceGroupedConvFwdMultipleD_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
     using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N<ELayout>({}, {}))>;
 
-    using ComputeDataType = ADataType;
-
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
index 23edf35e98..17bb4256ac 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp
@@ -13,6 +13,10 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+#ifdef CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
@@ -174,6 +178,42 @@ using device_grouped_conv_fwd_xdl_int8_instances = std::tuple<
     // clang-format on
     >;
 
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionForwardSpecialization ConvSpec>
+using device_grouped_conv_fwd_xdl_f16_comp_f8_instances = std::tuple<
+// clang-format off
+        //########################################|     NumDim|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|          Ds| EData|           A|           B|         CDE|    ConvForward|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer| ComputeType|
+        //########################################|    Spatial| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|    DataType|  Type| Elementwise| Elementwise| Elementwise| Specialization| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|            |
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|            |
+        //########################################|           |       |       |            |       |      |      |        |         |            |      |            |            |            |               |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |            |
+#ifdef CK_ENABLE_FP8
+        // generic instance
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               1,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>,
+        DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<NDimSpatial,ALayout,BLayout,    DsLayout,ELayout,   F16,   F16,     F32,      F16,    DsLayout,   F16, PassThrough, PassThrough, PassThrough,       ConvSpec, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8,         F8>
+#endif
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
index e6dbd349dd..888c00f900 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_forward.hpp
@@ -16,6 +16,7 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 namespace instance {
+
 #ifdef CK_ENABLE_BF16
 // grouped conv1d forward, GNWC/GKXC/GNWK
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
@@ -32,6 +33,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
@@ -47,6 +49,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
@@ -62,6 +65,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<1,
@@ -77,6 +81,71 @@ void add_device_grouped_conv1d_fwd_xdl_gnwc_gkxc_gnwk_int8_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              int8_t,
+                                                              int8_t,
+                                                              Empty_Tuple,
+                                                              int8_t,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
 #ifdef CK_ENABLE_BF16
 // grouped conv2d forward, GNHWC/GKYXC/GNHWK
 void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
@@ -93,6 +162,7 @@ void add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -108,6 +178,7 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -123,38 +194,7 @@ void add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#ifdef DL_KERNELS
-#ifdef CK_ENABLE_FP16
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-#endif
-#ifdef CK_ENABLE_FP32
-void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-#endif
-#endif
+
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -169,65 +209,8 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
-
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
-void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              GNHWC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              GNHWK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
-#ifdef DL_KERNELS
-void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F16,
-                                                              F16,
-                                                              Empty_Tuple,
-                                                              F16,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-#endif
 #endif
+
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -285,22 +268,7 @@ void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_oddc_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
-void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              F32,
-                                                              F32,
-                                                              Empty_Tuple,
-                                                              F32,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
 
-#endif
 // grouped conv2d forward, NHWGC/GKYXC/NHWGK
 #ifdef CK_ENABLE_BF16
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
@@ -317,6 +285,7 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -388,63 +357,7 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
-#ifdef CK_ENABLE_INT8
-void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
 
-void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
-void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_1x1s1p0_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-
-void add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_i8_oddc_instances(
-    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
-                                                              NHWGC,
-                                                              GKYXC,
-                                                              Empty_Tuple,
-                                                              NHWGK,
-                                                              int8_t,
-                                                              int8_t,
-                                                              Empty_Tuple,
-                                                              int8_t,
-                                                              PassThrough,
-                                                              PassThrough,
-                                                              PassThrough>>>& instances);
-#endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
@@ -460,6 +373,7 @@ void add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_BF16
 // grouped conv3d forward, GNDHWC/GKZYXC/GNDHWK
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
@@ -476,6 +390,7 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -547,6 +462,7 @@ void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -562,6 +478,7 @@ void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -633,6 +550,7 @@ void add_device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_BF16
 // grouped conv3d forward, NDHWGC/GKZYXC/NDHWGK
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
@@ -649,6 +567,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -663,7 +582,9 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -677,7 +598,9 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -691,7 +614,9 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instances
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -705,7 +630,9 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instanc
                                                               PassThrough,
                                                               PassThrough,
                                                               PassThrough>>>& instances);
+#endif
 
+#ifdef CK_ENABLE_FP16
 void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
                                                               NDHWGC,
@@ -720,6 +647,88 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP16
+void add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#ifdef CK_ENABLE_FP8
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              F8>>>& instances);
+#endif
+
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -735,6 +744,7 @@ void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
                                                               PassThrough,
                                                               PassThrough>>>& instances);
 #endif
+
 #ifdef CK_ENABLE_INT8
 void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
@@ -807,13 +817,79 @@ void add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instances(
                                                               PassThrough>>>& instances);
 #endif
 
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+
+#endif
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              NHWGC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              NHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+void add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<2,
+                                                              GNHWC,
+                                                              GKYXC,
+                                                              Empty_Tuple,
+                                                              GNHWK,
+                                                              F32,
+                                                              F32,
+                                                              Empty_Tuple,
+                                                              F32,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough>>>& instances);
+#endif
+
 template <ck::index_t NumDimSpatial,
           typename InLayout,
           typename WeiLayout,
           typename OutLayout,
           typename InDataType,
           typename WeiDataType,
-          typename OutDataType>
+          typename OutDataType,
+          typename ComputeType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD<
     NumDimSpatial,
     InLayout,
@@ -826,7 +902,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     OutDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>>
+    ck::tensor_operation::element_wise::PassThrough,
+    ComputeType>>
 {
     using DeviceOp = DeviceGroupedConvFwdMultipleD<NumDimSpatial,
                                                    InLayout,
@@ -839,7 +916,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                                    OutDataType,
                                                    ck::tensor_operation::element_wise::PassThrough,
                                                    ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough>;
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ComputeType>;
 
     static auto GetInstances()
     {
@@ -877,33 +955,46 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
-                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
+
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, GNHWC> &&
+                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, GNHWK>)
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
-#ifdef DL_KERNELS
-                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
-#endif
             }
 #endif
+
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f32_instances(op_ptrs);
+            }
+#endif
+
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
-#ifdef DL_KERNELS
-                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
-#endif
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1p0_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_1x1s1p0_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_f16_oddc_instances(op_ptrs);
             }
 #endif
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
+            {
+                add_device_grouped_conv2d_fwd_dl_gnhwc_gkyxc_gnhwk_f16_instances(op_ptrs);
+            }
+#endif
+
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
@@ -911,9 +1002,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 add_device_grouped_conv1d_fwd_xdl_gnhwc_gkyxc_gnhwk_bf16_instances(op_ptrs);
             }
 #endif
+
 #ifdef CK_ENABLE_INT8
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_instances(op_ptrs);
                 add_device_grouped_conv2d_fwd_wmma_gnhwc_gkyxc_gnhwk_i8_1x1p0_instances(op_ptrs);
@@ -922,33 +1014,43 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
-                          is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
+
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWGC> &&
+                     is_same_v<WeiLayout, GKYXC> && is_same_v<OutLayout, NHWGK>)
         {
+
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
-#ifdef DL_KERNELS
-                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
-#endif
             }
 #endif
+
+#if(defined(CK_ENABLE_FP32) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f32_instances(op_ptrs);
+            }
+#endif
+
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
             {
                 add_device_grouped_conv2d_fwd_xdl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
-#ifdef DL_KERNELS
-                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
-#endif
-                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
-                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1p0_instances(op_ptrs);
-                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_1x1s1p0_instances(op_ptrs);
-                add_device_grouped_conv2d_fwd_wmma_nhwgc_gkyxc_nhwgk_f16_oddc_instances(op_ptrs);
             }
 #endif
+
+#if(defined(CK_ENABLE_FP16) && defined(DL_KERNELS))
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
+            {
+                add_device_grouped_conv2d_fwd_dl_nhwgc_gkyxc_nhwgk_f16_instances(op_ptrs);
+            }
+#endif
+
 #ifdef CK_ENABLE_BF16
             if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                          is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
@@ -967,8 +1069,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
-                          is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
+
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, GNDHWC> &&
+                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, GNDHWK>)
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
@@ -1010,8 +1113,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
-                          is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
+
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWGC> &&
+                     is_same_v<WeiLayout, GKZYXC> && is_same_v<OutLayout, NDHWGK>)
         {
 #ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
@@ -1020,9 +1124,18 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(op_ptrs);
             }
 #endif
+
+#ifdef CK_ENABLE_FP8
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, ck::f8_t>)
+            {
+                add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances(
+                    op_ptrs);
+            }
+#endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                         is_same_v<OutDataType, half_t>)
+                         is_same_v<OutDataType, half_t> && is_same_v<ComputeType, half_t>)
             {
                 add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
                 add_device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(op_ptrs);
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
index 728e2d1c34..c3cc4cb054 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
@@ -4,10 +4,12 @@ add_instance_library(device_grouped_conv3d_fwd_instance
    xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
 
+
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
+   xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
 
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
new file mode 100644
index 0000000000..431dfed50a
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<3,
+                                                              NDHWGC,
+                                                              GKZYXC,
+                                                              Empty_Tuple,
+                                                              NDHWGK,
+                                                              F16,
+                                                              F16,
+                                                              Empty_Tuple,
+                                                              F16,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              PassThrough,
+                                                              F8>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_f16_comp_f8_instances<3,
+                                                          NDHWGC,
+                                                          GKZYXC,
+                                                          Empty_Tuple,
+                                                          NDHWGK,
+                                                          ConvFwdDefault>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_f16_comp_f8_instances<3,
+                                                          NDHWGC,
+                                                          GKZYXC,
+                                                          Empty_Tuple,
+                                                          NDHWGK,
+                                                          ConvFwd1x1P0>{});
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_fwd_xdl_f16_comp_f8_instances<3,
+                                                          NDHWGC,
+                                                          GKZYXC,
+                                                          Empty_Tuple,
+                                                          NDHWGK,
+                                                          ConvFwd1x1S1P0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 42facfc6b7433595e052c6c0b53138dae84530d7 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Wed, 4 Oct 2023 08:19:08 -0500
Subject: [PATCH 159/194] Add conv bwd weight fp16 comp bf8 fp8 op, instances
 and example (#945)

* Add f8 bf8 gemm example

* Add element-wise ops

* Add intrinsics

* Update reference calculation

* Add an additional type option for xdlops gemm

* Fix build process

* Add bf8 to buffer addressing

* Update blockwise op, split typeA and typeB

* Update for compatibility

* Uppdate naming to f8->fp8

* Update naming

* Format

* Update naming (#937)

* Add a client example

* Add computetypes to device and gridwise ops

* Add instances, update instance factory

* Format

* Fix a flag

* Add ckProfiler mode

* Fix typos

* Add an example

* Add bf8 generator

* add bf8 mfma; fixed type_convert for bf8

* move verfication ahead of timing

* Update reference calculation

* Fix reference

* Narrow down float init range

* Fix bf8 bf8 mfma

* Add bf8 @ fp8 mfma

* Update example

* Update instances

* Update profiler api

* Update for compatibility

* Format

* Remove extra example

* Clean up

* workaround convert

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../20_grouped_conv_bwd_weight/CMakeLists.txt |   6 +
 example/20_grouped_conv_bwd_weight/common.hpp |   6 +
 .../grouped_conv_bwd_weight_dl_fp16.cpp       |   9 ++
 .../grouped_conv_bwd_weight_xdl_bf16.cpp      |   9 ++
 .../grouped_conv_bwd_weight_xdl_fp16.cpp      |   9 ++
 ..._conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp |  86 ++++++++++++
 .../run_grouped_conv_bwd_weight_example.inc   |  39 +++---
 .../device/device_grouped_conv_bwd_weight.hpp |   4 +-
 ...e_grouped_conv_bwd_weight_xdl_cshuffle.hpp |  33 +++--
 .../element/unary_element_wise_operation.hpp  |   3 +-
 .../grid/gridwise_gemm_xdlops_bwd_weight.hpp  |  72 ++++++----
 .../tensor_operation/gpu/warp/xdlops_gemm.hpp | 132 +++++++++++++++++-
 include/ck/utility/amd_xdlops.hpp             | 130 +++++++++++++++++
 include/ck/utility/type_convert.hpp           |   2 +-
 .../cpu/reference_conv_bwd_weight.hpp         |  21 +--
 ...e_grouped_conv_bwd_weight_xdl_instance.hpp |  45 ++++++
 .../grouped_convolution_backward_weight.hpp   |  40 +++++-
 .../library/utility/host_tensor_generator.hpp |  35 +++++
 .../grouped_conv3d_bwd_weight/CMakeLists.txt  |   3 +-
 ...kzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp |  49 +++++++
 .../profile_grouped_conv_bwd_weight_impl.hpp  |   8 +-
 .../src/profile_grouped_conv_bwd_weight.cpp   |  61 +++++---
 22 files changed, 696 insertions(+), 106 deletions(-)
 create mode 100644 example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp

diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index ab8ed5c2bf..c83b384a01 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -11,6 +11,12 @@ foreach(gpu IN LISTS GPU_TARGETS)
    if(result EQUAL 0)
     add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
    endif()
+   if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+    add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
+    if(result EQUAL 0)
+      add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
+    endif()
+   endif()
    set(target 1)
  endif()
 endforeach()
diff --git a/example/20_grouped_conv_bwd_weight/common.hpp b/example/20_grouped_conv_bwd_weight/common.hpp
index 15727495f0..0e7eb00e4b 100644
--- a/example/20_grouped_conv_bwd_weight/common.hpp
+++ b/example/20_grouped_conv_bwd_weight/common.hpp
@@ -23,6 +23,12 @@
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
+#ifdef CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+#ifdef CK_ENABLE_BF8
+using BF8 = ck::bf8_t;
+#endif
 
 template <ck::index_t... Is>
 using S = ck::Sequence<Is...>;
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
index cb6d68e04a..dccde28aa0 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
@@ -65,6 +65,15 @@ using DeviceConvBwdWeightInstance = ck::tensor_operation::device::DeviceGroupedC
     5,                    // CThreadTransferSrcDstVectorDim
     4>;                   // CThreadTransferDstScalarPerVector
 
+template <ck::index_t NDimSpatial>
+using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
+                                                                                     InDataType,
+                                                                                     WeiDataType,
+                                                                                     OutDataType,
+                                                                                     InElementOp,
+                                                                                     WeiElementOp,
+                                                                                     OutElementOp>;
+
 #include "run_grouped_conv_bwd_weight_example.inc"
 
 int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
index 31e277b5c7..467209ef28 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
@@ -67,6 +67,15 @@ using DeviceConvBwdWeightInstance =
         S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
         128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
 
+template <ck::index_t NDimSpatial>
+using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
+                                                                                     InDataType,
+                                                                                     WeiDataType,
+                                                                                     OutDataType,
+                                                                                     InElementOp,
+                                                                                     WeiElementOp,
+                                                                                     OutElementOp>;
+
 #include "run_grouped_conv_bwd_weight_example.inc"
 
 int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
index 69c831cc54..8b751b0d98 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
@@ -66,6 +66,15 @@ using DeviceConvBwdWeightInstance =
         S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
         128 / (sizeof(WeiDataType) * CHAR_BIT)>; // CBlockTransferScalarPerVector_NWaveNPerXdl
 
+template <ck::index_t NDimSpatial>
+using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
+                                                                                     InDataType,
+                                                                                     WeiDataType,
+                                                                                     OutDataType,
+                                                                                     InElementOp,
+                                                                                     WeiElementOp,
+                                                                                     OutElementOp>;
+
 #include "run_grouped_conv_bwd_weight_example.inc"
 
 int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
new file mode 100644
index 0000000000..c678867e74
--- /dev/null
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp"
+
+using InDataType   = F16;
+using WeiDataType  = F16;
+using OutDataType  = F16;
+using AccDataType  = F32;
+using ComputeTypeA = BF8;
+using ComputeTypeB = F8;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = PassThrough;
+
+template <ck::index_t NDimSpatial>
+using DeviceConvBwdWeightInstance =
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Xdl_CShuffle<
+        NDimSpatial,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                      ck::tensor_layout::convolution::GNHWC,
+                                      ck::tensor_layout::convolution::GNDHWC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                      ck::tensor_layout::convolution::GKYXC,
+                                      ck::tensor_layout::convolution::GKZYXC>>,
+        ck::tuple_element_t<NDimSpatial - 1,
+                            ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                      ck::tensor_layout::convolution::GNHWK,
+                                      ck::tensor_layout::convolution::GNDHWK>>,
+        InDataType,           // InDataType
+        WeiDataType,          // WeiDataType
+        OutDataType,          // OutDataType
+        AccDataType,          // AccDataType
+        InElementOp,          // InElementwiseOperation
+        WeiElementOp,         // WeiElementwiseOperation
+        OutElementOp,         // OutElementwiseOperation
+        ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
+        256,                  // BlockSize
+        128,                  // MPerBlock
+        128,                  // NPerBlock
+        4,                    // K0PerBlock
+        8,                    // K1
+        32,                   // MPerXdl
+        32,                   // NPerXdl
+        2,                    // MXdlPerWave
+        2,                    // NXdlPerWave
+        S<1, 4, 16, 4>,       // ABlockTransferThreadClusterLengths_K0_M_K1
+        S<0, 3, 1, 2>,        // ABlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1, 3>,        // ABlockTransferSrcAccessOrder
+        2,                    // ABlockTransferSrcVectorDim
+        1,                    // ABlockTransferSrcScalarPerVector
+        1,                    // ABlockTransferDstScalarPerVector_K1
+        true,                 // ABlockLdsAddExtraM
+        S<1, 4, 16, 4>,       // BBlockTransferThreadClusterLengths_K0_N_K1
+        S<0, 3, 1, 2>,        // BBlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1, 3>,        // BBlockTransferSrcAccessOrder
+        2,                    // BBlockTransferSrcVectorDim
+        1,                    // BBlockTransferSrcScalarPerVector
+        1,                    // BBlockTransferDstScalarPerVector_K1
+        true,                 // BBlockLdsAddExtraN
+        1,                    // CShuffleMXdlPerWavePerShuffle
+        1,                    // CShuffleNXdlPerWavePerShuffle
+        S<1, 32, 1, 4>,       // CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock
+        2,                    // CBlockTransferScalarPerVector_NWaveNPerXdl
+        ComputeTypeA,         // ComputeTypeA
+        ComputeTypeB>;        // ComputeTypeB
+
+template <ck::index_t NDimSpatial>
+using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
+                                                                                     InDataType,
+                                                                                     WeiDataType,
+                                                                                     OutDataType,
+                                                                                     InElementOp,
+                                                                                     WeiElementOp,
+                                                                                     OutElementOp,
+                                                                                     ComputeTypeA,
+                                                                                     ComputeTypeB>;
+
+#include "run_grouped_conv_bwd_weight_example.inc"
+
+int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 34bca7dc86..755d6addd8 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -1,15 +1,6 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-template <ck::index_t NDimSpatial>
-using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
-                                                                                     InDataType,
-                                                                                     WeiDataType,
-                                                                                     OutDataType,
-                                                                                     InElementOp,
-                                                                                     WeiElementOp,
-                                                                                     OutElementOp>;
-
 template <ck::index_t NDimSpatial>
 bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
                                  const ck::utils::conv::ConvParam& conv_param)
@@ -46,8 +37,8 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
         out.GenerateTensorValue(GeneratorTensor_2<OutDataType>{-5, 5});
         break;
     default:
-        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 1.0});
-        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{-0.5, 0.5});
+        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{0.0, 0.2});
+        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{-0.1, 0.1});
     }
 
     DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
@@ -113,18 +104,7 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
         return true;
     }
 
-    float avg_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
-
-    std::size_t flop      = conv_param.GetFlops();
-    std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
-
-    float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
-
-    float gb_per_sec = num_btype / 1.E6 / avg_time;
-
-    std::cerr << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
-              << std::endl
-              << "DeviceOp: " << conv.GetTypeString() << std::endl;
+    invoker.Run(argument, StreamConfig{nullptr, false});
 
     if(config.do_verification)
     {
@@ -148,6 +128,19 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
         return ck::utils::check_err(wei_device_result.mData, wei_host_result.mData);
     }
 
+    float avg_time = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
+
+    std::size_t flop      = conv_param.GetFlops();
+    std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+    float tflops = static_cast<float>(flop) / 1.E9 / avg_time;
+
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+
+    std::cerr << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl
+              << "DeviceOp: " << conv.GetTypeString() << std::endl;
+
     return true;
 }
 
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
index ab9e6adb41..7296e4faaa 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp
@@ -20,7 +20,9 @@ template <ck::index_t NDimSpatial,
           typename OutDataType,
           typename InElementwiseOperation,
           typename WeiElementwiseOperation,
-          typename OutElementwiseOperation>
+          typename OutElementwiseOperation,
+          typename ComputeTypeA = InDataType,
+          typename ComputeTypeB = ComputeTypeA>
 struct DeviceGroupedConvBwdWeight : public BaseOperator
 {
     virtual std::unique_ptr<BaseArgument>
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
index 71c528e4c9..d4bba3bf64 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -48,7 +48,8 @@ struct ComputePtrOffsetOfStridedBatch
 } // namespace
 
 template <typename GridwiseGemm,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatC,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
@@ -64,8 +65,8 @@ __global__ void
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
         kernel_batched_gemm_xdlops_bwd_weight(
-            const FloatAB* __restrict__ p_a_grid,
-            const FloatAB* __restrict__ p_b_grid,
+            const FloatA* __restrict__ p_a_grid,
+            const FloatB* __restrict__ p_b_grid,
             FloatC* __restrict__ p_c_grid,
             const AElementwiseOperation a_element_op,
             const BElementwiseOperation b_element_op,
@@ -91,7 +92,7 @@ __global__ void
     const long_index_t c_batch_offset = __builtin_amdgcn_readfirstlane(
         static_cast<long_index_t>(compute_ptr_offset_of_batch.GetCPtrOffset(g_idx)));
 
-    __shared__ FloatAB p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatAB)];
+    __shared__ FloatA p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(FloatA)];
 
     GridwiseGemm::template Run<HasMainKBlockLoop>(p_a_grid + a_batch_offset,
                                                   p_b_grid + b_batch_offset,
@@ -163,7 +164,9 @@ template <ck::index_t NDimSpatial,
           index_t CShuffleMXdlPerWavePerShuffle,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-          index_t CBlockTransferScalarPerVector_NWaveNPerXdl>
+          index_t CBlockTransferScalarPerVector_NWaveNPerXdl,
+          typename ComputeTypeA = InDataType,
+          typename ComputeTypeB = ComputeTypeA>
 struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
     : public DeviceGroupedConvBwdWeight<NDimSpatial,
                                         InLayout,
@@ -174,7 +177,9 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                                         OutDataType,
                                         InElementwiseOperation,
                                         WeiElementwiseOperation,
-                                        OutElementwiseOperation>
+                                        OutElementwiseOperation,
+                                        ComputeTypeA,
+                                        ComputeTypeB>
 {
     using DeviceOp = DeviceGroupedConvBwdWeight_Xdl_CShuffle;
 
@@ -1045,7 +1050,8 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
     using GridwiseGemm = GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight<
         BlockSize,
-        ADataType, // TODO: distinguish A/B datatype
+        ADataType,
+        BDataType,
         AccDataType,
         CDataType,
         InMemoryDataOperationEnum::AtomicAdd,
@@ -1090,7 +1096,11 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         CBlockTransferScalarPerVector_NWaveNPerXdl,
         CBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
         true,
-        true>;
+        true,
+        1,
+        PipelineVersion::v1,
+        ComputeTypeA,
+        ComputeTypeB>;
 
     // Argument
     using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
@@ -1217,8 +1227,8 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         index_t M01_;
         index_t N01_;
 
-        InElementwiseOperation a_element_op_;
-        OutElementwiseOperation b_element_op_;
+        OutElementwiseOperation a_element_op_;
+        InElementwiseOperation b_element_op_;
         WeiElementwiseOperation c_element_op_;
 
         // for checking IsSupportedArgument()
@@ -1281,7 +1291,8 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
                 const auto kernel = kernel_batched_gemm_xdlops_bwd_weight<
                     GridwiseGemm,
-                    ADataType, // TODO: distiguish A/B datatype
+                    ADataType,
+                    BDataType,
                     CDataType,
                     OutElementwiseOperation,
                     InElementwiseOperation,
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 927574dfde..337117551f 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -185,7 +185,8 @@ struct PassThrough
     template <>
     __host__ __device__ void operator()<bf8_t, half_t>(bf8_t& y, const half_t& x) const
     {
-        y = type_convert<bf8_t>(x);
+        // to-do: fix half_t to bf8_t convert
+        y = ck::type_convert<bf8_t>(ck::type_convert<float>(x));
     }
 #endif
 };
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
index 8897ce5a10..06c87d1892 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp
@@ -139,7 +139,8 @@ __host__ __device__ constexpr auto make_merge_transform_v4_no_carry(const LowLen
 }
 
 template <typename GridwiseGemm,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatC,
           typename AGridDesc_B_K0_M_K1,
           typename BGridDesc_B_K0_N_K1,
@@ -153,8 +154,8 @@ __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_bwd_weight(const FloatAB* __restrict__ p_a_grid,
-                                      const FloatAB* __restrict__ p_b_grid,
+        kernel_gemm_xdlops_bwd_weight(const FloatA* __restrict__ p_a_grid,
+                                      const FloatB* __restrict__ p_b_grid,
                                       FloatC* __restrict__ p_c_grid,
                                       const AGridDesc_B_K0_M_K1 a_b_k0_m_k1_grid_desc,
                                       const BGridDesc_B_K0_N_K1 b_b_k0_n_k1_grid_desc,
@@ -181,21 +182,22 @@ __global__ void
                                                   c_element_op,
                                                   c_block_cluster_adaptor);
 #else
-    ignore                = p_a_grid;
-    ignore                = p_b_grid;
-    ignore                = p_c_grid;
-    ignore                = a_b_k0_m_k1_grid_desc;
-    ignore                = b_b_k0_n_k1_grid_desc;
-    ignore                = c_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore                = a_element_op;
-    ignore                = b_element_op;
-    ignore                = c_element_op;
-    ignore                = c_block_cluster_adaptor;
+    ignore               = p_a_grid;
+    ignore               = p_b_grid;
+    ignore               = p_c_grid;
+    ignore               = a_b_k0_m_k1_grid_desc;
+    ignore               = b_b_k0_n_k1_grid_desc;
+    ignore               = c_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore               = a_element_op;
+    ignore               = b_element_op;
+    ignore               = c_element_op;
+    ignore               = c_block_cluster_adaptor;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
 template <index_t BlockSize,
-          typename FloatAB,
+          typename FloatA,
+          typename FloatB,
           typename FloatAcc,
           typename FloatC,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation,
@@ -242,7 +244,9 @@ template <index_t BlockSize,
           bool ABlockLdsExtraM1Wrw      = false,
           bool BBlockLdsExtraN1Wrw      = false,
           index_t NumGemmKPrefetchStage = 1,
-          PipelineVersion PipelineVer   = PipelineVersion::v1>
+          PipelineVersion PipelineVer   = PipelineVersion::v1,
+          typename ComputeTypeA         = FloatA,
+          typename ComputeTypeB         = ComputeTypeA>
 struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
 {
     static constexpr auto I0 = Number<0>{};
@@ -265,11 +269,16 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
     // denorm test fix, required to work around fp16 mfma issue
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
-    // FloatABAdjusted -> FloatAB throughout this file
+    // FloatAAdjusted -> ComputeTypeA, FloatBAdjusted -> ComputeTypeB,
+    // throughout this file
 #if CK_WORKAROUND_DENORM_FIX
-    using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
+    using FloatAAdjusted =
+        conditional_t<is_same_v<ComputeTypeA, ck::half_t>, ck::bhalf_t, ComputeTypeA>;
+    using FloatBAdjusted =
+        conditional_t<is_same_v<ComputeTypeB, ck::half_t>, ck::bhalf_t, ComputeTypeB>;
 #else
-    using FloatABAdjusted = FloatAB;
+    using FloatAAdjusted = ComputeTypeA;
+    using FloatBAdjusted = ComputeTypeB;
 #endif
 
     // M0/M1/M1Padding
@@ -506,7 +515,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
         constexpr auto c_block_size =
             GetCBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock().GetElementSpaceSize();
 
-        return math::max((a_block_space_size + b_block_space_size) * sizeof(FloatAB),
+        return math::max((a_block_space_size * sizeof(FloatAAdjusted) +
+                          b_block_space_size * sizeof(FloatBAdjusted)),
                          c_block_size * sizeof(FloatC));
     }
 
@@ -610,8 +620,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
     using CBlockClusterAdaptor = decltype(MakeCBlockClusterAdaptor(CMNGridDesc{}, 1, 1, 1));
 
     template <bool HasMainKBlockLoop>
-    __device__ static void Run(const FloatAB* __restrict__ p_a_grid,
-                               const FloatAB* __restrict__ p_b_grid,
+    __device__ static void Run(const FloatA* __restrict__ p_a_grid,
+                               const FloatB* __restrict__ p_b_grid,
                                FloatC* __restrict__ p_c_grid,
                                void* __restrict__ p_shared,
                                const AGridDesc_B_K0_M_K1& a_b_k0_m_k1_grid_desc,
@@ -673,8 +683,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
                                                 Sequence<1, K0PerBlock, MPerBlock, K1>,
                                                 ABlockTransferThreadClusterLengths_K0_M_K1,
                                                 ABlockTransferThreadClusterArrangeOrder,
-                                                FloatAB,
-                                                FloatABAdjusted,
+                                                FloatA,
+                                                FloatAAdjusted,
                                                 decltype(a_b_k0_m_k1_grid_desc),
                                                 decltype(a_b_k0_m_k1_block_desc),
                                                 ABlockTransferSrcAccessOrder,
@@ -703,8 +713,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
                                                 Sequence<1, K0PerBlock, NPerBlock, K1>,
                                                 BBlockTransferThreadClusterLengths_K0_N_K1,
                                                 BBlockTransferThreadClusterArrangeOrder,
-                                                FloatAB,
-                                                FloatABAdjusted,
+                                                FloatB,
+                                                FloatBAdjusted,
                                                 decltype(b_b_k0_n_k1_grid_desc),
                                                 decltype(b_b_k0_n_k1_block_desc),
                                                 BBlockTransferSrcAccessOrder,
@@ -733,12 +743,14 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
         // sanity check
 
         constexpr index_t KPack =
-            math::max(K1, MfmaSelector<FloatABAdjusted, MPerXDL, NPerXDL>::selected_mfma.k_per_blk);
+            math::max(K1,
+                      MfmaSelector<FloatAAdjusted, MPerXDL, NPerXDL, FloatBAdjusted>::selected_mfma
+                          .k_per_blk);
 
         auto blockwise_gemm =
             BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_v1<BlockSize,
-                                                                FloatABAdjusted,
-                                                                FloatABAdjusted,
+                                                                FloatAAdjusted,
+                                                                FloatBAdjusted,
                                                                 FloatAcc,
                                                                 decltype(a_k0_m_k1_block_desc),
                                                                 decltype(b_k0_n_k1_block_desc),
@@ -758,10 +770,10 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
         constexpr auto b_block_slice_copy_step = make_multi_index(0, K0PerBlock, 0, 0);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<FloatABAdjusted*>(p_shared), a_k0_m_k1_block_desc.GetElementSpaceSize());
+            static_cast<FloatAAdjusted*>(p_shared), a_k0_m_k1_block_desc.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<FloatABAdjusted*>(p_shared) + a_block_space_size,
+            static_cast<FloatBAdjusted*>(p_shared) + a_block_space_size,
             b_k0_n_k1_block_desc.GetElementSpaceSize());
 
         // gridwise GEMM pipeline
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index 06b95aef21..c8e56fbc56 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -32,8 +32,12 @@ enum struct MfmaInstr
     mfma_f64_16x16x4f64,
     mfma_f32_32x32x16f8f8,
     mfma_f32_16x16x32f8f8,
+    mfma_f32_32x32x16bf8bf8,
+    mfma_f32_16x16x32bf8bf8,
     mfma_f32_32x32x16f8bf8,
-    mfma_f32_16x16x32f8bf8
+    mfma_f32_16x16x32f8bf8,
+    mfma_f32_32x32x16bf8f8,
+    mfma_f32_16x16x32bf8f8
 };
 
 template <MfmaInstr instr>
@@ -504,6 +508,52 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8f8>
 };
 #endif
 
+#if defined CK_ENABLE_BF8
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_32x32x16bf8bf8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_32x32x16bf8bf8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_16x16x32bf8bf8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_16x16x32bf8bf8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+#endif
+
 #if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
 template <>
 struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8bf8>
@@ -550,6 +600,52 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8bf8>
 };
 #endif
 
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_32x32x16bf8f8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 4;
+    static constexpr index_t num_regs_per_blk    = 16;
+    static constexpr index_t num_threads_per_blk = 32;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 2;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 32;
+    static constexpr index_t n_per_blk           = 32;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_32x32x16bf8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+
+template <>
+struct mfma_type<MfmaInstr::mfma_f32_16x16x32bf8f8>
+{
+    static constexpr index_t group_size          = 4;
+    static constexpr index_t num_groups_per_blk  = 1;
+    static constexpr index_t num_regs_per_blk    = 4;
+    static constexpr index_t num_threads_per_blk = 16;
+    static constexpr index_t wave_size           = 64;
+    static constexpr index_t num_input_blks      = 4;
+    static constexpr index_t num_output_blks     = 1;
+    static constexpr index_t m_per_blk           = 16;
+    static constexpr index_t n_per_blk           = 16;
+    static constexpr index_t k_per_blk           = 8;
+    static constexpr bool is_k_reduction         = true;
+
+    template <index_t MPerXdlops, index_t NPerXdlops, class FloatA, class FloatB, class FloatC>
+    __device__ void run(const FloatA& a, const FloatB& b, FloatC& reg_c) const
+    {
+        intrin_mfma_f32_16x16x32bf8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
+    }
+};
+#endif
+
 template <typename base_type,
           index_t MPerXdlops,
           index_t NPerXdlops,
@@ -710,6 +806,20 @@ struct MfmaSelector
     }
 #endif
 
+#if defined CK_ENABLE_BF8
+    template <>
+    static constexpr auto GetMfma<bf8_t, 32, 32>()
+    {
+        return MfmaInstr::mfma_f32_32x32x16bf8bf8;
+    }
+
+    template <>
+    static constexpr auto GetMfma<bf8_t, 16, 16>()
+    {
+        return MfmaInstr::mfma_f32_16x16x32bf8bf8;
+    }
+#endif
+
 #if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
     template <>
     static constexpr auto GetMfma<f8_t, 32, 32, bf8_t>()
@@ -724,6 +834,20 @@ struct MfmaSelector
     }
 #endif
 
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+    template <>
+    static constexpr auto GetMfma<bf8_t, 32, 32, f8_t>()
+    {
+        return MfmaInstr::mfma_f32_32x32x16bf8f8;
+    }
+
+    template <>
+    static constexpr auto GetMfma<bf8_t, 16, 16, f8_t>()
+    {
+        return MfmaInstr::mfma_f32_16x16x32bf8f8;
+    }
+#endif
+
     static constexpr auto selected_mfma =
         mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops, additional_type>()>{};
 
@@ -931,8 +1055,12 @@ struct XdlopsGemm
 #if defined CK_ENABLE_FP8
                 || is_same<base_type, f8_t>::value
 #endif
+#if defined CK_ENABLE_BF8
+                || is_same<base_type, bf8_t>::value
+#endif
 #if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-                || (is_same<base_type, f8_t>::value && is_same<additional_type, bf8_t>::value)
+                || (is_same<base_type, f8_t>::value && is_same<additional_type, bf8_t>::value) ||
+                (is_same<base_type, bf8_t>::value && is_same<additional_type, f8_t>::value)
 #endif
                 ,
             "base base_type must be double, float, half, bfloat16, int8_t, f8_t or bf8_t!");
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index 3768f92633..bf6241e46e 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -420,6 +420,71 @@ struct intrin_mfma_f32_16x16x32f8f8<16, 16>
 };
 #endif
 
+#if defined CK_ENABLE_BF8
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_32x32x16bf8bf8;
+
+template <>
+struct intrin_mfma_f32_32x32x16bf8bf8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const bf8x8_t& reg_a, const bf8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_f32_32x32x16_bf8_bf8(
+                bit_cast<long>(reg_a),
+                bit_cast<long>(reg_b),
+                reg_c.template AsType<float16_t>()[Number<0>{}],
+                0,
+                0,
+                0);
+#else
+        vector_type<bf8_t, 8> reg_a_v(reg_a);
+        vector_type<bf8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<bf8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<bf8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_32x32x2f32<32, 32>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_16x16x32bf8bf8;
+
+template <>
+struct intrin_mfma_f32_16x16x32bf8bf8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const bf8x8_t& reg_a, const bf8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f32_16x16x32_bf8_bf8(
+            bit_cast<long>(reg_a),
+            bit_cast<long>(reg_b),
+            reg_c.template AsType<float4_t>()[Number<0>{}],
+            0,
+            0,
+            0);
+#else
+        vector_type<bf8_t, 8> reg_a_v(reg_a);
+        vector_type<bf8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<bf8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<bf8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_16x16x4f32<16, 16>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+#endif
+
 #if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f32_32x32x16f8bf8;
@@ -484,5 +549,70 @@ struct intrin_mfma_f32_16x16x32f8bf8<16, 16>
     }
 };
 #endif
+
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_32x32x16bf8f8;
+
+template <>
+struct intrin_mfma_f32_32x32x16bf8f8<32, 32>
+{
+    template <class FloatC>
+    __device__ static void Run(const bf8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float16_t>()(Number<0>{}) =
+            __builtin_amdgcn_mfma_f32_32x32x16_bf8_fp8(
+                bit_cast<long>(reg_a),
+                bit_cast<long>(reg_b),
+                reg_c.template AsType<float16_t>()[Number<0>{}],
+                0,
+                0,
+                0);
+#else
+        vector_type<bf8_t, 8> reg_a_v(reg_a);
+        vector_type<f8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<bf8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<f8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_32x32x2f32<32, 32>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+
+template <index_t MPerWave, index_t NPerWave>
+struct intrin_mfma_f32_16x16x32bf8f8;
+
+template <>
+struct intrin_mfma_f32_16x16x32bf8f8<16, 16>
+{
+    template <class FloatC>
+    __device__ static void Run(const bf8x8_t& reg_a, const f8x8_t& reg_b, FloatC& reg_c)
+    {
+#if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+        reg_c.template AsType<float4_t>()(Number<0>{}) = __builtin_amdgcn_mfma_f32_16x16x32_bf8_fp8(
+            bit_cast<long>(reg_a),
+            bit_cast<long>(reg_b),
+            reg_c.template AsType<float4_t>()[Number<0>{}],
+            0,
+            0,
+            0);
+#else
+        vector_type<bf8_t, 8> reg_a_v(reg_a);
+        vector_type<f8_t, 8> reg_b_v(reg_b);
+
+        static_for<0, 8, 1>{}([&](auto k) {
+            float reg_a_f32 = type_convert<float>(reg_a_v.template AsType<bf8_t>()[Number<k>{}]);
+            float reg_b_f32 = type_convert<float>(reg_b_v.template AsType<f8_t>()[Number<k>{}]);
+
+            intrin_mfma_f32_16x16x4f32<16, 16>::Run(reg_a_f32, reg_b_f32, reg_c);
+        });
+#endif
+    }
+};
+#endif
 } // namespace ck
 #endif
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 935926070d..505de73ae2 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -221,7 +221,7 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
 {
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
-    return type_convert<f8_t>(type_convert<float>(x));
+    return type_convert<bf8_t>(type_convert<float>(x));
 #else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
index ec5df238ab..02ad7a033a 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_bwd_weight.hpp
@@ -25,6 +25,8 @@ template <ck::index_t NDimSpatial,
           typename InElementwiseOperation,
           typename WeiElementwiseOperation,
           typename OutElementwiseOperation,
+          typename ComputeTypeA                                                     = OutDataType,
+          typename ComputeTypeB                                                     = InDataType,
           typename std::enable_if<NDimSpatial >= 1 && NDimSpatial <= 3, bool>::type = false>
 struct ReferenceConvBwdWeight : public device::BaseOperator
 {
@@ -98,8 +100,8 @@ struct ReferenceConvBwdWeight : public device::BaseOperator
                             if(wi >= 0 &&
                                ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[3])
                             {
-                                float v_out;
-                                float v_in;
+                                ComputeTypeA v_out;
+                                ComputeTypeB v_in;
 
                                 arg.out_element_op_(
                                     v_out, ck::type_convert<float>(arg.output_(g, n, k, wo)));
@@ -107,7 +109,7 @@ struct ReferenceConvBwdWeight : public device::BaseOperator
                                 arg.in_element_op_(
                                     v_in, ck::type_convert<float>(arg.input_(g, n, c, wi)));
 
-                                v_acc += v_out * v_in;
+                                v_acc += type_convert<float>(v_out) * type_convert<float>(v_in);
                             }
                         }
                     }
@@ -158,8 +160,8 @@ struct ReferenceConvBwdWeight : public device::BaseOperator
                                    wi >= 0 &&
                                    ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[4])
                                 {
-                                    float v_out;
-                                    float v_in;
+                                    ComputeTypeA v_out;
+                                    ComputeTypeB v_in;
 
                                     arg.out_element_op_(
                                         v_out,
@@ -168,7 +170,7 @@ struct ReferenceConvBwdWeight : public device::BaseOperator
                                     arg.in_element_op_(
                                         v_in, ck::type_convert<float>(arg.input_(g, n, c, hi, wi)));
 
-                                    v_acc += v_out * v_in;
+                                    v_acc += type_convert<float>(v_out) * type_convert<float>(v_in);
                                 }
                             }
                         }
@@ -226,8 +228,8 @@ struct ReferenceConvBwdWeight : public device::BaseOperator
                                        ck::type_convert<std::size_t>(wi) <
                                            arg.input_.GetLengths()[5])
                                     {
-                                        float v_out;
-                                        float v_in;
+                                        ComputeTypeA v_out;
+                                        ComputeTypeB v_in;
 
                                         arg.out_element_op_(v_out,
                                                             ck::type_convert<float>(
@@ -237,7 +239,8 @@ struct ReferenceConvBwdWeight : public device::BaseOperator
                                                            ck::type_convert<float>(
                                                                arg.input_(g, n, c, di, hi, wi)));
 
-                                        v_acc += v_out * v_in;
+                                        v_acc +=
+                                            type_convert<float>(v_out) * type_convert<float>(v_in);
                                     }
                                 }
                             }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
index 6fc91565b9..096e0b1770 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp
@@ -19,6 +19,14 @@ using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
 
+#ifdef CK_ENABLE_FP8
+using F8 = ck::f8_t;
+#endif
+
+#ifdef CK_ENABLE_BF8
+using BF8 = ck::bf8_t;
+#endif
+
 using Empty_Tuple = ck::Tuple<>;
 
 template <ck::index_t... Is>
@@ -133,6 +141,43 @@ using device_grouped_conv_bwd_weight_xdl_c_shuffle_bf16_instances = std::tuple<
     // clang-format on
     >;
 
+template <ck::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_comp_bf8_f8_instances = std::tuple<
+// clang-format off
+        //#########################################|     Num| InLayout| WeiLayout| OutLayout| InData| WeiData| OutData| AccData|          In|         Wei|         Out|              ConvBackward| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer|   ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|   CBlockTransfer|  CBlockTransfer| Compute| Compute|
+        //#########################################|     Dim|         |          |          |   Type|    Type|    Type|    Type| Elementwise| Elementwise| Elementwise|                    Weight|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|    ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|   ClusterLengths| ScalarPerVector|   TypeA|   TypeB|
+        //#########################################| Spatial|         |          |          |       |        |        |        |   Operation|   Operation|   Operation|            Specialization|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|     ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| MBlock_MPerBlock|    NWaveNPerXdl|        |        |
+        //#########################################|        |         |          |          |       |        |        |        |            |            |            |                          |      |      |      |      |   |     |     |     |     |                |                 |               |               |               |               |          |                |               |               |              |               |               |          |            |            | NBlock_NPerBlock|                |        |        |
+#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+        // generic instance
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              2,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              2,              4,      true,           1,           1,   S<1, 16, 1, 4>,               2,       BF8,     F8>,   
+        // instance for small conv.K
+        // for fp16 conv.K and conv.C must be divisible by 2
+        // since half_t atomic_add require scalar_per_x_vector % 2 == 0
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              2,              1,      true,           1,           1,   S<1, 32, 1, 4>,               2,       BF8,     F8>,  
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              2,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>,
+
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 8>,               8,       BF8,     F8>,
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 32, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 8>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 8,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              1,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  8>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              1,      true,  S<1, 4, 16, 2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 32, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              4,      true,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              2,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>,   
+        DeviceGroupedConvBwdWeight_Xdl_CShuffle< NDimSpatial,  ALayout,   BLayout,   ELayout,    F16,     F16,     F16,     F32, PassThrough, PassThrough, PassThrough,                  ConvSpec,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 4,  4>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,              2,              8,              2,      true,  S<1, 4, 8,  2>,  S<0, 3, 1, 2>,  S<0, 2, 1, 3>,             2,              8,              4,      true,           1,           1,   S<1, 16, 1, 4>,               8,       BF8,     F8>
+#endif
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index c0a5b4bfe5..9f536f210d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -216,6 +216,21 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 #endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           BF8,
+                                                           F8>>>& instances);
+#endif
 
 #ifdef DL_KERNELS
 // dl
@@ -464,7 +479,9 @@ template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename InDataType,
           typename WeiDataType,
-          typename OutDataType>
+          typename OutDataType,
+          typename ComputeTypeA,
+          typename ComputeTypeB>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupedConvBwdWeight<
     NumDimSpatial,
     InLayout,
@@ -475,7 +492,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
     OutDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::PassThrough>>
+    ck::tensor_operation::element_wise::PassThrough,
+    ComputeTypeA,
+    ComputeTypeB>>
 {
     using DeviceOp = DeviceGroupedConvBwdWeight<NumDimSpatial,
                                                 InLayout,
@@ -486,7 +505,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                                                 OutDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::PassThrough>;
+                                                ck::tensor_operation::element_wise::PassThrough,
+                                                ComputeTypeA,
+                                                ComputeTypeB>;
 
     static auto GetInstances()
     {
@@ -706,7 +727,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
 #ifdef CK_ENABLE_FP16
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t>)
+                                  is_same_v<OutDataType, half_t> &&
+                                  is_same_v<ComputeTypeA, half_t> &&
+                                  is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
@@ -728,6 +751,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                                  is_same_v<OutDataType, half_t> &&
+                                  is_same_v<ComputeTypeA, bf8_t> && is_same_v<ComputeTypeB, f8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_f8_instances(
+                        op_ptrs);
+                }
 #endif
             }
         }
diff --git a/library/include/ck/library/utility/host_tensor_generator.hpp b/library/include/ck/library/utility/host_tensor_generator.hpp
index 07d3a35a18..6b531c4c6f 100644
--- a/library/include/ck/library/utility/host_tensor_generator.hpp
+++ b/library/include/ck/library/utility/host_tensor_generator.hpp
@@ -111,6 +111,22 @@ struct GeneratorTensor_2<ck::f8_t>
 };
 #endif
 
+#if defined CK_ENABLE_BF8
+template <>
+struct GeneratorTensor_2<ck::bf8_t>
+{
+    int min_value = 0;
+    int max_value = 1;
+
+    template <typename... Is>
+    ck::bf8_t operator()(Is...)
+    {
+        float tmp = (std::rand() % (max_value - min_value)) + min_value;
+        return ck::type_convert<ck::bf8_t>(tmp);
+    }
+};
+#endif
+
 template <typename T>
 struct GeneratorTensor_3
 {
@@ -162,6 +178,25 @@ struct GeneratorTensor_3<ck::f8_t>
 };
 #endif
 
+#if defined CK_ENABLE_BF8
+template <>
+struct GeneratorTensor_3<ck::bf8_t>
+{
+    float min_value = 0;
+    float max_value = 1;
+
+    template <typename... Is>
+    ck::bf8_t operator()(Is...)
+    {
+        float tmp = float(std::rand()) / float(RAND_MAX);
+
+        float fp32_tmp = min_value + tmp * (max_value - min_value);
+
+        return ck::type_convert<ck::bf8_t>(fp32_tmp);
+    }
+};
+#endif
+
 template <typename T>
 struct GeneratorTensor_4
 {
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
index 6bd9d6e646..65ea645a39 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
@@ -4,7 +4,8 @@ set(GROUPED_CONV3D_BWD_WEIGHT
     device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
     device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp)
+    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp)
 
 if(DL_KERNELS)
     list(APPEND GROUPED_CONV3D_BWD_WEIGHT
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp
new file mode 100644
index 0000000000..7f9493f602
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp
@@ -0,0 +1,49 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_xdl_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           BF8,
+                                                           F8>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_comp_bf8_f8_instances<
+            3,
+            NDHWGC,
+            GKZYXC,
+            NDHWGK,
+            ConvBwdWeightDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_xdl_c_shuffle_f16_comp_bf8_f8_instances<
+            3,
+            NDHWGC,
+            GKZYXC,
+            NDHWGK,
+            ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
index 48bf639a70..0300aed436 100644
--- a/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_conv_bwd_weight_impl.hpp
@@ -33,7 +33,9 @@ template <ck::index_t NDimSpatial,
           typename OutLayout,
           typename InDataType,
           typename WeiDataType,
-          typename OutDataType>
+          typename OutDataType,
+          typename ComputeTypeA = InDataType,
+          typename ComputeTypeB = ComputeTypeA>
 bool profile_grouped_conv_bwd_weight_impl(int do_verification,
                                           int init_method,
                                           bool do_log,
@@ -120,7 +122,9 @@ bool profile_grouped_conv_bwd_weight_impl(int do_verification,
                                                                               OutDataType,
                                                                               InElementOp,
                                                                               WeiElementOp,
-                                                                              OutElementOp>;
+                                                                              OutElementOp,
+                                                                              ComputeTypeA,
+                                                                              ComputeTypeB>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index be8a3230f7..9a9b5a0f74 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -20,9 +20,10 @@ enum struct ConvLayout
 
 enum struct ConvDataType
 {
-    F32_F32_F32,   // 0
-    F16_F16_F16,   // 1
-    BF16_F32_BF16, // 2
+    F32_F32_F32,       // 0
+    F16_F16_F16,       // 1
+    BF16_F32_BF16,     // 2
+    F16_F16_F16_BF8_F8 // 3
 };
 
 #define OP_NAME "grouped_conv_bwd_weight"
@@ -33,7 +34,8 @@ static void print_helper_msg()
     std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
               << "arg2: data type (0: Input fp32, Weight fp32, Output fp32\n"
               << "                 1: Input fp16, Weight fp16, Output fp16\n"
-              << "                 2: Input bf16, Weight fp32, Output bf16)\n"
+              << "                 2: Input bf16, Weight fp32, Output bf16\n"
+              << "                 3: Input fp16, Weight fp16, Output fp16, Gemm bf8@fp8)\n"
               << "arg3: tensor layout (0: Input[G, N, C, Hi, Wi], Weight[G, K, C, Y, X], Output[G, "
                  "N, K, Ho, Wo]\n"
               << "                     1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, "
@@ -82,6 +84,12 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
     using F32  = float;
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
+#ifdef CK_ENABLE_FP8
+    using F8 = ck::f8_t;
+#endif
+#ifdef CK_ENABLE_BF8
+    using BF8 = ck::bf8_t;
+#endif
 
     using namespace ck::tensor_layout::convolution;
 
@@ -95,7 +103,9 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
                        auto out_layout,
                        auto in_type,
                        auto wei_type,
-                       auto out_type) {
+                       auto out_type,
+                       auto compute_type_a,
+                       auto compute_type_b) {
         constexpr ck::index_t NDimSpatial = num_dim_spatial_tmp.value;
 
         using InLayout  = decltype(in_layout);
@@ -106,13 +116,18 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
         using WeiDataType = decltype(wei_type);
         using OutDataType = decltype(out_type);
 
+        using ComputeTypeA = decltype(compute_type_a);
+        using ComputeTypeB = decltype(compute_type_b);
+
         bool pass = ck::profiler::profile_grouped_conv_bwd_weight_impl<NDimSpatial,
                                                                        InLayout,
                                                                        WeiLayout,
                                                                        OutLayout,
                                                                        InDataType,
                                                                        WeiDataType,
-                                                                       OutDataType>(
+                                                                       OutDataType,
+                                                                       ComputeTypeA,
+                                                                       ComputeTypeB>(
             do_verification, init_method, do_log, time_kernel, params, split_k);
 
         return pass ? 0 : 1;
@@ -122,80 +137,84 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
-            return profile(I1, GNWC{}, GKXC{}, GNWK{}, F32{}, F32{}, F32{});
+            return profile(I1, GNWC{}, GKXC{}, GNWK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
         else if(data_type == ConvDataType::F16_F16_F16)
         {
-            return profile(I1, GNWC{}, GKXC{}, GNWK{}, F16{}, F16{}, F16{});
+            return profile(I1, GNWC{}, GKXC{}, GNWK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
         else if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
-            return profile(I1, GNWC{}, GKXC{}, GNWK{}, BF16{}, F32{}, BF16{});
+            return profile(I1, GNWC{}, GKXC{}, GNWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
     else if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
-            return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, F32{}, F32{}, F32{});
+            return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
         else if(data_type == ConvDataType::F16_F16_F16)
         {
-            return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, F16{}, F16{}, F16{});
+            return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
         else if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
-            return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, BF16{}, F32{}, BF16{});
+            return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
     else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
-            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F32{}, F32{}, F32{});
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
         else if(data_type == ConvDataType::F16_F16_F16)
         {
-            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F16{}, F16{}, F16{});
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
         else if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
-            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, BF16{}, F32{}, BF16{});
+            return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
     else if(num_dim_spatial == 3 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
-            return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, F32{}, F32{}, F32{});
+            return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
         else if(data_type == ConvDataType::F16_F16_F16)
         {
-            return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, F16{}, F16{}, F16{});
+            return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
         else if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
-            return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, BF16{}, F32{}, BF16{});
+            return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
     else if(num_dim_spatial == 3 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
-            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F32{}, F32{}, F32{});
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
         else if(data_type == ConvDataType::F16_F16_F16)
         {
-            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{});
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
         else if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
-            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, BF16{}, F32{}, BF16{});
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
+        }
+        else if(data_type == ConvDataType::F16_F16_F16_BF8_F8)
+        {
+            return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{}, BF8{}, F8{});
         }
     }
 

From 04f93aadb8fca26c141f70ab19710c460b46f6c0 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 4 Oct 2023 18:04:27 -0500
Subject: [PATCH 160/194] Grouped conv bwd data with fp16 input and bf8fp8 comp
 (#962)

* Add f8 bf8 gemm example

* Add element-wise ops

* Add intrinsics

* Update reference calculation

* Add an additional type option for xdlops gemm

* Fix build process

* Add bf8 to buffer addressing

* Update blockwise op, split typeA and typeB

* Update for compatibility

* Uppdate naming to f8->fp8

* Update naming

* Format

* Update naming (#937)

* Add a client example

* Add computetypes to device and gridwise ops

* Add instances, update instance factory

* Format

* Fix a flag

* Add ckProfiler mode

* Fix typos

* Add an example

* Add bf8 generator

* add bf8 mfma; fixed type_convert for bf8

* move verfication ahead of timing

* Update reference calculation

* Fix reference

* Narrow down float init range

* Fix bf8 bf8 mfma

* Add bf8 @ fp8 mfma

* Update example

* Update instances

* Update profiler api

* Update for compatibility

* Format

* Remove extra example

* Clean up

* workaround convert

* added instance of f16_bf8f8, and client example

* fixed mfma selector

* format

---------

Co-authored-by: Rostyslav Geyyer <rosty.geyyer@amd.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../10_grouped_conv2d_bwd_data/CMakeLists.txt |   2 -
 .../10_grouped_convnd_bwd_data/CMakeLists.txt |   8 +
 .../grouped_conv2d_bwd_data.cpp               |   0
 .../grouped_conv3d_bwd_data.cpp               | 205 +++++++++++++++++
 ..._conv3d_bwd_data_input_fp16_comp_bf8f8.cpp | 207 ++++++++++++++++++
 ...evice_grouped_conv_bwd_data_multiple_d.hpp |   4 +-
 ...nv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp |  18 +-
 .../gridwise_gemm_multiple_d_xdl_cshuffle.hpp |  34 +--
 ...ice_grouped_conv_bwd_data_xdl_instance.hpp |  39 ++++
 .../gpu/grouped_convolution_backward_data.hpp |  41 +++-
 .../grouped_conv3d_bwd_data/CMakeLists.txt    |   2 +-
 ..._ndhwgk_input_f16_comp_bf8_f8_instance.cpp |  52 +++++
 12 files changed, 582 insertions(+), 30 deletions(-)
 delete mode 100644 client_example/10_grouped_conv2d_bwd_data/CMakeLists.txt
 create mode 100644 client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
 rename client_example/{10_grouped_conv2d_bwd_data => 10_grouped_convnd_bwd_data}/grouped_conv2d_bwd_data.cpp (100%)
 create mode 100644 client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data.cpp
 create mode 100644 client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp

diff --git a/client_example/10_grouped_conv2d_bwd_data/CMakeLists.txt b/client_example/10_grouped_conv2d_bwd_data/CMakeLists.txt
deleted file mode 100644
index e564f3180d..0000000000
--- a/client_example/10_grouped_conv2d_bwd_data/CMakeLists.txt
+++ /dev/null
@@ -1,2 +0,0 @@
-add_executable(client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp)
-target_link_libraries(client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_operations)
diff --git a/client_example/10_grouped_convnd_bwd_data/CMakeLists.txt b/client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
new file mode 100644
index 0000000000..60543c7308
--- /dev/null
+++ b/client_example/10_grouped_convnd_bwd_data/CMakeLists.txt
@@ -0,0 +1,8 @@
+add_executable(client_grouped_conv2d_bwd_data grouped_conv2d_bwd_data.cpp)
+target_link_libraries(client_grouped_conv2d_bwd_data PRIVATE composable_kernel::device_operations)
+
+add_executable(client_grouped_conv3d_bwd_data grouped_conv3d_bwd_data.cpp)
+target_link_libraries(client_grouped_conv3d_bwd_data PRIVATE composable_kernel::device_operations)
+
+add_executable(client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp)
+target_link_libraries(client_grouped_conv3d_bwd_data_input_fp16_comp_bf8f8 PRIVATE composable_kernel::device_operations)
diff --git a/client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp b/client_example/10_grouped_convnd_bwd_data/grouped_conv2d_bwd_data.cpp
similarity index 100%
rename from client_example/10_grouped_conv2d_bwd_data/grouped_conv2d_bwd_data.cpp
rename to client_example/10_grouped_convnd_bwd_data/grouped_conv2d_bwd_data.cpp
diff --git a/client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data.cpp b/client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data.cpp
new file mode 100644
index 0000000000..d2f2ff41bc
--- /dev/null
+++ b/client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data.cpp
@@ -0,0 +1,205 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout    = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout   = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout   = ck::tensor_layout::convolution::NDHWGK;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 2;
+static constexpr ck::index_t N             = 16;
+static constexpr ck::index_t K             = 16;
+static constexpr ck::index_t C             = 16;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 14;
+static constexpr ck::index_t Hi            = 14;
+static constexpr ck::index_t Wi            = 14;
+static constexpr ck::index_t Do            = 14;
+static constexpr ck::index_t Ho            = 14;
+static constexpr ck::index_t Wo            = 14;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, C, Di, Hi, Wi};
+    std::array<ck::index_t, NumDimSpatial + 3> in_strides{
+        C, Di * Hi * Wi * G * C, 1, Hi * Wi * G * C, Wi * G * C, G * C};
+
+    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, C, Z, Y, X};
+    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{
+        K * Z * Y * X * C, Z * Y * X * C, 1, Y * X * C, X * C, C};
+
+    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, K, Do, Ho, Wo};
+    std::array<ck::index_t, NumDimSpatial + 3> out_strides{
+        K, Do * Ho * Wo * G * K, 1, Ho * Wo * G * K, Wo * G * K, G * K};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * G * N * Di * Hi * Wi * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Z * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Do * Ho * Wo * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NumDimSpatial,
+                                                                                     OutLayout,
+                                                                                     WeiLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     InLayout,
+                                                                                     OutDataType,
+                                                                                     WeiDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     InDataType,
+                                                                                     PassThrough,
+                                                                                     PassThrough,
+                                                                                     PassThrough>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop      = std::size_t(2) * G * N * K * C * Do * Ho * Wo * Y * X;
+            std::size_t num_bytes = sizeof(InDataType) * G * N * Di * Hi * Wi * C +
+                                    sizeof(WeiDataType) * G * K * Z * Y * X * C +
+                                    sizeof(OutDataType) * G * N * Do * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+}
diff --git a/client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp b/client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp
new file mode 100644
index 0000000000..2330228d1d
--- /dev/null
+++ b/client_example/10_grouped_convnd_bwd_data/grouped_conv3d_bwd_data_input_fp16_comp_bf8f8.cpp
@@ -0,0 +1,207 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iomanip>
+#include <iostream>
+#include <iterator>
+#include <numeric>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout    = ck::tensor_layout::convolution::NDHWGC;
+using WeiLayout   = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout   = ck::tensor_layout::convolution::NDHWGK;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr ck::index_t NumDimSpatial = 3;
+static constexpr ck::index_t G             = 2;
+static constexpr ck::index_t N             = 16;
+static constexpr ck::index_t K             = 16;
+static constexpr ck::index_t C             = 16;
+static constexpr ck::index_t Z             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Di            = 14;
+static constexpr ck::index_t Hi            = 14;
+static constexpr ck::index_t Wi            = 14;
+static constexpr ck::index_t Do            = 14;
+static constexpr ck::index_t Ho            = 14;
+static constexpr ck::index_t Wo            = 14;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main()
+{
+    std::array<ck::index_t, NumDimSpatial + 3> in_lengths{G, N, C, Di, Hi, Wi};
+    std::array<ck::index_t, NumDimSpatial + 3> in_strides{
+        C, Di * Hi * Wi * G * C, 1, Hi * Wi * G * C, Wi * G * C, G * C};
+
+    std::array<ck::index_t, NumDimSpatial + 3> wei_lengths{G, K, C, Z, Y, X};
+    std::array<ck::index_t, NumDimSpatial + 3> wei_strides{
+        K * Z * Y * X * C, Z * Y * X * C, 1, Y * X * C, X * C, C};
+
+    std::array<ck::index_t, NumDimSpatial + 3> out_lengths{G, N, K, Do, Ho, Wo};
+    std::array<ck::index_t, NumDimSpatial + 3> out_strides{
+        K, Do * Ho * Wo * G * K, 1, Ho * Wo * G * K, Wo * G * K, G * K};
+
+    std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1, 1};
+    std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1, 1};
+
+    SimpleDeviceMem in(sizeof(InDataType) * G * N * Di * Hi * Wi * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * G * K * Z * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Do * Ho * Wo * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<NumDimSpatial,
+                                                                                     OutLayout,
+                                                                                     WeiLayout,
+                                                                                     ck::Tuple<>,
+                                                                                     InLayout,
+                                                                                     OutDataType,
+                                                                                     WeiDataType,
+                                                                                     ck::Tuple<>,
+                                                                                     InDataType,
+                                                                                     PassThrough,
+                                                                                     PassThrough,
+                                                                                     PassThrough,
+                                                                                     ck::bf8_t,
+                                                                                     ck::f8_t>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop      = std::size_t(2) * G * N * K * C * Do * Ho * Wo * Y * X;
+            std::size_t num_bytes = sizeof(InDataType) * G * N * Di * Hi * Wi * C +
+                                    sizeof(WeiDataType) * G * K * Z * Y * X * C +
+                                    sizeof(OutDataType) * G * N * Do * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cerr << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    if(best_op_id < 0)
+    {
+        std::cerr << "no suitable instance" << std::endl;
+        return EXIT_FAILURE;
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(out.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        {},
+                                                        in.GetDeviceBuffer(),
+                                                        out_lengths,
+                                                        out_strides,
+                                                        wei_lengths,
+                                                        wei_strides,
+                                                        {},
+                                                        {},
+                                                        in_lengths,
+                                                        in_strides,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
index 7e4bca2bd6..2abf1d5a10 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_conv_bwd_data_multiple_d.hpp
@@ -29,7 +29,9 @@ template <ck::index_t NDimSpatial,
           typename EDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CDEElementwiseOperation>
+          typename CDEElementwiseOperation,
+          typename AComputeType = ADataType,
+          typename BComputeType = AComputeType>
 struct DeviceGroupedConvBwdDataMultipleD : public BaseOperator
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
index 37d7b0b1fc..88b859efbc 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp
@@ -198,7 +198,9 @@ template <index_t NDimSpatial,
           index_t CShuffleNXdlPerWavePerShuffle,
           typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CDEBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched = make_default_loop_scheduler()>
+          LoopScheduler LoopSched = make_default_loop_scheduler(),
+          typename AComputeType   = ADataType,
+          typename BComputeType   = AComputeType>
 struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
     : public DeviceGroupedConvBwdDataMultipleD<NDimSpatial,
                                                ALayout,    // output image
@@ -211,7 +213,9 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
                                                EDataType,  // input image
                                                AElementwiseOp,
                                                BElementwiseOp,
-                                               CDEElementwiseOp>
+                                               CDEElementwiseOp,
+                                               AComputeType,
+                                               BComputeType>
 {
     // TODO: Extend support for more spatial dimensions.
     static_assert(NDimSpatial == 2 || NDimSpatial == 3,
@@ -312,9 +316,9 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
 
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
-        ABDataType, // TODO: distinguish A/B datatype
-        ABDataType, // TODO: distinguish A/B datatype
-        ABDataType, // TODO: distinguish A/B datatype
+        ABDataType,
+        ABDataType,
+        AComputeType,
         AccDataType,
         CShuffleDataType,
         DsDataType,
@@ -354,7 +358,9 @@ struct DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1
         CShuffleNXdlPerWavePerShuffle,
         CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
         CDEBlockTransferScalarPerVector_NPerBlock,
-        LoopSched>;
+        LoopSched,
+        PipelineVersion::v1,
+        BComputeType>;
 
     template <typename Desc_K0_M_K1>
     static auto transform_k0_m_k1_to_m_k(const Desc_K0_M_K1& desc_k0_m_k1)
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
index c7ec435477..15c30a0dad 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_xdl_cshuffle.hpp
@@ -31,7 +31,7 @@ namespace ck {
 //   D0, D1, ... and E have the same layout
 template <typename ADataType,
           typename BDataType,
-          typename ComputeDataType_,
+          typename AComputeDataType_,
           typename AccDataType,
           typename CShuffleDataType,
           typename DsDataType,
@@ -72,7 +72,8 @@ template <typename ADataType,
           typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
           index_t CDEShuffleBlockTransferScalarPerVector_NPerBlock,
           LoopScheduler LoopSched,
-          PipelineVersion PipelineVer = PipelineVersion::v1>
+          PipelineVersion PipelineVer = PipelineVersion::v1,
+          typename BComputeDataType   = AComputeDataType_>
 struct GridwiseGemmMultipleD_xdl_cshuffle
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
@@ -100,10 +101,10 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         decltype(GridwiseGemmPipeline_Selector<PipelineVer, NumGemmKPrefetchStage, LoopSched>())>;
 
 #if CK_WORKAROUND_DENORM_FIX
-    using ComputeDataType =
-        conditional_t<is_same_v<ComputeDataType_, ck::half_t>, ck::bhalf_t, ComputeDataType_>;
+    using AComputeDataType =
+        conditional_t<is_same_v<AComputeDataType_, ck::half_t>, ck::bhalf_t, AComputeDataType_>;
 #else
-    using ComputeDataType = ComputeDataType_;
+    using AComputeDataType = AComputeDataType_;
 #endif
 
     __host__ __device__ static constexpr auto GetABlockDescriptor_AK0PerBlock_MPerBlock_AK1()
@@ -172,8 +173,8 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         constexpr auto c_block_size =
             c_shuffle_block_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize();
 
-        return math::max((a_block_space_size_aligned + b_block_space_size_aligned) *
-                             sizeof(ComputeDataType),
+        return math::max(a_block_space_size_aligned * sizeof(AComputeDataType) +
+                             b_block_space_size_aligned * sizeof(BComputeDataType),
                          c_block_size * sizeof(CShuffleDataType));
     }
 
@@ -502,7 +503,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                                                 ABlockTransferThreadClusterLengths_AK0_M_AK1,
                                                 ABlockTransferThreadClusterArrangeOrder,
                                                 ADataType,
-                                                ComputeDataType,
+                                                AComputeDataType,
                                                 decltype(a_grid_desc_ak0_m_ak1),
                                                 decltype(a_block_desc_ak0_m_ak1),
                                                 ABlockTransferSrcAccessOrder,
@@ -533,7 +534,7 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
                                                 BBlockTransferThreadClusterLengths_BK0_N_BK1,
                                                 BBlockTransferThreadClusterArrangeOrder,
                                                 BDataType,
-                                                ComputeDataType,
+                                                BComputeDataType,
                                                 decltype(b_grid_desc_bk0_n_bk1),
                                                 decltype(b_block_desc_bk0_n_bk1),
                                                 BBlockTransferSrcAccessOrder,
@@ -561,14 +562,15 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
         //     c_mtx[MPerBlock, NPerBlock] is distributed among threads, and saved in
         //       register
         // sanity check
-        constexpr index_t KPack =
-            math::max(math::lcm(AK1, BK1),
-                      MfmaSelector<ComputeDataType, MPerXdl, NPerXdl>::selected_mfma.k_per_blk);
+        constexpr index_t KPack = math::max(
+            math::lcm(AK1, BK1),
+            MfmaSelector<AComputeDataType, MPerXdl, NPerXdl, BComputeDataType>::selected_mfma
+                .k_per_blk);
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ComputeDataType,
-            ComputeDataType,
+            AComputeDataType,
+            BComputeDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
@@ -586,10 +588,10 @@ struct GridwiseGemmMultipleD_xdl_cshuffle
             a_block_desc_ak0_m_ak1.GetElementSpaceSize(), max_lds_align);
 
         auto a_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ComputeDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
+            static_cast<AComputeDataType*>(p_shared), a_block_desc_ak0_m_ak1.GetElementSpaceSize());
 
         auto b_block_buf = make_dynamic_buffer<AddressSpaceEnum::Lds>(
-            static_cast<ComputeDataType*>(p_shared) + a_block_space_size_aligned,
+            static_cast<BComputeDataType*>(p_shared) + a_block_space_size_aligned,
             b_block_desc_bk0_n_bk1.GetElementSpaceSize());
 
         constexpr auto a_block_slice_copy_step = make_multi_index(KPerBlock / AK1, 0, 0);
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
index 26acf4f5f7..ee86950992 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp
@@ -18,6 +18,8 @@ namespace instance {
 using BF16 = ck::bhalf_t;
 using F16  = ck::half_t;
 using F32  = float;
+using BF8  = ck::bf8_t;
+using F8   = ck::f8_t;
 
 using Empty_Tuple = ck::Tuple<>;
 
@@ -143,6 +145,43 @@ using device_grouped_conv_bwd_data_xdl_f32_instances =
         // clang-format on
         >;
 
+// f16_f16_f16_comp_f8
+template <index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename DsLayout,
+          typename ELayout,
+          ConvolutionBackwardDataSpecialization ConvSpec>
+using device_grouped_conv_bwd_data_xdl_input_fp16_comp_bf8f8_instances =
+    std::tuple<
+        // clang-format off
+         // ##############################################|       NDim| ALayout| BLayout|    DsLayout| ELayout| AData| BData| AccData| CShuffle|      DsData| EData| AElementwise| BElementwise| CDEElementwise| ConvolutionBackward| DoPad| DoPad|      NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer|    MXdl|    NXdl|    ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|    BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CShuffleMXdl| CShuffleNXdl|   CDEBlockTransfer| CDEBlockTransfer|
+         // ##############################################|    Spatial|        |        |            |        |  Type|  Type|    Type| DataType|        Type|  Type|    Operation|    Operation|      Operation|  DataSpecialization| GemmM| GemmN| PrefetchStage|  Size| Block| Block| Block|    |    |  XDL|  XDL| PerWave| PerWave|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraM|     ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar|    ExtraN|      PerWave|      PerWave|  _MBlock_MPerBlock|  ScalarPerVector|
+         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        | Lengths_AK0_M_AK1|   ArrangeOrder|               |               |      PerVector|  PerVector_AK1|          | Lengths_BK0_N_BK1|   ArrangeOrder|               |               |      PerVector|  PerVector_BK1|          |   PerShuffle|   PerShuffle|  _NBlock_NPerBlock|       _NPerBlock|
+         // ##############################################|           |        |        |            |        |      |      |        |         |            |      |             |             |               |                    |      |      |              |      |      |      |      |    |    |     |     |        |        |                  |               |               |               |               |               |          |                  |               |               |               |               |               |          |             |             |                   |                 |
+        // generic instance
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,            1,            1,     S<1, 16, 1, 4>,                1,  LoopScheduler::Default, BF8, F8>,
+        // instances for small conv.K and conv.C
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,            1,            1,     S<1, 32, 1, 4>,                1,  LoopScheduler::Default, BF8, F8>,
+
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   2,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              2,         0,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   256,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   256,    32,   8,   8,   32,   32,       2,       4,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,   128,    32,   8,   8,   32,   32,       4,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    64,   128,    32,   8,   8,   32,   32,       2,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    64,    32,   8,   8,   32,   32,       2,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,   128,    64,    32,   8,   8,   32,   32,       2,       1,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   256,    64,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,   128,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 32, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,   128,    32,   128,    32,   8,   8,   32,   32,       1,       2,       S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,       S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 8>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    64,    32,    32,   8,   8,   32,   32,       2,       1,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,            1,            1,     S<1, 16, 1, 4>,                4,  LoopScheduler::Default, BF8, F8>,
+        DeviceGroupedConvBwdDataMultipleD_Xdl_CShuffle_v1< NDimSpatial, ALayout, BLayout,    DsLayout, ELayout,   F16,   F16,     F32,      F32, Empty_Tuple,   F16,  PassThrough,  PassThrough,    PassThrough,            ConvSpec,  true,  true,             1,    64,    32,    64,    32,   8,   8,   32,   32,       1,       2,       S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,        S<4, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              8,              8,         1,            1,            1,     S<1, 16, 1, 4>,                8,  LoopScheduler::Default, BF8, F8>
+        // clang-format on
+        >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
index b259c640da..5cdaa48bec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_data.hpp
@@ -426,13 +426,32 @@ void add_device_grouped_conv3d_bwd_data_wmma_ndhwgk_gkzyxc_ndhwgc_i8_1x1s1p0_ins
                                                                   PassThrough,
                                                                   PassThrough>>>& instances);
 #endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_input_f16_comp_bf8f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  BF8,
+                                                                  F8>>>& instances);
+#endif
 template <ck::index_t NumDimSpatial,
           typename OutLayout,
           typename WeiLayout,
           typename InLayout,
           typename OutDataType,
           typename WeiDataType,
-          typename InDataType>
+          typename InDataType,
+          typename ComputeTypeA,
+          typename ComputeTypeB>
 struct DeviceOperationInstanceFactory<
     ck::tensor_operation::device::DeviceGroupedConvBwdDataMultipleD<
         NumDimSpatial,
@@ -446,7 +465,9 @@ struct DeviceOperationInstanceFactory<
         InDataType,
         ck::tensor_operation::element_wise::PassThrough,
         ck::tensor_operation::element_wise::PassThrough,
-        ck::tensor_operation::element_wise::PassThrough>>
+        ck::tensor_operation::element_wise::PassThrough,
+        ComputeTypeA,
+        ComputeTypeB>>
 {
     using DeviceOp =
         DeviceGroupedConvBwdDataMultipleD<NumDimSpatial,
@@ -460,7 +481,9 @@ struct DeviceOperationInstanceFactory<
                                           InDataType,
                                           ck::tensor_operation::element_wise::PassThrough,
                                           ck::tensor_operation::element_wise::PassThrough,
-                                          ck::tensor_operation::element_wise::PassThrough>;
+                                          ck::tensor_operation::element_wise::PassThrough,
+                                          ComputeTypeA,
+                                          ComputeTypeB>;
 
     static auto GetInstances()
     {
@@ -597,7 +620,8 @@ struct DeviceOperationInstanceFactory<
             {
 #ifdef CK_ENABLE_FP16
                 if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
-                             is_same_v<OutDataType, F16>)
+                             is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, F16> &&
+                             is_same_v<ComputeTypeB, F16>)
                 {
                     add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_f16_instances(
                         op_ptrs);
@@ -607,6 +631,15 @@ struct DeviceOperationInstanceFactory<
                         op_ptrs);
                 }
 #endif
+#if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
+                else if constexpr(is_same_v<InDataType, F16> && is_same_v<WeiDataType, F16> &&
+                                  is_same_v<OutDataType, F16> && is_same_v<ComputeTypeA, bf8_t> &&
+                                  is_same_v<ComputeTypeB, f8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_input_f16_comp_bf8f8_instances(
+                        op_ptrs);
+                }
+#endif
 #ifdef CK_ENABLE_FP32
                 else if constexpr(is_same_v<InDataType, F32> && is_same_v<WeiDataType, F32> &&
                                   is_same_v<OutDataType, F32>)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
index f44e97ed75..0927cf225d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
@@ -5,7 +5,7 @@ add_instance_library(device_grouped_conv3d_bwd_data_instance
    xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-
+   xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp
new file mode 100644
index 0000000000..d46be53ba4
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp
@@ -0,0 +1,52 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_data/device_grouped_conv_bwd_data_xdl_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+// Compilation parameters for out[n, di, hi, wi, g, c] * wei[g, k, z, y, x, c] = in[n, do, ho, wo,
+// g, k]
+void add_device_grouped_conv3d_bwd_data_xdl_ndhwgk_gkzyxc_ndhwgc_input_f16_comp_bf8f8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdDataMultipleD<3,
+                                                                  NDHWGK,
+                                                                  GKZYXC,
+                                                                  Empty_Tuple,
+                                                                  NDHWGC,
+                                                                  F16,
+                                                                  F16,
+                                                                  Empty_Tuple,
+                                                                  F16,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  PassThrough,
+                                                                  BF8,
+                                                                  F8>>>& instances)
+{
+    // 1. Default
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_data_xdl_input_fp16_comp_bf8f8_instances<3,
+                                                                         NDHWGK,
+                                                                         GKZYXC,
+                                                                         Empty_Tuple,
+                                                                         NDHWGC,
+                                                                         ConvBwdDataDefault>{});
+    // 2. Filter1x1Stride1Pad0
+    add_device_operation_instances(instances,
+                                   device_grouped_conv_bwd_data_xdl_input_fp16_comp_bf8f8_instances<
+                                       3,
+                                       NDHWGK,
+                                       GKZYXC,
+                                       Empty_Tuple,
+                                       NDHWGC,
+                                       ConvBwdDataFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From 570ff3ddbe52d6e1d5e89284d8f3456c0ba34c23 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Thu, 5 Oct 2023 11:41:01 -0500
Subject: [PATCH 161/194] remove example 60 (#963)

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/60_gemm_multiABD/CMakeLists.txt       |  10 -
 .../gemm_multiABD_xdl_fp16.cpp                | 361 ------------------
 2 files changed, 371 deletions(-)
 delete mode 100644 example/60_gemm_multiABD/CMakeLists.txt
 delete mode 100644 example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp

diff --git a/example/60_gemm_multiABD/CMakeLists.txt b/example/60_gemm_multiABD/CMakeLists.txt
deleted file mode 100644
index 9e2f70649f..0000000000
--- a/example/60_gemm_multiABD/CMakeLists.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
-set(target 0)
-foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-     add_example_executable(example_gemm_multiABD_xdl_fp16 gemm_multiABD_xdl_fp16.cpp)
-   set(target 1)
- endif()
-endforeach()
-endif()
diff --git a/example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp b/example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp
deleted file mode 100644
index 23d41d7cc4..0000000000
--- a/example/60_gemm_multiABD/gemm_multiABD_xdl_fp16.cpp
+++ /dev/null
@@ -1,361 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/utility/check_err.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using ADataType        = F16;
-using BDataType        = F16;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DDataType        = F16;
-using EDataType        = F16;
-
-using ALayout = Row;
-using BLayout = Col;
-using DLayout = Row;
-using ELayout = Row;
-
-struct AddScale
-{
-    static constexpr auto I0 = ck::Number<0>{};
-    static constexpr auto I1 = ck::Number<1>{};
-    static constexpr auto I2 = ck::Number<2>{};
-    static constexpr auto I3 = ck::Number<3>{};
-
-    __host__ __device__ constexpr void
-    operator()(ck::half4_t& a, const ck::half4_t& a0, const ck::half4_t& a1) const
-    {
-        const auto a0_v_t = ck::vector_type<ck::half_t, 4>{a0};
-        const auto a1_v_t = ck::vector_type<ck::half_t, 4>{a1};
-
-        auto r_v_t = ck::vector_type<ck::half_t, 4>{};
-
-        r_v_t.AsType<ck::half_t>()(I0) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I0] + a1_v_t.AsType<ck::half_t>()[I0]);
-        r_v_t.AsType<ck::half_t>()(I1) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I1] + a1_v_t.AsType<ck::half_t>()[I1]);
-        r_v_t.AsType<ck::half_t>()(I2) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I2] + a1_v_t.AsType<ck::half_t>()[I2]);
-        r_v_t.AsType<ck::half_t>()(I3) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I3] + a1_v_t.AsType<ck::half_t>()[I3]);
-
-        a = r_v_t.AsType<ck::half4_t>()[I0];
-    }
-
-    __host__ __device__ constexpr void
-    operator()(ck::half_t& a, const ck::half_t& a0, const ck::half_t& a1) const
-    {
-        a = scale * (a0 + a1);
-    }
-
-    static constexpr ck::index_t vec_len = 4;
-
-    float scale = 1.0;
-};
-
-struct AlphaBetaAdd
-{
-    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
-
-    template <typename E, typename C, typename D>
-    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
-
-    template <>
-    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
-        ck::half_t& e, const float& c, const ck::half_t& d) const
-    {
-        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
-    };
-
-    float alpha_;
-    float beta_;
-};
-
-using AElementOp   = AddScale;
-using BElementOp   = PassThrough;
-using CDEElementOp = AlphaBetaAdd;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleABD_Xdl_CShuffle<
-    ck::Tuple<ALayout, ALayout>,
-    ck::Tuple<BLayout>,
-    ck::Tuple<DLayout>,
-    ELayout,
-    ck::Tuple<ADataType, ADataType>,
-    ck::Tuple<BDataType>,
-    AccDataType,
-    CShuffleDataType,
-    ck::Tuple<DDataType>,
-    EDataType,
-    AElementOp,
-    BElementOp,
-    CDEElementOp,
-    GemmSpec,
-    1,
-    256,
-    256,
-    128,
-    32,
-    8,
-    8,
-    32,
-    32,
-    4,
-    2,
-    S<4, 64, 1>,
-    S<1, 0, 2>,
-    S<1, 0, 2>,
-    2,
-    8,
-    8,
-    1,
-    S<4, 64, 1>,
-    S<1, 0, 2>,
-    S<1, 0, 2>,
-    2,
-    8,
-    8,
-    1,
-    1,
-    1,
-    S<1, 32, 1, 8>,
-    8>;
-
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // GEMM shape
-    ck::index_t M = 3840;
-    ck::index_t N = 4096;
-    ck::index_t K = 4096;
-
-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideD = 4096;
-    ck::index_t StrideE = 4096;
-
-    float alpha = 1.0f;
-    float beta  = 1.0f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 6)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        alpha = std::stof(argv[4]);
-        beta  = std::stof(argv[5]);
-    }
-    else if(argc == 13)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        M = std::stoi(argv[4]);
-        N = std::stoi(argv[5]);
-        K = std::stoi(argv[6]);
-
-        StrideA = std::stoi(argv[7]);
-        StrideB = std::stoi(argv[8]);
-        StrideD = std::stoi(argv[9]);
-        StrideE = std::stoi(argv[10]);
-
-        alpha = std::stof(argv[11]);
-        beta  = std::stof(argv[12]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
-               "beta\n");
-        exit(0);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            using namespace ck::literals;
-
-            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
-            {
-                return HostTensorDescriptor({row, col}, {stride, 1_uz});
-            }
-            else
-            {
-                return HostTensorDescriptor({row, col}, {1_uz, stride});
-            }
-        };
-
-    Tensor<ADataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
-    Tensor<ADataType> a1_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
-    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
-    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
-    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
-    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
-
-    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
-    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
-    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a0_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
-        break;
-    default:
-        a0_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
-    }
-
-    DeviceMem a0_device_buf(sizeof(ADataType) * a0_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem a1_device_buf(sizeof(ADataType) * a1_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a0_device_buf.ToDevice(a0_m_k.mData.data());
-    a1_device_buf.ToDevice(a1_m_k.mData.data());
-    b_device_buf.ToDevice(b_k_n.mData.data());
-    d_device_buf.ToDevice(d_m_n.mData.data());
-    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
-
-    auto a_element_op   = AElementOp{0.2};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{alpha, beta};
-
-    // do GEMM
-    auto device_op = DeviceOpInstance{};
-    auto invoker   = device_op.MakeInvoker();
-    auto argument =
-        device_op.MakeArgument(std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
-                                                          a1_device_buf.GetDeviceBuffer()},
-                               std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
-                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
-                               e_device_buf.GetDeviceBuffer(),
-                               M,
-                               N,
-                               K,
-                               std::array<ck::index_t, 2>{StrideA, StrideA},
-                               std::array<ck::index_t, 1>{StrideB},
-                               std::array<ck::index_t, 1>{StrideD},
-                               StrideE,
-                               a_element_op,
-                               b_element_op,
-                               cde_element_op);
-
-    if(!device_op.IsSupportedArgument(argument))
-    {
-        throw std::runtime_error(
-            "wrong! device_gemm with the specified compilation parameters does "
-            "not support this GEMM problem");
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    std::size_t flop = std::size_t(2) * M * N * K;
-    std::size_t num_btype =
-        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
-              << std::endl;
-
-    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_m_n({M, N});
-
-        Tensor<ADataType> a_m_k({M, K});
-
-        for(int m = 0; m < M; ++m)
-        {
-            for(int k = 0; k < K; ++k)
-            {
-                a_element_op(a_m_k(m, k), a0_m_k(m, k), a1_m_k(m, k));
-            }
-        }
-
-        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
-                                                                                BDataType,
-                                                                                CShuffleDataType,
-                                                                                AccDataType,
-                                                                                PassThrough,
-                                                                                BElementOp,
-                                                                                PassThrough>;
-        auto ref_gemm               = ReferenceGemmInstance{};
-        auto ref_invoker            = ref_gemm.MakeInvoker();
-
-        auto ref_argument =
-            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, b_element_op, PassThrough{});
-
-        ref_invoker.Run(ref_argument);
-
-        for(int m = 0; m < M; ++m)
-        {
-            for(int n = 0; n < N; ++n)
-            {
-                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
-            }
-        }
-
-        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
-
-        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}

From 4daedf8ca56f3bd93481708bd9d762045839ec20 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 5 Oct 2023 14:58:23 -0700
Subject: [PATCH 162/194] Revert "Add support for mixed precision in
 contraction scale and bilinear" (#967)

* Revert "Add support for mixed precision in contraction scale and bilinear (#936)"

This reverts commit f07485060e23b871872640a5b1d39fe05bd53a5d.

* revert commits #957 and #960
---
 example/26_contraction/CMakeLists.txt         |  48 -
 example/26_contraction/common_instances.hpp   | 183 ----
 ...raction_bilinear_xdl_bf16_compute_fp32.cpp |  86 --
 ...raction_bilinear_xdl_fp16_compute_fp32.cpp |  86 --
 .../contraction_bilinear_xdl_fp32.cpp         | 319 +++++--
 ...raction_bilinear_xdl_fp32_compute_bf16.cpp |  86 --
 ...raction_bilinear_xdl_fp32_compute_fp16.cpp |  86 --
 .../contraction_bilinear_xdl_fp64.cpp         | 319 +++++--
 ...raction_bilinear_xdl_fp64_compute_fp32.cpp |  86 --
 ...ontraction_scale_xdl_bf16_compute_fp32.cpp |  85 --
 ...ontraction_scale_xdl_fp16_compute_fp32.cpp |  85 --
 .../contraction_scale_xdl_fp32.cpp            | 303 +++++--
 ...ontraction_scale_xdl_fp32_compute_bf16.cpp |  85 --
 ...ontraction_scale_xdl_fp32_compute_fp16.cpp |  85 --
 .../contraction_scale_xdl_fp64.cpp            | 303 +++++--
 ...ontraction_scale_xdl_fp64_compute_fp32.cpp |  85 --
 .../run_contraction_bilinear_example.inc      | 234 -----
 .../run_contraction_scale_example.inc         | 217 -----
 example/60_gemm_multi_ABD/CMakeLists.txt      |  10 -
 .../gemm_multi_ABD_xdl_fp16.cpp               | 362 --------
 .../61_contraction_multi_ABD/CMakeLists.txt   |  10 -
 .../contraction_multi_ABD_xdl_fp16.cpp        | 329 -------
 .../device_contraction_multiple_abd.hpp       |  61 --
 .../device/device_contraction_multiple_d.hpp  |   3 +-
 ..._contraction_multiple_abd_xdl_cshuffle.hpp | 846 ------------------
 ...ce_contraction_multiple_d_xdl_cshuffle.hpp |   6 +-
 .../element/binary_element_wise_operation.hpp |  19 -
 .../element/unary_element_wise_operation.hpp  |  26 -
 ...ridwise_gemm_multiple_abd_xdl_cshuffle.hpp |   3 +-
 .../cpu/reference_contraction.hpp             |  16 +-
 .../device_operation_instance_factory.hpp     |   2 -
 .../device_contraction_instance.hpp           | 292 ------
 .../gpu/contraction_bilinear.hpp              | 424 +--------
 .../gpu/contraction_scale.hpp                 | 420 +--------
 .../gpu/contraction_bilinear/CMakeLists.txt   |  44 +-
 ...16_bf16_bf16_compute_f32_kknn_instance.cpp |  57 --
 ...16_bf16_bf16_compute_f32_knnn_instance.cpp |  57 --
 ...16_bf16_bf16_compute_f32_mknn_instance.cpp |  57 --
 ...16_bf16_bf16_compute_f32_mnnn_instance.cpp |  57 --
 ..._f16_f16_f16_compute_f32_kknn_instance.cpp |  57 --
 ..._f16_f16_f16_compute_f32_knnn_instance.cpp |  57 --
 ..._f16_f16_f16_compute_f32_mknn_instance.cpp |  57 --
 ..._f16_f16_f16_compute_f32_mnnn_instance.cpp |  57 --
 ...f32_f32_f32_compute_bf16_kknn_instance.cpp |  57 --
 ...f32_f32_f32_compute_bf16_knnn_instance.cpp |  57 --
 ...f32_f32_f32_compute_bf16_mknn_instance.cpp |  57 --
 ...f32_f32_f32_compute_bf16_mnnn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_f16_kknn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_f16_knnn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_f16_mknn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_f16_mnnn_instance.cpp |  57 --
 ..._shuffle_f32_f32_f32_f32_kknn_instance.cpp |  52 +-
 ..._shuffle_f32_f32_f32_f32_knnn_instance.cpp |  55 +-
 ..._shuffle_f32_f32_f32_f32_mknn_instance.cpp |  55 +-
 ..._shuffle_f32_f32_f32_f32_mnnn_instance.cpp |  55 +-
 ..._f64_f64_f64_compute_f32_kknn_instance.cpp |  57 --
 ..._f64_f64_f64_compute_f32_knnn_instance.cpp |  57 --
 ..._f64_f64_f64_compute_f32_mknn_instance.cpp |  57 --
 ..._f64_f64_f64_compute_f32_mnnn_instance.cpp |  57 --
 ..._shuffle_f64_f64_f64_f64_kknn_instance.cpp |  49 +-
 ..._shuffle_f64_f64_f64_f64_knnn_instance.cpp |  49 +-
 ..._shuffle_f64_f64_f64_f64_mknn_instance.cpp |  49 +-
 ..._shuffle_f64_f64_f64_f64_mnnn_instance.cpp |  49 +-
 .../gpu/contraction_scale/CMakeLists.txt      |  32 +-
 ...f16_bf16_bf16_compute_f32_kkn_instance.cpp |  57 --
 ...f16_bf16_bf16_compute_f32_knn_instance.cpp |  57 --
 ...f16_bf16_bf16_compute_f32_mkn_instance.cpp |  57 --
 ...f16_bf16_bf16_compute_f32_mnn_instance.cpp |  57 --
 ...e_f16_f16_f16_compute_f32_kkn_instance.cpp |  57 --
 ...e_f16_f16_f16_compute_f32_knn_instance.cpp |  57 --
 ...e_f16_f16_f16_compute_f32_mkn_instance.cpp |  57 --
 ...e_f16_f16_f16_compute_f32_mnn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_bf16_kkn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_bf16_knn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_bf16_mkn_instance.cpp |  57 --
 ..._f32_f32_f32_compute_bf16_mnn_instance.cpp |  57 --
 ...e_f32_f32_f32_compute_f16_kkn_instance.cpp |  57 --
 ...e_f32_f32_f32_compute_f16_knn_instance.cpp |  57 --
 ...e_f32_f32_f32_compute_f16_mkn_instance.cpp |  57 --
 ...e_f32_f32_f32_compute_f16_mnn_instance.cpp |  57 --
 ...xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp |  56 +-
 ...xdl_c_shuffle_f32_f32_f32_knn_instance.cpp |  59 +-
 ...xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp |  59 +-
 ...xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp |  59 +-
 ...e_f64_f64_f64_compute_f32_kkn_instance.cpp |  57 --
 ...e_f64_f64_f64_compute_f32_knn_instance.cpp |  57 --
 ...e_f64_f64_f64_compute_f32_mkn_instance.cpp |  57 --
 ...e_f64_f64_f64_compute_f32_mnn_instance.cpp |  57 --
 ...xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp |  49 +-
 ...xdl_c_shuffle_f64_f64_f64_knn_instance.cpp |  49 +-
 ...xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp |  49 +-
 ...xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp |  49 +-
 profiler/README.md                            |  24 +-
 .../profiler/profile_contraction_impl.hpp     |  62 +-
 .../profiler/profile_contraction_utils.hpp    |  14 +-
 profiler/src/profile_contraction_bilinear.cpp | 219 ++---
 profiler/src/profile_contraction_scale.cpp    | 215 ++---
 test/contraction/test_contraction.cpp         | 145 ++-
 .../test_contraction_interface.cpp            |  10 +-
 99 files changed, 1961 insertions(+), 7536 deletions(-)
 delete mode 100644 example/26_contraction/common_instances.hpp
 delete mode 100644 example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp
 delete mode 100644 example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp
 delete mode 100644 example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp
 delete mode 100644 example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp
 delete mode 100644 example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp
 delete mode 100644 example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp
 delete mode 100644 example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp
 delete mode 100644 example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp
 delete mode 100644 example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp
 delete mode 100644 example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp
 delete mode 100644 example/26_contraction/run_contraction_bilinear_example.inc
 delete mode 100644 example/26_contraction/run_contraction_scale_example.inc
 delete mode 100644 example/60_gemm_multi_ABD/CMakeLists.txt
 delete mode 100644 example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
 delete mode 100644 example/61_contraction_multi_ABD/CMakeLists.txt
 delete mode 100644 example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
 delete mode 100644 include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
 delete mode 100644 include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
 delete mode 100644 library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
 delete mode 100644 library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp

diff --git a/example/26_contraction/CMakeLists.txt b/example/26_contraction/CMakeLists.txt
index 1a0489ce9c..167a9f147d 100644
--- a/example/26_contraction/CMakeLists.txt
+++ b/example/26_contraction/CMakeLists.txt
@@ -1,52 +1,4 @@
-add_custom_target(example_contraction)
-add_custom_target(example_contraction_scale)
-add_custom_target(example_contraction_bilinear)
-
-# FP32
 add_example_executable(example_contraction_bilinear_xdl_fp32 contraction_bilinear_xdl_fp32.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32)
-
 add_example_executable(example_contraction_scale_xdl_fp32 contraction_scale_xdl_fp32.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32)
-
-add_example_executable(example_contraction_bilinear_xdl_fp32_compute_bf16 contraction_bilinear_xdl_fp32_compute_bf16.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_bf16)
-
-add_example_executable(example_contraction_scale_xdl_fp32_compute_bf16 contraction_scale_xdl_fp32_compute_bf16.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_bf16)
-
-add_example_executable(example_contraction_bilinear_xdl_fp32_compute_fp16 contraction_bilinear_xdl_fp32_compute_fp16.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp32_compute_fp16)
-
-add_example_executable(example_contraction_scale_xdl_fp32_compute_fp16 contraction_scale_xdl_fp32_compute_fp16.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp32_compute_fp16)
-
-# FP64
 add_example_executable(example_contraction_bilinear_xdl_fp64 contraction_bilinear_xdl_fp64.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64)
-
 add_example_executable(example_contraction_scale_xdl_fp64 contraction_scale_xdl_fp64.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64)
-
-add_example_executable(example_contraction_bilinear_xdl_fp64_compute_fp32 contraction_bilinear_xdl_fp64_compute_fp32.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp64_compute_fp32)
-
-add_example_executable(example_contraction_scale_xdl_fp64_compute_fp32 contraction_scale_xdl_fp64_compute_fp32.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp64_compute_fp32)
-
-# FP16
-add_example_executable(example_contraction_bilinear_xdl_fp16_compute_fp32 contraction_bilinear_xdl_fp16_compute_fp32.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_fp16_compute_fp32)
-
-add_example_executable(example_contraction_scale_xdl_fp16_compute_fp32 contraction_scale_xdl_fp16_compute_fp32.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_fp16_compute_fp32)
-
-# BF16
-add_example_executable(example_contraction_bilinear_xdl_bf16_compute_fp32 contraction_bilinear_xdl_bf16_compute_fp32.cpp)
-add_dependencies(example_contraction_bilinear example_contraction_bilinear_xdl_bf16_compute_fp32)
-
-add_example_executable(example_contraction_scale_xdl_bf16_compute_fp32 contraction_scale_xdl_bf16_compute_fp32.cpp)
-add_dependencies(example_contraction_scale example_contraction_scale_xdl_bf16_compute_fp32)
-
-add_dependencies(example_contraction example_contraction_scale)
-add_dependencies(example_contraction example_contraction_bilinear)
diff --git a/example/26_contraction/common_instances.hpp b/example/26_contraction/common_instances.hpp
deleted file mode 100644
index eb2fa5ce73..0000000000
--- a/example/26_contraction/common_instances.hpp
+++ /dev/null
@@ -1,183 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
-
-using F16  = ck::half_t;
-using BF16 = ck::bhalf_t;
-using F32  = float;
-using F64  = double;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-// Generic instances for fp32, fp16 and bf16 data types.
-// clang-format off
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceKK_Generic = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
-
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceKN_Generic = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
-
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceMK_Generic = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
-
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceMN_Generic = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
-
-// Fp64 instances.
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceKK_FP64 = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
-
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceKN_FP64 = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
-
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceMK_FP64 = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
-
-template <ck::index_t NumDimM,
-          ck::index_t NumDimN,
-          ck::index_t NumDimK,
-          typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementOp,
-          typename BElementOp,
-          typename CDEElementOp>
-using DeviceOpInstanceMN_FP64 = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK|     AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |      Type|      Type|        Type|         DataType|       Type|      Type|            Data| Elementwise| Elementwise|  Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |          |          |            |                 |           |          |            Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |          |          |            |                 |           |          |                |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK, ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
-
-// clang-format on
diff --git a/example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp
deleted file mode 100644
index 77ec5669a7..0000000000
--- a/example/26_contraction/contraction_bilinear_xdl_bf16_compute_fp32.cpp
+++ /dev/null
@@ -1,86 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = BF16;
-using BDataType        = BF16;
-using AccDataType      = F32;
-using CShuffleDataType = BF16;
-using DDataType        = BF16;
-using DsDataType       = ck::Tuple<DDataType>;
-using EDataType        = BF16;
-using ComputeDataType  = F32;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
-
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKNN;
-
-#include "run_contraction_bilinear_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp
deleted file mode 100644
index 25c4d9eac4..0000000000
--- a/example/26_contraction/contraction_bilinear_xdl_fp16_compute_fp32.cpp
+++ /dev/null
@@ -1,86 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F16;
-using BDataType        = F16;
-using AccDataType      = F32;
-using CShuffleDataType = F16;
-using DDataType        = F16;
-using DsDataType       = ck::Tuple<DDataType>;
-using EDataType        = F16;
-using ComputeDataType  = F32;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
-
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKNN;
-
-#include "run_contraction_bilinear_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
index d440e7394e..78522160c8 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp32.cpp
@@ -1,10 +1,29 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
 #include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "common_instances.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F32 = float;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 using ADataType        = F32;
 using BDataType        = F32;
@@ -13,7 +32,6 @@ using CShuffleDataType = F32;
 using DDataType        = F32;
 using DsDataType       = ck::Tuple<DDataType>;
 using EDataType        = F32;
-using ComputeDataType  = F32;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -23,64 +41,253 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
 
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
+// clang-format off
+using DeviceOpInstanceKKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
 
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
+using DeviceOpInstanceKNNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
 
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
+using DeviceOpInstanceMKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
+
+using DeviceOpInstanceMNNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
+// clang-format on
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-#include "run_contraction_bilinear_example.inc"
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
 
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float alpha = 1.f;
+    float beta  = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 28)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        d_ms_ns_lengths = {M0, M1, N0, N1};
+        d_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
+
+        alpha = std::stof(argv[26]);
+        beta  = std::stof(argv[27]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 7: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
+        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg26 to 27: alpha, beta\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop      = std::size_t(2) * M * N * K;
+    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1),
+                                       d_ms_ns(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp
deleted file mode 100644
index 5b748949bd..0000000000
--- a/example/26_contraction/contraction_bilinear_xdl_fp32_compute_bf16.cpp
+++ /dev/null
@@ -1,86 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F32;
-using BDataType        = F32;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DDataType        = F32;
-using DsDataType       = ck::Tuple<DDataType>;
-using EDataType        = F32;
-using ComputeDataType  = BF16;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
-
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKNN;
-
-#include "run_contraction_bilinear_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp b/example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp
deleted file mode 100644
index 334cd615d1..0000000000
--- a/example/26_contraction/contraction_bilinear_xdl_fp32_compute_fp16.cpp
+++ /dev/null
@@ -1,86 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F32;
-using BDataType        = F32;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DDataType        = F32;
-using DsDataType       = ck::Tuple<DDataType>;
-using EDataType        = F32;
-using ComputeDataType  = F16;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
-
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                        NumDimN,
-                                                        NumDimK,
-                                                        ADataType,
-                                                        BDataType,
-                                                        AccDataType,
-                                                        CShuffleDataType,
-                                                        DsDataType,
-                                                        EDataType,
-                                                        ComputeDataType,
-                                                        AElementOp,
-                                                        BElementOp,
-                                                        CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKNN;
-
-#include "run_contraction_bilinear_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
index 86ea7e0212..6cceed5bc1 100644
--- a/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_bilinear_xdl_fp64.cpp
@@ -1,10 +1,29 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
 #include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "common_instances.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F64 = double;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 using ADataType        = F64;
 using BDataType        = F64;
@@ -13,7 +32,6 @@ using CShuffleDataType = F64;
 using DDataType        = F64;
 using DsDataType       = ck::Tuple<DDataType>;
 using EDataType        = F64;
-using ComputeDataType  = F64;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -23,64 +41,253 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
 
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
+// clang-format off
+using DeviceOpInstanceKKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
 
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
+using DeviceOpInstanceKNNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
 
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
+using DeviceOpInstanceMKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceMNNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+// clang-format on
 
 using DeviceOpInstance = DeviceOpInstanceKKNN;
 
-#include "run_contraction_bilinear_example.inc"
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
 
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float alpha = 1.f;
+    float beta  = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 28)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        d_ms_ns_lengths = {M0, M1, N0, N1};
+        d_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
+
+        alpha = std::stof(argv[26]);
+        beta  = std::stof(argv[27]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 7: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
+        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg26 to 27: alpha, beta\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop      = std::size_t(2) * M * N * K;
+    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
+                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1),
+                                       d_ms_ns(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp b/example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp
deleted file mode 100644
index b0a2a9a10f..0000000000
--- a/example/26_contraction/contraction_bilinear_xdl_fp64_compute_fp32.cpp
+++ /dev/null
@@ -1,86 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F64;
-using BDataType        = F64;
-using AccDataType      = F32;
-using CShuffleDataType = F64;
-using DDataType        = F64;
-using DsDataType       = ck::Tuple<DDataType>;
-using EDataType        = F64;
-using ComputeDataType  = F32;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Bilinear;
-
-using DeviceOpInstanceKKNN = DeviceOpInstanceKK_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
-
-using DeviceOpInstanceKNNN = DeviceOpInstanceKN_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
-
-using DeviceOpInstanceMKNN = DeviceOpInstanceMK_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
-
-using DeviceOpInstanceMNNN = DeviceOpInstanceMN_FP64<NumDimM,
-                                                     NumDimN,
-                                                     NumDimK,
-                                                     ADataType,
-                                                     BDataType,
-                                                     AccDataType,
-                                                     CShuffleDataType,
-                                                     DsDataType,
-                                                     EDataType,
-                                                     ComputeDataType,
-                                                     AElementOp,
-                                                     BElementOp,
-                                                     CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKNN;
-
-#include "run_contraction_bilinear_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_bilinear_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp b/example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp
deleted file mode 100644
index 32652fc6f3..0000000000
--- a/example/26_contraction/contraction_scale_xdl_bf16_compute_fp32.cpp
+++ /dev/null
@@ -1,85 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = BF16;
-using BDataType        = BF16;
-using AccDataType      = F32;
-using CShuffleDataType = BF16;
-using DsDataType       = ck::Tuple<>;
-using EDataType        = BF16;
-using ComputeDataType  = F32;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Scale;
-
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKN;
-
-#include "run_contraction_scale_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp
deleted file mode 100644
index a4797e2400..0000000000
--- a/example/26_contraction/contraction_scale_xdl_fp16_compute_fp32.cpp
+++ /dev/null
@@ -1,85 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F16;
-using BDataType        = F16;
-using AccDataType      = F32;
-using CShuffleDataType = F16;
-using DsDataType       = ck::Tuple<>;
-using EDataType        = F16;
-using ComputeDataType  = F32;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Scale;
-
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKN;
-
-#include "run_contraction_scale_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp32.cpp
index fe984b29d6..1574f5d18f 100644
--- a/example/26_contraction/contraction_scale_xdl_fp32.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp32.cpp
@@ -1,10 +1,29 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
 #include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "common_instances.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F32 = float;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 using ADataType        = F32;
 using BDataType        = F32;
@@ -12,7 +31,6 @@ using AccDataType      = F32;
 using CShuffleDataType = F32;
 using DsDataType       = ck::Tuple<>;
 using EDataType        = F32;
-using ComputeDataType  = F32;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -22,64 +40,237 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Scale;
 
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
+// clang-format off
+using DeviceOpInstanceKKN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
 
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
+using DeviceOpInstanceKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
 
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
+using DeviceOpInstanceMKN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>;
+
+using DeviceOpInstanceMNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F32,   F32,     F32,      F32, DsDataType,   F32,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>;
+// clang-format on
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-#include "run_contraction_scale_example.inc"
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
 
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float scale = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 23)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        scale = std::stof(argv[22]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg22: scale\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{scale};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 0>{},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp b/example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp
deleted file mode 100644
index 83c8fec179..0000000000
--- a/example/26_contraction/contraction_scale_xdl_fp32_compute_bf16.cpp
+++ /dev/null
@@ -1,85 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F32;
-using BDataType        = F32;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DsDataType       = ck::Tuple<>;
-using EDataType        = F32;
-using ComputeDataType  = BF16;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Scale;
-
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKN;
-
-#include "run_contraction_scale_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp b/example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp
deleted file mode 100644
index 27e42e8203..0000000000
--- a/example/26_contraction/contraction_scale_xdl_fp32_compute_fp16.cpp
+++ /dev/null
@@ -1,85 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F32;
-using BDataType        = F32;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DsDataType       = ck::Tuple<>;
-using EDataType        = F32;
-using ComputeDataType  = F16;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Scale;
-
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_Generic<NumDimM,
-                                                       NumDimN,
-                                                       NumDimK,
-                                                       ADataType,
-                                                       BDataType,
-                                                       AccDataType,
-                                                       CShuffleDataType,
-                                                       DsDataType,
-                                                       EDataType,
-                                                       ComputeDataType,
-                                                       AElementOp,
-                                                       BElementOp,
-                                                       CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKN;
-
-#include "run_contraction_scale_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/contraction_scale_xdl_fp64.cpp b/example/26_contraction/contraction_scale_xdl_fp64.cpp
index 37374780b6..3dacc70887 100644
--- a/example/26_contraction/contraction_scale_xdl_fp64.cpp
+++ b/example/26_contraction/contraction_scale_xdl_fp64.cpp
@@ -1,10 +1,29 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
 #include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "common_instances.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/numeric.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F64 = double;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 using ADataType        = F64;
 using BDataType        = F64;
@@ -12,7 +31,6 @@ using AccDataType      = F64;
 using CShuffleDataType = F64;
 using DsDataType       = ck::Tuple<>;
 using EDataType        = F64;
-using ComputeDataType  = F64;
 
 static constexpr ck::index_t NumDimM = 2;
 static constexpr ck::index_t NumDimN = 2;
@@ -22,64 +40,237 @@ using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
 using CDEElementOp = ck::tensor_operation::element_wise::Scale;
 
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
 
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
+// clang-format off
+using DeviceOpInstanceKKN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
 
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
+using DeviceOpInstanceKNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              2,              2,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
 
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
+using DeviceOpInstanceMKN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              2,              2,         1,           1,           1,              S<1, 16, 1, 16>,               1>;
+
+using DeviceOpInstanceMNN = ck::tensor_operation::device::
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|     DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|       Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |           |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |           |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle< NumDimM, NumDimN, NumDimK,   F64,   F64,     F64,      F64, DsDataType,   F64,   AElementOp,  BElementOp, CDEElementOp,       GemmSpec,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>;
+// clang-format on
 
 using DeviceOpInstance = DeviceOpInstanceKKN;
 
-#include "run_contraction_scale_example.inc"
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
 
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
+    // A[M0, M1, K0, K1]
+    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
+
+    float scale = 1.f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 23)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        const ck::index_t M0 = std::stoi(argv[4]);
+        const ck::index_t M1 = std::stoi(argv[5]);
+
+        const ck::index_t N0 = std::stoi(argv[6]);
+        const ck::index_t N1 = std::stoi(argv[7]);
+
+        const ck::index_t K0 = std::stoi(argv[8]);
+        const ck::index_t K1 = std::stoi(argv[9]);
+
+        a_ms_ks_lengths = {M0, M1, K0, K1};
+        a_ms_ks_strides = {
+            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
+
+        b_ns_ks_lengths = {N0, N1, K0, K1};
+        b_ns_ks_strides = {
+            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
+
+        e_ms_ns_lengths = {M0, M1, N0, N1};
+        e_ms_ns_strides = {
+            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
+
+        scale = std::stof(argv[22]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
+        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
+        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
+        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
+        printf("arg22: scale\n");
+        exit(0);
+    }
+
+    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a_device_buf.ToDevice(a_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{scale};
+
+    // device operation
+    auto op       = DeviceOpInstance{};
+    auto invoker  = op.MakeInvoker();
+    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
+                                    b_device_buf.GetDeviceBuffer(),
+                                    std::array<const void*, 0>{},
+                                    e_device_buf.GetDeviceBuffer(),
+                                    a_ms_ks_lengths,
+                                    a_ms_ks_strides,
+                                    b_ns_ks_lengths,
+                                    b_ns_ks_strides,
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    std::array<std::vector<ck::index_t>, 0>{},
+                                    e_ms_ns_lengths,
+                                    e_ms_ns_strides,
+                                    a_element_op,
+                                    b_element_op,
+                                    cde_element_op);
+
+    if(!op.IsSupportedArgument(argument))
+    {
+        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
+
+        return 0;
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    ck::index_t M =
+        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+    ck::index_t N = ck::accumulate_n<ck::index_t>(
+        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+    ck::index_t K = ck::accumulate_n<ck::index_t>(
+        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << op.GetTypeString() << std::endl;
+
+    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      ADataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      AElementOp,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp b/example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp
deleted file mode 100644
index 5dbeb23154..0000000000
--- a/example/26_contraction/contraction_scale_xdl_fp64_compute_fp32.cpp
+++ /dev/null
@@ -1,85 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "common_instances.hpp"
-
-using ADataType        = F64;
-using BDataType        = F64;
-using AccDataType      = F32;
-using CShuffleDataType = F64;
-using DsDataType       = ck::Tuple<>;
-using EDataType        = F64;
-using ComputeDataType  = F32;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-using AElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using BElementOp   = ck::tensor_operation::element_wise::PassThrough;
-using CDEElementOp = ck::tensor_operation::element_wise::Scale;
-
-using DeviceOpInstanceKKN = DeviceOpInstanceKK_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
-
-using DeviceOpInstanceKNN = DeviceOpInstanceKN_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
-
-using DeviceOpInstanceMKN = DeviceOpInstanceMK_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
-
-using DeviceOpInstanceMNN = DeviceOpInstanceMN_FP64<NumDimM,
-                                                    NumDimN,
-                                                    NumDimK,
-                                                    ADataType,
-                                                    BDataType,
-                                                    AccDataType,
-                                                    CShuffleDataType,
-                                                    DsDataType,
-                                                    EDataType,
-                                                    ComputeDataType,
-                                                    AElementOp,
-                                                    BElementOp,
-                                                    CDEElementOp>;
-
-using DeviceOpInstance = DeviceOpInstanceKKN;
-
-#include "run_contraction_scale_example.inc"
-
-int main(int argc, char* argv[]) { return run_contraction_scale_example(argc, argv); }
diff --git a/example/26_contraction/run_contraction_bilinear_example.inc b/example/26_contraction/run_contraction_bilinear_example.inc
deleted file mode 100644
index 7b9a01e0ae..0000000000
--- a/example/26_contraction/run_contraction_bilinear_example.inc
+++ /dev/null
@@ -1,234 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <cstdlib>
-#include <iostream>
-#include <string>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/numeric.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-
-int run_contraction_bilinear_example(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // A[M0, M1, K0, K1]
-    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
-    // D[M0, M1, N0, N1]
-    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> d_ms_ns_strides{524288, 4096, 128, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
-
-    float alpha = 1.f;
-    float beta  = 1.f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 28)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        const ck::index_t M0 = std::stoi(argv[4]);
-        const ck::index_t M1 = std::stoi(argv[5]);
-
-        const ck::index_t N0 = std::stoi(argv[6]);
-        const ck::index_t N1 = std::stoi(argv[7]);
-
-        const ck::index_t K0 = std::stoi(argv[8]);
-        const ck::index_t K1 = std::stoi(argv[9]);
-
-        a_ms_ks_lengths = {M0, M1, K0, K1};
-        a_ms_ks_strides = {
-            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
-
-        b_ns_ks_lengths = {N0, N1, K0, K1};
-        b_ns_ks_strides = {
-            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
-
-        d_ms_ns_lengths = {M0, M1, N0, N1};
-        d_ms_ns_strides = {
-            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
-
-        e_ms_ns_lengths = {M0, M1, N0, N1};
-        e_ms_ns_strides = {
-            std::stoi(argv[22]), std::stoi(argv[23]), std::stoi(argv[24]), std::stoi(argv[25])};
-
-        alpha = std::stof(argv[26]);
-        beta  = std::stof(argv[27]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
-        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
-        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
-        printf("arg18 to 21: Stride_D_M0, Stride_D_M1, Stride_D_N0, Stride_D_N1\n");
-        printf("arg22 to 25: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
-        printf("arg26 to 27: alpha, beta\n");
-        exit(0);
-    }
-
-    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-    d_device_buf.ToDevice(d_ms_ns.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{alpha, beta};
-
-    // device operation
-    auto op       = DeviceOpInstance{};
-    auto invoker  = op.MakeInvoker();
-    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
-                                    b_device_buf.GetDeviceBuffer(),
-                                    std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
-                                    e_device_buf.GetDeviceBuffer(),
-                                    a_ms_ks_lengths,
-                                    a_ms_ks_strides,
-                                    b_ns_ks_lengths,
-                                    b_ns_ks_strides,
-                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
-                                    std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
-                                    e_ms_ns_lengths,
-                                    e_ms_ns_strides,
-                                    a_element_op,
-                                    b_element_op,
-                                    cde_element_op);
-
-    if(!op.IsSupportedArgument(argument))
-    {
-        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    ck::index_t M =
-        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-    ck::index_t N = ck::accumulate_n<ck::index_t>(
-        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-    ck::index_t K = ck::accumulate_n<ck::index_t>(
-        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-    std::size_t flop      = std::size_t(2) * M * N * K;
-    std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
-                            sizeof(DDataType) * M * N + sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
-              << op.GetTypeString() << std::endl;
-
-    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      ComputeDataType,
-                                                                      AElementOp,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1),
-                                       d_ms_ns(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
diff --git a/example/26_contraction/run_contraction_scale_example.inc b/example/26_contraction/run_contraction_scale_example.inc
deleted file mode 100644
index 65ca182da2..0000000000
--- a/example/26_contraction/run_contraction_scale_example.inc
+++ /dev/null
@@ -1,217 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <cstdlib>
-#include <iostream>
-#include <string>
-#include <vector>
-
-#include "ck/ck.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/numeric.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-
-int run_contraction_scale_example(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // A[M0, M1, K0, K1]
-    std::vector<ck::index_t> a_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a_ms_ks_strides{524288, 4096, 128, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{524288, 4096, 128, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{524288, 4096, 128, 1};
-
-    float scale = 1.f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 23)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        const ck::index_t M0 = std::stoi(argv[4]);
-        const ck::index_t M1 = std::stoi(argv[5]);
-
-        const ck::index_t N0 = std::stoi(argv[6]);
-        const ck::index_t N1 = std::stoi(argv[7]);
-
-        const ck::index_t K0 = std::stoi(argv[8]);
-        const ck::index_t K1 = std::stoi(argv[9]);
-
-        a_ms_ks_lengths = {M0, M1, K0, K1};
-        a_ms_ks_strides = {
-            std::stoi(argv[10]), std::stoi(argv[11]), std::stoi(argv[12]), std::stoi(argv[13])};
-
-        b_ns_ks_lengths = {N0, N1, K0, K1};
-        b_ns_ks_strides = {
-            std::stoi(argv[14]), std::stoi(argv[15]), std::stoi(argv[16]), std::stoi(argv[17])};
-
-        e_ms_ns_lengths = {M0, M1, N0, N1};
-        e_ms_ns_strides = {
-            std::stoi(argv[18]), std::stoi(argv[19]), std::stoi(argv[20]), std::stoi(argv[21])};
-
-        scale = std::stof(argv[22]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 9: M0, M1, N0, N1, K0, K1\n");
-        printf("arg10 to 13: Stride_A_M0, Stride_A_M1, Stride_A_K0, Stride_A_K1\n");
-        printf("arg14 to 17: Stride_B_N0, Stride_B_N1, Stride_B_K0, Stride_B_K1\n");
-        printf("arg18 to 21: Stride_E_M0, Stride_E_M1, Stride_E_N0, Stride_E_N1\n");
-        printf("arg22: scale\n");
-        exit(0);
-    }
-
-    Tensor<ADataType> a_ms_ks(a_ms_ks_lengths, a_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a_ms_ks: " << a_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a_ms_ks.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a_device_buf(sizeof(ADataType) * a_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a_device_buf.ToDevice(a_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{scale};
-
-    // device operation
-    auto op       = DeviceOpInstance{};
-    auto invoker  = op.MakeInvoker();
-    auto argument = op.MakeArgument(a_device_buf.GetDeviceBuffer(),
-                                    b_device_buf.GetDeviceBuffer(),
-                                    std::array<const void*, 0>{},
-                                    e_device_buf.GetDeviceBuffer(),
-                                    a_ms_ks_lengths,
-                                    a_ms_ks_strides,
-                                    b_ns_ks_lengths,
-                                    b_ns_ks_strides,
-                                    std::array<std::vector<ck::index_t>, 0>{},
-                                    std::array<std::vector<ck::index_t>, 0>{},
-                                    e_ms_ns_lengths,
-                                    e_ms_ns_strides,
-                                    a_element_op,
-                                    b_element_op,
-                                    cde_element_op);
-
-    if(!op.IsSupportedArgument(argument))
-    {
-        std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
-
-        return 0;
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    ck::index_t M =
-        ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-    ck::index_t N = ck::accumulate_n<ck::index_t>(
-        e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-    ck::index_t K = ck::accumulate_n<ck::index_t>(
-        a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-    std::size_t flop = std::size_t(2) * M * N * K;
-    std::size_t num_btype =
-        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
-              << op.GetTypeString() << std::endl;
-
-    e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      ADataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      ComputeDataType,
-                                                                      AElementOp,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, a_element_op, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
diff --git a/example/60_gemm_multi_ABD/CMakeLists.txt b/example/60_gemm_multi_ABD/CMakeLists.txt
deleted file mode 100644
index 0c6ce7ae87..0000000000
--- a/example/60_gemm_multi_ABD/CMakeLists.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
-set(target 0)
-foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-     add_example_executable(example_gemm_multi_ABD_xdl_fp16 gemm_multi_ABD_xdl_fp16.cpp)
-   set(target 1)
- endif()
-endforeach()
-endif()
diff --git a/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
deleted file mode 100644
index 90558364a4..0000000000
--- a/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
+++ /dev/null
@@ -1,362 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/utility/check_err.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using ADataType        = F16;
-using BDataType        = F16;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DDataType        = F16;
-using EDataType        = F16;
-
-using ALayout = Row;
-using BLayout = Col;
-using DLayout = Row;
-using ELayout = Row;
-
-struct AddScale
-{
-    static constexpr auto I0 = ck::Number<0>{};
-    static constexpr auto I1 = ck::Number<1>{};
-    static constexpr auto I2 = ck::Number<2>{};
-    static constexpr auto I3 = ck::Number<3>{};
-
-    __host__ __device__ constexpr void
-    operator()(ck::half4_t& a, const ck::half4_t& a0, const ck::half4_t& a1) const
-    {
-        const auto a0_v_t = ck::vector_type<ck::half_t, 4>{a0};
-        const auto a1_v_t = ck::vector_type<ck::half_t, 4>{a1};
-
-        auto r_v_t = ck::vector_type<ck::half_t, 4>{};
-
-        r_v_t.AsType<ck::half_t>()(I0) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I0] + a1_v_t.AsType<ck::half_t>()[I0]);
-        r_v_t.AsType<ck::half_t>()(I1) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I1] + a1_v_t.AsType<ck::half_t>()[I1]);
-        r_v_t.AsType<ck::half_t>()(I2) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I2] + a1_v_t.AsType<ck::half_t>()[I2]);
-        r_v_t.AsType<ck::half_t>()(I3) =
-            scale * (a0_v_t.AsType<ck::half_t>()[I3] + a1_v_t.AsType<ck::half_t>()[I3]);
-
-        a = r_v_t.AsType<ck::half4_t>()[I0];
-    }
-
-    __host__ __device__ constexpr void
-    operator()(ck::half_t& a, const ck::half_t& a0, const ck::half_t& a1) const
-    {
-        a = scale * (a0 + a1);
-    }
-
-    // this attribute will force copy_function applying element_wise with vector_type
-    static constexpr ck::index_t vec_len = 4;
-
-    float scale = 1.0;
-};
-
-struct AlphaBetaAdd
-{
-    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
-
-    template <typename E, typename C, typename D>
-    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
-
-    template <>
-    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
-        ck::half_t& e, const float& c, const ck::half_t& d) const
-    {
-        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
-    };
-
-    float alpha_;
-    float beta_;
-};
-
-using AElementOp   = AddScale;
-using BElementOp   = PassThrough;
-using CDEElementOp = AlphaBetaAdd;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleABD_Xdl_CShuffle<
-    ck::Tuple<ALayout, ALayout>,
-    ck::Tuple<BLayout>,
-    ck::Tuple<DLayout>,
-    ELayout,
-    ck::Tuple<ADataType, ADataType>,
-    ck::Tuple<BDataType>,
-    AccDataType,
-    CShuffleDataType,
-    ck::Tuple<DDataType>,
-    EDataType,
-    AElementOp,
-    BElementOp,
-    CDEElementOp,
-    GemmSpec,
-    1,
-    256,
-    256,
-    128,
-    32,
-    8,
-    8,
-    32,
-    32,
-    4,
-    2,
-    S<4, 64, 1>,
-    S<1, 0, 2>,
-    S<1, 0, 2>,
-    2,
-    8,
-    8,
-    1,
-    S<4, 64, 1>,
-    S<1, 0, 2>,
-    S<1, 0, 2>,
-    2,
-    8,
-    8,
-    1,
-    1,
-    1,
-    S<1, 32, 1, 8>,
-    8>;
-
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    // GEMM shape
-    ck::index_t M = 3840;
-    ck::index_t N = 4096;
-    ck::index_t K = 4096;
-
-    ck::index_t StrideA = 4096;
-    ck::index_t StrideB = 4096;
-    ck::index_t StrideD = 4096;
-    ck::index_t StrideE = 4096;
-
-    float alpha = 1.0f;
-    float beta  = 1.0f;
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else if(argc == 6)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        alpha = std::stof(argv[4]);
-        beta  = std::stof(argv[5]);
-    }
-    else if(argc == 13)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-
-        M = std::stoi(argv[4]);
-        N = std::stoi(argv[5]);
-        K = std::stoi(argv[6]);
-
-        StrideA = std::stoi(argv[7]);
-        StrideB = std::stoi(argv[8]);
-        StrideD = std::stoi(argv[9]);
-        StrideE = std::stoi(argv[10]);
-
-        alpha = std::stof(argv[11]);
-        beta  = std::stof(argv[12]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
-               "beta\n");
-        exit(0);
-    }
-
-    auto f_host_tensor_descriptor =
-        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
-            using namespace ck::literals;
-
-            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
-            {
-                return HostTensorDescriptor({row, col}, {stride, 1_uz});
-            }
-            else
-            {
-                return HostTensorDescriptor({row, col}, {1_uz, stride});
-            }
-        };
-
-    Tensor<ADataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
-    Tensor<ADataType> a1_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
-    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
-    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
-    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
-    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
-
-    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
-    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
-    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
-    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
-    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a0_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
-        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
-        break;
-    default:
-        a0_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
-    }
-
-    DeviceMem a0_device_buf(sizeof(ADataType) * a0_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem a1_device_buf(sizeof(ADataType) * a1_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
-    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
-
-    a0_device_buf.ToDevice(a0_m_k.mData.data());
-    a1_device_buf.ToDevice(a1_m_k.mData.data());
-    b_device_buf.ToDevice(b_k_n.mData.data());
-    d_device_buf.ToDevice(d_m_n.mData.data());
-    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
-
-    auto a_element_op   = AElementOp{0.2};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{alpha, beta};
-
-    // do GEMM
-    auto device_op = DeviceOpInstance{};
-    auto invoker   = device_op.MakeInvoker();
-    auto argument =
-        device_op.MakeArgument(std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
-                                                          a1_device_buf.GetDeviceBuffer()},
-                               std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
-                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
-                               e_device_buf.GetDeviceBuffer(),
-                               M,
-                               N,
-                               K,
-                               std::array<ck::index_t, 2>{StrideA, StrideA},
-                               std::array<ck::index_t, 1>{StrideB},
-                               std::array<ck::index_t, 1>{StrideD},
-                               StrideE,
-                               a_element_op,
-                               b_element_op,
-                               cde_element_op);
-
-    if(!device_op.IsSupportedArgument(argument))
-    {
-        throw std::runtime_error(
-            "wrong! device_gemm with the specified compilation parameters does "
-            "not support this GEMM problem");
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    std::size_t flop = std::size_t(2) * M * N * K;
-    std::size_t num_btype =
-        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
-
-    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-    float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
-              << std::endl;
-
-    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
-
-    if(do_verification)
-    {
-        Tensor<CShuffleDataType> c_m_n({M, N});
-
-        Tensor<ADataType> a_m_k({M, K});
-
-        for(int m = 0; m < M; ++m)
-        {
-            for(int k = 0; k < K; ++k)
-            {
-                a_element_op(a_m_k(m, k), a0_m_k(m, k), a1_m_k(m, k));
-            }
-        }
-
-        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
-                                                                                BDataType,
-                                                                                CShuffleDataType,
-                                                                                AccDataType,
-                                                                                PassThrough,
-                                                                                BElementOp,
-                                                                                PassThrough>;
-        auto ref_gemm               = ReferenceGemmInstance{};
-        auto ref_invoker            = ref_gemm.MakeInvoker();
-
-        auto ref_argument =
-            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, b_element_op, PassThrough{});
-
-        ref_invoker.Run(ref_argument);
-
-        for(int m = 0; m < M; ++m)
-        {
-            for(int n = 0; n < N; ++n)
-            {
-                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
-            }
-        }
-
-        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
-
-        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
diff --git a/example/61_contraction_multi_ABD/CMakeLists.txt b/example/61_contraction_multi_ABD/CMakeLists.txt
deleted file mode 100644
index 0420fdb9c3..0000000000
--- a/example/61_contraction_multi_ABD/CMakeLists.txt
+++ /dev/null
@@ -1,10 +0,0 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
-set(target 0)
-foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-     add_example_executable(example_contraction_multi_ABD_xdl_fp16 contraction_multi_ABD_xdl_fp16.cpp)
-   set(target 1)
- endif()
-endforeach()
-endif()
diff --git a/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
deleted file mode 100644
index 5a274d485e..0000000000
--- a/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
+++ /dev/null
@@ -1,329 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-#include <numeric>
-#include <initializer_list>
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-
-#include "ck/library/utility/device_memory.hpp"
-#include "ck/library/utility/host_tensor.hpp"
-#include "ck/library/utility/host_tensor_generator.hpp"
-#include "ck/library/utility/literals.hpp"
-#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
-#include "ck/library/utility/check_err.hpp"
-#include "ck/library/utility/numeric.hpp"
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-using F16 = ck::half_t;
-using F32 = float;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-
-using A0DataType       = F16;
-using A1DataType       = F32;
-using BDataType        = F16;
-using AccDataType      = F32;
-using CShuffleDataType = F32;
-using DDataType        = F16;
-using EDataType        = F16;
-
-static constexpr ck::index_t NumDimM = 2;
-static constexpr ck::index_t NumDimN = 2;
-static constexpr ck::index_t NumDimK = 2;
-
-struct AlphaBetaAdd
-{
-    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
-
-    template <typename E, typename C, typename D>
-    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
-
-    template <>
-    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
-        ck::half_t& e, const float& c, const ck::half_t& d) const
-    {
-        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
-    };
-
-    float alpha_;
-    float beta_;
-};
-
-struct Multiply
-{
-    __host__ __device__ constexpr void
-    operator()(ck::half_t& a, const ck::half_t& a0, const float& a1) const
-    {
-        a = ck::type_convert<ck::half_t>(ck::type_convert<float>(a0) * a1);
-    }
-};
-
-using AElementOp   = Multiply;
-using BElementOp   = PassThrough;
-using CDEElementOp = AlphaBetaAdd;
-
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-using DeviceOpInstance = ck::tensor_operation::device::DeviceContractionMultipleABD_Xdl_CShuffle<
-    NumDimM,
-    NumDimN,
-    NumDimK,
-    ck::Tuple<A0DataType, A1DataType>,
-    ck::Tuple<BDataType>,
-    AccDataType,
-    CShuffleDataType,
-    ck::Tuple<DDataType>,
-    EDataType,
-    AElementOp,
-    BElementOp,
-    CDEElementOp,
-    GemmSpec,
-    1,
-    256,
-    256,
-    128,
-    32,
-    8,
-    8,
-    32,
-    32,
-    4,
-    2,
-    S<4, 64, 1>,
-    S<1, 0, 2>,
-    S<1, 0, 2>,
-    2,
-    8,
-    8,
-    1,
-    S<4, 64, 1>,
-    S<1, 0, 2>,
-    S<1, 0, 2>,
-    2,
-    8,
-    8,
-    1,
-    1,
-    1,
-    S<1, 32, 1, 8>,
-    8>;
-
-int main(int argc, char* argv[])
-{
-    bool do_verification = true;
-    int init_method      = 1;
-    bool time_kernel     = false;
-
-    float alpha = 1.0f;
-    float beta  = 1.0f;
-
-    // A0[M0, M1, K0, K1]
-    std::vector<ck::index_t> a0_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a0_ms_ks_strides{128 * 32 * 64, 32 * 64, 64, 1};
-    // A1[M1, K1] -> A1[M0, M1, K0, K1]
-    std::vector<ck::index_t> a1_ms_ks_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> a1_ms_ks_strides{0, 64, 0, 1};
-    // B[N0, N1, K0, K1]
-    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
-    std::vector<ck::index_t> b_ns_ks_strides{64 * 32 * 64, 32 * 64, 64, 1};
-    // D[M0, M1, N0, N1]
-    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> d_ms_ns_strides{128 * 32 * 64, 32 * 64, 64, 1};
-    // E[M0, M1, N0, N1]
-    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
-    std::vector<ck::index_t> e_ms_ns_strides{128 * 32 * 64, 32 * 64, 64, 1};
-
-    if(argc == 1)
-    {
-        // use default case
-    }
-    else if(argc == 4)
-    {
-        do_verification = std::stoi(argv[1]);
-        init_method     = std::stoi(argv[2]);
-        time_kernel     = std::stoi(argv[3]);
-    }
-    else
-    {
-        printf("arg1: verification (0=no, 1=yes)\n");
-        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
-        printf("arg3: time kernel (0=no, 1=yes)\n");
-        exit(0);
-    }
-
-    Tensor<A0DataType> a0_ms_ks(a0_ms_ks_lengths, a0_ms_ks_strides);
-    Tensor<A1DataType> a1_ms_ks(a1_ms_ks_lengths, a1_ms_ks_strides);
-    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
-    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-    std::cout << "a0_ms_ks: " << a0_ms_ks.mDesc << std::endl;
-    std::cout << "a1_ms_ks: " << a1_ms_ks.mDesc << std::endl;
-    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
-    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
-    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
-
-    switch(init_method)
-    {
-    case 0: break;
-    case 1:
-        a0_ms_ks.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-5, 5});
-        a1_ms_ks.GenerateTensorValue(GeneratorTensor_2<A1DataType>{-5, 5});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
-        break;
-    default:
-        a0_ms_ks.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
-        a1_ms_ks.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0.0, 1.0});
-        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
-        break;
-    }
-
-    DeviceMem a0_device_buf(sizeof(A0DataType) * a0_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem a1_device_buf(sizeof(A1DataType) * a1_ms_ks.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
-    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
-    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
-
-    a0_device_buf.ToDevice(a0_ms_ks.mData.data());
-    a1_device_buf.ToDevice(a1_ms_ks.mData.data());
-    b_device_buf.ToDevice(b_ns_ks.mData.data());
-    d_device_buf.ToDevice(d_ms_ns.mData.data());
-
-    // set zero
-    e_device_buf.SetZero();
-
-    auto a_element_op   = AElementOp{};
-    auto b_element_op   = BElementOp{};
-    auto cde_element_op = CDEElementOp{alpha, beta};
-
-    // do GEMM
-    auto device_op = DeviceOpInstance{};
-    auto invoker   = device_op.MakeInvoker();
-    auto argument  = device_op.MakeArgument(
-        std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
-                                   a1_device_buf.GetDeviceBuffer()},
-        std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
-        std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
-        e_device_buf.GetDeviceBuffer(),
-        std::array<std::vector<ck::index_t>, 2>{a0_ms_ks_lengths, a1_ms_ks_lengths},
-        std::array<std::vector<ck::index_t>, 2>{a0_ms_ks_strides, a1_ms_ks_strides},
-        std::array<std::vector<ck::index_t>, 1>{b_ns_ks_lengths},
-        std::array<std::vector<ck::index_t>, 1>{b_ns_ks_strides},
-        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
-        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
-        e_ms_ns_lengths,
-        e_ms_ns_strides,
-        a_element_op,
-        b_element_op,
-        cde_element_op);
-
-    if(!device_op.IsSupportedArgument(argument))
-    {
-        throw std::runtime_error(
-            "wrong! device_contraction with the specified compilation parameters does "
-            "not support this problem");
-    }
-
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
-
-    if(time_kernel)
-    {
-        ck::index_t M =
-            ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
-
-        ck::index_t N = ck::accumulate_n<ck::index_t>(
-            e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
-
-        ck::index_t K = ck::accumulate_n<ck::index_t>(
-            a0_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
-
-        std::size_t flop = std::size_t(2) * M * N * K;
-        std::size_t num_btype =
-            sizeof(A0DataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
-
-        float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
-
-        float gb_per_sec = num_btype / 1.E6 / ave_time;
-
-        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                  << " GB/s" << std::endl;
-    }
-
-    if(do_verification)
-    {
-
-        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
-
-        Tensor<A0DataType> a_ms_ks(a0_ms_ks_lengths, a0_ms_ks_strides);
-
-        for(size_t m0 = 0; m0 < a_ms_ks.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < a_ms_ks.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t k0 = 0; k0 < a_ms_ks.mDesc.GetLengths()[2]; ++k0)
-                {
-                    for(size_t k1 = 0; k1 < a_ms_ks.mDesc.GetLengths()[3]; ++k1)
-                    {
-                        a_element_op(a_ms_ks(m0, m1, k0, k1),
-                                     a0_ms_ks(m0, m1, k0, k1),
-                                     a1_ms_ks(m0, m1, k0, k1));
-                    }
-                }
-            }
-        }
-
-        using ReferenceOpInstance =
-            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
-                                                                      NumDimN,
-                                                                      NumDimK,
-                                                                      A0DataType,
-                                                                      BDataType,
-                                                                      CShuffleDataType,
-                                                                      AccDataType,
-                                                                      A0DataType,
-                                                                      PassThrough,
-                                                                      BElementOp>;
-
-        auto ref_op      = ReferenceOpInstance{};
-        auto ref_invoker = ref_op.MakeInvoker();
-
-        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
-        auto ref_argument =
-            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, PassThrough{}, b_element_op);
-
-        ref_invoker.Run(ref_argument);
-
-        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
-        {
-            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
-            {
-                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
-                {
-                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
-                    {
-                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
-                                       c_ms_ns_host_result(m0, m1, n0, n1),
-                                       d_ms_ns(m0, m1, n0, n1));
-                    }
-                }
-            }
-        }
-
-        e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
-
-        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
-    }
-
-    return 0;
-}
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
deleted file mode 100644
index 3116d40474..0000000000
--- a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
+++ /dev/null
@@ -1,61 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <array>
-
-#include "ck/tensor_operation/gpu/device/device_base.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-// GEMM:
-//   input : A0[M0, M1, ... K0, K1, ...], ...
-//   input : B0[N0, N1, ... K0, K1, ...], ...
-//   input : D0[M0, M1, ... N0, N1, ...], D1[M0, M1, ... N0, N1, ...], ...
-//   output : E[M0, M1, ... N0, N1, ...]
-//   C = a_op(A) * b_op(B)
-//   E = cde_op(C, D0, D1, ...)
-// Assume:
-//   D0, D1, ... and E have the same layout
-template <index_t NumDimM,
-          index_t NumDimN,
-          index_t NumDimK,
-          typename AsDataType,
-          typename BsDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation>
-struct DeviceContractionMultipleABD : public BaseOperator
-{
-    static constexpr index_t NumATensor = AsDataType::Size();
-    static constexpr index_t NumBTensor = BsDataType::Size();
-    static constexpr index_t NumDTensor = DsDataType::Size();
-
-    virtual std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
-                        std::array<const void*, NumBTensor> p_bs,
-                        std::array<const void*, NumDTensor> p_ds,
-                        void* p_e,
-                        const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
-                        const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
-                        const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
-                        const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
-                        const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
-                        const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
-                        const std::vector<index_t>& e_ms_ns_length,
-                        const std::vector<index_t>& e_ms_ns_stride,
-                        AElementwiseOperation a_element_op,
-                        BElementwiseOperation b_element_op,
-                        CDEElementwiseOperation cde_element_op) = 0;
-
-    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
-};
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
index dffb51d0ba..118ade8978 100644
--- a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp
@@ -33,8 +33,7 @@ template <index_t NumDimM,
           typename EDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          typename ComputeDataType = ADataType>
+          typename CDEElementwiseOperation>
 struct DeviceContractionMultipleD : public BaseOperator
 {
     static constexpr index_t NumDTensor = DsDataType::Size();
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
deleted file mode 100644
index a4c6767434..0000000000
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
+++ /dev/null
@@ -1,846 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#pragma once
-
-#include <iostream>
-#include <sstream>
-
-#include "ck/utility/common_header.hpp"
-#include "ck/tensor_description/tensor_descriptor.hpp"
-#include "ck/tensor_description/tensor_descriptor_helper.hpp"
-#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
-#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp"
-#include "ck/host_utility/device_prop.hpp"
-#include "ck/host_utility/kernel_launch.hpp"
-
-namespace ck {
-
-template <typename GridwiseGemm,
-          typename AsPointer,
-          typename BsPointer,
-          typename DsPointer,
-          typename EDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          typename AsGridDesc_AK0_M_AK1,
-          typename BsGridDesc_BK0_N_BK1,
-          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
-          typename Block2ETileMap,
-          bool HasMainKBlockLoop>
-__global__ void
-#if CK_USE_LAUNCH_BOUNDS
-    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
-#endif
-        kernel_contraction_multiple_abd_xdl_cshuffle(
-            AsPointer p_as_grid,
-            BsPointer p_bs_grid,
-            DsPointer p_ds_grid,
-            EDataType* __restrict__ p_e_grid,
-            const AElementwiseOperation a_element_op,
-            const BElementwiseOperation b_element_op,
-            const CDEElementwiseOperation cde_element_op,
-            const AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1,
-            const BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1,
-            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                ds_grid_desc_mblock_mperblock_nblock_nperblock,
-            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
-                e_grid_desc_mblock_mperblock_nblock_nperblock,
-            const Block2ETileMap block_2_etile_map)
-{
-#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
-    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
-    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
-
-    GridwiseGemm::template Run<HasMainKBlockLoop>(p_as_grid,
-                                                  p_bs_grid,
-                                                  p_ds_grid,
-                                                  p_e_grid,
-                                                  p_shared,
-                                                  a_element_op,
-                                                  b_element_op,
-                                                  cde_element_op,
-                                                  as_grid_desc_ak0_m_ak1,
-                                                  bs_grid_desc_bk0_n_bk1,
-                                                  ds_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  e_grid_desc_mblock_mperblock_nblock_nperblock,
-                                                  block_2_etile_map);
-#else
-    ignore = p_as_grid;
-    ignore = p_bs_grid;
-    ignore = p_ds_grid;
-    ignore = p_e_grid;
-    ignore = a_element_op;
-    ignore = b_element_op;
-    ignore = cde_element_op;
-    ignore = as_grid_desc_ak0_m_ak1;
-    ignore = bs_grid_desc_bk0_n_bk1;
-    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
-    ignore = block_2_etile_map;
-#endif
-}
-
-} // namespace ck
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-
-// GEMM:
-//   input : A[M, K]
-//   input : B[N, K]
-//   input : D0[M, N], D1[M, N], ...
-//   output : E[M, N]
-//   C = a_op(A) * b_op(B)
-//   E = cde_op(C, D0, D1, ...)
-// Assume:
-//   D0, D1, ... and E have the same layout
-template <index_t NumDimM,
-          index_t NumDimN,
-          index_t NumDimK,
-          typename AsDataType,
-          typename BsDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename AElementwiseOperation,
-          typename BElementwiseOperation,
-          typename CDEElementwiseOperation,
-          GemmSpecialization GemmSpec,
-          index_t NumGemmKPrefetchStage,
-          index_t BlockSize,
-          index_t MPerBlock,
-          index_t NPerBlock,
-          index_t KPerBlock,
-          index_t AK1,
-          index_t BK1,
-          index_t MPerXDL,
-          index_t NPerXDL,
-          index_t MXdlPerWave,
-          index_t NXdlPerWave,
-          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
-          typename ABlockTransferThreadClusterArrangeOrder,
-          typename ABlockTransferSrcAccessOrder,
-          index_t ABlockTransferSrcVectorDim,
-          index_t ABlockTransferSrcScalarPerVector,
-          index_t ABlockTransferDstScalarPerVector_AK1,
-          index_t ABlockLdsExtraM,
-          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
-          typename BBlockTransferThreadClusterArrangeOrder,
-          typename BBlockTransferSrcAccessOrder,
-          index_t BBlockTransferSrcVectorDim,
-          index_t BBlockTransferSrcScalarPerVector,
-          index_t BBlockTransferDstScalarPerVector_BK1,
-          index_t BBlockLdsExtraN,
-          index_t CShuffleMXdlPerWavePerShuffle,
-          index_t CShuffleNXdlPerWavePerShuffle,
-          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-          index_t CDEBlockTransferScalarPerVector_NPerBlock,
-          LoopScheduler LoopSched     = make_default_loop_scheduler(),
-          PipelineVersion PipelineVer = PipelineVersion::v1>
-struct DeviceContractionMultipleABD_Xdl_CShuffle
-    : public DeviceContractionMultipleABD<NumDimM,
-                                          NumDimN,
-                                          NumDimK,
-                                          AsDataType,
-                                          BsDataType,
-                                          DsDataType,
-                                          EDataType,
-                                          AElementwiseOperation,
-                                          BElementwiseOperation,
-                                          CDEElementwiseOperation>
-{
-    using DeviceOp = DeviceContractionMultipleABD_Xdl_CShuffle;
-
-    static constexpr index_t NumATensor = AsDataType::Size();
-    static constexpr index_t NumBTensor = BsDataType::Size();
-    static constexpr index_t NumDTensor = DsDataType::Size();
-
-    static constexpr auto I0 = Number<0>{};
-    static constexpr auto I1 = Number<1>{};
-    static constexpr auto I2 = Number<2>{};
-    static constexpr auto I3 = Number<3>{};
-
-    using ComputeDataType = EDataType;
-
-    // GridwiseGemm
-    using GridwiseGemm = GridwiseGemmMultipleABD_xdl_cshuffle<
-        AsDataType,
-        BsDataType,
-        ComputeDataType,
-        AccDataType,
-        CShuffleDataType,
-        DsDataType,
-        EDataType,
-        AElementwiseOperation,
-        BElementwiseOperation,
-        CDEElementwiseOperation,
-        InMemoryDataOperationEnum::Set,
-        NumGemmKPrefetchStage,
-        BlockSize,
-        MPerBlock,
-        NPerBlock,
-        KPerBlock,
-        AK1,
-        BK1,
-        MPerXDL,
-        NPerXDL,
-        MXdlPerWave,
-        NXdlPerWave,
-        ABlockTransferThreadClusterLengths_AK0_M_AK1,
-        ABlockTransferThreadClusterArrangeOrder,
-        ABlockTransferSrcAccessOrder,
-        ABlockTransferSrcVectorDim,
-        ABlockTransferSrcScalarPerVector,
-        ABlockTransferDstScalarPerVector_AK1,
-        false,
-        ABlockLdsExtraM,
-        BBlockTransferThreadClusterLengths_BK0_N_BK1,
-        BBlockTransferThreadClusterArrangeOrder,
-        BBlockTransferSrcAccessOrder,
-        BBlockTransferSrcVectorDim,
-        BBlockTransferSrcScalarPerVector,
-        BBlockTransferDstScalarPerVector_BK1,
-        false,
-        BBlockLdsExtraN,
-        CShuffleMXdlPerWavePerShuffle,
-        CShuffleNXdlPerWavePerShuffle,
-        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
-        CDEBlockTransferScalarPerVector_NPerBlock,
-        LoopSched,
-        PipelineVer>;
-
-    static constexpr auto matrix_padder =
-        ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
-            MPerBlock, NPerBlock, KPerBlock};
-
-    static auto MakeAGridDescriptor_M_K(const std::vector<index_t>& a_ms_ks_lengths_,
-                                        const std::vector<index_t>& a_ms_ks_strides_)
-    {
-        assert(a_ms_ks_lengths_.size() == NumDimM + NumDimK &&
-               a_ms_ks_strides_.size() == NumDimM + NumDimK);
-
-        const auto to_tuple = [&](auto& vec, auto num) {
-            return generate_tuple([&](auto i) { return vec[i]; }, num);
-        };
-
-        const auto a_ms_ks_lengths = to_tuple(a_ms_ks_lengths_, Number<NumDimM + NumDimK>{});
-        const auto a_ms_ks_strides = to_tuple(a_ms_ks_strides_, Number<NumDimM + NumDimK>{});
-
-        // dimension Ids for M0, M1, ...
-        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
-
-        // dimension Ids for K0, K1, ...
-        constexpr auto kDimIds =
-            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimK, 1>::type{};
-
-        // lengths for M0, M1, ...
-        const auto mLengths = get_container_subset(a_ms_ks_lengths, mDimIds);
-
-        // lengths for K0, K1, ...
-        const auto kLengths = get_container_subset(a_ms_ks_lengths, kDimIds);
-
-        // naive tensor A[M0, M1, M2, ..., K0, K1, K2...]
-        const auto a_grid_desc_ms_ks =
-            make_naive_tensor_descriptor(a_ms_ks_lengths, a_ms_ks_strides);
-
-        // transformed tensor A[MRaw = M0 * M1 * M2 * ... , KRaw = K0 * K1 * K2 * ...]
-        const auto a_grid_desc_mraw_kraw = transform_tensor_descriptor(
-            a_grid_desc_ms_ks,
-            make_tuple(make_merge_transform(mLengths), make_merge_transform(kLengths)),
-            make_tuple(mDimIds, kDimIds),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
-    }
-
-    __host__ __device__ static auto
-    MakeAsGridDescriptor_M_K(const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_lengths,
-                             const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_strides)
-    {
-        return generate_tuple(
-            [&](auto i) {
-                return MakeAGridDescriptor_M_K(as_ms_ks_lengths[i], as_ms_ks_strides[i]);
-            },
-            Number<NumATensor>{});
-    }
-
-    // Assume: B[N0, N1, N2, ..., K0, K1, K2, ...]
-    static auto MakeBGridDescriptor_N_K(const std::vector<index_t>& b_ns_ks_lengths_,
-                                        const std::vector<index_t>& b_ns_ks_strides_)
-    {
-        assert(b_ns_ks_lengths_.size() == NumDimN + NumDimK &&
-               b_ns_ks_strides_.size() == NumDimN + NumDimK);
-
-        const auto to_tuple = [&](auto& vec, auto num) {
-            return generate_tuple([&](auto i) { return vec[i]; }, num);
-        };
-
-        const auto b_ns_ks_lengths = to_tuple(b_ns_ks_lengths_, Number<NumDimN + NumDimK>{});
-        const auto b_ns_ks_strides = to_tuple(b_ns_ks_strides_, Number<NumDimN + NumDimK>{});
-
-        // dimension Ids for N0, N1, ...
-        constexpr auto nDimIds = typename arithmetic_sequence_gen<0, NumDimN, 1>::type{};
-
-        // dimension Ids for K0, K1, ...
-        constexpr auto kDimIds =
-            typename arithmetic_sequence_gen<NumDimN, NumDimN + NumDimK, 1>::type{};
-
-        // lengths for K0, K1, ...
-        const auto kLengths = get_container_subset(b_ns_ks_lengths, kDimIds);
-
-        // lengths for N0, N1, ...
-        const auto nLengths = get_container_subset(b_ns_ks_lengths, nDimIds);
-
-        // naive tensor B[N0, N1, N2, ..., K0, K1, K2, ...]
-        const auto b_grid_desc_ns_ks =
-            make_naive_tensor_descriptor(b_ns_ks_lengths, b_ns_ks_strides);
-
-        // transformed tensor B[NRaw = N0 * N1 * N2 * ..., KRaw = K0 * K1 * K2 * ...]
-        const auto b_grid_desc_nraw_kraw = transform_tensor_descriptor(
-            b_grid_desc_ns_ks,
-            make_tuple(make_merge_transform(nLengths), make_merge_transform(kLengths)),
-            make_tuple(nDimIds, kDimIds),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
-    }
-
-    __host__ __device__ static auto
-    MakeBsGridDescriptor_N_K(const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_lengths,
-                             const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_strides)
-    {
-        return generate_tuple(
-            [&](auto i) {
-                return MakeBGridDescriptor_N_K(bs_ns_ks_lengths[i], bs_ns_ks_strides[i]);
-            },
-            Number<NumBTensor>{});
-    }
-
-    // assume E[M0, M1, M2, ..., N0, N1, N2...]
-    static auto MakeEGridDescriptor_M_N(const std::vector<index_t>& e_ms_ns_lengths_,
-                                        const std::vector<index_t>& e_ms_ns_strides_)
-    {
-        assert(e_ms_ns_lengths_.size() == NumDimM + NumDimN &&
-               e_ms_ns_strides_.size() == NumDimM + NumDimN);
-
-        const auto to_tuple = [&](auto& vec, auto num) {
-            return generate_tuple([&](auto i) { return vec[i]; }, num);
-        };
-
-        const auto e_ms_ns_lengths = to_tuple(e_ms_ns_lengths_, Number<NumDimM + NumDimN>{});
-        const auto e_ms_ns_strides = to_tuple(e_ms_ns_strides_, Number<NumDimM + NumDimN>{});
-
-        // dimension Ids for M0, M1, ...
-        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
-
-        // dimension Ids for N0, N1, ...
-        constexpr auto nDimIds =
-            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimN, 1>::type{};
-
-        // lengths for M0, M1, ...
-        const auto mLengths = get_container_subset(e_ms_ns_lengths, mDimIds);
-
-        // lengths for K0, K1, ...
-        const auto nLengths = get_container_subset(e_ms_ns_lengths, nDimIds);
-
-        // naive tensor E[M0, M1, M2, ..., N0, N1, N2...]
-        const auto e_grid_desc_ms_ns =
-            make_naive_tensor_descriptor(e_ms_ns_lengths, e_ms_ns_strides);
-
-        // transformed tensor E[MRaw = M0 * M1 * M2 * ... , NRaw = N0 * N1 * N2 * ...]
-        const auto e_grid_desc_mraw_nraw = transform_tensor_descriptor(
-            e_grid_desc_ms_ns,
-            make_tuple(make_merge_transform(mLengths), make_merge_transform(nLengths)),
-            make_tuple(mDimIds, nDimIds),
-            make_tuple(Sequence<0>{}, Sequence<1>{}));
-
-        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
-    }
-
-    static auto
-    MakeDsGridDescriptor_M_N(const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_lengths,
-                             const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_strides)
-    {
-        return generate_tuple(
-            [&](auto i) {
-                return MakeEGridDescriptor_M_N(ds_ms_ns_lengths[i], ds_ms_ns_strides[i]);
-            },
-            Number<NumDTensor>{});
-    }
-
-    // desc for problem definition
-    using AsGridDesc_M_K = remove_cvref_t<decltype(MakeAsGridDescriptor_M_K({}, {}))>;
-    using BsGridDesc_N_K = remove_cvref_t<decltype(MakeBsGridDescriptor_N_K({}, {}))>;
-    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
-    using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N({}, {}))>;
-
-    // desc for blockwise copy
-    using AsGridDesc_AK0_M_AK1 =
-        remove_cvref_t<decltype(GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(AsGridDesc_M_K{}))>;
-    using BsGridDesc_BK0_N_BK1 =
-        remove_cvref_t<decltype(GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(BsGridDesc_N_K{}))>;
-    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
-        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-            DsGridDesc_M_N{}))>;
-    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
-        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-            EGridDesc_M_N{}))>;
-
-    // block-to-e-tile map
-    using Block2ETileMap =
-        remove_cvref_t<decltype(GridwiseGemm::MakeBlock2ETileMap(EGridDesc_M_N{}))>;
-
-    // Argument
-    struct Argument : public BaseArgument
-    {
-        Argument(std::array<const void*, NumATensor> p_as_grid,
-                 std::array<const void*, NumBTensor> p_bs_grid,
-                 std::array<const void*, NumDTensor> p_ds_grid,
-                 void* p_e_grid,
-                 const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
-                 const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
-                 const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
-                 const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
-                 const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
-                 const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
-                 const std::vector<index_t>& e_ms_ns_length,
-                 const std::vector<index_t>& e_ms_ns_stride,
-                 AElementwiseOperation a_element_op,
-                 BElementwiseOperation b_element_op,
-                 CDEElementwiseOperation cde_element_op)
-            : p_as_grid_{},
-              p_bs_grid_{},
-              p_ds_grid_{},
-              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
-              as_grid_desc_m_k_{},
-              bs_grid_desc_n_k_{},
-              ds_grid_desc_m_n_{},
-              e_grid_desc_m_n_{MakeEGridDescriptor_M_N(e_ms_ns_length, e_ms_ns_stride)},
-              as_grid_desc_ak0_m_ak1_{},
-              bs_grid_desc_bk0_n_bk1_{},
-              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              e_grid_desc_mblock_mperblock_nblock_nperblock_{},
-              block_2_etile_map_{GridwiseGemm::MakeBlock2ETileMap(e_grid_desc_m_n_)},
-              a_element_op_{a_element_op},
-              b_element_op_{b_element_op},
-              cde_element_op_{cde_element_op}
-        {
-            // populate pointer, desc for As
-            static_for<0, NumATensor, 1>{}([&](auto i) {
-                // using ALayout   = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
-                using ADataType = remove_cvref_t<tuple_element_t<i.value, AsDataType>>;
-
-                // A pointer
-                p_as_grid_(i) = static_cast<const ADataType*>(p_as_grid[i]);
-
-                // A desc
-                as_grid_desc_m_k_(i) =
-                    MakeAGridDescriptor_M_K(a_ms_ks_lengths[i], a_ms_ks_strides[i]);
-            });
-
-            // populate pointer, desc for Bs
-            static_for<0, NumBTensor, 1>{}([&](auto i) {
-                // using BLayout   = remove_cvref_t<tuple_element_t<i.value, BsLayout>>;
-                using BDataType = remove_cvref_t<tuple_element_t<i.value, BsDataType>>;
-
-                // B pointer
-                p_bs_grid_(i) = static_cast<const BDataType*>(p_bs_grid[i]);
-
-                // B desc
-                bs_grid_desc_n_k_(i) =
-                    MakeBGridDescriptor_N_K(b_ns_ks_lengths[i], b_ns_ks_strides[i]);
-            });
-
-            // populate pointer, desc for Ds
-            static_for<0, NumDTensor, 1>{}([&](auto i) {
-                // using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
-                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
-
-                // D pointer
-                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
-
-                // D desc
-                ds_grid_desc_m_n_(i) =
-                    MakeEGridDescriptor_M_N(d_ms_ns_lengths[i], d_ms_ns_strides[i]);
-            });
-
-            // populate desc for Ds/E
-            if(GridwiseGemm::CheckValidity(as_grid_desc_m_k_,
-                                           bs_grid_desc_n_k_,
-                                           ds_grid_desc_m_n_,
-                                           e_grid_desc_m_n_,
-                                           block_2_etile_map_))
-            {
-                as_grid_desc_ak0_m_ak1_ =
-                    GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_);
-
-                bs_grid_desc_bk0_n_bk1_ =
-                    GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_);
-
-                ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        ds_grid_desc_m_n_);
-
-                e_grid_desc_mblock_mperblock_nblock_nperblock_ =
-                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
-                        e_grid_desc_m_n_);
-            }
-
-            // for sanity check of vector memory access
-            for(index_t i = 0; i < NumATensor; ++i)
-            {
-                a_mz_stride_[i] = a_ms_ks_strides[i][NumDimM - 1];
-                a_kz_stride_[i] = a_ms_ks_strides[i][NumDimM + NumDimK - 1];
-            }
-
-            for(index_t i = 0; i < NumBTensor; ++i)
-            {
-                b_nz_stride_[i] = b_ns_ks_strides[i][NumDimN - 1];
-                b_kz_stride_[i] = b_ns_ks_strides[i][NumDimN + NumDimK - 1];
-            }
-
-            for(index_t i = 0; i < NumDTensor; ++i)
-            {
-                ds_nz_stride_[i] = d_ms_ns_strides[i][NumDimM + NumDimN - 1];
-            }
-
-            e_nz_stride_ = e_ms_ns_stride[NumDimM + NumDimN - 1];
-        }
-
-        // pointers
-        typename GridwiseGemm::AsGridPointer p_as_grid_;
-        typename GridwiseGemm::BsGridPointer p_bs_grid_;
-        typename GridwiseGemm::DsGridPointer p_ds_grid_;
-        EDataType* p_e_grid_;
-
-        // tensor descriptors for problem definiton
-        AsGridDesc_M_K as_grid_desc_m_k_;
-        BsGridDesc_N_K bs_grid_desc_n_k_;
-        DsGridDesc_M_N ds_grid_desc_m_n_;
-        EGridDesc_M_N e_grid_desc_m_n_;
-
-        // tensor descriptors for block/thread-wise copy
-        AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1_;
-        BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1_;
-        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
-            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
-        EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock e_grid_desc_mblock_mperblock_nblock_nperblock_;
-
-        // block-to-e-tile map
-        Block2ETileMap block_2_etile_map_;
-
-        // element-wise op
-        AElementwiseOperation a_element_op_;
-        BElementwiseOperation b_element_op_;
-        CDEElementwiseOperation cde_element_op_;
-
-        // Strides for the last M/N/K dimensions of A/B/Ds/E
-        //   for sanity check of vector load/store
-        std::array<index_t, NumATensor> a_mz_stride_;
-        std::array<index_t, NumATensor> a_kz_stride_;
-
-        std::array<index_t, NumBTensor> b_nz_stride_;
-        std::array<index_t, NumBTensor> b_kz_stride_;
-
-        std::array<index_t, NumDTensor> ds_nz_stride_;
-        index_t e_nz_stride_;
-    };
-
-    // Invoker
-    struct Invoker : public BaseInvoker
-    {
-        using Argument = DeviceOp::Argument;
-
-        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
-        {
-            if(!GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
-                                            arg.bs_grid_desc_n_k_,
-                                            arg.ds_grid_desc_m_n_,
-                                            arg.e_grid_desc_m_n_,
-                                            arg.block_2_etile_map_))
-            {
-                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
-            }
-
-            const index_t grid_size =
-                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
-
-            auto launch_kernel = [&](auto has_main_k_block_loop) {
-                constexpr bool has_main_loop = has_main_k_block_loop.value;
-
-                const auto kernel = kernel_contraction_multiple_abd_xdl_cshuffle<
-                    GridwiseGemm,
-                    typename GridwiseGemm::AsGridPointer,
-                    typename GridwiseGemm::BsGridPointer,
-                    typename GridwiseGemm::DsGridPointer,
-                    EDataType,
-                    AElementwiseOperation,
-                    BElementwiseOperation,
-                    CDEElementwiseOperation,
-                    DeviceOp::AsGridDesc_AK0_M_AK1,
-                    DeviceOp::BsGridDesc_BK0_N_BK1,
-                    DeviceOp::DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
-                    DeviceOp::EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
-                    DeviceOp::Block2ETileMap,
-                    has_main_loop>;
-
-                return launch_and_time_kernel(stream_config,
-                                              kernel,
-                                              dim3(grid_size),
-                                              dim3(BlockSize),
-                                              0,
-                                              arg.p_as_grid_,
-                                              arg.p_bs_grid_,
-                                              arg.p_ds_grid_,
-                                              arg.p_e_grid_,
-                                              arg.a_element_op_,
-                                              arg.b_element_op_,
-                                              arg.cde_element_op_,
-                                              arg.as_grid_desc_ak0_m_ak1_,
-                                              arg.bs_grid_desc_bk0_n_bk1_,
-                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
-                                              arg.block_2_etile_map_);
-            };
-
-            const auto K = arg.as_grid_desc_m_k_[I0].GetLength(I1);
-
-            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
-            {
-                return launch_kernel(integral_constant<bool, true>{});
-            }
-            else
-            {
-                return launch_kernel(integral_constant<bool, false>{});
-            }
-        }
-
-        // polymorphic
-        float Run(const BaseArgument* p_arg,
-                  const StreamConfig& stream_config = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
-        }
-    };
-
-    static bool IsSupportedArgument(const Argument& arg)
-    {
-        if(!ck::is_xdl_supported())
-        {
-            return false;
-        }
-
-        // check vector load/store
-        {
-            bool all_valid = true;
-
-            static_for<0, NumATensor, 1>{}([&](auto i) {
-                // vector memory access of A: could be on M or AK1 dimension
-                if constexpr(ABlockTransferSrcVectorDim == 1)
-                {
-                    if(!(arg.a_mz_stride_[i] == 1 && arg.as_grid_desc_ak0_m_ak1_[i].GetLength(I1) %
-                                                             ABlockTransferSrcScalarPerVector ==
-                                                         0))
-                    {
-                        all_valid = false;
-                    }
-                }
-                else
-                {
-                    if(!(arg.a_kz_stride_[i] == 1 && arg.as_grid_desc_ak0_m_ak1_[i].GetLength(I2) %
-                                                             ABlockTransferSrcScalarPerVector ==
-                                                         0))
-                    {
-                        all_valid = false;
-                    }
-                }
-            });
-
-            // vector memory access of B: could be on N or BK1 dimension
-            static_for<0, NumBTensor, 1>{}([&](auto i) {
-                if constexpr(BBlockTransferSrcVectorDim == 1)
-                {
-                    if(!(arg.b_nz_stride_[i] == 1 && arg.bs_grid_desc_bk0_n_bk1_[i].GetLength(I1) %
-                                                             BBlockTransferSrcScalarPerVector ==
-                                                         0))
-                    {
-                        all_valid = false;
-                    }
-                }
-                else
-                {
-                    if(!(arg.b_kz_stride_[i] == 1 && arg.bs_grid_desc_bk0_n_bk1_[i].GetLength(I2) %
-                                                             BBlockTransferSrcScalarPerVector ==
-                                                         0))
-                    {
-                        all_valid = false;
-                    }
-                }
-            });
-
-            // check vector load of Ds
-            static_for<0, NumDTensor, 1>{}([&](auto i) {
-                if(!(arg.ds_nz_stride_[i] == 1 &&
-                     arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_[i].GetLength(I3) %
-                             CDEBlockTransferScalarPerVector_NPerBlock ==
-                         0))
-                {
-                    all_valid = false;
-                }
-            });
-
-            // vector memory access of E: always on NPerBlock dimension
-            if(!(arg.e_nz_stride_ == 1 &&
-                 arg.e_grid_desc_mblock_mperblock_nblock_nperblock_.GetLength(I3) %
-                         CDEBlockTransferScalarPerVector_NPerBlock ==
-                     0))
-            {
-                all_valid = false;
-            }
-
-            if(!all_valid)
-            {
-                return false;
-            }
-        }
-
-        return GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
-                                           arg.bs_grid_desc_n_k_,
-                                           arg.ds_grid_desc_m_n_,
-                                           arg.e_grid_desc_m_n_,
-                                           arg.block_2_etile_map_);
-    }
-
-    // polymorphic
-    bool IsSupportedArgument(const BaseArgument* p_arg) override
-    {
-        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
-    }
-
-    static auto MakeArgument(std::array<const void*, NumATensor> p_as,
-                             std::array<const void*, NumBTensor> p_bs,
-                             std::array<const void*, NumDTensor> p_ds,
-                             void* p_e,
-                             const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
-                             const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
-                             const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
-                             const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
-                             const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
-                             const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
-                             const std::vector<index_t>& e_ms_ns_length,
-                             const std::vector<index_t>& e_ms_ns_stride,
-                             AElementwiseOperation a_element_op,
-                             BElementwiseOperation b_element_op,
-                             CDEElementwiseOperation cde_element_op)
-    {
-        return Argument{p_as,
-                        p_bs,
-                        p_ds,
-                        p_e,
-                        a_ms_ks_lengths,
-                        a_ms_ks_strides,
-                        b_ns_ks_lengths,
-                        b_ns_ks_strides,
-                        d_ms_ns_lengths,
-                        d_ms_ns_strides,
-                        e_ms_ns_length,
-                        e_ms_ns_stride,
-                        a_element_op,
-                        b_element_op,
-                        cde_element_op};
-    }
-
-    static auto MakeInvoker() { return Invoker{}; }
-
-    // polymorphic
-    std::unique_ptr<BaseArgument>
-    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
-                        std::array<const void*, NumBTensor> p_bs,
-                        std::array<const void*, NumDTensor> p_ds,
-                        void* p_e,
-                        const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_lengths,
-                        const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_strides,
-                        const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_lengths,
-                        const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_strides,
-                        const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_lengths,
-                        const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_strides,
-                        const std::vector<index_t>& e_ms_ns_length,
-                        const std::vector<index_t>& e_ms_ns_stride,
-                        AElementwiseOperation a_element_op,
-                        BElementwiseOperation b_element_op,
-                        CDEElementwiseOperation cde_element_op) override
-    {
-        return std::make_unique<Argument>(p_as,
-                                          p_bs,
-                                          p_ds,
-                                          p_e,
-                                          as_ms_ks_lengths,
-                                          as_ms_ks_strides,
-                                          bs_ns_ks_lengths,
-                                          bs_ns_ks_strides,
-                                          ds_ms_ns_lengths,
-                                          ds_ms_ns_strides,
-                                          e_ms_ns_length,
-                                          e_ms_ns_stride,
-                                          a_element_op,
-                                          b_element_op,
-                                          cde_element_op);
-    }
-
-    // polymorphic
-    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
-    {
-        return std::make_unique<Invoker>(Invoker{});
-    }
-
-    // polymorphic
-    std::string GetTypeString() const override
-    {
-        auto str = std::stringstream();
-
-        std::map<LoopScheduler, std::string> LoopSchedToString{
-            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
-
-        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
-                                                                       {PipelineVersion::v2, "v2"}};
-
-        // clang-format off
-        str << "DeviceContractionMultipleABD_Xdl_CShuffle"
-            << "<"
-            << BlockSize << ", "
-            << MPerBlock << ", "
-            << NPerBlock << ", "
-            << KPerBlock << ", "
-            << AK1 << ", "
-            << BK1 << ", "
-            << MPerXDL << ", "
-            << NPerXDL << ", "
-            << MXdlPerWave << ", "
-            << NXdlPerWave << ", "
-            << ABlockTransferSrcScalarPerVector << ", "
-            << BBlockTransferSrcScalarPerVector << ", "
-            << CShuffleMXdlPerWavePerShuffle << ", "
-            << CShuffleNXdlPerWavePerShuffle << ", "
-            << getGemmSpecializationString(GemmSpec)
-            << ">"
-            << " LoopScheduler: "
-            << LoopSchedToString[LoopSched] << ", "
-            << "PipelineVersion: "
-            << PipelineVersionToString[PipelineVer];
-        // clang-format on
-
-        return str.str();
-    }
-};
-
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
index db119178b7..5ae06836fa 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp
@@ -112,7 +112,6 @@ template <index_t NumDimM,
           typename CShuffleDataType,
           typename DsDataType,
           typename EDataType,
-          typename ComputeDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           typename CDEElementwiseOperation,
@@ -157,8 +156,7 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
                                         EDataType,
                                         AElementwiseOperation,
                                         BElementwiseOperation,
-                                        CDEElementwiseOperation,
-                                        ComputeDataType>
+                                        CDEElementwiseOperation>
 {
     using DeviceOp = DeviceContractionMultipleD_Xdl_CShuffle;
 
@@ -312,6 +310,8 @@ struct DeviceContractionMultipleD_Xdl_CShuffle
     using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({{}}, {{}}))>;
     using EGridDesc_M_N  = decltype(MakeEGridDescriptor_M_N({}, {}));
 
+    using ComputeDataType = ADataType;
+
     // GridwiseGemm
     using GridwiseGemm = GridwiseGemmMultipleD_xdl_cshuffle<
         ADataType, // TODO: distinguish A/B datatype
diff --git a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
index 1b9e37b74a..9fe0931cba 100644
--- a/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
@@ -186,25 +186,6 @@ struct Bilinear
         y = type_convert<half_t>(alpha_ * x0 + beta_ * ck::type_convert<float>(x1));
     };
 
-    template <>
-    __host__ __device__ constexpr void
-    operator()<bhalf_t, bhalf_t, bhalf_t>(bhalf_t& y, const bhalf_t& x0, const bhalf_t& x1) const
-    {
-        const float x0_tmp = type_convert<float>(x0);
-        const float x1_tmp = type_convert<float>(x1);
-        const float y_tmp  = alpha_ * x0_tmp + beta_ * x1_tmp;
-        y                  = type_convert<bhalf_t>(y_tmp);
-    };
-
-    template <>
-    __host__ __device__ constexpr void
-    operator()<bhalf_t, float, bhalf_t>(bhalf_t& y, const float& x0, const bhalf_t& x1) const
-    {
-        const float x1_tmp = ck::type_convert<float>(x1);
-        const float y_tmp  = alpha_ * x0 + beta_ * x1_tmp;
-        y                  = y_tmp;
-    };
-
     template <>
     __host__ __device__ constexpr void operator()<std::int8_t, std::int32_t, std::int8_t>(
         std::int8_t& y, const std::int32_t& x0, const std::int8_t& x1) const
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 337117551f..0b0e731c60 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -33,12 +33,6 @@ struct PassThrough
         y = type_convert<float>(x);
     }
 
-    template <>
-    __host__ __device__ void operator()<double, float>(double& y, const float& x) const
-    {
-        y = type_convert<double>(x);
-    }
-
     template <>
     __host__ __device__ void operator()<float, float>(float& y, const float& x) const
     {
@@ -75,12 +69,6 @@ struct PassThrough
         y = type_convert<bhalf_t>(x);
     }
 
-    template <>
-    __host__ __device__ void operator()<float, bhalf_t>(float& y, const bhalf_t& x) const
-    {
-        y = type_convert<float>(x);
-    }
-
     template <>
     __host__ __device__ void operator()<bhalf_t, half_t>(bhalf_t& y, const half_t& x) const
     {
@@ -243,20 +231,6 @@ struct Scale
     template <typename Y, typename X>
     __host__ __device__ void operator()(Y& y, const X& x) const;
 
-    template <>
-    __host__ __device__ void operator()<half_t, half_t>(half_t& y, const half_t& x) const
-    {
-        y = ck::type_convert<half_t>(scale_) * x;
-    };
-
-    template <>
-    __host__ __device__ void operator()<bhalf_t, bhalf_t>(bhalf_t& y, const bhalf_t& x) const
-    {
-        const float x_tmp = ck::type_convert<float>(x);
-        const float y_tmp = scale_ * x_tmp;
-        y                 = ck::type_convert<bhalf_t>(y_tmp);
-    };
-
     template <>
     __host__ __device__ void operator()<float, float>(float& y, const float& x) const
     {
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
index 6321de73a9..10a0842dac 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
@@ -428,7 +428,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
             [&](auto i) {
                 using ALayout = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
 
-                return MakeAGridDescriptor_M_K<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
+                return MakeAGridDescriptor_M_N<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
             },
             Number<NumATensor>{});
     }
@@ -657,7 +657,6 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
             ComputeDataType,
-            ComputeDataType,
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
index 527dac6d3b..92a8c82a6e 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_contraction.hpp
@@ -23,7 +23,6 @@ template <ck::index_t NumDimM,
           typename BDataType,
           typename CDataType,
           typename AccDataType,
-          typename ComputeDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
           ck::enable_if_t<NumDimM == 2 && NumDimN == 2 && NumDimK == 2, bool> = false>
@@ -70,24 +69,19 @@ struct ReferenceContraction_M2_N2_K2 : public ck::tensor_operation::device::Base
                 {
                     for(ck::index_t k1 = 0; k1 < K1; ++k1)
                     {
-                        // Simulate the possible casting when ComputeDataType is different than the
-                        // A/B data types
-                        ComputeDataType v_a_compute_input =
-                            ck::type_convert<ComputeDataType>(arg.a_ms_ks_(m0, m1, k0, k1));
-                        ComputeDataType v_b_compute_input =
-                            ck::type_convert<ComputeDataType>(arg.b_ns_ks_(n0, n1, k0, k1));
-
                         AccDataType v_a;
                         AccDataType v_b;
 
-                        arg.a_element_op_(v_a, ck::type_convert<AccDataType>(v_a_compute_input));
-                        arg.b_element_op_(v_b, ck::type_convert<AccDataType>(v_b_compute_input));
+                        arg.a_element_op_(
+                            v_a, ck::type_convert<const AccDataType>(arg.a_ms_ks_(m0, m1, k0, k1)));
+                        arg.b_element_op_(
+                            v_b, ck::type_convert<const AccDataType>(arg.b_ns_ks_(n0, n1, k0, k1)));
 
                         v_acc += v_a * v_b;
                     }
                 }
 
-                arg.c_ms_ns_(m0, m1, n0, n1) = ck::type_convert<CDataType>(v_acc);
+                arg.c_ms_ns_(m0, m1, n0, n1) = v_acc;
             };
 
             make_ParallelTensorFunctor(f_ms_ns,
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index f6fb3f9eee..ea11fd2e1a 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -29,8 +29,6 @@ using BF8 = ck::bf8_t;
 
 using Empty_Tuple = ck::Tuple<>;
 
-using BF16_Tuple = ck::Tuple<BF16>;
-
 using F16_Tuple     = ck::Tuple<F16>;
 using F16_F16_Tuple = ck::Tuple<F16, F16>;
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp
deleted file mode 100644
index 96f4d4e30e..0000000000
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp
+++ /dev/null
@@ -1,292 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-#pragma once
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-using F16  = ck::half_t;
-using BF16 = ck::bhalf_t;
-using F32  = float;
-using F64  = double;
-
-using F16_Tuple   = ck::Tuple<F16>;
-using BF16_Tuple  = ck::Tuple<BF16>;
-using F32_Tuple   = ck::Tuple<F32>;
-using F64_Tuple   = ck::Tuple<F64>;
-using Empty_Tuple = ck::Tuple<>;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
-using Scale       = ck::tensor_operation::element_wise::Scale;
-
-template <ck::index_t... Is>
-using S = ck::Sequence<Is...>;
-
-static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_kk_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_kn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,      
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,      
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,      
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_mk_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_mn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_f64_kk_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_f64_kn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_f64_mk_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
-
-template <typename ADataType,
-          typename BDataType,
-          typename AccDataType,
-          typename CShuffleDataType,
-          typename DsDataType,
-          typename EDataType,
-          typename ComputeDataType,
-          typename AElementwiseOp,
-          typename BElementwiseOp,
-          typename CDEElementwiseOp>
-using device_contraction_f64_mn_instance = std::tuple<
-    // clang-format off
-        //#####################################| NumDimM| NumDimN| NumDimK|      AData|     BData|     AccData|         CShuffle|     DsData|     EData|         Compute|              A|              B|              CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //#####################################|        |        |        |       Type|      Type|        Type|         DataType|       Type|      Type|            Data|    Elementwise|    Elementwise|      Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //#####################################|        |        |        |           |          |            |                 |           |          |            Type|      Operation|      Operation|        Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //#####################################|        |        |        |           |          |            |                 |           |          |                |               |               |                 |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
-        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,  ADataType, BDataType, AccDataType, CShuffleDataType, DsDataType, EDataType, ComputeDataType, AElementwiseOp, BElementwiseOp, CDEElementwiseOp, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
-    // clang-format on
-    >;
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
index d06cab1196..bd3af891ec 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_bilinear.hpp
@@ -17,6 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #ifdef CK_ENABLE_FP32
+// float
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -27,8 +28,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
+                                                           Bilinear>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -40,8 +40,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
+                                                           Bilinear>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -53,8 +52,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
+                                                           Bilinear>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,115 +64,10 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances);
-#endif // CK_ENABLE_FP32
-
+                                                           Bilinear>>>& instances);
+#endif
 #ifdef CK_ENABLE_FP64
+// double
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -185,8 +78,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances);
+                                                           Bilinear>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -198,8 +90,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances);
+                                                           Bilinear>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -211,8 +102,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances);
+                                                           Bilinear>>>& instances);
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -224,170 +114,8 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-#endif // CK_ENABLE_FP64
-
-#ifdef CK_ENABLE_FP16
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-#endif // CK_ENABLE_FP16
-
-#ifdef CK_ENABLE_BF16
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances);
-#endif // CK_ENABLE_FP16
-
+                                                           Bilinear>>>& instances);
+#endif
 // Contraction + Bilinear
 template <index_t NumDimM,
           index_t NumDimN,
@@ -395,8 +123,7 @@ template <index_t NumDimM,
           typename ADataType,
           typename BDataType,
           typename DDataType,
-          typename EDataType,
-          typename ComputeDataType>
+          typename EDataType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContractionMultipleD<
     NumDimM,
     NumDimN,
@@ -407,8 +134,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     EDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::Bilinear,
-    ComputeDataType>>
+    ck::tensor_operation::element_wise::Bilinear>>
 {
     using DeviceOp = DeviceContractionMultipleD<NumDimM,
                                                 NumDimN,
@@ -419,125 +145,45 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                                                 EDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::Bilinear,
-                                                ComputeDataType>;
+                                                ck::tensor_operation::element_wise::Bilinear>;
 
     static auto GetInstances()
     {
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 #ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
-                     is_same_v<EDataType, float>)
+                     is_same_v<DDataType, float> && is_same_v<EDataType, float>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
-                        op_ptrs);
-                }
-                else if constexpr(is_same_v<ComputeDataType, ck::half_t>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
-                        op_ptrs);
-                }
-                else if constexpr(is_same_v<ComputeDataType, ck::bhalf_t>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
-                        op_ptrs);
-                }
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
+                    op_ptrs);
             }
         }
-#endif // CK_ENABLE_FP32
+#endif
 #ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
-                     is_same_v<EDataType, double>)
+                     is_same_v<DDataType, double> && is_same_v<EDataType, double>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                if constexpr(is_same_v<ComputeDataType, double>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
-                        op_ptrs);
-                }
-                else if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
-                        op_ptrs);
-                }
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
+                    op_ptrs);
+                add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
+                    op_ptrs);
             }
         }
-#endif // CK_ENABLE_FP64
-#ifdef CK_ENABLE_FP16
-        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
-                     is_same_v<EDataType, ck::half_t>)
-        {
-            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
-            {
-                if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
-                        op_ptrs);
-                }
-            }
-        }
-#endif // CK_ENABLE_FP16
-#ifdef CK_ENABLE_BF16
-        if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
-                     is_same_v<EDataType, ck::bhalf_t>)
-        {
-            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
-            {
-                if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
-                        op_ptrs);
-                    add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
-                        op_ptrs);
-                }
-            }
-        }
-#endif // CK_ENABLE_BF16
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
index 8e994d61ae..4dd735d062 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/contraction_scale.hpp
@@ -17,6 +17,7 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 #ifdef CK_ENABLE_FP32
+// float
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -27,8 +28,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
+                                                           Scale>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -40,8 +40,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
+                                                           Scale>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -53,8 +52,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
+                                                           Scale>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -66,115 +64,10 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances);
-#endif // CK_ENABLE_FP32
-
+                                                           Scale>>>& instances);
+#endif
 #ifdef CK_ENABLE_FP64
+// double
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
                                                            2,
@@ -185,8 +78,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances);
+                                                           Scale>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -198,8 +90,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances);
+                                                           Scale>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -211,8 +102,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances);
+                                                           Scale>>>& instances);
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -224,178 +114,15 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-#endif // CK_ENABLE_FP64
-
-#ifdef CK_ENABLE_FP16
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-#endif // CK_ENABLE_FP16
-
-#ifdef CK_ENABLE_BF16
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances);
-#endif // CK_ENABLE_FP16
-
+                                                           Scale>>>& instances);
+#endif
 // Contraction + Scale
 template <index_t NumDimM,
           index_t NumDimN,
           index_t NumDimK,
           typename ADataType,
           typename BDataType,
-          typename EDataType,
-          typename ComputeDataType>
+          typename EDataType>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContractionMultipleD<
     NumDimM,
     NumDimN,
@@ -406,8 +133,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
     EDataType,
     ck::tensor_operation::element_wise::PassThrough,
     ck::tensor_operation::element_wise::PassThrough,
-    ck::tensor_operation::element_wise::Scale,
-    ComputeDataType>>
+    ck::tensor_operation::element_wise::Scale>>
 {
     using DeviceOp = DeviceContractionMultipleD<NumDimM,
                                                 NumDimN,
@@ -418,8 +144,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
                                                 EDataType,
                                                 ck::tensor_operation::element_wise::PassThrough,
                                                 ck::tensor_operation::element_wise::PassThrough,
-                                                ck::tensor_operation::element_wise::Scale,
-                                                ComputeDataType>;
+                                                ck::tensor_operation::element_wise::Scale>;
 
     static auto GetInstances()
     {
@@ -430,113 +155,34 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceContra
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
-                        op_ptrs);
-                }
-                else if constexpr(is_same_v<ComputeDataType, ck::half_t>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
-                        op_ptrs);
-                }
-                else if constexpr(is_same_v<ComputeDataType, ck::bhalf_t>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
-                        op_ptrs);
-                }
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
+                    op_ptrs);
             }
         }
-#endif // CK_ENABLE_FP32
+#endif
 #ifdef CK_ENABLE_FP64
         if constexpr(is_same_v<ADataType, double> && is_same_v<BDataType, double> &&
                      is_same_v<EDataType, double>)
         {
             if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
             {
-                if constexpr(is_same_v<ComputeDataType, double>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
-                        op_ptrs);
-                }
-                else if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
-                        op_ptrs);
-                }
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
+                    op_ptrs);
+                add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
+                    op_ptrs);
             }
         }
-#endif // CK_ENABLE_FP64
-#ifdef CK_ENABLE_FP16
-        if constexpr(is_same_v<ADataType, ck::half_t> && is_same_v<BDataType, ck::half_t> &&
-                     is_same_v<EDataType, ck::half_t>)
-        {
-            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
-            {
-                if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
-                        op_ptrs);
-                }
-            }
-        }
-#endif // CK_ENABLE_FP16
-#ifdef CK_ENABLE_BF16
-        if constexpr(is_same_v<ADataType, ck::bhalf_t> && is_same_v<BDataType, ck::bhalf_t> &&
-                     is_same_v<EDataType, ck::bhalf_t>)
-        {
-            if constexpr(NumDimM == 2 && NumDimN == 2 && NumDimK == 2)
-            {
-                if constexpr(is_same_v<ComputeDataType, float>)
-                {
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
-                        op_ptrs);
-                    add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
-                        op_ptrs);
-                }
-            }
-        }
-#endif // CK_ENABLE_BF16
+#endif
         return op_ptrs;
     }
 };
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
index 87a6bbba4e..ee9c419e1a 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/CMakeLists.txt
@@ -1,42 +1,14 @@
 set(DEVICE_CONTRACTION_BILINEAR_INSTANCES)
-
-# FP32
+#float
 list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp)
 
-list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp)
-
-list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp)
-
-# FP64
+#double
 list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
-
-list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp)
-
-# FP16
-list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp)
-
-# BF16
-list(APPEND DEVICE_CONTRACTION_BILINEAR_INSTANCES device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
-                                                    device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp)
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+                                                  device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp)
 
 add_instance_library(device_contraction_bilinear_instance ${DEVICE_CONTRACTION_BILINEAR_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
deleted file mode 100644
index 56dc1d2c71..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance =
-    device_contraction_kk_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   BF16_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_kknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
deleted file mode 100644
index 7b8163814e..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance =
-    device_contraction_kn_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   BF16_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_knnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
deleted file mode 100644
index ba202202f9..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance =
-    device_contraction_mk_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   BF16_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp
deleted file mode 100644
index b6acab1c18..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance =
-    device_contraction_mn_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   BF16_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           BF16_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_bf16_compute_f32_mnnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
deleted file mode 100644
index 1bd3495b36..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance =
-    device_contraction_kk_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   F16_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_kknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
deleted file mode 100644
index 6817546d3d..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance =
-    device_contraction_kn_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   F16_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_knnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
deleted file mode 100644
index cedace2cb2..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance =
-    device_contraction_mk_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   F16_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp
deleted file mode 100644
index b8383e5268..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance =
-    device_contraction_mn_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   F16_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           F16_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_f16_compute_f32_mnnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
deleted file mode 100644
index 2e73bdb804..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance =
-    device_contraction_kk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_kknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
deleted file mode 100644
index ae69db03ac..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance =
-    device_contraction_kn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_knnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
deleted file mode 100644
index fc6b6003f7..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance =
-    device_contraction_mk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp
deleted file mode 100644
index b81984acc4..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance =
-    device_contraction_mn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_bf16_mnnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
deleted file mode 100644
index 8da6a79f0a..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance =
-    device_contraction_kk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_kknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
deleted file mode 100644
index b05c443e85..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance =
-    device_contraction_kn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_knnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
deleted file mode 100644
index c2cbb9f25f..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance =
-    device_contraction_mk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp
deleted file mode 100644
index 887eaad7eb..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance =
-    device_contraction_mn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           F32_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_compute_f16_mnnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
index 9a3c71acb5..5587db77e0 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,40 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F32       = float;
+using F32_Tuple = ck::Tuple<F32>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance =
-    device_contraction_kk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +66,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_kknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
index 5659d477d4..26262855ea 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,43 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F32       = float;
+using F32_Tuple = ck::Tuple<F32>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance =
-    device_contraction_kn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +69,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_knnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
index 7cbe1e34be..befc0dcd10 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,43 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F32       = float;
+using F32_Tuple = ck::Tuple<F32>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance =
-    device_contraction_mk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +69,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mknn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
index e26dcf394b..e45b47cf94 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,43 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F32       = float;
+using F32_Tuple = ck::Tuple<F32>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance =
-    device_contraction_mn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   F32_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, F32_Tuple,   F32,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +69,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_f32_mnnn
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
deleted file mode 100644
index 74c5c9c653..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance =
-    device_contraction_f64_kk_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_kknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
deleted file mode 100644
index 8a2705f722..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance =
-    device_contraction_f64_kn_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_knnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
deleted file mode 100644
index 47924a843d..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance =
-    device_contraction_f64_mk_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mknn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp
deleted file mode 100644
index d6c684e2b0..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance =
-    device_contraction_f64_mn_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
-
-void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           F64_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Bilinear,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_compute_f32_mnnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
index 75bd8e3f4a..f437a227d5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance =
-    device_contraction_f64_kk_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_kknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
index c515f37916..13fdbeb35c 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance =
-    device_contraction_f64_kn_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_knnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
index ff9fa61e0b..95ef8c4929 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance =
-    device_contraction_f64_mk_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mknn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
index f274c45b58..290f81d7c9 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_bilinear/device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64       = double;
+using F64_Tuple = ck::Tuple<F64>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Bilinear    = ck::tensor_operation::element_wise::Bilinear;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance =
-    device_contraction_f64_mn_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       F64_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Bilinear>;
+using device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|    DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|      Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |          |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |          |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, F64_Tuple,   F64,  PassThrough, PassThrough,    Bilinear, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_bilinear_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_f64_mnnn
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Bilinear,
-                                                           F64>>>& instances)
+                                                           Bilinear>>>& instances)
 {
     add_device_operation_instances(
         instances,
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
index a0918d9d3f..673fae3373 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/CMakeLists.txt
@@ -1,43 +1,15 @@
 set(DEVICE_CONTRACTION_SCALE_INSTANCES)
-
-# FP32
+#float
 list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp)
 
-list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp)
-
-list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp)
-
-# FP64
+#double
 list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
                                                 device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp)
 
-list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp)
-
-# FP16
-list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp)
-
-# BF16
-list(APPEND DEVICE_CONTRACTION_SCALE_INSTANCES device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
-                                                device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp)
-
 add_instance_library(device_contraction_scale_instance ${DEVICE_CONTRACTION_SCALE_INSTANCES})
 
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
deleted file mode 100644
index ace4d4a330..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance =
-    device_contraction_kk_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   Empty_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_kkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
deleted file mode 100644
index d2a1d45f7a..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance =
-    device_contraction_kn_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   Empty_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_knn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
deleted file mode 100644
index aa4e4df9d9..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance =
-    device_contraction_mk_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   Empty_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp
deleted file mode 100644
index a59bce0a1f..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance =
-    device_contraction_mn_instance<BF16,
-                                   BF16,
-                                   F32,
-                                   BF16,
-                                   Empty_Tuple,
-                                   BF16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           BF16,
-                                                           BF16,
-                                                           Empty_Tuple,
-                                                           BF16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_bf16_bf16_bf16_compute_f32_mnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
deleted file mode 100644
index 876acd3f69..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance =
-    device_contraction_kk_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   Empty_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_kkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
deleted file mode 100644
index 96e961d3bb..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance =
-    device_contraction_kn_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   Empty_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_knn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
deleted file mode 100644
index 69cae90889..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance =
-    device_contraction_mk_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   Empty_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp
deleted file mode 100644
index e6ac98273a..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance =
-    device_contraction_mn_instance<F16,
-                                   F16,
-                                   F32,
-                                   F16,
-                                   Empty_Tuple,
-                                   F16,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F16,
-                                                           F16,
-                                                           Empty_Tuple,
-                                                           F16,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f16_f16_f16_compute_f32_mnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
deleted file mode 100644
index da1df4c43e..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance =
-    device_contraction_kk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_kkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
deleted file mode 100644
index 21c4e87fb1..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance =
-    device_contraction_kn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_knn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
deleted file mode 100644
index 9d22f8feff..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance =
-    device_contraction_mk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp
deleted file mode 100644
index 8e861f74c9..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance =
-    device_contraction_mn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   BF16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           BF16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_bf16_mnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
deleted file mode 100644
index 390b5526e3..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance =
-    device_contraction_kk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_kkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
deleted file mode 100644
index 385a13cc25..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance =
-    device_contraction_kn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_knn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
deleted file mode 100644
index 1afd432bac..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance =
-    device_contraction_mk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp
deleted file mode 100644
index bfb4d4630f..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance =
-    device_contraction_mn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F16,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F32,
-                                                           F32,
-                                                           Empty_Tuple,
-                                                           F32,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F16>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_compute_f16_mnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
index 51bc64eaad..16fd1cb407 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,40 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance =
-    device_contraction_kk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
+using F32         = float;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
+// k/k/n are the fast changing dimension for A/B/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    32,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    32,    16,   4,   4,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    32,    64,    16,   4,   4,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,              S<1,  8, 1,  8>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +66,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_kkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
index 4cd28d52dd..ff37bf7cce 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,43 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance =
-    device_contraction_kn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
+using F32         = float;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
+// k/n/n are the fast changing dimension for A/B/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   1,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   1,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   1,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32, PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +69,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_knn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
index 82bba62715..8a1f6f9334 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,43 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance =
-    device_contraction_mk_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
+using F32         = float;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
+// m/k/n are the fast changing dimension for A/B/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   4,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   4,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   4,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   4,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              4,              4,         1,           1,           1,               S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +69,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
index 06bf4f34b5..d333f59726 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,43 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance =
-    device_contraction_mn_instance<F32,
-                                   F32,
-                                   F32,
-                                   F32,
-                                   Empty_Tuple,
-                                   F32,
-                                   F32,
-                                   PassThrough,
-                                   PassThrough,
-                                   Scale>;
+using F32         = float;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] = E[m0, m1, n0, n1]
+// m/n/n are the fast changing dimension for A/B/E
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   1,   1,   32,   32,    2,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   256,    16,   4,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   1,   1,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1,  8>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   32,   32,    2,    2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   4,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         1,           1,           1,              S<1,  8, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   32,   32,    2,    1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   4,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   32,   32,    1,    2,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              1,         0,           1,           1,              S<1, 16, 1, 16>,               4>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F32,   F32,     F32,      F32, Empty_Tuple,   F32,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   4,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              4,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              4,         1,           1,           1,              S<1, 16, 1, 16>,               4>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +69,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instanc
                                                            F32,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f32_f32_f32_mnn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
deleted file mode 100644
index c047153bd9..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance =
-    device_contraction_f64_kk_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_kkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
deleted file mode 100644
index ee33c56223..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance =
-    device_contraction_f64_kn_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_knn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
deleted file mode 100644
index b62d4bb2aa..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance =
-    device_contraction_f64_mk_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mkn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp
deleted file mode 100644
index 346f91beb0..0000000000
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance.cpp
+++ /dev/null
@@ -1,57 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-// This (ifndef) is a hack to use customized behavior for buffer load rather than using default
-// setting Don't use this hack unless absolutely necessary!
-// FIXME: make the behavior of buffer load a configurable (template) parameter of each device op
-#define CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK 1
-
-#include <cstdlib>
-
-#include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
-#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
-namespace ck {
-namespace tensor_operation {
-namespace device {
-namespace instance {
-
-// A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
-// m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance =
-    device_contraction_f64_mn_instance<F64,
-                                       F64,
-                                       F32,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F32,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
-
-void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance(
-    std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
-                                                           2,
-                                                           2,
-                                                           F64,
-                                                           F64,
-                                                           Empty_Tuple,
-                                                           F64,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           Scale,
-                                                           F32>>>& instances)
-{
-    add_device_operation_instances(
-        instances,
-        device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_compute_f32_mnn_instance{});
-}
-
-} // namespace instance
-} // namespace device
-} // namespace tensor_operation
-} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
index 30cc3b9652..4c87b51a9e 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance =
-    device_contraction_f64_kk_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    32,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    64,    32,    16,   2,   2,   16,   16,    4,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,    64,    32,    64,    16,   2,   2,   16,   16,    2,    4,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,              S<1,  8, 1,  8>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_kkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
index 0a36ef7ffd..fd3f57c6b6 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // k/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance =
-    device_contraction_f64_kn_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   1,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   1,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_knn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
index 617332c2c7..1e53f0b2f5 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/k/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance =
-    device_contraction_f64_mk_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|  CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|          _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|          _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                              |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   2,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   2,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   2,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              1,              1,         1,           1,           1,               S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mkn_instance{});
diff --git a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
index 1df7f31201..d02d146a9b 100644
--- a/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/contraction_scale/device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance.cpp
@@ -9,9 +9,11 @@
 #include <cstdlib>
 
 #include "ck/ck.hpp"
-#include "ck/tensor_operation/gpu/device/device_contraction_multiple_d.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
-#include "ck/library/tensor_operation_instance/gpu/contraction/device_contraction_instance.hpp"
+
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
 
 namespace ck {
@@ -19,19 +21,37 @@ namespace tensor_operation {
 namespace device {
 namespace instance {
 
+using F64         = double;
+using Empty_Tuple = ck::Tuple<>;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+using Scale       = ck::tensor_operation::element_wise::Scale;
+
+static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1]
 // m/n/n/n are the fast changing dimension for A/B/D/E
-using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance =
-    device_contraction_f64_mn_instance<F64,
-                                       F64,
-                                       F64,
-                                       F64,
-                                       Empty_Tuple,
-                                       F64,
-                                       F64,
-                                       PassThrough,
-                                       PassThrough,
-                                       Scale>;
+using device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance = std::tuple<
+    // clang-format off
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|      DsData| EData|            A|           B|         CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|        Type|  Type|  Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |            |      |    Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |            |      |             |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   1,   1,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,   128,    64,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1,  8>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   1,   1,   16,   16,    4,    4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   128,    64,   128,    16,   2,   2,   16,   16,    4,    4,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1,  8, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   1,   1,   16,   16,    4,    2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,   128,    64,    16,   2,   2,   16,   16,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   1,   1,   16,   16,    2,    4,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         0,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         0,           1,           1,              S<1, 16, 1, 16>,               1>,
+        DeviceContractionMultipleD_Xdl_CShuffle<       2,       2,       2,   F64,   F64,     F64,      F64, Empty_Tuple,   F64,  PassThrough, PassThrough,       Scale, GemmMNKPadding,        1,   256,    64,   128,    16,   2,   2,   16,   16,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              1,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              1,         1,           1,           1,              S<1, 16, 1, 16>,               1>
+    // clang-format on
+    >;
 
 void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance(
     std::vector<std::unique_ptr<DeviceContractionMultipleD<2,
@@ -43,8 +63,7 @@ void add_device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instanc
                                                            F64,
                                                            PassThrough,
                                                            PassThrough,
-                                                           Scale,
-                                                           F64>>>& instances)
+                                                           Scale>>>& instances)
 {
     add_device_operation_instances(
         instances, device_contraction_scale_m2_n2_k2_xdl_c_shuffle_f64_f64_f64_mnn_instance{});
diff --git a/profiler/README.md b/profiler/README.md
index b7b5be420b..d081fc33e9 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -50,23 +50,21 @@ Best Perf: 1.42509 ms, 102.988 TFlops, 234.086 GB/s
 ## Profile contraction kernels
 ```bash
 #arg1: tensor operation (contraction_bilinear=CONTRACTION+Bilinear)
-#arg2: data type (0: fp32; 1: f64; 2: f16; 3: bf16)
-#arg3: compute data type (0: fp32; 1: f64; 2: f16; 3: bf16)
-#arg4: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
+#arg2: data type (0: fp32; 1: f64)\n"
+#arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
 #                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
 #                     2: A[k0, k1, m0, m1] * B[k0, k1, n0, n1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1];
 #                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + D[m0, m1, n0, n1] = E[m0, m1, n0, n1])
-#arg5: verification (0: no; 1: yes)
-#arg6: initialization (0: no init; 1: integer value; 2: decimal value)
-#arg7: print tensor value (0: no; 1: yes)
-#arg8: time kernel (0: no, 1: yes)
-#arg9: alpha
-#arg10: beta
-#arg11 to 16: M0, M1, N0, N1, K0, K1
-#arg17 to 32: Strides for A, B, D and E (skip for default)
+#arg4: verification (0: no; 1: yes)
+#arg5: initialization (0: no init; 1: integer value; 2: decimal value)
+#arg6: print tensor value (0: no; 1: yes)
+#arg7: time kernel (0: no, 1: yes)
+#arg8 and arg9: alpha and beta
+#arg10 to 15: M0, M1, N0, N1, K0, K1
+#arg16 to 31: Strides for A, B, D and E (skip for default)
 
-################                   op  datatype  compute_datatype  layout  verify  init  log  time  alpha  beta  M0  M1  N0  N1  K0  K1
-./bin/ckProfiler contraction_bilinear         0                 0       1       0     0    0     1    1.0   1.0 128 128 128 128 128 128
+################                   op  datatype  layout  verify  init  log  time  alpha  beta  M0  M1  N0  N1  K0  K1
+./bin/ckProfiler contraction_bilinear         0       1       0     0    0     1    1.0   1.0 128 128 128 128 128 128
 ```
 
 Result (MI100)
diff --git a/profiler/include/profiler/profile_contraction_impl.hpp b/profiler/include/profiler/profile_contraction_impl.hpp
index f6e4b3f395..660cc3f9e5 100644
--- a/profiler/include/profiler/profile_contraction_impl.hpp
+++ b/profiler/include/profiler/profile_contraction_impl.hpp
@@ -31,14 +31,10 @@ namespace profiler {
 using Bilinear = ck::tensor_operation::element_wise::Bilinear;
 using Scale    = ck::tensor_operation::element_wise::Scale;
 
-using F32 = float;
-using F64 = double;
-
 template <typename ALayout,
           typename BLayout,
           typename CDELayout,
           typename DataType,
-          typename ComputeDataType,
           typename DTupleDataType,
           typename CDElementOp>
 int profile_contraction_impl(ck::index_t do_verification,
@@ -49,10 +45,10 @@ int profile_contraction_impl(ck::index_t do_verification,
                              const std::vector<ck::index_t>& M,
                              const std::vector<ck::index_t>& N,
                              const std::vector<ck::index_t>& K,
-                             const std::vector<ck::index_t>& StridesA, // [M0, M1, K0, K1]
-                             const std::vector<ck::index_t>& StridesB, // [N0, N1, K0, K1]
-                             const std::vector<ck::index_t>& StridesE, // [M0, M1, N0, N1]
-                             const std::vector<ck::index_t>& StridesD) // [M0, M1, N0, N1]
+                             const std::vector<ck::index_t>& StridesA,
+                             const std::vector<ck::index_t>& StridesB,
+                             const std::vector<ck::index_t>& StridesE,
+                             const std::vector<ck::index_t>& StridesD)
 {
     bool pass = true;
 
@@ -67,13 +63,13 @@ int profile_contraction_impl(ck::index_t do_verification,
     };
 
     Tensor<DataType> a_m_k(f_host_tensor_descriptor(M, K, StridesA));
-    Tensor<DataType> b_n_k(f_host_tensor_descriptor(N, K, StridesB));
+    Tensor<DataType> b_k_n(f_host_tensor_descriptor(K, N, StridesB));
     Tensor<DataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
     Tensor<DataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StridesE));
     Tensor<DataType> d_m_n(f_host_tensor_descriptor(M, N, StridesD));
 
     std::cout << "a_m_k: " << a_m_k.mDesc << std::endl;
-    std::cout << "b_n_k: " << b_n_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
     std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
     std::cout << "e_m_n: " << e_m_n_device_result.mDesc << std::endl;
 
@@ -82,12 +78,12 @@ int profile_contraction_impl(ck::index_t do_verification,
     case 0: break;
     case 1:
         a_m_k.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
-        b_n_k.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
         d_m_n.GenerateTensorValue(GeneratorTensor_2<DataType>{-5, 5});
         break;
     default:
         a_m_k.GenerateTensorValue(GeneratorTensor_3<DataType>{0.0, 1.0});
-        b_n_k.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
         d_m_n.GenerateTensorValue(GeneratorTensor_3<DataType>{-0.5, 0.5});
     }
 
@@ -95,12 +91,12 @@ int profile_contraction_impl(ck::index_t do_verification,
     using BElementOp = ck::tensor_operation::element_wise::PassThrough;
 
     DeviceMem a_device_buf(sizeof(DataType) * a_m_k.mDesc.GetElementSpaceSize());
-    DeviceMem b_device_buf(sizeof(DataType) * b_n_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(DataType) * b_k_n.mDesc.GetElementSpaceSize());
     DeviceMem e_device_buf(sizeof(DataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
     DeviceMem d_device_buf(sizeof(DataType) * d_m_n.mDesc.GetElementSpaceSize());
 
     a_device_buf.ToDevice(a_m_k.mData.data());
-    b_device_buf.ToDevice(b_n_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
     e_device_buf.SetZero();
     d_device_buf.ToDevice(d_m_n.mData.data());
 
@@ -122,8 +118,7 @@ int profile_contraction_impl(ck::index_t do_verification,
                                                                               DataType,
                                                                               AElementOp,
                                                                               BElementOp,
-                                                                              CDElementOp,
-                                                                              ComputeDataType>;
+                                                                              CDElementOp>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -131,9 +126,6 @@ int profile_contraction_impl(ck::index_t do_verification,
 
     std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
 
-    using AccDataType =
-        typename std::conditional<std::is_same<ComputeDataType, F64>::value, F64, F32>::type;
-
     // Run reference op
     if(do_verification)
     {
@@ -144,8 +136,7 @@ int profile_contraction_impl(ck::index_t do_verification,
                                                                       DataType,
                                                                       DataType,
                                                                       DataType,
-                                                                      AccDataType,
-                                                                      ComputeDataType,
+                                                                      DataType,
                                                                       AElementOp,
                                                                       BElementOp>;
 
@@ -155,7 +146,7 @@ int profile_contraction_impl(ck::index_t do_verification,
         Tensor<DataType> c_m_n_host_result(f_host_tensor_descriptor(M, N, StridesE));
 
         auto ref_argument =
-            ref_op.MakeArgument(a_m_k, b_n_k, c_m_n_host_result, a_element_op, b_element_op);
+            ref_op.MakeArgument(a_m_k, b_k_n, c_m_n_host_result, a_element_op, b_element_op);
 
         ref_invoker.Run(ref_argument);
 
@@ -281,29 +272,8 @@ int profile_contraction_impl(ck::index_t do_verification,
             {
                 e_device_buf.FromDevice(e_m_n_device_result.mData.data());
 
-                // Both the kernel and the reference use `AccDataType`, so an absolute error of both
-                // of them is bounded by `nelems_k * std::numeric_limits<AccDataType>::epsilon()`.
-                // Comparing one to another can result in an absolute error as high as twice that
-                // value.
-                double threshold = 2 * nelems_k * std::numeric_limits<AccDataType>::epsilon();
-                // Handle the possible casting error of either AccDataType -> DataType or
-                // DataType -> ComputeDataType.
-                // TODO: Add a generic solution for calculating thresholds in CK.
-                if constexpr(ck::is_same_v<DataType, ck::bhalf_t> ||
-                             ck::is_same_v<ComputeDataType, ck::bhalf_t>)
-                {
-                    const double epsilon = std::pow(2, -7);
-                    // Maximum relative casting error when rounding to zero.
-                    threshold += epsilon * 2;
-                }
-                else if constexpr(ck::is_same_v<DataType, ck::half_t> ||
-                                  ck::is_same_v<ComputeDataType, ck::half_t>)
-                {
-                    const double epsilon = std::pow(2, -10);
-                    // Maximum relative casting error when rounding to zero.
-                    threshold += epsilon * 2;
-                }
-
+                float threshold =
+                    static_cast<DataType>(nelems_k) * std::numeric_limits<DataType>::epsilon();
                 pass = pass & ck::utils::check_err(e_m_n_device_result,
                                                    e_m_n_host_result,
                                                    "Error: incorrect results!",
@@ -313,7 +283,7 @@ int profile_contraction_impl(ck::index_t do_verification,
                 if(do_log)
                 {
                     LogRangeAsType<float>(std::cout << "a : ", a_m_k.mData, ",") << std::endl;
-                    LogRangeAsType<float>(std::cout << "b: ", b_n_k.mData, ",") << std::endl;
+                    LogRangeAsType<float>(std::cout << "b: ", b_k_n.mData, ",") << std::endl;
                     LogRangeAsType<float>(std::cout << "c_host  : ", e_m_n_host_result.mData, ",")
                         << std::endl;
                     LogRangeAsType<float>(std::cout << "c_device: ", e_m_n_device_result.mData, ",")
diff --git a/profiler/include/profiler/profile_contraction_utils.hpp b/profiler/include/profiler/profile_contraction_utils.hpp
index 05ec7daf9d..076bbd4559 100644
--- a/profiler/include/profiler/profile_contraction_utils.hpp
+++ b/profiler/include/profiler/profile_contraction_utils.hpp
@@ -23,18 +23,8 @@ enum struct ContractionMatrixLayout
 
 enum struct ContractionDataType
 {
-    F32_F32_F32_F32,     // 0
-    F64_F64_F64_F64,     // 1
-    F16_F16_F16_F16,     // 2
-    BF16_BF16_BF16_BF16, // 3
-};
-
-enum struct ContractionComputeDataType
-{
-    F32 = 0,
-    F64,
-    F16,
-    BF16,
+    F32_F32_F32_F32, // 0
+    F64_F64_F64_F64, // 1
 };
 
 inline void collect_index_params(char* argv[],
diff --git a/profiler/src/profile_contraction_bilinear.cpp b/profiler/src/profile_contraction_bilinear.cpp
index 8cb34ccc6c..6ed1841204 100644
--- a/profiler/src/profile_contraction_bilinear.cpp
+++ b/profiler/src/profile_contraction_bilinear.cpp
@@ -17,9 +17,8 @@
 static void print_helper_msg()
 {
     std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
-              << "arg2: data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
-              << "arg3: compute data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
-              << "arg4: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+              << "arg2: data type (0: fp32; 1: f64)\n"
+              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
@@ -27,42 +26,40 @@ static void print_helper_msg()
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
-              << "arg5: verification (0: no; 1: yes)\n"
-              << "arg6: initialization (0: no init; 1: integer value; 2: decimal "
+              << "arg4: verification (0: no; 1: yes)\n"
+              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
               << "value)\n"
-              << "arg7: print tensor value (0: no; 1: yes)\n"
-              << "arg8: time kernel (0: no, 1: yes)\n"
-              << "arg9: alpha\n"
-              << "arg10: beta\n"
-              << "arg11 to 16: M0, M1, N0, N1, K0, K1\n"
-              << "arg17 to 32: Strides for A, B, D and E (skip for default)\n"
+              << "arg6: print tensor value (0: no; 1: yes)\n"
+              << "arg7: time kernel (0: no, 1: yes)\n"
+              << "arg8 and arg9: alpha and beta\n"
+              << "arg10 to 15: M0, M1, N0, N1, K0, K1\n"
+              << "arg16 to 31: Strides for A, B, D and E (skip for default)\n"
               << std::endl;
 }
 
 int profile_contraction_bilinear(int argc, char* argv[])
 {
-    const bool default_strides = argc == 17;
+    const bool default_strides = argc == 16;
 
-    if(argc != 33 && argc != 17)
+    if(argc != 32 && argc != 16)
     {
         print_helper_msg();
         exit(1);
     }
 
     const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
-    const auto compute_data_type  = static_cast<ContractionComputeDataType>(std::stoi(argv[3]));
-    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[4]));
-    const bool do_verification    = std::stoi(argv[5]);
-    const ck::index_t init_method = std::stoi(argv[6]);
-    const bool do_log             = std::stoi(argv[7]);
-    const bool time_kernel        = std::stoi(argv[8]);
-    const float alpha             = std::stof(argv[9]);
-    const float beta              = std::stof(argv[10]);
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification    = std::stoi(argv[4]);
+    const ck::index_t init_method = std::stoi(argv[5]);
+    const bool do_log             = std::stoi(argv[6]);
+    const bool time_kernel        = std::stoi(argv[7]);
+    const float alpha             = std::stof(argv[8]);
+    const float beta              = std::stof(argv[9]);
 
     std::vector<ck::index_t> M;
     std::vector<ck::index_t> N;
     std::vector<ck::index_t> K;
-    const ck::index_t dims_arg_num = 11;
+    const ck::index_t dims_arg_num = 10;
     collect_index_params(argv, M, dims_arg_num, 2);
     collect_index_params(argv, N, dims_arg_num + 2, 2);
     collect_index_params(argv, K, dims_arg_num + 4, 2);
@@ -79,130 +76,90 @@ int profile_contraction_bilinear(int argc, char* argv[])
         collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
     }
 
-    using F16  = ck::half_t;
-    using BF16 = ck::bhalf_t;
-    using F32  = float;
-    using F64  = double;
+    using F32 = float;
+    using F64 = double;
 
-    auto profile =
-        [&](auto a_layout, auto b_layout, auto cde_layout, auto type, auto compute_type) {
-            using ALayout   = decltype(a_layout);
-            using BLayout   = decltype(b_layout);
-            using CDELayout = decltype(cde_layout);
+    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
+        using ALayout   = decltype(a_layout);
+        using BLayout   = decltype(b_layout);
+        using CDELayout = decltype(cde_layout);
 
-            using DataType        = decltype(type);
-            using ComputeDataType = decltype(compute_type);
+        using DataType = decltype(type);
 
-            if(default_strides)
-            {
-                assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
-                assign_default_strides(b_layout, StridesB, {N[0], N[1], K[0], K[1]});
-                assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
-                assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
-            }
-            bool pass = ck::profiler::profile_contraction_impl<ALayout,
-                                                               BLayout,
-                                                               CDELayout,
-                                                               DataType,
-                                                               ComputeDataType,
-                                                               ck::Tuple<DataType>,
-                                                               Bilinear>(do_verification,
-                                                                         init_method,
-                                                                         do_log,
-                                                                         time_kernel,
-                                                                         Bilinear{alpha, beta},
-                                                                         M,
-                                                                         N,
-                                                                         K,
-                                                                         StridesA,
-                                                                         StridesB,
-                                                                         StridesE,
-                                                                         StridesD);
-
-            return pass;
-        };
-
-    auto run_profile_for_datatype = [&](auto type, auto compute_type) {
-        if(layout == ContractionMatrixLayout::MK_KN_MN_MN)
+        if(default_strides)
         {
-            return profile(Row{}, Row{}, Row{}, type, compute_type);
+            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
         }
-        else if(layout == ContractionMatrixLayout::MK_NK_MN_MN)
-        {
-            return profile(Row{}, Col{}, Row{}, type, compute_type);
-        }
-        else if(layout == ContractionMatrixLayout::KM_KN_MN_MN)
-        {
-            return profile(Col{}, Row{}, Row{}, type, compute_type);
-        }
-        else if(layout == ContractionMatrixLayout::KM_NK_MN_MN)
-        {
-            return profile(Col{}, Col{}, Row{}, type, compute_type);
-        }
-        return false;
+        bool pass = ck::profiler::profile_contraction_impl<ALayout,
+                                                           BLayout,
+                                                           CDELayout,
+                                                           DataType,
+                                                           ck::Tuple<DataType>,
+                                                           Bilinear>(do_verification,
+                                                                     init_method,
+                                                                     do_log,
+                                                                     time_kernel,
+                                                                     Bilinear{alpha, beta},
+                                                                     M,
+                                                                     N,
+                                                                     K,
+                                                                     StridesA,
+                                                                     StridesB,
+                                                                     StridesE,
+                                                                     StridesD);
+
+        return pass;
     };
 
-    if(data_type == ContractionDataType::F32_F32_F32_F32)
+    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+       layout == ContractionMatrixLayout::MK_KN_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(F32{}, F32{});
-        }
-        else if(compute_data_type == ContractionComputeDataType::F16)
-        {
-            return run_profile_for_datatype(F32{}, F16{});
-        }
-        else if(compute_data_type == ContractionComputeDataType::BF16)
-        {
-            return run_profile_for_datatype(F32{}, BF16{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Row{}, Row{}, Row{}, F32{});
     }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64)
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F64)
-        {
-            return run_profile_for_datatype(F64{}, F64{});
-        }
-        else if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(F64{}, F32{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Row{}, Col{}, Row{}, F32{});
     }
-    else if(data_type == ContractionDataType::F16_F16_F16_F16)
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(F16{}, F32{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Col{}, Row{}, Row{}, F32{});
     }
-    else if(data_type == ContractionDataType::BF16_BF16_BF16_BF16)
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(BF16{}, F32{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Col{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F64{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
     }
-    return 1;
 }
 
 REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_bilinear);
diff --git a/profiler/src/profile_contraction_scale.cpp b/profiler/src/profile_contraction_scale.cpp
index ca9c199988..6784b916f6 100644
--- a/profiler/src/profile_contraction_scale.cpp
+++ b/profiler/src/profile_contraction_scale.cpp
@@ -17,9 +17,8 @@
 static void print_helper_msg()
 {
     std::cout << "arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n"
-              << "arg2: data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
-              << "arg3: compute data type (0: fp32; 1: f64; 2: f16; 3: bf16)\n"
-              << "arg4: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
+              << "arg2: data type (0: fp32; 1: f64)\n"
+              << "arg3: matrix layout (0: A[m0, m1, k0, k1] * B[k0, k1, n0, n1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     1: A[m0, m1, k0, k1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
@@ -27,40 +26,39 @@ static void print_helper_msg()
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1];\n"
               << "                     3: A[k0, k1, m0, m1] * B[n0, n1, k0, k1] + "
                  "D[m0, m1, n0, n1] = E[m0, m1, n0, n1])\n"
-              << "arg5: verification (0: no; 1: yes)\n"
-              << "arg6: initialization (0: no init; 1: integer value; 2: decimal "
+              << "arg4: verification (0: no; 1: yes)\n"
+              << "arg5: initialization (0: no init; 1: integer value; 2: decimal "
               << "value)\n"
-              << "arg7: print tensor value (0: no; 1: yes)\n"
-              << "arg8: time kernel (0: no, 1: yes)\n"
-              << "arg9: alpha\n"
-              << "arg10 to 15: M0, M1, N0, N1, K0, K1\n"
-              << "arg16 to 31: Strides for A, B, D and E (skip for default)\n"
+              << "arg6: print tensor value (0: no; 1: yes)\n"
+              << "arg7: time kernel (0: no, 1: yes)\n"
+              << "arg8: alpha\n"
+              << "arg9 to 14: M0, M1, N0, N1, K0, K1\n"
+              << "arg15 to 30: Strides for A, B, D and E (skip for default)\n"
               << std::endl;
 }
 
 int profile_contraction_scale(int argc, char* argv[])
 {
-    const bool default_strides = argc == 16;
+    const bool default_strides = argc == 15;
 
-    if(argc != 32 && argc != 16)
+    if(argc != 31 && argc != 15)
     {
         print_helper_msg();
         exit(1);
     }
 
     const auto data_type          = static_cast<ContractionDataType>(std::stoi(argv[2]));
-    const auto compute_data_type  = static_cast<ContractionComputeDataType>(std::stoi(argv[3]));
-    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[4]));
-    const bool do_verification    = std::stoi(argv[5]);
-    const ck::index_t init_method = std::stoi(argv[6]);
-    const bool do_log             = std::stoi(argv[7]);
-    const bool time_kernel        = std::stoi(argv[8]);
-    const float alpha             = std::stof(argv[9]);
+    const auto layout             = static_cast<ContractionMatrixLayout>(std::stoi(argv[3]));
+    const bool do_verification    = std::stoi(argv[4]);
+    const ck::index_t init_method = std::stoi(argv[5]);
+    const bool do_log             = std::stoi(argv[6]);
+    const bool time_kernel        = std::stoi(argv[7]);
+    const float alpha             = std::stof(argv[8]);
 
     std::vector<ck::index_t> M;
     std::vector<ck::index_t> N;
     std::vector<ck::index_t> K;
-    const ck::index_t dims_arg_num = 10;
+    const ck::index_t dims_arg_num = 9;
     collect_index_params(argv, M, dims_arg_num, 2);
     collect_index_params(argv, N, dims_arg_num + 2, 2);
     collect_index_params(argv, K, dims_arg_num + 4, 2);
@@ -77,131 +75,88 @@ int profile_contraction_scale(int argc, char* argv[])
         collect_index_params(argv, StridesD, dims_arg_num + 18, 4);
     }
 
-    using F16  = ck::half_t;
-    using BF16 = ck::bhalf_t;
-    using F32  = float;
-    using F64  = double;
+    using F32 = float;
+    using F64 = double;
 
-    auto profile =
-        [&](auto a_layout, auto b_layout, auto cde_layout, auto type, auto compute_type) {
-            using ALayout   = decltype(a_layout);
-            using BLayout   = decltype(b_layout);
-            using CDELayout = decltype(cde_layout);
+    auto profile = [&](auto a_layout, auto b_layout, auto cde_layout, auto type) {
+        using ALayout   = decltype(a_layout);
+        using BLayout   = decltype(b_layout);
+        using CDELayout = decltype(cde_layout);
 
-            using DataType        = decltype(type);
-            using ComputeDataType = decltype(compute_type);
+        using DataType = decltype(type);
 
-            if(default_strides)
-            {
-                assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
-                assign_default_strides(b_layout, StridesB, {N[0], N[1], K[0], K[1]});
-                assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
-                assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
-            }
-
-            bool pass = ck::profiler::profile_contraction_impl<ALayout,
-                                                               BLayout,
-                                                               CDELayout,
-                                                               DataType,
-                                                               ComputeDataType,
-                                                               ck::Tuple<>,
-                                                               Scale>(do_verification,
-                                                                      init_method,
-                                                                      do_log,
-                                                                      time_kernel,
-                                                                      Scale{alpha},
-                                                                      M,
-                                                                      N,
-                                                                      K,
-                                                                      StridesA,
-                                                                      StridesB,
-                                                                      StridesE,
-                                                                      StridesD);
-
-            return pass;
-        };
-
-    auto run_profile_for_datatype = [&](auto type, auto compute_type) {
-        if(layout == ContractionMatrixLayout::MK_KN_MN_MN)
+        if(default_strides)
         {
-            return profile(Row{}, Row{}, Row{}, type, compute_type);
+            assign_default_strides(a_layout, StridesA, {M[0], M[1], K[0], K[1]});
+            assign_default_strides(b_layout, StridesB, {K[0], K[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesE, {M[0], M[1], N[0], N[1]});
+            assign_default_strides(cde_layout, StridesD, {M[0], M[1], N[0], N[1]});
         }
-        else if(layout == ContractionMatrixLayout::MK_NK_MN_MN)
-        {
-            return profile(Row{}, Col{}, Row{}, type, compute_type);
-        }
-        else if(layout == ContractionMatrixLayout::KM_KN_MN_MN)
-        {
-            return profile(Col{}, Row{}, Row{}, type, compute_type);
-        }
-        else if(layout == ContractionMatrixLayout::KM_NK_MN_MN)
-        {
-            return profile(Col{}, Col{}, Row{}, type, compute_type);
-        }
-        return false;
+
+        bool pass = ck::profiler::
+            profile_contraction_impl<ALayout, BLayout, CDELayout, DataType, ck::Tuple<>, Scale>(
+                do_verification,
+                init_method,
+                do_log,
+                time_kernel,
+                Scale{alpha},
+                M,
+                N,
+                K,
+                StridesA,
+                StridesB,
+                StridesE,
+                StridesD);
+
+        return pass;
     };
 
-    if(data_type == ContractionDataType::F32_F32_F32_F32)
+    if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+       layout == ContractionMatrixLayout::MK_KN_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(F32{}, F32{});
-        }
-        else if(compute_data_type == ContractionComputeDataType::F16)
-        {
-            return run_profile_for_datatype(F32{}, F16{});
-        }
-        else if(compute_data_type == ContractionComputeDataType::BF16)
-        {
-            return run_profile_for_datatype(F32{}, BF16{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Row{}, Row{}, Row{}, F32{});
     }
-    else if(data_type == ContractionDataType::F64_F64_F64_F64)
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F64)
-        {
-            return run_profile_for_datatype(F64{}, F64{});
-        }
-        else if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(F64{}, F32{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Row{}, Col{}, Row{}, F32{});
     }
-    else if(data_type == ContractionDataType::F16_F16_F16_F16)
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(F16{}, F32{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Col{}, Row{}, Row{}, F32{});
     }
-    else if(data_type == ContractionDataType::BF16_BF16_BF16_BF16)
+    else if(data_type == ContractionDataType::F32_F32_F32_F32 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
     {
-        if(compute_data_type == ContractionComputeDataType::F32)
-        {
-            return run_profile_for_datatype(BF16{}, F32{});
-        }
-        else
-        {
-            std::cout << "Incorrect combination of data type and compute data type." << std::endl;
-            return 1;
-        }
+        return profile(Col{}, Col{}, Row{}, F32{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_KN_MN_MN)
+    {
+        return profile(Row{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::MK_NK_MN_MN)
+    {
+        return profile(Row{}, Col{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_KN_MN_MN)
+    {
+        return profile(Col{}, Row{}, Row{}, F64{});
+    }
+    else if(data_type == ContractionDataType::F64_F64_F64_F64 &&
+            layout == ContractionMatrixLayout::KM_NK_MN_MN)
+    {
+        return profile(Col{}, Col{}, Row{}, F64{});
+    }
+    else
+    {
+        std::cout << "this data_type & layout is not implemented" << std::endl;
+
+        return 1;
     }
-    return 1;
 }
 
 REGISTER_PROFILER_OPERATION(OP_NAME, OP_DESC, profile_contraction_scale);
diff --git a/test/contraction/test_contraction.cpp b/test/contraction/test_contraction.cpp
index 958d5be38c..c86b849235 100644
--- a/test/contraction/test_contraction.cpp
+++ b/test/contraction/test_contraction.cpp
@@ -10,12 +10,9 @@
 #include <gtest/gtest.h>
 
 #include "profiler/profile_contraction_impl.hpp"
-#include "profiler/profile_contraction_utils.hpp"
 
-using F16  = ck::half_t;
-using BF16 = ck::bhalf_t;
-using F32  = float;
-using F64  = double;
+using F32 = float;
+using F64 = double;
 
 using Row = ck::tensor_layout::gemm::RowMajor;
 using Col = ck::tensor_layout::gemm::ColumnMajor;
@@ -23,49 +20,49 @@ using Col = ck::tensor_layout::gemm::ColumnMajor;
 using Bilinear = ck::tensor_operation::element_wise::Bilinear;
 using Scale    = ck::tensor_operation::element_wise::Scale;
 
-struct Dimensions
+struct MemoryParams
 {
     std::vector<ck::index_t> M;
     std::vector<ck::index_t> N;
     std::vector<ck::index_t> K;
+    std::vector<ck::index_t> StridesA;
+    std::vector<ck::index_t> StridesB;
+    std::vector<ck::index_t> StridesC;
+    std::vector<ck::index_t> StridesD;
 };
 
 template <typename Tuple>
 class TestContraction : public ::testing::Test
 {
     protected:
-    using ALayout         = std::tuple_element_t<0, Tuple>;
-    using BLayout         = std::tuple_element_t<1, Tuple>;
-    using CDLayout        = std::tuple_element_t<2, Tuple>;
-    using DataType        = std::tuple_element_t<3, Tuple>;
-    using DTupleDataType  = std::tuple_element_t<4, Tuple>;
-    using ComputeDataType = std::tuple_element_t<5, Tuple>;
-    using CDElementOp     = std::tuple_element_t<6, Tuple>;
+    using ALayout        = std::tuple_element_t<0, Tuple>;
+    using BLayout        = std::tuple_element_t<1, Tuple>;
+    using CDLayout       = std::tuple_element_t<2, Tuple>;
+    using DataType       = std::tuple_element_t<3, Tuple>;
+    using DTupleDataType = std::tuple_element_t<4, Tuple>;
+    using CDElementOp    = std::tuple_element_t<5, Tuple>;
 
-    std::vector<Dimensions> dimension_list = {{{32, 32}, {32, 32}, {32, 32}},
-                                              {{16, 16}, {32, 32}, {16, 16}}};
+    std::vector<MemoryParams> list_of_memory_params = {{{32, 32},
+                                                        {32, 32},
+                                                        {32, 32},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1},
+                                                        {32768, 1024, 32, 1}},
+                                                       {{16, 16},
+                                                        {32, 32},
+                                                        {16, 16},
+                                                        {4096, 256, 16, 1},
+                                                        {16, 1, 8192, 256},
+                                                        {16384, 1024, 32, 1},
+                                                        {16384, 1024, 32, 1}}};
 
-    std::vector<ck::index_t> init_methods = {1, 2};
+    std::vector<ck::index_t> init_methods = {0, 1, 2};
     std::unique_ptr<CDElementOp> p_cd_element_op;
-
     void Run()
     {
-        for(auto& dimension_params : dimension_list)
+        for(auto& memory_params : list_of_memory_params)
         {
-            std::vector<ck::index_t> StridesA;
-            std::vector<ck::index_t> StridesB;
-            std::vector<ck::index_t> StridesC;
-            std::vector<ck::index_t> StridesD;
-
-            const auto& M = dimension_params.M;
-            const auto& N = dimension_params.N;
-            const auto& K = dimension_params.K;
-
-            assign_default_strides(ALayout{}, StridesA, {M[0], M[1], K[0], K[1]});
-            assign_default_strides(BLayout{}, StridesB, {N[0], N[1], K[0], K[1]});
-            assign_default_strides(CDLayout{}, StridesC, {M[0], M[1], N[0], N[1]});
-            assign_default_strides(CDLayout{}, StridesD, {M[0], M[1], N[0], N[1]});
-
             for(const ck::index_t init_method : init_methods)
             {
                 bool pass =
@@ -73,20 +70,19 @@ class TestContraction : public ::testing::Test
                                                            BLayout,
                                                            CDLayout,
                                                            DataType,
-                                                           ComputeDataType,
                                                            DTupleDataType,
                                                            CDElementOp>(true /*do_verification*/,
                                                                         init_method,
                                                                         false /*do_logs*/,
                                                                         false /*time_kernel*/,
                                                                         *p_cd_element_op,
-                                                                        dimension_params.M,
-                                                                        dimension_params.N,
-                                                                        dimension_params.K,
-                                                                        StridesA,
-                                                                        StridesB,
-                                                                        StridesC,
-                                                                        StridesD);
+                                                                        memory_params.M,
+                                                                        memory_params.N,
+                                                                        memory_params.K,
+                                                                        memory_params.StridesA,
+                                                                        memory_params.StridesB,
+                                                                        memory_params.StridesC,
+                                                                        memory_params.StridesD);
                 EXPECT_TRUE(pass);
             }
         }
@@ -103,18 +99,24 @@ class TestContractionBilinear : public TestContraction<Tuple>
 {
 };
 
-#define ALL_LAYOUT_COMBINATIONS(dt, tuple_dt, compute_dt, op)    \
-    std::tuple<Row, Row, Row, dt, tuple_dt, compute_dt, op>,     \
-        std::tuple<Row, Col, Row, dt, tuple_dt, compute_dt, op>, \
-        std::tuple<Col, Row, Row, dt, tuple_dt, compute_dt, op>, \
-        std::tuple<Col, Col, Row, dt, tuple_dt, compute_dt, op>
-
 using BilinearKernelTypes =
-    ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F32, Bilinear),
-                     ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F64, Bilinear)>;
+    ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Row, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Col, Row, F32, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Row, Col, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Row, Row, F64, ck::Tuple<F32>, Bilinear>,
+                     std::tuple<Col, Col, Row, F64, ck::Tuple<F32>, Bilinear>>;
 
-using ScaleKernelTypes = ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F32, Scale),
-                                          ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F64, Scale)>;
+using ScaleKernelTypes = ::testing::Types<std::tuple<Row, Row, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Col, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Row, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Col, Row, F32, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Row, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Row, Col, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Row, Row, F64, ck::Tuple<>, Scale>,
+                                          std::tuple<Col, Col, Row, F64, ck::Tuple<>, Scale>>;
 
 TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes);
 TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
@@ -134,46 +136,3 @@ TYPED_TEST(TestContractionScale, scale)
     this->p_cd_element_op = std::make_unique<Scale>(0.5f);
     this->Run();
 }
-
-template <typename Tuple>
-class TestContractionScaleMixedPrecision : public TestContraction<Tuple>
-{
-};
-
-template <typename Tuple>
-class TestContractionBilinearMixedPrecision : public TestContraction<Tuple>
-{
-};
-
-using BilinearKernelTypesMixedPrecision =
-    ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F16, Bilinear),
-                     ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, BF16, Bilinear),
-                     ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F32, Bilinear),
-                     ALL_LAYOUT_COMBINATIONS(F16, ck::Tuple<F16>, F32, Bilinear),
-                     ALL_LAYOUT_COMBINATIONS(BF16, ck::Tuple<BF16>, F32, Bilinear)>;
-
-using ScaleKernelTypesMixedPrecision =
-    ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F16, Scale),
-                     ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, BF16, Scale),
-                     ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F32, Scale),
-                     ALL_LAYOUT_COMBINATIONS(F16, ck::Tuple<>, F32, Scale),
-                     ALL_LAYOUT_COMBINATIONS(BF16, ck::Tuple<>, F32, Scale)>;
-
-TYPED_TEST_SUITE(TestContractionBilinearMixedPrecision, BilinearKernelTypesMixedPrecision);
-TYPED_TEST_SUITE(TestContractionScaleMixedPrecision, ScaleKernelTypesMixedPrecision);
-
-TYPED_TEST(TestContractionBilinearMixedPrecision, bilinear)
-{
-    this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
-    this->Run();
-    this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
-    this->Run();
-}
-
-TYPED_TEST(TestContractionScaleMixedPrecision, scale)
-{
-    this->p_cd_element_op = std::make_unique<Scale>(1.f);
-    this->Run();
-    this->p_cd_element_op = std::make_unique<Scale>(0.5f);
-    this->Run();
-}
diff --git a/test/contraction/test_contraction_interface.cpp b/test/contraction/test_contraction_interface.cpp
index e0f547445e..12a307f5da 100644
--- a/test/contraction/test_contraction_interface.cpp
+++ b/test/contraction/test_contraction_interface.cpp
@@ -34,11 +34,11 @@ class ContractionInstanceWrapper
     static constexpr ck::index_t NumDim = 2;
     // clang-format off
     using ContractionDeviceInstance = ck::tensor_operation::device::
-        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|         DsData| EData| Compute|           A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer|             ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer|              BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|                  CBlockTransfer|
-        //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|    Data| Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
-        //#####################################|        |        |        |      |      |        |         |               |      |    Type|   Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
-        //#####################################|        |        |        |      |      |        |         |               |      |        |            |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
-        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,     F32,        Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
+        //#####################################| NumDimM| NumDimN| NumDimK| AData| BData| AccData| CShuffle|         DsData| EData|            A|           B|          CDE|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer|             ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer|              BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|                  CBlockTransfer|
+        //#####################################|        |        |        |  Type|  Type|    Type| DataType|           Type|  Type|  Elementwise| Elementwise|  Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|               SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl|                 ScalarPerVector|
+        //#####################################|        |        |        |      |      |        |         |               |      |    Operation|   Operation|    Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |                           |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|                   _NWaveNPerXdl|
+        //#####################################|        |        |        |      |      |        |         |               |      |             |            |             |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |                           |               |               |          |                |               |               |                           |               |               |          |            |            |                             |                                |
+        DeviceContractionMultipleD_Xdl_CShuffle<  NumDim,  NumDim,  NumDim,   F32,   F32,     F32,      F32, ck::Tuple<F32>,   F32,         Pass,        Pass,     Bilinear,       GemmSpec,        1,   256,   256,   128,    16,   4,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, ABlockTransferSrcVectorDim,              4,              4,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>, BBlockTransferSrcVectorDim,              4,              4,         1,           1,           1,              S<1, 16, 1, 16>, CDEBlockTransferScalarPerVector>;
     // clang-format on
 
     bool isSupported(std::vector<ck::index_t>& ADims,

From 5913609168e6f691185f1b7b214e517d67dd5075 Mon Sep 17 00:00:00 2001
From: Lauren Wrubleski <Liam.Wrubleski@amd.com>
Date: Thu, 5 Oct 2023 15:58:58 -0600
Subject: [PATCH 163/194] Replace CMake `return` from later CMake (#970)

---
 example/CMakeLists.txt                                 | 10 +++++-----
 .../src/tensor_operation_instance/gpu/CMakeLists.txt   |  4 ++--
 test/CMakeLists.txt                                    |  8 ++++----
 3 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/example/CMakeLists.txt b/example/CMakeLists.txt
index d030c53453..7f8704f281 100644
--- a/example/CMakeLists.txt
+++ b/example/CMakeLists.txt
@@ -30,7 +30,7 @@ function(add_example_executable EXAMPLE_NAME FILE_NAME)
                 set(test 0)
                 break()
             elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
                 NOT(source MATCHES type OR source MATCHES type1))
                     #if filename contains a type which doesn't match any selected type, mark it for removal
                     set(test 1)
@@ -59,7 +59,7 @@ function(add_example_executable EXAMPLE_NAME FILE_NAME)
         set(result 0)
     endif()
     #message("add_example returns ${result}")
-    return(PROPAGATE result)
+    set(result ${result} PARENT_SCOPE)
 endfunction(add_example_executable EXAMPLE_NAME)
 
 function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
@@ -87,7 +87,7 @@ function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
                     set(test 0)
                     break()
                 elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
                   NOT(source MATCHES type OR source MATCHES type1))
                     #if filename contains a type which doesn't match any selected type, mark it for removal
                     set(test 1)
@@ -96,7 +96,7 @@ function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
         if(test EQUAL 1)
             message("removing example ${source} ")
             list(REMOVE_ITEM FILE_NAME "${source}")
-        endif()    
+        endif()
     endforeach()
     endif()
     foreach(source IN LISTS FILE_NAME)
@@ -114,7 +114,7 @@ function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
         set(result 0)
     endif()
     #message("add_example returns ${result}")
-    return(PROPAGATE result)
+    set(result ${result} PARENT_SCOPE)
 endfunction(add_example_executable_no_testing EXAMPLE_NAME)
 
 # add all example subdir
diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
index a0478c9f09..9c1f72eca3 100644
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
@@ -24,7 +24,7 @@ function(add_instance_library INSTANCE_NAME)
                         set(test 0)
                         break()
                 elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
                   NOT(source MATCHES type OR source MATCHES type1))
                         #if filename contains a type which doesn't match any selected type, mark it for removal
                         set(test 1)
@@ -51,7 +51,7 @@ function(add_instance_library INSTANCE_NAME)
         set(result 0)
     endif()
     #message("add_instance_library returns ${result}")
-    return(PROPAGATE result)
+    set(result ${result} PARENT_SCOPE)
 endfunction(add_instance_library INSTANCE_NAME)
 
 
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 07dd675afc..4b2c7bbf38 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -32,7 +32,7 @@ function(add_test_executable TEST_NAME)
                 set(test 0)
                 break()
             elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
                 NOT(source MATCHES type OR source MATCHES type1))
                     #if filename contains a type which doesn't match any selected type, mark it for removal
                     set(test 1)
@@ -61,7 +61,7 @@ function(add_test_executable TEST_NAME)
         set(result 0)
     endif()
     #message("add_test returns ${result}")
-    return(PROPAGATE result)
+    set(result ${result} PARENT_SCOPE)
 endfunction(add_test_executable TEST_NAME)
 
 include(GoogleTest)
@@ -91,7 +91,7 @@ function(add_gtest_executable TEST_NAME)
                 set(test 0)
                 break()
             elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND 
+                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
                 NOT(source MATCHES type OR source MATCHES type1))
                     #if filename contains a type which doesn't match any selected type, mark it for removal
                     set(test 1)
@@ -123,7 +123,7 @@ function(add_gtest_executable TEST_NAME)
         set(result 0)
     endif()
     #message("add_gtest returns ${result}")
-    return(PROPAGATE result)
+    set(result ${result} PARENT_SCOPE)
 endfunction(add_gtest_executable TEST_NAME)
 
 add_subdirectory(magic_number_division)

From ac9595a9f118a023e248eaffcfa5c324f36fd081 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 10 Oct 2023 10:30:26 -0500
Subject: [PATCH 164/194] Fixed f8_gemm NaN (#975)

* workaround nan problem by changing output to fp16

* enable f8/bf8 gemm tests on MI200

* workaround f16 to f8 conversion

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/01_gemm/CMakeLists.txt                     | 14 +++++---------
 example/01_gemm/gemm_xdl_fp8.cpp                   |  6 +++---
 example/01_gemm/gemm_xdl_fp8_bf8.cpp               |  6 +++---
 .../gpu/element/unary_element_wise_operation.hpp   |  3 +--
 include/ck/utility/type_convert.hpp                |  8 ++++++--
 5 files changed, 18 insertions(+), 19 deletions(-)

diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index e0124e57f7..703d0e3834 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -66,21 +66,17 @@ endif()
 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
 
 
-if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-  add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
-  if(result EQUAL 0)
+add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
-  endif()
 endif()
 
-if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-  add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
-  if(result EQUAL 0)
+add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
+if(result EQUAL 0)
     add_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
-  endif()
 endif()
 
 add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp)
 if(result EQUAL 0)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
+    add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
 endif()
diff --git a/example/01_gemm/gemm_xdl_fp8.cpp b/example/01_gemm/gemm_xdl_fp8.cpp
index 1015926777..2d4df3fc13 100644
--- a/example/01_gemm/gemm_xdl_fp8.cpp
+++ b/example/01_gemm/gemm_xdl_fp8.cpp
@@ -7,9 +7,9 @@
 
 using ADataType        = ck::f8_t;
 using BDataType        = ck::f8_t;
-using CDataType        = ck::f8_t;
+using CDataType        = ck::half_t;
 using AccDataType      = float;
-using CShuffleDataType = ck::f8_t;
+using CShuffleDataType = float;
 
 using ALayout = Row;
 using BLayout = Col;
@@ -27,7 +27,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
 // ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
 // ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
 // ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16>;
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,               8>;
 // clang-format on
 
 using ReferenceGemmInstance = ck::tensor_operation::host::
diff --git a/example/01_gemm/gemm_xdl_fp8_bf8.cpp b/example/01_gemm/gemm_xdl_fp8_bf8.cpp
index 0d69b7a90f..b54df8ff3d 100644
--- a/example/01_gemm/gemm_xdl_fp8_bf8.cpp
+++ b/example/01_gemm/gemm_xdl_fp8_bf8.cpp
@@ -7,9 +7,9 @@
 
 using ADataType        = ck::f8_t;
 using BDataType        = ck::bf8_t;
-using CDataType        = ck::f8_t;
+using CDataType        = ck::half_t;
 using AccDataType      = float;
-using CShuffleDataType = ck::f8_t;
+using CShuffleDataType = float;
 
 using ALayout = Row;
 using BLayout = Col;
@@ -31,7 +31,7 @@ using DeviceGemmInstance = ck::tensor_operation::device::DeviceGemm_Xdl_CShuffle
 // ######|        |        |        |      Type|      Type|      Type|        Type|         DataType| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
 // ######|        |        |        |          |          |          |            |                 |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
 // ######|        |        |        |          |          |          |            |                 |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,              16,  LoopSched, PipelineVer, ComputeTypeA, ComputeTypeB>;
+         < ALayout, BLayout, CLayout, ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp,  CElementOp,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 64, 1, 4>,               8,  LoopSched, PipelineVer, ComputeTypeA, ComputeTypeB>;
 // clang-format on
 
 using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 0b0e731c60..05cd5c5b8d 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -173,8 +173,7 @@ struct PassThrough
     template <>
     __host__ __device__ void operator()<bf8_t, half_t>(bf8_t& y, const half_t& x) const
     {
-        // to-do: fix half_t to bf8_t convert
-        y = ck::type_convert<bf8_t>(ck::type_convert<float>(x));
+        y = ck::type_convert<bf8_t>(x);
     }
 #endif
 };
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 505de73ae2..8b70a6bfb4 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -344,7 +344,7 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return f8_convert_sr<f8_t>(type_convert<float>(x));
-#else
+#elif 0
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
@@ -353,6 +353,8 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#else
+    return type_convert<f8_t>(type_convert<float>(x));
 #endif
 }
 #endif
@@ -393,7 +395,7 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return f8_convert_sr<f8_t>(type_convert<float>(x));
-#else
+#elif 0
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
@@ -403,6 +405,8 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#else
+    return type_convert<bf8_t>(type_convert<float>(x));
 #endif
 }
 #endif

From 98c80714758b022b59c0130b1d946e23ec96cfb9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 10 Oct 2023 21:48:07 +0200
Subject: [PATCH 165/194] Fix MNKPadding in gridwise_gemm_xdlops_v2r3 (#981)

---
 .../gpu/grid/gridwise_gemm_xdlops_v2r3.hpp    |   3 +-
 test/batched_gemm/CMakeLists.txt              |  18 +-
 test/batched_gemm/batched_gemm_bf16.cpp       | 114 -----------
 test/batched_gemm/batched_gemm_fp16.cpp       | 114 -----------
 test/batched_gemm/batched_gemm_fp32.cpp       | 114 -----------
 test/batched_gemm/batched_gemm_int8.cpp       | 114 -----------
 test/batched_gemm/test_batched_gemm.cpp       | 185 ++++++++++++++++++
 7 files changed, 189 insertions(+), 473 deletions(-)
 delete mode 100644 test/batched_gemm/batched_gemm_bf16.cpp
 delete mode 100644 test/batched_gemm/batched_gemm_fp16.cpp
 delete mode 100644 test/batched_gemm/batched_gemm_fp32.cpp
 delete mode 100644 test/batched_gemm/batched_gemm_int8.cpp
 create mode 100644 test/batched_gemm/test_batched_gemm.cpp

diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
index 54760b0556..2c42ac9345 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp
@@ -945,7 +945,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3_ext
             }
         }();
 
-        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding)
+        if constexpr(GemmSpec == tensor_operation::device::GemmSpecialization::MNPadding ||
+                     GemmSpec == tensor_operation::device::GemmSpecialization::MNKPadding)
         {
             return transform_tensor_descriptor(c_grid_desc_m_n,
                                                make_tuple(make_right_pad_transform(M, MPad - M),
diff --git a/test/batched_gemm/CMakeLists.txt b/test/batched_gemm/CMakeLists.txt
index d169c135ca..9482821b68 100644
--- a/test/batched_gemm/CMakeLists.txt
+++ b/test/batched_gemm/CMakeLists.txt
@@ -2,22 +2,8 @@ list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_test_executable(test_batched_gemm_fp16 batched_gemm_fp16.cpp)
-   if(result EQUAL 0)
-      target_link_libraries(test_batched_gemm_fp16 PRIVATE utility device_batched_gemm_instance)
-   endif()
-   add_test_executable(test_batched_gemm_fp32 batched_gemm_fp32.cpp)
-   if(result EQUAL 0)
-      target_link_libraries(test_batched_gemm_fp32 PRIVATE utility device_batched_gemm_instance)
-   endif()
-   add_test_executable(test_batched_gemm_bf16 batched_gemm_bf16.cpp)
-   if(result EQUAL 0)
-      target_link_libraries(test_batched_gemm_bf16 PRIVATE utility device_batched_gemm_instance)
-   endif()
-   add_test_executable(test_batched_gemm_int8 batched_gemm_int8.cpp)
-   if(result EQUAL 0)
-      target_link_libraries(test_batched_gemm_int8 PRIVATE utility device_batched_gemm_instance)
-   endif()
+   add_gtest_executable(test_batched_gemm test_batched_gemm.cpp)
+   target_link_libraries(test_batched_gemm PRIVATE utility device_batched_gemm_instance)
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/batched_gemm/batched_gemm_bf16.cpp b/test/batched_gemm/batched_gemm_bf16.cpp
deleted file mode 100644
index 5d12a1e956..0000000000
--- a/test/batched_gemm/batched_gemm_bf16.cpp
+++ /dev/null
@@ -1,114 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-
-#include "profiler/profile_batched_gemm_impl.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
-
-namespace {
-using ADataType = ck::bhalf_t;
-using BDataType = ck::bhalf_t;
-using CDataType = ck::bhalf_t;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-} // namespace
-
-int main()
-{
-    int M          = 256;
-    int N          = 256;
-    int K          = 128;
-    int BatchCount = 3;
-
-    bool pass = true;
-
-    using namespace ck::tensor_operation::device;
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
-
-    std::cout << "test BatchedGEMM bf16: " << (pass ? "Pass" : "Fail") << std::endl;
-    return pass ? 0 : 1;
-}
diff --git a/test/batched_gemm/batched_gemm_fp16.cpp b/test/batched_gemm/batched_gemm_fp16.cpp
deleted file mode 100644
index a2b61d951a..0000000000
--- a/test/batched_gemm/batched_gemm_fp16.cpp
+++ /dev/null
@@ -1,114 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-
-#include "profiler/profile_batched_gemm_impl.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
-
-namespace {
-using ADataType = ck::half_t;
-using BDataType = ck::half_t;
-using CDataType = ck::half_t;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-} // namespace
-
-int main()
-{
-    int M          = 512;
-    int N          = 256;
-    int K          = 128;
-    int BatchCount = 3;
-
-    bool pass = true;
-
-    using namespace ck::tensor_operation::device;
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
-
-    std::cout << "test BatchedGEMM fp16: " << (pass ? "Pass" : "Fail") << std::endl;
-    return pass ? 0 : 1;
-}
diff --git a/test/batched_gemm/batched_gemm_fp32.cpp b/test/batched_gemm/batched_gemm_fp32.cpp
deleted file mode 100644
index 2b18d166e6..0000000000
--- a/test/batched_gemm/batched_gemm_fp32.cpp
+++ /dev/null
@@ -1,114 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-
-#include "profiler/profile_batched_gemm_impl.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
-
-namespace {
-using ADataType = float;
-using BDataType = float;
-using CDataType = float;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-} // namespace
-
-int main()
-{
-    int M          = 256;
-    int N          = 256;
-    int K          = 128;
-    int BatchCount = 3;
-
-    bool pass = true;
-
-    using namespace ck::tensor_operation::device;
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
-
-    std::cout << "test BatchedGEMM fp32: " << (pass ? "Pass" : "Fail") << std::endl;
-    return pass ? 0 : 1;
-}
diff --git a/test/batched_gemm/batched_gemm_int8.cpp b/test/batched_gemm/batched_gemm_int8.cpp
deleted file mode 100644
index f607eaa84b..0000000000
--- a/test/batched_gemm/batched_gemm_int8.cpp
+++ /dev/null
@@ -1,114 +0,0 @@
-// SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
-
-#include <iostream>
-
-#include "profiler/profile_batched_gemm_impl.hpp"
-
-#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
-
-namespace {
-using ADataType = int8_t;
-using BDataType = int8_t;
-using CDataType = int8_t;
-
-using Row = ck::tensor_layout::gemm::RowMajor;
-using Col = ck::tensor_layout::gemm::ColumnMajor;
-
-using PassThrough = ck::tensor_operation::element_wise::PassThrough;
-} // namespace
-
-int main()
-{
-    int M          = 256;
-    int N          = 256;
-    int K          = 128;
-    int BatchCount = 3;
-
-    bool pass = true;
-
-    using namespace ck::tensor_operation::device;
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Row,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Row,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Row,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Row,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
-
-    pass = pass && ck::profiler::profile_batched_gemm_impl<ADataType,
-                                                           BDataType,
-                                                           CDataType,
-                                                           Col,
-                                                           Col,
-                                                           Row,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           PassThrough,
-                                                           DeviceBatchedGemm<Col,
-                                                                             Col,
-                                                                             Row,
-                                                                             ADataType,
-                                                                             BDataType,
-                                                                             CDataType,
-                                                                             PassThrough,
-                                                                             PassThrough,
-                                                                             PassThrough>>(
-                       true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
-
-    std::cout << "test BatchedGEMM int8: " << (pass ? "Pass" : "Fail") << std::endl;
-    return pass ? 0 : 1;
-}
diff --git a/test/batched_gemm/test_batched_gemm.cpp b/test/batched_gemm/test_batched_gemm.cpp
new file mode 100644
index 0000000000..f9bb626ce5
--- /dev/null
+++ b/test/batched_gemm/test_batched_gemm.cpp
@@ -0,0 +1,185 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include <gtest/gtest.h>
+
+#include "profiler/profile_batched_gemm_impl.hpp"
+
+#include "ck/library/tensor_operation_instance/gpu/batched_gemm.hpp"
+
+struct GemmParams
+{
+    ck::index_t M;
+    ck::index_t N;
+    ck::index_t K;
+    ck::index_t BatchCount;
+};
+
+class TestBatchedGemm : public ::testing::Test
+{
+    protected:
+    using Row = ck::tensor_layout::gemm::RowMajor;
+    using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+    using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+    std::vector<GemmParams> params;
+
+    template <typename DataType>
+    void Run()
+    {
+        using namespace ck::tensor_operation::device;
+
+        bool pass = true;
+        for(auto& param : params)
+        {
+            const auto M          = param.M;
+            const auto N          = param.N;
+            const auto K          = param.K;
+            const auto BatchCount = param.BatchCount;
+
+            pass =
+                pass && ck::profiler::profile_batched_gemm_impl<DataType,
+                                                                DataType,
+                                                                DataType,
+                                                                Row,
+                                                                Row,
+                                                                Row,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                DeviceBatchedGemm<Row,
+                                                                                  Row,
+                                                                                  Row,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  PassThrough,
+                                                                                  PassThrough,
+                                                                                  PassThrough>>(
+                            true, 1, false, 1, M, N, K, K, N, N, M * K, K * N, M * N, BatchCount);
+
+            pass =
+                pass && ck::profiler::profile_batched_gemm_impl<DataType,
+                                                                DataType,
+                                                                DataType,
+                                                                Row,
+                                                                Col,
+                                                                Row,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                DeviceBatchedGemm<Row,
+                                                                                  Col,
+                                                                                  Row,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  PassThrough,
+                                                                                  PassThrough,
+                                                                                  PassThrough>>(
+                            true, 1, false, 1, M, N, K, K, K, N, M * K, K * N, M * N, BatchCount);
+
+            pass =
+                pass && ck::profiler::profile_batched_gemm_impl<DataType,
+                                                                DataType,
+                                                                DataType,
+                                                                Col,
+                                                                Row,
+                                                                Row,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                DeviceBatchedGemm<Col,
+                                                                                  Row,
+                                                                                  Row,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  PassThrough,
+                                                                                  PassThrough,
+                                                                                  PassThrough>>(
+                            true, 1, false, 1, M, N, K, M, N, N, M * K, K * N, M * N, BatchCount);
+
+            pass =
+                pass && ck::profiler::profile_batched_gemm_impl<DataType,
+                                                                DataType,
+                                                                DataType,
+                                                                Col,
+                                                                Col,
+                                                                Row,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                PassThrough,
+                                                                DeviceBatchedGemm<Col,
+                                                                                  Col,
+                                                                                  Row,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  DataType,
+                                                                                  PassThrough,
+                                                                                  PassThrough,
+                                                                                  PassThrough>>(
+                            true, 1, false, 1, M, N, K, M, K, N, M * K, K * N, M * N, BatchCount);
+        }
+        EXPECT_TRUE(pass);
+    }
+};
+
+#ifdef CK_ENABLE_INT8
+TEST_F(TestBatchedGemm, i8)
+{
+    this->params.push_back({64, 64, 64, 2});
+    this->params.push_back({64, 64, 64, 1});
+    this->params.push_back({60, 60, 60, 2});
+    this->params.push_back({68, 68, 68, 2});
+    this->params.push_back({40, 40, 40, 2});
+    this->params.push_back({256, 256, 128, 3});
+    this->template Run<int8_t>();
+}
+#endif
+
+#ifdef CK_ENABLE_BF16
+TEST_F(TestBatchedGemm, bf16)
+{
+    this->params.push_back({64, 64, 64, 2});
+    this->params.push_back({64, 64, 64, 1});
+    this->params.push_back({60, 60, 60, 2});
+    this->params.push_back({68, 68, 68, 2});
+    this->params.push_back({40, 40, 40, 2});
+    this->params.push_back({256, 256, 128, 3});
+    this->template Run<ck::bhalf_t>();
+}
+#endif
+
+#ifdef CK_ENABLE_FP16
+TEST_F(TestBatchedGemm, fp16)
+{
+    this->params.push_back({64, 64, 64, 2});
+    this->params.push_back({64, 64, 64, 1});
+    this->params.push_back({60, 60, 60, 2});
+    this->params.push_back({68, 68, 68, 2});
+    this->params.push_back({40, 40, 40, 2});
+    this->params.push_back({256, 256, 128, 3});
+    this->template Run<ck::half_t>();
+}
+#endif
+
+#ifdef CK_ENABLE_FP32
+TEST_F(TestBatchedGemm, fp32)
+{
+    this->params.push_back({64, 64, 64, 2});
+    this->params.push_back({64, 64, 64, 1});
+    this->params.push_back({60, 60, 60, 2});
+    this->params.push_back({68, 68, 68, 2});
+    this->params.push_back({40, 40, 40, 2});
+    this->params.push_back({256, 256, 128, 3});
+    this->template Run<float>();
+}
+#endif

From a4f72a314a85732ed67d5aa8d1088d207a7e0e61 Mon Sep 17 00:00:00 2001
From: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Date: Wed, 11 Oct 2023 05:21:15 +0200
Subject: [PATCH 166/194] Grouped Gemm with looping over the tiles. (#788)

* Introduce LocalBlockToCTileMap.

* Change the signature of CalculateBottomIndex() function which now does
not accept any argument. The B2C map which is already passed as an
argument to the kernel Run function is calculating block's local id
already outside at kernel entry point __global__ function.
The LocalB2C map stores as members local block ID.

* Use LocalBlockToCTile map in device ops.

* First draft of tile loop work distribution.

* Fix typo.

* Simplify kernel arguments.

Calculate descriptors & B2C maps on the device.

* Use looping kernel.

* Fix B2C constructor.

* Fix Navi21 errors.

* Calculate tile start/end in device kernel.

* Change Run API to accept user provided workspace buffer.

* Add new line at EOF.

* Move Gemm KernelArguments to device op interface.

* Remove unused code.

* Update API.

* Launch grid size which is min of occupancy vs tile count

* Get back to use constant memory for gemm descriptors.

* Remove unused code.

* Add default virtual method implementation.

* Update comments to conform with doxygen style.

* Fix doc style and unused parameters.

* Add thread cluster lengths to kernel name.

* Remove old splitk impl and replace it with tile looping one.

* Modify instances.

* set KPerBlock to 64
* maximize wherever possible vector load size.

* Fix instances cluster lengths.

* Change comment style.

* Use 128b store where possible in instances.

* Update test cases, since KPerBlock has doubled.

* Update output stream operator for Sequence.

* Add pipeline version to GroupedGEMM device op type string.

* Fix pipeline version type logging.

* Fix input tensors type after merge.

* Fix compiler error.

* Fix output stream operator for Pipeline version.

* Store using 128b.

* Set of instances with kpb 32/64

* Limit number of instances

* Remove commented out instances.

* Fix function name.

* Limit the number of instances.

Add pipline version to the regular instances

* Change thr cluster layout for reading B tensor.

* disabled failed instances

---------

Co-authored-by: Adam Osewski <aosewski@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../17_grouped_gemm_fastgelu/CMakeLists.txt   |   2 +-
 .../gpu/device/device_grouped_gemm_splitk.hpp |  74 ++-
 .../gpu/device/device_softmax.hpp             |  32 +-
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |   2 +-
 ..._conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp |  32 +-
 ...ice_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp |  32 +-
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |  32 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp | 580 +++++++++++-------
 .../gpu/device/impl/device_softmax_impl.hpp   |  36 +-
 .../gpu/grid/block_to_ctile_map.hpp           |  91 ++-
 .../grid/gridwise_gemm_pipeline_selector.hpp  |  18 +
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  74 ++-
 include/ck/utility/sequence.hpp               |  11 +
 .../cpu/reference_conv_fwd.hpp                |  42 +-
 ...gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp |  68 +-
 ...l_splitk_f16_f16_f16_mk_kn_mn_instance.cpp |  56 +-
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |  86 ++-
 ...l_splitk_f16_f16_f16_mk_nk_mn_instance.cpp |  42 +-
 ...16_f16_f16_mk_nk_mn_irregular_instance.cpp |  49 +-
 script/cmake-ck-dev.sh                        |   4 +-
 .../test_grouped_gemm_ut_cases.inc            |  22 +-
 21 files changed, 861 insertions(+), 524 deletions(-)

diff --git a/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt b/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
index 659e6769d8..06c581b706 100644
--- a/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
+++ b/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
@@ -1,2 +1,2 @@
 add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
-target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations)
\ No newline at end of file
+target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations)
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
index 06d180d30f..eb5447c8fa 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
@@ -8,6 +8,57 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
+///
+/// @brief      Structure representing single GEMM problem arguments.
+///
+///             The pointer to the vector of those structures is passed
+///             to the GroupedGEMM entry point kernel.
+///
+struct GroupedGemmKernelArguments
+{
+    __host__ __device__ GroupedGemmKernelArguments(const void* p_a_grid_,
+                                                   const void* p_b_grid_,
+                                                   void* p_c_grid_,
+                                                   index_t M_,
+                                                   index_t N_,
+                                                   index_t K_,
+                                                   index_t StrideA_,
+                                                   index_t StrideB_,
+                                                   index_t StrideC_)
+        : p_a_grid{p_a_grid_},
+          p_b_grid{p_b_grid_},
+          p_c_grid{p_c_grid_},
+          M{M_},
+          N{N_},
+          K{K_},
+          StrideA{StrideA_},
+          StrideB{StrideB_},
+          StrideC{StrideC_}
+    {
+    }
+
+    const void* p_a_grid;
+    const void* p_b_grid;
+    void* p_c_grid;
+    index_t M;
+    index_t N;
+    index_t K;
+    index_t StrideA;
+    index_t StrideB;
+    index_t StrideC;
+
+    void Print() const
+    {
+        std::cout << "arg {"
+                  << "M:" << M << ", "
+                  << "N:" << N << ", "
+                  << "K:" << K << ", "
+                  << "SA:" << StrideA << ", "
+                  << "SB:" << StrideB << ", "
+                  << "SC:" << StrideC << "}" << std::endl;
+    }
+};
+
 template <typename ALayout,
           typename BLayout,
           typename DsLayout,
@@ -31,7 +82,28 @@ struct DeviceGroupedGemmSplitK : public DeviceGroupedGemm<ALayout,
                                                           BElementwiseOperation,
                                                           CElementwiseOperation>
 {
-    virtual void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const = 0;
+    //----------------------------------------------------------------------------------------------
+    /// @brief      Sets the k batch size.
+    ///
+    /// @param      p_arg   Pointer to the Argument we're going to change.
+    /// @param[in]  kbatch  The kbatch value.
+    ///
+    virtual void SetKBatchSize([[maybe_unused]] BaseArgument* p_arg,
+                               [[maybe_unused]] index_t kbatch) const
+    {
+    }
+
+    //----------------------------------------------------------------------------------------------
+    /// @brief      Sets the device kernel arguments pointer.
+    ///
+    /// @param      p_arg              The pointer to the Argument we're going to update.
+    /// @param[in]  p_dev_kernel_args  The pointer to the device memory which contains kernel
+    ///                                arguments.
+    ///
+    virtual void SetDeviceKernelArgs([[maybe_unused]] BaseArgument* p_arg,
+                                     [[maybe_unused]] const void* p_dev_kernel_args) const
+    {
+    }
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/device/device_softmax.hpp b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
index 1902fd09ee..68c29a5340 100644
--- a/include/ck/tensor_operation/gpu/device/device_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
@@ -22,22 +22,22 @@ template <typename InDataType,
           index_t NumReduceDim>
 struct DeviceSoftmax : public BaseOperator
 {
-    //
-    // @brief      Makes a pointer to Argument class.
-    //
-    // @param[in]  inLengths           Input tensor extent(s) from high to low dimension
-    // @param[in]  inStrides           Input tensor stride(s) from high to low dimension
-    // @param[in]  reduceDims          The dimension(s) the normalization operation is applied
-    // @param[in]  alpha               double type value
-    // @param[in]  beta                double type value
-    // @param[in]  in_dev              Typeless const pointer in device memory storing the input
-    //                                 tensor
-    // @param      out_dev             Typeless pointer in device memory storing the output tensor
-    // @param[in]  in_elementwise_op   The input elementwise operation.
-    // @param[in]  acc_elementwise_op  The accumulation elementwise operation.
-    //
-    // @return     Unique pointer to the Argument class.
-    //
+    ///
+    /// @brief      Makes a pointer to Argument class.
+    ///
+    /// @param[in]  inLengths           Input tensor extent(s) from high to low dimension
+    /// @param[in]  inStrides           Input tensor stride(s) from high to low dimension
+    /// @param[in]  reduceDims          The dimension(s) the normalization operation is applied
+    /// @param[in]  alpha               double type value
+    /// @param[in]  beta                double type value
+    /// @param[in]  in_dev              Typeless const pointer in device memory storing the input
+    ///                                 tensor
+    /// @param      out_dev             Typeless pointer in device memory storing the output tensor
+    /// @param[in]  in_elementwise_op   The input elementwise operation.
+    /// @param[in]  acc_elementwise_op  The accumulation elementwise operation.
+    ///
+    /// @return     Unique pointer to the Argument class.
+    ///
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const std::vector<index_t> inLengths,
                         const std::vector<index_t> inStrides,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index e7365daed2..fa7bc7837a 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -168,7 +168,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                      stream_config.stream_id_));
 
                 ave_time = launch_and_time_kernel(
-                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg, b2c_map);
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
             };
 
             if(has_main_k0_block_loop)
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 8cfdd04e55..4120516416 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -157,22 +157,22 @@ __global__ void
 }
 } // namespace
 
-//
-// @brief      Device Convolution operation.
-//
-// Supports:
-//  @li         Forward convolution with up to 3 spatial dimentions
-//  @li         Input tensor in GNWC data format
-//  @li         Weight tensor in GKXC data format
-//  @li         Output tensor in GNWK data format
-//
-// 1D:
-// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
-// 2D:
-// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
-// 3D:
-// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
-//
+///
+/// @brief      Device Convolution operation.
+///
+/// Supports:
+///  @li         Forward convolution with up to 3 spatial dimentions
+///  @li         Input tensor in GNWC data format
+///  @li         Weight tensor in GKXC data format
+///  @li         Output tensor in GNWK data format
+///
+/// 1D:
+/// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+/// 2D:
+/// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+/// 3D:
+/// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+///
 template <index_t NDimSpatial,
           typename ADataType,
           typename BDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
index f18fbcfe4b..46b4157991 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -154,22 +154,22 @@ __global__ void
 
 } // namespace
 
-//
-// @brief      Device Convolution operation.
-//
-// Supports:
-//  @li         Forward convolution with up to 3 spatial dimentions
-//  @li         Input tensor in GNWC data format
-//  @li         Weight tensor in GKXC data format
-//  @li         Output tensor in GNWK data format
-//
-// 1D:
-// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
-// 2D:
-// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
-// 3D:
-// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
-//
+///
+/// @brief      Device Convolution operation.
+///
+/// Supports:
+///  @li         Forward convolution with up to 3 spatial dimentions
+///  @li         Input tensor in GNWC data format
+///  @li         Weight tensor in GKXC data format
+///  @li         Output tensor in GNWK data format
+///
+/// 1D:
+/// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+/// 2D:
+/// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+/// 3D:
+/// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+///
 template <
     index_t NDimSpatial,
     typename ADataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index f4b8d66ecf..2a6f2f4940 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -150,22 +150,22 @@ __global__ void
 
 } // namespace
 
-//
-// @brief      Device Convolution operation.
-//
-// Supports:
-//  @li         Forward convolution with up to 3 spatial dimentions
-//  @li         Input tensor in GNWC data format
-//  @li         Weight tensor in GKXC data format
-//  @li         Output tensor in GNWK data format
-//
-// 1D:
-// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
-// 2D:
-// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
-// 3D:
-// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
-//
+///
+/// @brief      Device Convolution operation.
+///
+/// Supports:
+///  @li         Forward convolution with up to 3 spatial dimentions
+///  @li         Input tensor in GNWC data format
+///  @li         Weight tensor in GKXC data format
+///  @li         Output tensor in GNWK data format
+///
+/// 1D:
+/// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+/// 2D:
+/// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+/// 3D:
+/// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+///
 template <index_t NDimSpatial,
           typename ALayout,
           typename BLayout,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index 41445aeaf0..bad699b013 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -5,11 +5,13 @@
 
 #include <iostream>
 #include <sstream>
+#include <tuple>
 
 #include "ck/ck.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/hip_check_error.hpp"
+#include "ck/host_utility/stream_utility.hpp"
 #include "ck/utility/common_header.hpp"
 #include "ck/utility/tuple.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
@@ -23,8 +25,28 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
+///
+/// @brief      Entry point kernel for device-wide Grouped GEMM operation.
+///
+/// @param[in]  gemm_descs_const            The pointer to the array of GEMM descriptor structures.
+/// @param[in]  tile_count                  The overall number of output tiles we divided all groups
+///                                         into.
+/// @param[in]  k_batch                     The number of batches we split the K dimension into.
+///
+/// @tparam     GridwiseGemm                The specific GridwiseGEMM algorithm implementation.
+/// @tparam     GemmDesc                    The structure holding all necessary descriptors and
+///                                         other data needed for groupd gemm calculation and work
+///                                         distribution.
+/// @tparam     HasMainKBlockLoop           Flag indicating whether all GEMM problem configurations
+///                                         need to loop over tiles in K dimension.
+/// @tparam     CGlobalMemoryDataOperation  The functor used to store data in output C matrix.
+///                                         In example could be: AtomicAdd or Store.
+///
 template <typename GridwiseGemm,
           typename GemmDesc,
+          typename FloatA,
+          typename FloatB,
+          typename FloatC,
           bool HasMainKBlockLoop,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation>
 __global__ void
@@ -32,42 +54,99 @@ __global__ void
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
         kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
-                                       const index_t group_count)
+                                       const index_t tile_count,
+                                       const index_t k_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
     __shared__ uint8_t p_shared[shared_size];
 
-    const index_t block_id = get_block_1d_id();
+    index_t tile_id         = get_block_1d_id();
+    const index_t grid_size = get_grid_size();
     const auto gemm_desc_ptr =
         reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
 
-    index_t left     = 0;
-    index_t right    = group_count;
-    index_t group_id = index_t((left + right) / 2);
-    while((!(block_id >= gemm_desc_ptr[group_id].block_start_ &&
-             block_id < gemm_desc_ptr[group_id].block_end_)) &&
-          left <= right)
+    static constexpr index_t MPerBlock = GridwiseGemm::GetMPerBlock();
+    static constexpr index_t NPerBlock = GridwiseGemm::GetNPerBlock();
+    static constexpr index_t B2E_M01   = 8;
+
+    using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
+    using Block2ETileMapKSplit =
+        BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
+
+    index_t group_id = 0;
+    index_t offset   = 0;
+
+    auto M                = gemm_desc_ptr[group_id].M;
+    auto N                = gemm_desc_ptr[group_id].N;
+    auto StrideC          = gemm_desc_ptr[group_id].StrideC;
+    auto c_grid_desc_m_n  = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
+    auto b2c_tile_map     = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
+    index_t grid_size_grp = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+    index_t gemm_tile_id_start = 0;
+    index_t gemm_tile_id_end   = grid_size_grp;
+
+    while(tile_id < tile_count)
     {
-        if(block_id < gemm_desc_ptr[group_id].block_start_)
+        // Find corresponding GEMM group for out tile
+        while(!(tile_id >= gemm_tile_id_start && tile_id < gemm_tile_id_end))
         {
-            right = group_id;
+            offset += grid_size_grp;
+            group_id++;
+
+            M               = gemm_desc_ptr[group_id].M;
+            N               = gemm_desc_ptr[group_id].N;
+            StrideC         = gemm_desc_ptr[group_id].StrideC;
+            c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
+            b2c_tile_map    = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
+            grid_size_grp   = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+            gemm_tile_id_start = offset;
+            gemm_tile_id_end   = offset + grid_size_grp;
         }
-        else
-        {
-            left = group_id;
-        }
-        group_id = index_t((left + right) / 2);
+
+        const auto p_a_grid = reinterpret_cast<const FloatA*>(gemm_desc_ptr[group_id].p_a_grid);
+        const auto p_b_grid = reinterpret_cast<const FloatB*>(gemm_desc_ptr[group_id].p_b_grid);
+        const auto p_c_grid = reinterpret_cast<FloatC*>(gemm_desc_ptr[group_id].p_c_grid);
+
+        const auto K       = gemm_desc_ptr[group_id].K;
+        const auto StrideA = gemm_desc_ptr[group_id].StrideA;
+        const auto StrideB = gemm_desc_ptr[group_id].StrideB;
+
+        const auto MPadded = GridwiseGemm::CalculateMPadded(M);
+        const auto NPadded = GridwiseGemm::CalculateNPadded(N);
+        const auto KPadded = GridwiseGemm::CalculateKPadded(K, k_batch);
+        const auto K0      = GridwiseGemm::CalculateK0(K, k_batch);
+
+        LocalBlockToCTileMap<Block2ETileMapKSplit> local_b2c{b2c_tile_map, tile_id - offset};
+        GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
+            p_a_grid,
+            p_b_grid,
+            p_c_grid,
+            M,
+            N,
+            K,
+            StrideA,
+            StrideB,
+            StrideC,
+            MPadded,
+            NPadded,
+            KPadded,
+            K0,
+            k_batch,
+            static_cast<void*>(p_shared),
+            local_b2c);
+
+        tile_id += grid_size;
     }
 
-    GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
-        gemm_desc_ptr[group_id].karg_,
-        static_cast<void*>(p_shared),
-        gemm_desc_ptr[group_id].block_2_ctile_map_);
 #else
     ignore = gemm_descs_const;
-    ignore = group_count;
+    ignore = tile_count;
+    ignore = k_batch;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -186,33 +265,13 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         LoopSched,
         PipelineVer>;
 
-    using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
+    using CGridDesc_M_N   = typename GridwiseGemm::CGridDesc_M_N;
+    using GridwiseGemmArg = typename GridwiseGemm::Argument;
+    using KernelArguments = GroupedGemmKernelArguments;
     using Block2ETileMapKSplit =
         BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
     // Block2CTileMap configuration parameter.
-    static constexpr index_t B2E_M01 = 8;
-    using GroupedGemmBlock2ETileMap  = OffsettedBlockToCTileMap<Block2ETileMapKSplit>;
-    using KernelArgument             = typename GridwiseGemm::Argument;
-
-    struct GemmTransKernelArg
-    {
-        KernelArgument karg_;
-        GroupedGemmBlock2ETileMap block_2_ctile_map_;
-        index_t block_start_, block_end_;
-
-        GemmTransKernelArg() = default;
-        GemmTransKernelArg(KernelArgument&& karg,
-                           GroupedGemmBlock2ETileMap&& b2c_map,
-                           index_t block_start,
-                           index_t block_end)
-            : karg_{karg},
-              block_2_ctile_map_{b2c_map},
-              block_start_{block_start},
-              block_end_{block_end}
-        {
-        }
-    };
-
+    static constexpr index_t B2E_M01       = 8;
     static constexpr index_t DefaultKBatch = 1;
 
     // Argument
@@ -225,7 +284,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                  std::vector<GemmDesc>& gemm_descs)
             : Argument(p_As, p_Bs, p_Es, gemm_descs, DefaultKBatch)
         {
-            // TODO: use occupancy api to calculate appropriate batch size.
         }
 
         Argument(std::vector<const void*>& p_As,
@@ -233,9 +291,8 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                  std::vector<void*>& p_Es,
                  std::vector<GemmDesc>& gemm_descs,
                  index_t kbatch)
-            : K_BATCH{kbatch}
+            : K_BATCH{kbatch}, group_count_{0}, skipped_group_count_{0}, grid_size_{0}
         {
-            grid_size_   = 0;
             group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
 
             if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
@@ -247,8 +304,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
             gemm_kernel_args_.reserve(group_count_);
 
-            skipped_group_count_ = 0;
-
             for(std::size_t i = 0; i < gemm_descs.size(); ++i)
             {
                 const index_t M = gemm_descs[i].M_;
@@ -265,51 +320,29 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 const index_t stride_b = gemm_descs[i].stride_B_;
                 const index_t stride_c = gemm_descs[i].stride_C_;
 
-                const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
-                const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
-                const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
-                const index_t k0       = GridwiseGemm::CalculateK0(K, K_BATCH);
-
                 const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
 
-                const auto local_b2c_tile_map =
-                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
                 const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
-
-                const index_t block_start = grid_size_;
-                const index_t block_end   = grid_size_ + grid_size_grp;
-
                 grid_size_ += grid_size_grp;
 
-                // block-to-e-tile map
-                auto grouped_block_2_ctile_map =
-                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
-
-                auto karg = KernelArgument{type_convert<const ADataType*>(p_As[i]),
-                                           type_convert<const BDataType*>(p_Bs[i]),
-                                           type_convert<EDataType*>(p_Es[i]),
-                                           M,
-                                           N,
-                                           K,
-                                           stride_a,
-                                           stride_b,
-                                           stride_c,
-                                           m_padded,
-                                           n_padded,
-                                           k_padded,
-                                           k0,
-                                           K_BATCH};
-
-                gemm_kernel_args_.emplace_back(
-                    std::move(karg), std::move(grouped_block_2_ctile_map), block_start, block_end);
+                gemm_kernel_args_.emplace_back(type_convert<const ADataType*>(p_As[i]),
+                                               type_convert<const BDataType*>(p_Bs[i]),
+                                               type_convert<EDataType*>(p_Es[i]),
+                                               M,
+                                               N,
+                                               K,
+                                               stride_a,
+                                               stride_b,
+                                               stride_c);
             }
         }
 
-        /**
-         * @brief      Recalculate group grid size for all gemms and update B2C maps.
-         *
-         * @param[in]  kbatch  The new splitK parameter value.
-         */
+        ///
+        /// @brief      Set new kbatch value.
+        ///
+        /// @param[in]  kbatch  The new splitK parameter value.
+        ///
         void UpdateKBatch(index_t kbatch)
         {
             K_BATCH    = kbatch;
@@ -318,33 +351,14 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
             for(std::size_t i = 0; i < gemm_kernel_args_.size(); ++i)
             {
 
-                auto& karg = gemm_kernel_args_[i].karg_;
-
-                const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
-                const index_t k0       = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
+                auto& gemm_arg = gemm_kernel_args_[i];
 
                 const auto c_grid_desc_m_n =
-                    GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
+                    GridwiseGemm::MakeCGridDescriptor_M_N(gemm_arg.M, gemm_arg.N, gemm_arg.StrideC);
 
-                const auto local_b2c_tile_map =
-                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
                 const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
-
-                const index_t block_start = grid_size_;
-                const index_t block_end   = grid_size_ + grid_size_grp;
-
                 grid_size_ += grid_size_grp;
-
-                // block-to-e-tile map
-                auto grouped_block_2_ctile_map =
-                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
-
-                karg.KPadded                            = k_padded;
-                karg.K0                                 = k0;
-                karg.k_batch                            = K_BATCH;
-                gemm_kernel_args_[i].block_2_ctile_map_ = grouped_block_2_ctile_map;
-                gemm_kernel_args_[i].block_start_       = block_start;
-                gemm_kernel_args_[i].block_end_         = block_end;
             }
         }
 
@@ -352,31 +366,167 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         index_t K_BATCH;
         index_t group_count_;
         index_t skipped_group_count_;
-
-        std::vector<GemmTransKernelArg> gemm_kernel_args_;
+        // The overall number of output tiles to be processed.
         index_t grid_size_;
+        const void* p_dev_gemm_args_;
+
+        std::vector<KernelArguments> gemm_kernel_args_;
     };
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
+        // The oversubscription factor for the number of blocks that can simultaneously reside on
+        // GPU.
+        static constexpr int BLOCK_SUBSCRIPTION_FACTOR = 1;
+        static constexpr int BLOCK_WAVES               = BlockSize / get_warp_size();
+        static constexpr int CU_SIMDS                  = 4;
+        // Assume we want to have at most 2 waves per SIMD
+        static constexpr int CU_BLOCKS = math::integer_divide_floor(2 * CU_SIMDS, BLOCK_WAVES);
+
+        ///
+        /// @brief      Launch Grouped Gemm kernel.
+        ///
+        /// @note       This function overload is using user provided device buffer for kernel
+        ///             arguments.
+        ///
+        /// @param[in]  arg            The structure containing kernel arguments (in host memory).
+        /// @param[in]  dev_gemm_args  The point to device memory with kernel arguments.
+        /// @param[in]  stream_config  The device stream configuration.
+        ///
+        /// @return     The average kernel execution time (if time measurement is enabled.)
+        ///
+        float Run(const Argument& arg,
+                  const void* dev_gemm_args,
+                  const StreamConfig& stream_config = StreamConfig{})
+        {
+            auto [all_have_kbatch_gt_one, all_have_main_k0_block_loop] =
+                CheckArgument(arg, stream_config);
+
+            if(dev_gemm_args == nullptr)
+            {
+                std::ostringstream err;
+                err << "The gemm arguments workspace buffer is not allocated!"
+                    << " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
+                throw std::runtime_error(err.str());
+            }
+
+            if(all_have_kbatch_gt_one)
+            {
+                for(const auto& gemm_arg : arg.gemm_kernel_args_)
+                {
+                    hip_check_error(hipMemsetAsync(gemm_arg.p_c_grid,
+                                                   0,
+                                                   gemm_arg.M * gemm_arg.N * sizeof(EDataType),
+                                                   stream_config.stream_id_));
+                }
+            }
+
+            float ave_time = 0;
+
+            if(all_have_main_k0_block_loop)
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, true>(
+                        arg, dev_gemm_args, stream_config);
+                }
+                else
+                {
+                    ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, true>(
+                        arg, dev_gemm_args, stream_config);
+                }
+            }
+            else
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, false>(
+                        arg, dev_gemm_args, stream_config);
+                }
+                else
+                {
+                    ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, false>(
+                        arg, dev_gemm_args, stream_config);
+                }
+            }
+
+            return ave_time;
+        }
+
+        ///
+        /// @brief      Launch Grouped Gemm kernel.
+        ///
+        /// @note       This function overload is using device workspace buffer for kernel
+        ///             arguments. The user should call @see GetWorkSpaceSize and @see
+        ///             SetWorkSpacePointer on arg parameter to properly allocate this buffer.
+        ///
+        /// @param[in]  arg            The structure containing kernel arguments (in host memory).
+        /// @param[in]  stream_config  The device stream configuration.
+        ///
+        /// @return     The average kernel execution time (if time measurement is enabled.)
+        ///
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-            index_t K0                       = arg.gemm_kernel_args_[0].karg_.K0;
-            bool all_have_kbatch_gt_one      = arg.gemm_kernel_args_[0].karg_.k_batch > 1;
+            if(arg.p_workspace_ != nullptr)
+            {
+                hip_check_error(
+                    hipMemcpyWithStream(arg.p_workspace_,
+                                        arg.gemm_kernel_args_.data(),
+                                        arg.gemm_kernel_args_.size() * sizeof(KernelArguments),
+                                        hipMemcpyHostToDevice,
+                                        stream_config.stream_id_));
+            }
+            else
+            {
+                std::ostringstream err;
+                err << "The gemm arguments workspace buffer is not allocated!"
+                    << " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
+                throw std::runtime_error(err.str());
+            }
+
+            return Run(arg, arg.p_workspace_, stream_config);
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+
+        private:
+        auto CheckArgument(const Argument& arg, const StreamConfig& stream_config) const
+        {
+            index_t K0 = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[0].K, arg.K_BATCH);
+            bool all_have_kbatch_gt_one      = arg.K_BATCH > 1;
             bool all_have_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
 
             for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
             {
-                const auto& karg = arg.gemm_kernel_args_[i].karg_;
+                const auto& gemm_arg = arg.gemm_kernel_args_[i];
                 if(stream_config.log_level_ > 0)
                 {
-                    karg.Print();
+                    gemm_arg.Print();
                 }
 
-                auto kbatch = karg.k_batch;
+                // Currently all groups use same kbatch value.
+                auto kbatch = arg.K_BATCH;
+                K0          = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[i].K, arg.K_BATCH);
 
-                if(!GridwiseGemm::CheckValidity(karg))
+                if(!GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
+                                                                nullptr,
+                                                                nullptr,
+                                                                gemm_arg.M,
+                                                                gemm_arg.N,
+                                                                gemm_arg.K,
+                                                                gemm_arg.StrideA,
+                                                                gemm_arg.StrideB,
+                                                                gemm_arg.StrideC,
+                                                                0, // MPadded
+                                                                0, // NPadded
+                                                                0, // KPadded
+                                                                K0,
+                                                                kbatch}))
                 {
                     std::ostringstream err;
                     err << "Group id: " << i << " has invalid GridwiseGemm settings!" << __FILE__
@@ -384,7 +534,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                     throw std::runtime_error(err.str());
                 }
 
-                K0 = karg.K0;
                 bool not_all_have_main_k0_block_loop_same =
                     all_have_main_k0_block_loop xor GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
                 bool not_all_have_kbatch_value_same = all_have_kbatch_gt_one xor (kbatch > 1);
@@ -402,99 +551,75 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                     std::ostringstream err;
                     err << "Not all gemms have same kbatch value (=1 or >1)! "
                         << "group [" << i << "], kbatch: " << kbatch
-                        << ", group [0], kbatch: " << arg.gemm_kernel_args_[0].karg_.k_batch
-                        << " in " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
+                        << ", group [0], kbatch: " << arg.K_BATCH << " in " << __FILE__ << ":"
+                        << __LINE__ << ", in function: " << __func__;
                     throw std::runtime_error(err.str());
                 }
             }
-
-            hip_check_error(
-                hipMemcpyWithStream(arg.p_workspace_,
-                                    arg.gemm_kernel_args_.data(),
-                                    arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
-                                    hipMemcpyHostToDevice,
-                                    stream_config.stream_id_));
-
-            float ave_time = 0;
-
-            const auto Run = [&](const auto& kernel) {
-                if(all_have_kbatch_gt_one)
-                {
-                    for(const auto& trans_arg : arg.gemm_kernel_args_)
-                    {
-                        const auto& karg = trans_arg.karg_;
-                        hip_check_error(hipMemsetAsync(karg.p_c_grid,
-                                                       0,
-                                                       karg.M * karg.N * sizeof(EDataType),
-                                                       stream_config.stream_id_));
-                    }
-                }
-
-                ave_time =
-                    launch_and_time_kernel(stream_config,
-                                           kernel,
-                                           dim3(arg.grid_size_),
-                                           dim3(BlockSize),
-                                           0,
-                                           cast_pointer_to_constant_address_space(arg.p_workspace_),
-                                           arg.gemm_kernel_args_.size());
-            };
-
-            if(all_have_main_k0_block_loop)
-            {
-                if(all_have_kbatch_gt_one)
-                {
-                    const auto kernel =
-                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
-                                                       GemmTransKernelArg,
-                                                       true,
-                                                       InMemoryDataOperationEnum::AtomicAdd>;
-
-                    Run(kernel);
-                }
-                else
-                {
-                    const auto kernel =
-                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
-                                                       GemmTransKernelArg,
-                                                       true,
-                                                       InMemoryDataOperationEnum::Set>;
-
-                    Run(kernel);
-                }
-            }
-            else
-            {
-                if(all_have_kbatch_gt_one)
-                {
-                    const auto kernel =
-                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
-                                                       GemmTransKernelArg,
-                                                       false,
-                                                       InMemoryDataOperationEnum::AtomicAdd>;
-
-                    Run(kernel);
-                }
-                else
-                {
-                    const auto kernel =
-                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
-                                                       GemmTransKernelArg,
-                                                       false,
-                                                       InMemoryDataOperationEnum::Set>;
-
-                    Run(kernel);
-                }
-            }
-
-            return ave_time;
+            return std::make_tuple(all_have_kbatch_gt_one, all_have_main_k0_block_loop);
         }
 
-        // polymorphic
-        float Run(const BaseArgument* p_arg,
-                  const StreamConfig& stream_config = StreamConfig{}) override
+        template <InMemoryDataOperationEnum CGlobalMemoryDataOperation, bool HasMainKBlockLoop>
+        float DispatchKernel(const Argument& arg,
+                             const void* dev_gemm_args,
+                             const StreamConfig& stream_config) const
         {
-            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+            const auto kernel = kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                               KernelArguments,
+                                                               ADataType,
+                                                               BDataType,
+                                                               EDataType,
+                                                               HasMainKBlockLoop,
+                                                               CGlobalMemoryDataOperation>;
+            return LaunchKernel(kernel, arg, dev_gemm_args, stream_config);
+        }
+
+        template <typename KernelFunction>
+        int CalculateMaxOccupancyGridSize(const KernelFunction& kernel,
+                                          const StreamConfig& stream_config) const
+        {
+            // Calculate max number of workgroups that can simultaneously reside on the CU.
+            int num_blocks                = 0;
+            size_t dyn_shared_mem_per_blk = 0;
+            hip_check_error(hipOccupancyMaxActiveBlocksPerMultiprocessor(
+                &num_blocks, kernel, BlockSize, dyn_shared_mem_per_blk));
+
+            int cu_count = getAvailableComputeUnitCount(stream_config);
+
+            if(stream_config.log_level_ > 0)
+            {
+                std::cout << "MaxActiveBlocksPerCU: " << num_blocks
+                          << ", available CUs count: " << cu_count << ", occup. grid size: "
+                          << ck::math::min(num_blocks, CU_BLOCKS) * cu_count *
+                                 BLOCK_SUBSCRIPTION_FACTOR
+                          << std::endl;
+            }
+
+            return cu_count * ck::math::min(num_blocks, CU_BLOCKS) * BLOCK_SUBSCRIPTION_FACTOR;
+        }
+
+        template <typename KernelFunction>
+        float LaunchKernel(const KernelFunction& kernel,
+                           const Argument& arg,
+                           const void* dev_gemm_args,
+                           const StreamConfig& stream_config) const
+        {
+            int max_occupancy_grid_size = CalculateMaxOccupancyGridSize(kernel, stream_config);
+
+            // We launch the smaller number of workgroups from acutally needed tiles and the
+            // number of workgroups that maximize the GPU occupancy. That is because for some tile
+            // configuration the first is smaller than the latter. Launching too many workgroups
+            // mean some of them will have to iterate through all gemm problem descriptors just to
+            // find out they have nothing to do which is of course waste of GPU cycles.
+            return launch_and_time_kernel(
+                stream_config,
+                kernel,
+                dim3(ck::math::min(arg.grid_size_, max_occupancy_grid_size)),
+                dim3(BlockSize),
+                0,
+                cast_pointer_to_constant_address_space(dev_gemm_args),
+                arg.grid_size_,
+                arg.K_BATCH);
         }
     };
 
@@ -506,11 +631,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
     static bool IsSupportedArgument(const Argument& arg)
     {
-        if(!ck::is_xdl_supported())
-        {
-            return false;
-        }
-
         if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
             arg.skipped_group_count_) != arg.group_count_)
         {
@@ -525,14 +645,28 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         bool supported = true;
         for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
         {
-            const auto& a        = arg.gemm_kernel_args_[i].karg_;
-            bool group_arg_valid = GridwiseGemm::CheckValidity(a);
+            const auto& gemm_arg = arg.gemm_kernel_args_[i];
+            const auto K0        = GridwiseGemm::CalculateK0(gemm_arg.K, arg.K_BATCH);
+            bool group_arg_valid = GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
+                                                                               nullptr,
+                                                                               nullptr,
+                                                                               gemm_arg.M,
+                                                                               gemm_arg.N,
+                                                                               gemm_arg.K,
+                                                                               gemm_arg.StrideA,
+                                                                               gemm_arg.StrideB,
+                                                                               gemm_arg.StrideC,
+                                                                               0, // MPadded
+                                                                               0, // NPadded
+                                                                               0, // KPadded
+                                                                               K0,
+                                                                               arg.K_BATCH});
             if(not group_arg_valid)
             {
 #if DEBUG_LOG
                 std::cout << "[" << __func__ << "] group id: " << i
                           << " has invalid GridwiseGemm settings!" << std::endl;
-                a.Print();
+                gemm_arg.Print();
 #endif // DEBUG_LOG
             }
             supported = supported && group_arg_valid;
@@ -540,7 +674,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         return supported;
     }
 
-    // polymorphic
     bool IsSupportedArgument(const BaseArgument* p_arg) override
     {
         return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
@@ -560,7 +693,6 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
     static auto MakeInvoker() { return Invoker{}; }
 
-    // polymorphic
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(std::vector<const void*>& p_As,
                         std::vector<const void*>& p_Bs,
@@ -574,19 +706,17 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         return std::make_unique<Argument>(p_As, p_Bs, p_Es, gemm_descs);
     }
 
-    // polymorphic
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
     {
         return std::make_unique<Invoker>(Invoker{});
     }
 
-    // polymorphic
     std::string GetTypeString() const override
     {
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceGroupedGemm_XdlSplitK"
+        str << "DeviceGroupedGemm_XdlSplitKTileLoop"
             << "<"
             << std::string(ALayout::name)[0] << ","
             << std::string(BLayout::name)[0] << ","
@@ -605,7 +735,9 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
             << BBlockTransferSrcScalarPerVector << ", "
             << CShuffleMXdlPerWavePerShuffle << ", "
             << CShuffleNXdlPerWavePerShuffle << ", "
-            << getGemmSpecializationString(GemmSpec)
+            << ABlockTransferThreadClusterLengths_K0_M_K1{} << ", "
+            << getGemmSpecializationString(GemmSpec) << ", "
+            << PipelineVer
             << ">";
         // clang-format on
 
@@ -615,16 +747,24 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
     size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
     {
         return dynamic_cast<const Argument*>(p_arg)->gemm_kernel_args_.size() *
-               sizeof(GemmTransKernelArg);
+               sizeof(KernelArguments);
     }
 
     static void SetKBatchSize(Argument& arg, index_t kbatch) { arg.UpdateKBatch(kbatch); }
+    static void SetDeviceKernelArgs(Argument& arg, const void* p_dev_kernel_args)
+    {
+        arg.p_dev_gemm_args_ = p_dev_kernel_args;
+    }
 
-    // polymorphic
     void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const override
     {
         return SetKBatchSize(*dynamic_cast<Argument*>(p_arg), kbatch);
     }
+
+    void SetDeviceKernelArgs(BaseArgument* p_arg, const void* p_dev_kernel_args) const override
+    {
+        return SetDeviceKernelArgs(*dynamic_cast<Argument*>(p_arg), p_dev_kernel_args);
+    }
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
index 8eff9d2415..6da1769d7f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
@@ -348,24 +348,24 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
                         acc_elementwise_op};
     };
 
-    //
-    // @brief      Makes a pointer to Argument class.
-    //
-    // @param[in]  inLengths           Input tensor extent(s) from high to low dimension
-    // @param[in]  inStrides           Input tensor stride(s) from high to low dimension
-    // @param[in]  reduceDims          The dimension(s) the normalization operation is applied
-    // @param[in]  alpha               Typeless pointer in host memory storing the alpha scaling
-    //                                 value as type AccDataType
-    // @param[in]  beta                Typeless pointer in host memory storing the beta scaling
-    //                                 value as type AccDataType
-    // @param[in]  in_dev              Typeless const pointer in device memory storing the input
-    //                                 tensor
-    // @param      out_dev             Typeless pointer in device memory storing the output tensor
-    // @param[in]  in_elementwise_op   The input elementwise operation.
-    // @param[in]  acc_elementwise_op  The accumulation elementwise operation.
-    //
-    // @return     Unique pointer to the Argument class.
-    //
+    ///
+    /// @brief      Makes a pointer to Argument class.
+    ///
+    /// @param[in]  inLengths           Input tensor extent(s) from high to low dimension
+    /// @param[in]  inStrides           Input tensor stride(s) from high to low dimension
+    /// @param[in]  reduceDims          The dimension(s) the normalization operation is applied
+    /// @param[in]  alpha               Typeless pointer in host memory storing the alpha scaling
+    ///                                 value as type AccDataType
+    /// @param[in]  beta                Typeless pointer in host memory storing the beta scaling
+    ///                                 value as type AccDataType
+    /// @param[in]  in_dev              Typeless const pointer in device memory storing the input
+    ///                                 tensor
+    /// @param      out_dev             Typeless pointer in device memory storing the output tensor
+    /// @param[in]  in_elementwise_op   The input elementwise operation.
+    /// @param[in]  acc_elementwise_op  The accumulation elementwise operation.
+    ///
+    /// @return     Unique pointer to the Argument class.
+    ///
     std::unique_ptr<BaseArgument> MakeArgumentPointer(const std::vector<index_t> inLengths,
                                                       const std::vector<index_t> inStrides,
                                                       const std::vector<int> reduceDims,
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index 7bb47e9d3c..82d1dc179d 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -271,7 +271,8 @@ struct BlockToCTileMap_KSplit_M00_N0_M01Adapt
     {
     }
 
-    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+    __host__ __device__ constexpr index_t
+    CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
     {
         const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I0), MPerBlock);
         const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I1), NPerBlock);
@@ -624,24 +625,36 @@ struct OffsettedBlockToCTileMap
     index_t block_start_;
 };
 
-/**
- * @brief      Simple tile mapping which creates 3D grid of block of threads.
- *
- * @paragraph  Description
- *             This Block-to-C-tile-map creates a 3D grid (n_blocks, m_blocks, z_blocks) of thread
- *             blocks. The first 2D are regular 2D tiles created by division of output GEMM
- *             dimenions by corresponding tile size. The third dimension (Z) is a k-split dimension,
- *             which denotes the number of blocks we use to divide work on GEMM K dimension onto.
- *
- * @tparam     MPerBlock  Output block tile size in M dimension.
- * @tparam     NPerBlock  Output block tile size in N dimension.
- */
+///
+/// @brief      Simple tile mapping which creates 3D grid of block of threads.
+///
+/// @paragraph  Description
+///             This Block-to-C-tile-map creates a 3D grid (n_blocks, m_blocks, z_blocks) of thread
+///             blocks. The first 2D are regular 2D tiles created by division of output GEMM
+///             dimenions by corresponding tile size. The third dimension (Z) is a k-split
+///             dimension, which denotes the number of blocks we use to divide work on GEMM K
+///             dimension onto.
+///
+/// @tparam     MPerBlock  Output block tile size in M dimension.
+/// @tparam     NPerBlock  Output block tile size in N dimension.
+///
 template <index_t MPerBlock, index_t NPerBlock>
 struct BlockToCTileMap_3DGrid_KSplit
 {
-
     __host__ __device__ BlockToCTileMap_3DGrid_KSplit() = default;
 
+    ///
+    /// @brief      Constructs a new instance.
+    ///
+    /// @param[in]  top_idx  Swallow blockIdx.
+    ///
+    /// @tparam     TopIdx   The type of block index.
+    ///
+    template <typename TopIdx>
+    __host__ __device__ BlockToCTileMap_3DGrid_KSplit([[maybe_unused]] TopIdx top_idx)
+    {
+    }
+
     __host__ __device__ constexpr auto
     CalculateGridSize(index_t M, index_t N, index_t k_split) const
     {
@@ -652,8 +665,7 @@ struct BlockToCTileMap_3DGrid_KSplit
         return std::make_tuple(N0, M0, k_split);
     }
 
-    template <typename TopIdx>
-    __device__ constexpr auto CalculateBottomIndex(const TopIdx&) const
+    __device__ constexpr auto CalculateBottomIndex() const
     {
         return make_tuple(blockIdx.z, blockIdx.y, blockIdx.x);
     }
@@ -672,6 +684,53 @@ struct BlockToCTileMap_3DGrid_KSplit
     }
 };
 
+///
+/// @brief      Block to CTile Map which foster external mechanism for setting up local block id.
+///
+///             In example this type can be easily used to implement tile looping work distribution
+///             scheme.
+///
+/// @tparam     UnderlyingBlockToCTileMap  The type of the local tile mapp.
+///
+template <typename UnderlyingBlockToCTileMap>
+struct LocalBlockToCTileMap
+{
+    using underlying_type = UnderlyingBlockToCTileMap;
+
+    __host__ __device__ LocalBlockToCTileMap(UnderlyingBlockToCTileMap block_to_ctile_map,
+                                             index_t local_id)
+        : block_to_ctile_map_{block_to_ctile_map}, local_block_id_{local_id}
+    {
+    }
+
+    __host__ __device__ constexpr auto CalculateBottomIndex() const
+    {
+        return block_to_ctile_map_.CalculateBottomIndex(make_multi_index(local_block_id_));
+    }
+
+    template <typename CTileIdx, typename CTileDim>
+    __host__ __device__ bool ValidCTileIndex(const CTileIdx& c_tile_idx,
+                                             const CTileDim& c_tile_dim) const
+    {
+        return block_to_ctile_map_.ValidCTileIndex(c_tile_idx, c_tile_dim);
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ bool CheckValidity(const CGridDesc_M_N& c_grid_desc_m_n) const
+    {
+        return block_to_ctile_map_.CheckValidity(c_grid_desc_m_n);
+    }
+
+    template <typename CGridDesc_M_N>
+    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+    {
+        return block_to_ctile_map_.CalculateGridSize(c_grid_desc_m_n);
+    }
+
+    UnderlyingBlockToCTileMap block_to_ctile_map_;
+    index_t local_block_id_;
+};
+
 enum StreamKReductionStrategy
 {
     Atomic = 0, // sk block use atomic to do reduction
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
index f760feb2ed..b34b07d53a 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
@@ -4,6 +4,8 @@
 #pragma once
 
 #include <iostream>
+#include <ostream>
+#include <string>
 
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp"
@@ -42,4 +44,20 @@ constexpr auto GridwiseGemmPipeline_Selector()
     }
 }
 
+inline std::string getPipelineVersionString(const PipelineVersion& pv)
+{
+    switch(pv)
+    {
+    case PipelineVersion::v1: return "PipelineVersion::v1";
+    case PipelineVersion::v2: return "PipelineVersion::v2";
+    default: return "Unrecognized pipeline version!";
+    }
+}
+
 } // namespace ck
+
+inline std::ostream& operator<<(std::ostream& os, const ck::PipelineVersion pv)
+{
+    os << ck::getPipelineVersionString(pv);
+    return os;
+}
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index f9744d7167..43a7768451 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -27,8 +27,7 @@ __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg,
-                                             const Block2CTileMap& b2c_map)
+        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
@@ -36,11 +35,12 @@ __global__ void
 
     __shared__ uint8_t p_shared[shared_size];
 
+    Block2CTileMap b2c_map{get_block_1d_id()};
+
     GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
         karg, static_cast<void*>(p_shared), b2c_map);
 #else
     ignore = karg;
-    ignore = b2c_map;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -541,15 +541,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
     }
 
-    // return block_id to C matrix tile idx (m0, n0) mapping
-    template <typename CGridDesc>
-    __host__ __device__ static constexpr auto MakeCBlockClusterAdaptor(
-        const CGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
-    {
-        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc>(
-            c_m_n_grid_desc, 8, KBatch);
-    }
-
     __host__ __device__ static constexpr auto
     GetCBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
     {
@@ -575,18 +566,28 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     template <bool HasMainKBlockLoop,
               InMemoryDataOperationEnum CGlobalMemoryDataOperation,
               typename Block2CTileMap>
-    __device__ static void Run(const Argument& karg,
+    __device__ static void Run(const FloatA* p_a_grid,
+                               const FloatB* p_b_grid,
+                               FloatC* p_c_grid,
+                               index_t M,
+                               index_t N,
+                               index_t K,
+                               index_t StrideA,
+                               index_t StrideB,
+                               index_t StrideC,
+                               index_t MPadded,
+                               index_t NPadded,
+                               index_t KPadded,
+                               index_t K0,
+                               index_t k_batch,
                                void* __restrict__ p_shared_block,
                                const Block2CTileMap& block_2_ctile_map)
     {
-        const FloatA* p_a_grid           = karg.p_a_grid;
-        const FloatB* p_b_grid           = karg.p_b_grid;
-        FloatC* p_c_grid                 = karg.p_c_grid;
-        const auto a_b_k0_m_k1_grid_desc = MakeAGridDescriptor_KBatch_K0_M_K1(
-            karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
-        const auto b_b_k0_n_k1_grid_desc = MakeBGridDescriptor_KBatch_K0_N_K1(
-            karg.K, karg.NPadded, karg.N, karg.StrideB, karg.k_batch, karg.K0, karg.KPadded);
-        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
+        const auto a_b_k0_m_k1_grid_desc =
+            MakeAGridDescriptor_KBatch_K0_M_K1(M, MPadded, K, StrideA, k_batch, K0, KPadded);
+        const auto b_b_k0_n_k1_grid_desc =
+            MakeBGridDescriptor_KBatch_K0_N_K1(K, NPadded, N, StrideB, k_batch, K0, KPadded);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(M, N, StrideC);
 
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
@@ -602,8 +603,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
 
         // divide block work by [KBatch, M, N]
-        const auto block_work_idx =
-            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
+        const auto block_work_idx = block_2_ctile_map.CalculateBottomIndex();
 
         if(!block_2_ctile_map.ValidCTileIndex(
                block_work_idx,
@@ -1010,6 +1010,34 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
     }
 
+    template <bool HasMainKBlockLoop,
+              InMemoryDataOperationEnum CGlobalMemoryDataOperation,
+              typename Block2CTileMap>
+    __device__ static void Run(const Argument& karg,
+                               void* __restrict__ p_shared_block,
+                               const Block2CTileMap& block_2_ctile_map)
+    {
+        Run<HasMainKBlockLoop, CGlobalMemoryDataOperation, Block2CTileMap>(karg.p_a_grid,
+                                                                           karg.p_b_grid,
+                                                                           karg.p_c_grid,
+                                                                           karg.M,
+                                                                           karg.N,
+                                                                           karg.K,
+                                                                           karg.StrideA,
+                                                                           karg.StrideB,
+                                                                           karg.StrideC,
+                                                                           karg.MPadded,
+                                                                           karg.NPadded,
+                                                                           karg.KPadded,
+                                                                           karg.K0,
+                                                                           karg.k_batch,
+                                                                           p_shared_block,
+                                                                           block_2_ctile_map);
+    }
+
+    static constexpr auto GetMPerBlock() { return MPerBlock; }
+    static constexpr auto GetNPerBlock() { return NPerBlock; }
+
     static std::string GetTypeString()
     {
         auto str = std::stringstream();
diff --git a/include/ck/utility/sequence.hpp b/include/ck/utility/sequence.hpp
index d6bfb2eba1..63422a10e9 100644
--- a/include/ck/utility/sequence.hpp
+++ b/include/ck/utility/sequence.hpp
@@ -897,3 +897,14 @@ template <index_t NSize, index_t I>
 using uniform_sequence_gen_t = typename uniform_sequence_gen<NSize, I>::type;
 
 } // namespace ck
+
+template <ck::index_t... Is>
+std::ostream& operator<<(std::ostream& os, const ck::Sequence<Is...>)
+{
+    using S = ck::Sequence<Is...>;
+    os << "{";
+    ck::static_for<0, S::Size() - ck::Number<1>{}, 1>{}(
+        [&](auto i) { os << S::At(i).value << ", "; });
+    os << S::At(S::Size() - ck::Number<1>{}).value << "}";
+    return os;
+}
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
index 8f4182a231..2466092124 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
@@ -14,27 +14,27 @@ namespace ck {
 namespace tensor_operation {
 namespace host {
 
-//
-// @brief      Reference implementation for forward convolution.
-//
-// @paragraph
-//             Tensor descriptor in GNCHW/GKCXY/GNKHW dimensional order
-//             Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
-//             as long as dimensions in tensor descriptor is in GNCHW order
-//
-// @tparam     InDataType               Input tensor data type.
-// @tparam     WeiDataType              Weights tensor data type.
-// @tparam     OutDataType              Output tensor data type.
-// @tparam     InElementwiseOperation   Functor for input tensor elementwise
-//                                      operation.
-// @tparam     WeiElementwiseOperation  Functor for weights tensor elementwise
-//                                      operation.
-// @tparam     NDimSpatial  Number of spatial dimensions.
-//
-// input descriptor in [G, N, C, Do, Ho, Wo] order
-// weight descriptor in [G, K, C, Z, Y, X] order
-// output descriptor in [G, N, K, Di, Hi, Wi] order
-// phyiscal layout is irrelavent
+///
+/// @brief      Reference implementation for forward convolution.
+///
+/// @paragraph
+///             Tensor descriptor in GNCHW/GKCXY/GNKHW dimensional order
+///             Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
+///             as long as dimensions in tensor descriptor is in GNCHW order
+///
+/// @tparam     InDataType               Input tensor data type.
+/// @tparam     WeiDataType              Weights tensor data type.
+/// @tparam     OutDataType              Output tensor data type.
+/// @tparam     InElementwiseOperation   Functor for input tensor elementwise
+///                                      operation.
+/// @tparam     WeiElementwiseOperation  Functor for weights tensor elementwise
+///                                      operation.
+/// @tparam     NDimSpatial  Number of spatial dimensions.
+///
+/// input descriptor in [G, N, C, Do, Ho, Wo] order
+/// weight descriptor in [G, K, C, Z, Y, X] order
+/// output descriptor in [G, N, K, Di, Hi, Wi] order
+/// phyiscal layout is irrelavent
 template <ck::index_t NDimSpatial,
           typename InDataType,
           typename WeiDataType,
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index c829e8863d..f1bed87ad1 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -33,26 +33,26 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 // a[m, k] * b[k, n] = e[m, n]
 using device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
-        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+    //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+    //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+    //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
     // clang-format on
     >;
 
@@ -60,20 +60,20 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 
 using device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
     // clang-format off
-        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+    //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+    //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+    //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+    //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index 8642562fa3..a2762b45c2 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -33,27 +33,41 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 // a[m, k] * b[k, n] = e[m, n]
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
-        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        // Currently AK1 must equal BK1 !
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     PipelineVersion|
+    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                    |
+    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                    |
+    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                    |
+    // Pipline v1
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+
+    // Pipline v2
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index 90223fd9bd..fc028aa30b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -31,57 +31,41 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
     // clang-format off
-        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        // Currently AK1 must equal BK1 !
-     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
-     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-        
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
+    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     PipelineVersion|
+    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                    |
+    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                    |
+    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                    |
+    // Pipline v1
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+    
+    // Pipline v2
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index aa6365cd98..729111d054 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -32,23 +32,31 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 // a[m, k] * b[n, k] = e[m, n]
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // clang-format off
-        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
+    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         PipelineVersion|
+    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                        |
+    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                        |
+    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                        |
+    // Pipeline v1
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v1>,
+    // Pipeline v2
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v2>
+
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
index f4460b360b..14c82de360 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -31,30 +31,31 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
     // clang-format off
-        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
-        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
+    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         PipelineVersion|
+    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                        |
+    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                        |
+    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                        |
+    // Pipeline v1
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v1>,
+    
+    // Pipeline v2
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
+  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
+    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v2>
     // clang-format on
     >;
 
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index da9e92b862..7f812792aa 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -12,7 +12,9 @@ cmake
 -save-temps=$PWD"                                                                                 \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=ON                                                                                   \
--D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
+-D GPU_TARGETS="gfx90a"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
+
+#-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
diff --git a/test/grouped_gemm/test_grouped_gemm_ut_cases.inc b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
index d94d140d97..0ab7c3245e 100644
--- a/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
+++ b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
@@ -4,7 +4,7 @@ TEST_P(RRR_F16_F16_F16, TinyCases)
 {
     const std::vector<int> Ms{0, 1};
     constexpr int N = 768;
-    constexpr int K = 544;
+    constexpr int K = 1088;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -17,9 +17,9 @@ TEST_P(RRR_F16_F16_F16, TinyCases)
 
 TEST_P(RRR_F16_F16_F16, SmallCases)
 {
-    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
+    const std::vector<int> Ms{2, 3, 4, 5};
     constexpr int N = 768;
-    constexpr int K = 544;
+    constexpr int K = 1088;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -34,7 +34,7 @@ TEST_P(RRR_F16_F16_F16, MidCases)
 {
     const std::vector<int> Ms{167, 183, 177, 153, 139, 204};
     constexpr int N = 768;
-    constexpr int K = 544;
+    constexpr int K = 1088;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -49,7 +49,7 @@ TEST_P(RRR_F16_F16_F16, Regular)
 {
     const std::vector<int> Ms{64, 128, 256};
     constexpr int N = 768;
-    constexpr int K = 320;
+    constexpr int K = 640;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -79,7 +79,7 @@ TEST_P(RCR_F16_F16_F16, TinyCases)
 {
     const std::vector<int> Ms{0, 1};
     constexpr int N = 768;
-    constexpr int K = 544;
+    constexpr int K = 1088;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -91,9 +91,9 @@ TEST_P(RCR_F16_F16_F16, TinyCases)
 
 TEST_P(RCR_F16_F16_F16, SmallCases)
 {
-    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
+    const std::vector<int> Ms{2, 3, 4, 5};
     constexpr int N = 768;
-    constexpr int K = 544;
+    constexpr int K = 1088;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -123,7 +123,7 @@ TEST_P(RCR_F16_F16_F16, Regular)
 {
     const std::vector<int> Ms{32, 64, 128, 256};
     constexpr int N = 768;
-    constexpr int K = 320;
+    constexpr int K = 640;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -151,9 +151,9 @@ TEST_P(RCR_F16_F16_F16, MNKPadded)
 
 TEST_P(RRR_F16_F16_F16_LargeK, TestLargeKBatch)
 {
-    const std::vector<int> Ms{188, 210};
+    const std::vector<int> Ms{127, 150, 188, 210};
     constexpr int N = 768;
-    constexpr int K = 4096;
+    constexpr int K = 8192;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);

From c99323be6e4a5c610e33493d2c30be73e4d0891c Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 11 Oct 2023 14:27:29 -0500
Subject: [PATCH 167/194] Revert "Grouped Gemm with looping over the tiles.
 (#788)" (#982)

This reverts commit a4f72a314a85732ed67d5aa8d1088d207a7e0e61.
---
 .../17_grouped_gemm_fastgelu/CMakeLists.txt   |   2 +-
 .../gpu/device/device_grouped_gemm_splitk.hpp |  74 +--
 .../gpu/device/device_softmax.hpp             |  32 +-
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |   2 +-
 ..._conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp |  32 +-
 ...ice_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp |  32 +-
 ...ouped_conv_fwd_multiple_d_xdl_cshuffle.hpp |  32 +-
 ...evice_grouped_gemm_xdl_splitk_cshuffle.hpp | 566 +++++++-----------
 .../gpu/device/impl/device_softmax_impl.hpp   |  36 +-
 .../gpu/grid/block_to_ctile_map.hpp           |  91 +--
 .../grid/gridwise_gemm_pipeline_selector.hpp  |  18 -
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  74 +--
 include/ck/utility/sequence.hpp               |  11 -
 .../cpu/reference_conv_fwd.hpp                |  42 +-
 ...gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp |  68 +--
 ...l_splitk_f16_f16_f16_mk_kn_mn_instance.cpp |  56 +-
 ...16_f16_f16_mk_kn_mn_irregular_instance.cpp |  86 +--
 ...l_splitk_f16_f16_f16_mk_nk_mn_instance.cpp |  42 +-
 ...16_f16_f16_mk_nk_mn_irregular_instance.cpp |  49 +-
 script/cmake-ck-dev.sh                        |   4 +-
 .../test_grouped_gemm_ut_cases.inc            |  22 +-
 21 files changed, 517 insertions(+), 854 deletions(-)

diff --git a/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt b/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
index 06c581b706..659e6769d8 100644
--- a/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
+++ b/client_example/17_grouped_gemm_fastgelu/CMakeLists.txt
@@ -1,2 +1,2 @@
 add_executable(client_grouped_gemm_fastgelu grouped_gemm_fastgelu.cpp)
-target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations)
+target_link_libraries(client_grouped_gemm_fastgelu PRIVATE composable_kernel::device_operations)
\ No newline at end of file
diff --git a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
index eb5447c8fa..06d180d30f 100644
--- a/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_grouped_gemm_splitk.hpp
@@ -8,57 +8,6 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-///
-/// @brief      Structure representing single GEMM problem arguments.
-///
-///             The pointer to the vector of those structures is passed
-///             to the GroupedGEMM entry point kernel.
-///
-struct GroupedGemmKernelArguments
-{
-    __host__ __device__ GroupedGemmKernelArguments(const void* p_a_grid_,
-                                                   const void* p_b_grid_,
-                                                   void* p_c_grid_,
-                                                   index_t M_,
-                                                   index_t N_,
-                                                   index_t K_,
-                                                   index_t StrideA_,
-                                                   index_t StrideB_,
-                                                   index_t StrideC_)
-        : p_a_grid{p_a_grid_},
-          p_b_grid{p_b_grid_},
-          p_c_grid{p_c_grid_},
-          M{M_},
-          N{N_},
-          K{K_},
-          StrideA{StrideA_},
-          StrideB{StrideB_},
-          StrideC{StrideC_}
-    {
-    }
-
-    const void* p_a_grid;
-    const void* p_b_grid;
-    void* p_c_grid;
-    index_t M;
-    index_t N;
-    index_t K;
-    index_t StrideA;
-    index_t StrideB;
-    index_t StrideC;
-
-    void Print() const
-    {
-        std::cout << "arg {"
-                  << "M:" << M << ", "
-                  << "N:" << N << ", "
-                  << "K:" << K << ", "
-                  << "SA:" << StrideA << ", "
-                  << "SB:" << StrideB << ", "
-                  << "SC:" << StrideC << "}" << std::endl;
-    }
-};
-
 template <typename ALayout,
           typename BLayout,
           typename DsLayout,
@@ -82,28 +31,7 @@ struct DeviceGroupedGemmSplitK : public DeviceGroupedGemm<ALayout,
                                                           BElementwiseOperation,
                                                           CElementwiseOperation>
 {
-    //----------------------------------------------------------------------------------------------
-    /// @brief      Sets the k batch size.
-    ///
-    /// @param      p_arg   Pointer to the Argument we're going to change.
-    /// @param[in]  kbatch  The kbatch value.
-    ///
-    virtual void SetKBatchSize([[maybe_unused]] BaseArgument* p_arg,
-                               [[maybe_unused]] index_t kbatch) const
-    {
-    }
-
-    //----------------------------------------------------------------------------------------------
-    /// @brief      Sets the device kernel arguments pointer.
-    ///
-    /// @param      p_arg              The pointer to the Argument we're going to update.
-    /// @param[in]  p_dev_kernel_args  The pointer to the device memory which contains kernel
-    ///                                arguments.
-    ///
-    virtual void SetDeviceKernelArgs([[maybe_unused]] BaseArgument* p_arg,
-                                     [[maybe_unused]] const void* p_dev_kernel_args) const
-    {
-    }
+    virtual void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const = 0;
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/device/device_softmax.hpp b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
index 68c29a5340..1902fd09ee 100644
--- a/include/ck/tensor_operation/gpu/device/device_softmax.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_softmax.hpp
@@ -22,22 +22,22 @@ template <typename InDataType,
           index_t NumReduceDim>
 struct DeviceSoftmax : public BaseOperator
 {
-    ///
-    /// @brief      Makes a pointer to Argument class.
-    ///
-    /// @param[in]  inLengths           Input tensor extent(s) from high to low dimension
-    /// @param[in]  inStrides           Input tensor stride(s) from high to low dimension
-    /// @param[in]  reduceDims          The dimension(s) the normalization operation is applied
-    /// @param[in]  alpha               double type value
-    /// @param[in]  beta                double type value
-    /// @param[in]  in_dev              Typeless const pointer in device memory storing the input
-    ///                                 tensor
-    /// @param      out_dev             Typeless pointer in device memory storing the output tensor
-    /// @param[in]  in_elementwise_op   The input elementwise operation.
-    /// @param[in]  acc_elementwise_op  The accumulation elementwise operation.
-    ///
-    /// @return     Unique pointer to the Argument class.
-    ///
+    //
+    // @brief      Makes a pointer to Argument class.
+    //
+    // @param[in]  inLengths           Input tensor extent(s) from high to low dimension
+    // @param[in]  inStrides           Input tensor stride(s) from high to low dimension
+    // @param[in]  reduceDims          The dimension(s) the normalization operation is applied
+    // @param[in]  alpha               double type value
+    // @param[in]  beta                double type value
+    // @param[in]  in_dev              Typeless const pointer in device memory storing the input
+    //                                 tensor
+    // @param      out_dev             Typeless pointer in device memory storing the output tensor
+    // @param[in]  in_elementwise_op   The input elementwise operation.
+    // @param[in]  acc_elementwise_op  The accumulation elementwise operation.
+    //
+    // @return     Unique pointer to the Argument class.
+    //
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const std::vector<index_t> inLengths,
                         const std::vector<index_t> inStrides,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index fa7bc7837a..e7365daed2 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -168,7 +168,7 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                      stream_config.stream_id_));
 
                 ave_time = launch_and_time_kernel(
-                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg);
+                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg, b2c_map);
             };
 
             if(has_main_k0_block_loop)
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
index 4120516416..8cfdd04e55 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
@@ -157,22 +157,22 @@ __global__ void
 }
 } // namespace
 
-///
-/// @brief      Device Convolution operation.
-///
-/// Supports:
-///  @li         Forward convolution with up to 3 spatial dimentions
-///  @li         Input tensor in GNWC data format
-///  @li         Weight tensor in GKXC data format
-///  @li         Output tensor in GNWK data format
-///
-/// 1D:
-/// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
-/// 2D:
-/// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
-/// 3D:
-/// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
-///
+//
+// @brief      Device Convolution operation.
+//
+// Supports:
+//  @li         Forward convolution with up to 3 spatial dimentions
+//  @li         Input tensor in GNWC data format
+//  @li         Weight tensor in GKXC data format
+//  @li         Output tensor in GNWK data format
+//
+// 1D:
+// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+// 2D:
+// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+// 3D:
+// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+//
 template <index_t NDimSpatial,
           typename ADataType,
           typename BDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
index 46b4157991..f18fbcfe4b 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
@@ -154,22 +154,22 @@ __global__ void
 
 } // namespace
 
-///
-/// @brief      Device Convolution operation.
-///
-/// Supports:
-///  @li         Forward convolution with up to 3 spatial dimentions
-///  @li         Input tensor in GNWC data format
-///  @li         Weight tensor in GKXC data format
-///  @li         Output tensor in GNWK data format
-///
-/// 1D:
-/// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
-/// 2D:
-/// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
-/// 3D:
-/// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
-///
+//
+// @brief      Device Convolution operation.
+//
+// Supports:
+//  @li         Forward convolution with up to 3 spatial dimentions
+//  @li         Input tensor in GNWC data format
+//  @li         Weight tensor in GKXC data format
+//  @li         Output tensor in GNWK data format
+//
+// 1D:
+// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+// 2D:
+// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+// 3D:
+// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+//
 template <
     index_t NDimSpatial,
     typename ADataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
index 2a6f2f4940..f4b8d66ecf 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp
@@ -150,22 +150,22 @@ __global__ void
 
 } // namespace
 
-///
-/// @brief      Device Convolution operation.
-///
-/// Supports:
-///  @li         Forward convolution with up to 3 spatial dimentions
-///  @li         Input tensor in GNWC data format
-///  @li         Weight tensor in GKXC data format
-///  @li         Output tensor in GNWK data format
-///
-/// 1D:
-/// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
-/// 2D:
-/// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
-/// 3D:
-/// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
-///
+//
+// @brief      Device Convolution operation.
+//
+// Supports:
+//  @li         Forward convolution with up to 3 spatial dimentions
+//  @li         Input tensor in GNWC data format
+//  @li         Weight tensor in GKXC data format
+//  @li         Output tensor in GNWK data format
+//
+// 1D:
+// out[N, Wo, K] = in[N, Wi, C] * wei[K, X, C]
+// 2D:
+// out[N, Ho, Wo, K] = in[N, Hi, Wi, C] * wei[K, Y, X, C]
+// 3D:
+// out[N, Do, Ho, Wo, K] = in[N, Di, Hi, Wi, C] * wei[K, Z, Y, X, C]
+//
 template <index_t NDimSpatial,
           typename ALayout,
           typename BLayout,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
index bad699b013..41445aeaf0 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_splitk_cshuffle.hpp
@@ -5,13 +5,11 @@
 
 #include <iostream>
 #include <sstream>
-#include <tuple>
 
 #include "ck/ck.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 #include "ck/host_utility/hip_check_error.hpp"
-#include "ck/host_utility/stream_utility.hpp"
 #include "ck/utility/common_header.hpp"
 #include "ck/utility/tuple.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"
@@ -25,28 +23,8 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-///
-/// @brief      Entry point kernel for device-wide Grouped GEMM operation.
-///
-/// @param[in]  gemm_descs_const            The pointer to the array of GEMM descriptor structures.
-/// @param[in]  tile_count                  The overall number of output tiles we divided all groups
-///                                         into.
-/// @param[in]  k_batch                     The number of batches we split the K dimension into.
-///
-/// @tparam     GridwiseGemm                The specific GridwiseGEMM algorithm implementation.
-/// @tparam     GemmDesc                    The structure holding all necessary descriptors and
-///                                         other data needed for groupd gemm calculation and work
-///                                         distribution.
-/// @tparam     HasMainKBlockLoop           Flag indicating whether all GEMM problem configurations
-///                                         need to loop over tiles in K dimension.
-/// @tparam     CGlobalMemoryDataOperation  The functor used to store data in output C matrix.
-///                                         In example could be: AtomicAdd or Store.
-///
 template <typename GridwiseGemm,
           typename GemmDesc,
-          typename FloatA,
-          typename FloatB,
-          typename FloatC,
           bool HasMainKBlockLoop,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation>
 __global__ void
@@ -54,99 +32,42 @@ __global__ void
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
         kernel_grouped_gemm_xdl_splitk(const void CK_CONSTANT_ADDRESS_SPACE* gemm_descs_const,
-                                       const index_t tile_count,
-                                       const index_t k_batch)
+                                       const index_t group_count)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
-
     constexpr index_t shared_size = GridwiseGemm::GetSharedMemoryNumberOfByte();
     __shared__ uint8_t p_shared[shared_size];
 
-    index_t tile_id         = get_block_1d_id();
-    const index_t grid_size = get_grid_size();
+    const index_t block_id = get_block_1d_id();
     const auto gemm_desc_ptr =
         reinterpret_cast<const GemmDesc*>(cast_pointer_to_generic_address_space(gemm_descs_const));
 
-    static constexpr index_t MPerBlock = GridwiseGemm::GetMPerBlock();
-    static constexpr index_t NPerBlock = GridwiseGemm::GetNPerBlock();
-    static constexpr index_t B2E_M01   = 8;
-
-    using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
-    using Block2ETileMapKSplit =
-        BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
-
-    index_t group_id = 0;
-    index_t offset   = 0;
-
-    auto M                = gemm_desc_ptr[group_id].M;
-    auto N                = gemm_desc_ptr[group_id].N;
-    auto StrideC          = gemm_desc_ptr[group_id].StrideC;
-    auto c_grid_desc_m_n  = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
-    auto b2c_tile_map     = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
-    index_t grid_size_grp = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
-
-    index_t gemm_tile_id_start = 0;
-    index_t gemm_tile_id_end   = grid_size_grp;
-
-    while(tile_id < tile_count)
+    index_t left     = 0;
+    index_t right    = group_count;
+    index_t group_id = index_t((left + right) / 2);
+    while((!(block_id >= gemm_desc_ptr[group_id].block_start_ &&
+             block_id < gemm_desc_ptr[group_id].block_end_)) &&
+          left <= right)
     {
-        // Find corresponding GEMM group for out tile
-        while(!(tile_id >= gemm_tile_id_start && tile_id < gemm_tile_id_end))
+        if(block_id < gemm_desc_ptr[group_id].block_start_)
         {
-            offset += grid_size_grp;
-            group_id++;
-
-            M               = gemm_desc_ptr[group_id].M;
-            N               = gemm_desc_ptr[group_id].N;
-            StrideC         = gemm_desc_ptr[group_id].StrideC;
-            c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, StrideC);
-            b2c_tile_map    = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, k_batch};
-            grid_size_grp   = b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
-
-            gemm_tile_id_start = offset;
-            gemm_tile_id_end   = offset + grid_size_grp;
+            right = group_id;
         }
-
-        const auto p_a_grid = reinterpret_cast<const FloatA*>(gemm_desc_ptr[group_id].p_a_grid);
-        const auto p_b_grid = reinterpret_cast<const FloatB*>(gemm_desc_ptr[group_id].p_b_grid);
-        const auto p_c_grid = reinterpret_cast<FloatC*>(gemm_desc_ptr[group_id].p_c_grid);
-
-        const auto K       = gemm_desc_ptr[group_id].K;
-        const auto StrideA = gemm_desc_ptr[group_id].StrideA;
-        const auto StrideB = gemm_desc_ptr[group_id].StrideB;
-
-        const auto MPadded = GridwiseGemm::CalculateMPadded(M);
-        const auto NPadded = GridwiseGemm::CalculateNPadded(N);
-        const auto KPadded = GridwiseGemm::CalculateKPadded(K, k_batch);
-        const auto K0      = GridwiseGemm::CalculateK0(K, k_batch);
-
-        LocalBlockToCTileMap<Block2ETileMapKSplit> local_b2c{b2c_tile_map, tile_id - offset};
-        GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
-            p_a_grid,
-            p_b_grid,
-            p_c_grid,
-            M,
-            N,
-            K,
-            StrideA,
-            StrideB,
-            StrideC,
-            MPadded,
-            NPadded,
-            KPadded,
-            K0,
-            k_batch,
-            static_cast<void*>(p_shared),
-            local_b2c);
-
-        tile_id += grid_size;
+        else
+        {
+            left = group_id;
+        }
+        group_id = index_t((left + right) / 2);
     }
 
+    GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
+        gemm_desc_ptr[group_id].karg_,
+        static_cast<void*>(p_shared),
+        gemm_desc_ptr[group_id].block_2_ctile_map_);
 #else
     ignore = gemm_descs_const;
-    ignore = tile_count;
-    ignore = k_batch;
+    ignore = group_count;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -265,13 +186,33 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         LoopSched,
         PipelineVer>;
 
-    using CGridDesc_M_N   = typename GridwiseGemm::CGridDesc_M_N;
-    using GridwiseGemmArg = typename GridwiseGemm::Argument;
-    using KernelArguments = GroupedGemmKernelArguments;
+    using CGridDesc_M_N = typename GridwiseGemm::CGridDesc_M_N;
     using Block2ETileMapKSplit =
         BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc_M_N>;
     // Block2CTileMap configuration parameter.
-    static constexpr index_t B2E_M01       = 8;
+    static constexpr index_t B2E_M01 = 8;
+    using GroupedGemmBlock2ETileMap  = OffsettedBlockToCTileMap<Block2ETileMapKSplit>;
+    using KernelArgument             = typename GridwiseGemm::Argument;
+
+    struct GemmTransKernelArg
+    {
+        KernelArgument karg_;
+        GroupedGemmBlock2ETileMap block_2_ctile_map_;
+        index_t block_start_, block_end_;
+
+        GemmTransKernelArg() = default;
+        GemmTransKernelArg(KernelArgument&& karg,
+                           GroupedGemmBlock2ETileMap&& b2c_map,
+                           index_t block_start,
+                           index_t block_end)
+            : karg_{karg},
+              block_2_ctile_map_{b2c_map},
+              block_start_{block_start},
+              block_end_{block_end}
+        {
+        }
+    };
+
     static constexpr index_t DefaultKBatch = 1;
 
     // Argument
@@ -284,6 +225,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                  std::vector<GemmDesc>& gemm_descs)
             : Argument(p_As, p_Bs, p_Es, gemm_descs, DefaultKBatch)
         {
+            // TODO: use occupancy api to calculate appropriate batch size.
         }
 
         Argument(std::vector<const void*>& p_As,
@@ -291,8 +233,9 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                  std::vector<void*>& p_Es,
                  std::vector<GemmDesc>& gemm_descs,
                  index_t kbatch)
-            : K_BATCH{kbatch}, group_count_{0}, skipped_group_count_{0}, grid_size_{0}
+            : K_BATCH{kbatch}
         {
+            grid_size_   = 0;
             group_count_ = ck::type_convert<ck::index_t>(gemm_descs.size());
 
             if(!(group_count_ == ck::type_convert<ck::index_t>(p_As.size()) &&
@@ -304,6 +247,8 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
             gemm_kernel_args_.reserve(group_count_);
 
+            skipped_group_count_ = 0;
+
             for(std::size_t i = 0; i < gemm_descs.size(); ++i)
             {
                 const index_t M = gemm_descs[i].M_;
@@ -320,29 +265,51 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                 const index_t stride_b = gemm_descs[i].stride_B_;
                 const index_t stride_c = gemm_descs[i].stride_C_;
 
+                const index_t m_padded = GridwiseGemm::CalculateMPadded(M);
+                const index_t n_padded = GridwiseGemm::CalculateNPadded(N);
+                const index_t k_padded = GridwiseGemm::CalculateKPadded(K, K_BATCH);
+                const index_t k0       = GridwiseGemm::CalculateK0(K, K_BATCH);
+
                 const auto c_grid_desc_m_n = GridwiseGemm::MakeCGridDescriptor_M_N(M, N, stride_c);
 
-                auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                const auto local_b2c_tile_map =
+                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
                 const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+                const index_t block_start = grid_size_;
+                const index_t block_end   = grid_size_ + grid_size_grp;
+
                 grid_size_ += grid_size_grp;
 
-                gemm_kernel_args_.emplace_back(type_convert<const ADataType*>(p_As[i]),
-                                               type_convert<const BDataType*>(p_Bs[i]),
-                                               type_convert<EDataType*>(p_Es[i]),
-                                               M,
-                                               N,
-                                               K,
-                                               stride_a,
-                                               stride_b,
-                                               stride_c);
+                // block-to-e-tile map
+                auto grouped_block_2_ctile_map =
+                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
+
+                auto karg = KernelArgument{type_convert<const ADataType*>(p_As[i]),
+                                           type_convert<const BDataType*>(p_Bs[i]),
+                                           type_convert<EDataType*>(p_Es[i]),
+                                           M,
+                                           N,
+                                           K,
+                                           stride_a,
+                                           stride_b,
+                                           stride_c,
+                                           m_padded,
+                                           n_padded,
+                                           k_padded,
+                                           k0,
+                                           K_BATCH};
+
+                gemm_kernel_args_.emplace_back(
+                    std::move(karg), std::move(grouped_block_2_ctile_map), block_start, block_end);
             }
         }
 
-        ///
-        /// @brief      Set new kbatch value.
-        ///
-        /// @param[in]  kbatch  The new splitK parameter value.
-        ///
+        /**
+         * @brief      Recalculate group grid size for all gemms and update B2C maps.
+         *
+         * @param[in]  kbatch  The new splitK parameter value.
+         */
         void UpdateKBatch(index_t kbatch)
         {
             K_BATCH    = kbatch;
@@ -351,14 +318,33 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
             for(std::size_t i = 0; i < gemm_kernel_args_.size(); ++i)
             {
 
-                auto& gemm_arg = gemm_kernel_args_[i];
+                auto& karg = gemm_kernel_args_[i].karg_;
+
+                const index_t k_padded = GridwiseGemm::CalculateKPadded(karg.K, K_BATCH);
+                const index_t k0       = GridwiseGemm::CalculateK0(karg.K, K_BATCH);
 
                 const auto c_grid_desc_m_n =
-                    GridwiseGemm::MakeCGridDescriptor_M_N(gemm_arg.M, gemm_arg.N, gemm_arg.StrideC);
+                    GridwiseGemm::MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
 
-                auto local_b2c_tile_map = Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
+                const auto local_b2c_tile_map =
+                    Block2ETileMapKSplit{c_grid_desc_m_n, B2E_M01, K_BATCH};
                 const index_t grid_size_grp = local_b2c_tile_map.CalculateGridSize(c_grid_desc_m_n);
+
+                const index_t block_start = grid_size_;
+                const index_t block_end   = grid_size_ + grid_size_grp;
+
                 grid_size_ += grid_size_grp;
+
+                // block-to-e-tile map
+                auto grouped_block_2_ctile_map =
+                    GroupedGemmBlock2ETileMap(local_b2c_tile_map, block_start);
+
+                karg.KPadded                            = k_padded;
+                karg.K0                                 = k0;
+                karg.k_batch                            = K_BATCH;
+                gemm_kernel_args_[i].block_2_ctile_map_ = grouped_block_2_ctile_map;
+                gemm_kernel_args_[i].block_start_       = block_start;
+                gemm_kernel_args_[i].block_end_         = block_end;
             }
         }
 
@@ -366,167 +352,31 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         index_t K_BATCH;
         index_t group_count_;
         index_t skipped_group_count_;
-        // The overall number of output tiles to be processed.
-        index_t grid_size_;
-        const void* p_dev_gemm_args_;
 
-        std::vector<KernelArguments> gemm_kernel_args_;
+        std::vector<GemmTransKernelArg> gemm_kernel_args_;
+        index_t grid_size_;
     };
 
     // Invoker
     struct Invoker : public BaseInvoker
     {
-        // The oversubscription factor for the number of blocks that can simultaneously reside on
-        // GPU.
-        static constexpr int BLOCK_SUBSCRIPTION_FACTOR = 1;
-        static constexpr int BLOCK_WAVES               = BlockSize / get_warp_size();
-        static constexpr int CU_SIMDS                  = 4;
-        // Assume we want to have at most 2 waves per SIMD
-        static constexpr int CU_BLOCKS = math::integer_divide_floor(2 * CU_SIMDS, BLOCK_WAVES);
-
-        ///
-        /// @brief      Launch Grouped Gemm kernel.
-        ///
-        /// @note       This function overload is using user provided device buffer for kernel
-        ///             arguments.
-        ///
-        /// @param[in]  arg            The structure containing kernel arguments (in host memory).
-        /// @param[in]  dev_gemm_args  The point to device memory with kernel arguments.
-        /// @param[in]  stream_config  The device stream configuration.
-        ///
-        /// @return     The average kernel execution time (if time measurement is enabled.)
-        ///
-        float Run(const Argument& arg,
-                  const void* dev_gemm_args,
-                  const StreamConfig& stream_config = StreamConfig{})
-        {
-            auto [all_have_kbatch_gt_one, all_have_main_k0_block_loop] =
-                CheckArgument(arg, stream_config);
-
-            if(dev_gemm_args == nullptr)
-            {
-                std::ostringstream err;
-                err << "The gemm arguments workspace buffer is not allocated!"
-                    << " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
-                throw std::runtime_error(err.str());
-            }
-
-            if(all_have_kbatch_gt_one)
-            {
-                for(const auto& gemm_arg : arg.gemm_kernel_args_)
-                {
-                    hip_check_error(hipMemsetAsync(gemm_arg.p_c_grid,
-                                                   0,
-                                                   gemm_arg.M * gemm_arg.N * sizeof(EDataType),
-                                                   stream_config.stream_id_));
-                }
-            }
-
-            float ave_time = 0;
-
-            if(all_have_main_k0_block_loop)
-            {
-                if(all_have_kbatch_gt_one)
-                {
-                    ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, true>(
-                        arg, dev_gemm_args, stream_config);
-                }
-                else
-                {
-                    ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, true>(
-                        arg, dev_gemm_args, stream_config);
-                }
-            }
-            else
-            {
-                if(all_have_kbatch_gt_one)
-                {
-                    ave_time = DispatchKernel<InMemoryDataOperationEnum::AtomicAdd, false>(
-                        arg, dev_gemm_args, stream_config);
-                }
-                else
-                {
-                    ave_time = DispatchKernel<InMemoryDataOperationEnum::Set, false>(
-                        arg, dev_gemm_args, stream_config);
-                }
-            }
-
-            return ave_time;
-        }
-
-        ///
-        /// @brief      Launch Grouped Gemm kernel.
-        ///
-        /// @note       This function overload is using device workspace buffer for kernel
-        ///             arguments. The user should call @see GetWorkSpaceSize and @see
-        ///             SetWorkSpacePointer on arg parameter to properly allocate this buffer.
-        ///
-        /// @param[in]  arg            The structure containing kernel arguments (in host memory).
-        /// @param[in]  stream_config  The device stream configuration.
-        ///
-        /// @return     The average kernel execution time (if time measurement is enabled.)
-        ///
         float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
         {
-            if(arg.p_workspace_ != nullptr)
-            {
-                hip_check_error(
-                    hipMemcpyWithStream(arg.p_workspace_,
-                                        arg.gemm_kernel_args_.data(),
-                                        arg.gemm_kernel_args_.size() * sizeof(KernelArguments),
-                                        hipMemcpyHostToDevice,
-                                        stream_config.stream_id_));
-            }
-            else
-            {
-                std::ostringstream err;
-                err << "The gemm arguments workspace buffer is not allocated!"
-                    << " In " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
-                throw std::runtime_error(err.str());
-            }
-
-            return Run(arg, arg.p_workspace_, stream_config);
-        }
-
-        float Run(const BaseArgument* p_arg,
-                  const StreamConfig& stream_config = StreamConfig{}) override
-        {
-            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
-        }
-
-        private:
-        auto CheckArgument(const Argument& arg, const StreamConfig& stream_config) const
-        {
-            index_t K0 = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[0].K, arg.K_BATCH);
-            bool all_have_kbatch_gt_one      = arg.K_BATCH > 1;
+            index_t K0                       = arg.gemm_kernel_args_[0].karg_.K0;
+            bool all_have_kbatch_gt_one      = arg.gemm_kernel_args_[0].karg_.k_batch > 1;
             bool all_have_main_k0_block_loop = GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
 
             for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
             {
-                const auto& gemm_arg = arg.gemm_kernel_args_[i];
+                const auto& karg = arg.gemm_kernel_args_[i].karg_;
                 if(stream_config.log_level_ > 0)
                 {
-                    gemm_arg.Print();
+                    karg.Print();
                 }
 
-                // Currently all groups use same kbatch value.
-                auto kbatch = arg.K_BATCH;
-                K0          = GridwiseGemm::CalculateK0(arg.gemm_kernel_args_[i].K, arg.K_BATCH);
+                auto kbatch = karg.k_batch;
 
-                if(!GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
-                                                                nullptr,
-                                                                nullptr,
-                                                                gemm_arg.M,
-                                                                gemm_arg.N,
-                                                                gemm_arg.K,
-                                                                gemm_arg.StrideA,
-                                                                gemm_arg.StrideB,
-                                                                gemm_arg.StrideC,
-                                                                0, // MPadded
-                                                                0, // NPadded
-                                                                0, // KPadded
-                                                                K0,
-                                                                kbatch}))
+                if(!GridwiseGemm::CheckValidity(karg))
                 {
                     std::ostringstream err;
                     err << "Group id: " << i << " has invalid GridwiseGemm settings!" << __FILE__
@@ -534,6 +384,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                     throw std::runtime_error(err.str());
                 }
 
+                K0 = karg.K0;
                 bool not_all_have_main_k0_block_loop_same =
                     all_have_main_k0_block_loop xor GridwiseGemm::CalculateHasMainK0BlockLoop(K0);
                 bool not_all_have_kbatch_value_same = all_have_kbatch_gt_one xor (kbatch > 1);
@@ -551,75 +402,99 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
                     std::ostringstream err;
                     err << "Not all gemms have same kbatch value (=1 or >1)! "
                         << "group [" << i << "], kbatch: " << kbatch
-                        << ", group [0], kbatch: " << arg.K_BATCH << " in " << __FILE__ << ":"
-                        << __LINE__ << ", in function: " << __func__;
+                        << ", group [0], kbatch: " << arg.gemm_kernel_args_[0].karg_.k_batch
+                        << " in " << __FILE__ << ":" << __LINE__ << ", in function: " << __func__;
                     throw std::runtime_error(err.str());
                 }
             }
-            return std::make_tuple(all_have_kbatch_gt_one, all_have_main_k0_block_loop);
-        }
 
-        template <InMemoryDataOperationEnum CGlobalMemoryDataOperation, bool HasMainKBlockLoop>
-        float DispatchKernel(const Argument& arg,
-                             const void* dev_gemm_args,
-                             const StreamConfig& stream_config) const
-        {
-            const auto kernel = kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
-                                                               KernelArguments,
-                                                               ADataType,
-                                                               BDataType,
-                                                               EDataType,
-                                                               HasMainKBlockLoop,
-                                                               CGlobalMemoryDataOperation>;
-            return LaunchKernel(kernel, arg, dev_gemm_args, stream_config);
-        }
+            hip_check_error(
+                hipMemcpyWithStream(arg.p_workspace_,
+                                    arg.gemm_kernel_args_.data(),
+                                    arg.gemm_kernel_args_.size() * sizeof(GemmTransKernelArg),
+                                    hipMemcpyHostToDevice,
+                                    stream_config.stream_id_));
 
-        template <typename KernelFunction>
-        int CalculateMaxOccupancyGridSize(const KernelFunction& kernel,
-                                          const StreamConfig& stream_config) const
-        {
-            // Calculate max number of workgroups that can simultaneously reside on the CU.
-            int num_blocks                = 0;
-            size_t dyn_shared_mem_per_blk = 0;
-            hip_check_error(hipOccupancyMaxActiveBlocksPerMultiprocessor(
-                &num_blocks, kernel, BlockSize, dyn_shared_mem_per_blk));
+            float ave_time = 0;
 
-            int cu_count = getAvailableComputeUnitCount(stream_config);
+            const auto Run = [&](const auto& kernel) {
+                if(all_have_kbatch_gt_one)
+                {
+                    for(const auto& trans_arg : arg.gemm_kernel_args_)
+                    {
+                        const auto& karg = trans_arg.karg_;
+                        hip_check_error(hipMemsetAsync(karg.p_c_grid,
+                                                       0,
+                                                       karg.M * karg.N * sizeof(EDataType),
+                                                       stream_config.stream_id_));
+                    }
+                }
 
-            if(stream_config.log_level_ > 0)
+                ave_time =
+                    launch_and_time_kernel(stream_config,
+                                           kernel,
+                                           dim3(arg.grid_size_),
+                                           dim3(BlockSize),
+                                           0,
+                                           cast_pointer_to_constant_address_space(arg.p_workspace_),
+                                           arg.gemm_kernel_args_.size());
+            };
+
+            if(all_have_main_k0_block_loop)
             {
-                std::cout << "MaxActiveBlocksPerCU: " << num_blocks
-                          << ", available CUs count: " << cu_count << ", occup. grid size: "
-                          << ck::math::min(num_blocks, CU_BLOCKS) * cu_count *
-                                 BLOCK_SUBSCRIPTION_FACTOR
-                          << std::endl;
+                if(all_have_kbatch_gt_one)
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       true,
+                                                       InMemoryDataOperationEnum::AtomicAdd>;
+
+                    Run(kernel);
+                }
+                else
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       true,
+                                                       InMemoryDataOperationEnum::Set>;
+
+                    Run(kernel);
+                }
+            }
+            else
+            {
+                if(all_have_kbatch_gt_one)
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       false,
+                                                       InMemoryDataOperationEnum::AtomicAdd>;
+
+                    Run(kernel);
+                }
+                else
+                {
+                    const auto kernel =
+                        kernel_grouped_gemm_xdl_splitk<GridwiseGemm,
+                                                       GemmTransKernelArg,
+                                                       false,
+                                                       InMemoryDataOperationEnum::Set>;
+
+                    Run(kernel);
+                }
             }
 
-            return cu_count * ck::math::min(num_blocks, CU_BLOCKS) * BLOCK_SUBSCRIPTION_FACTOR;
+            return ave_time;
         }
 
-        template <typename KernelFunction>
-        float LaunchKernel(const KernelFunction& kernel,
-                           const Argument& arg,
-                           const void* dev_gemm_args,
-                           const StreamConfig& stream_config) const
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
         {
-            int max_occupancy_grid_size = CalculateMaxOccupancyGridSize(kernel, stream_config);
-
-            // We launch the smaller number of workgroups from acutally needed tiles and the
-            // number of workgroups that maximize the GPU occupancy. That is because for some tile
-            // configuration the first is smaller than the latter. Launching too many workgroups
-            // mean some of them will have to iterate through all gemm problem descriptors just to
-            // find out they have nothing to do which is of course waste of GPU cycles.
-            return launch_and_time_kernel(
-                stream_config,
-                kernel,
-                dim3(ck::math::min(arg.grid_size_, max_occupancy_grid_size)),
-                dim3(BlockSize),
-                0,
-                cast_pointer_to_constant_address_space(dev_gemm_args),
-                arg.grid_size_,
-                arg.K_BATCH);
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
         }
     };
 
@@ -631,6 +506,11 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
         if((ck::type_convert<ck::index_t>(arg.gemm_kernel_args_.size()) +
             arg.skipped_group_count_) != arg.group_count_)
         {
@@ -645,28 +525,14 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         bool supported = true;
         for(std::size_t i = 0; i < arg.gemm_kernel_args_.size(); ++i)
         {
-            const auto& gemm_arg = arg.gemm_kernel_args_[i];
-            const auto K0        = GridwiseGemm::CalculateK0(gemm_arg.K, arg.K_BATCH);
-            bool group_arg_valid = GridwiseGemm::CheckValidity(GridwiseGemmArg{nullptr,
-                                                                               nullptr,
-                                                                               nullptr,
-                                                                               gemm_arg.M,
-                                                                               gemm_arg.N,
-                                                                               gemm_arg.K,
-                                                                               gemm_arg.StrideA,
-                                                                               gemm_arg.StrideB,
-                                                                               gemm_arg.StrideC,
-                                                                               0, // MPadded
-                                                                               0, // NPadded
-                                                                               0, // KPadded
-                                                                               K0,
-                                                                               arg.K_BATCH});
+            const auto& a        = arg.gemm_kernel_args_[i].karg_;
+            bool group_arg_valid = GridwiseGemm::CheckValidity(a);
             if(not group_arg_valid)
             {
 #if DEBUG_LOG
                 std::cout << "[" << __func__ << "] group id: " << i
                           << " has invalid GridwiseGemm settings!" << std::endl;
-                gemm_arg.Print();
+                a.Print();
 #endif // DEBUG_LOG
             }
             supported = supported && group_arg_valid;
@@ -674,6 +540,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         return supported;
     }
 
+    // polymorphic
     bool IsSupportedArgument(const BaseArgument* p_arg) override
     {
         return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
@@ -693,6 +560,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
 
     static auto MakeInvoker() { return Invoker{}; }
 
+    // polymorphic
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(std::vector<const void*>& p_As,
                         std::vector<const void*>& p_Bs,
@@ -706,17 +574,19 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
         return std::make_unique<Argument>(p_As, p_Bs, p_Es, gemm_descs);
     }
 
+    // polymorphic
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
     {
         return std::make_unique<Invoker>(Invoker{});
     }
 
+    // polymorphic
     std::string GetTypeString() const override
     {
         auto str = std::stringstream();
 
         // clang-format off
-        str << "DeviceGroupedGemm_XdlSplitKTileLoop"
+        str << "DeviceGroupedGemm_XdlSplitK"
             << "<"
             << std::string(ALayout::name)[0] << ","
             << std::string(BLayout::name)[0] << ","
@@ -735,9 +605,7 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
             << BBlockTransferSrcScalarPerVector << ", "
             << CShuffleMXdlPerWavePerShuffle << ", "
             << CShuffleNXdlPerWavePerShuffle << ", "
-            << ABlockTransferThreadClusterLengths_K0_M_K1{} << ", "
-            << getGemmSpecializationString(GemmSpec) << ", "
-            << PipelineVer
+            << getGemmSpecializationString(GemmSpec)
             << ">";
         // clang-format on
 
@@ -747,24 +615,16 @@ struct DeviceGroupedGemmXdlSplitKCShuffle : public DeviceGroupedGemmSplitK<ALayo
     size_t GetWorkSpaceSize(const BaseArgument* p_arg) const override
     {
         return dynamic_cast<const Argument*>(p_arg)->gemm_kernel_args_.size() *
-               sizeof(KernelArguments);
+               sizeof(GemmTransKernelArg);
     }
 
     static void SetKBatchSize(Argument& arg, index_t kbatch) { arg.UpdateKBatch(kbatch); }
-    static void SetDeviceKernelArgs(Argument& arg, const void* p_dev_kernel_args)
-    {
-        arg.p_dev_gemm_args_ = p_dev_kernel_args;
-    }
 
+    // polymorphic
     void SetKBatchSize(BaseArgument* p_arg, index_t kbatch) const override
     {
         return SetKBatchSize(*dynamic_cast<Argument*>(p_arg), kbatch);
     }
-
-    void SetDeviceKernelArgs(BaseArgument* p_arg, const void* p_dev_kernel_args) const override
-    {
-        return SetDeviceKernelArgs(*dynamic_cast<Argument*>(p_arg), p_dev_kernel_args);
-    }
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
index 6da1769d7f..8eff9d2415 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_softmax_impl.hpp
@@ -348,24 +348,24 @@ struct DeviceSoftmaxImpl : public DeviceSoftmax<InDataType,
                         acc_elementwise_op};
     };
 
-    ///
-    /// @brief      Makes a pointer to Argument class.
-    ///
-    /// @param[in]  inLengths           Input tensor extent(s) from high to low dimension
-    /// @param[in]  inStrides           Input tensor stride(s) from high to low dimension
-    /// @param[in]  reduceDims          The dimension(s) the normalization operation is applied
-    /// @param[in]  alpha               Typeless pointer in host memory storing the alpha scaling
-    ///                                 value as type AccDataType
-    /// @param[in]  beta                Typeless pointer in host memory storing the beta scaling
-    ///                                 value as type AccDataType
-    /// @param[in]  in_dev              Typeless const pointer in device memory storing the input
-    ///                                 tensor
-    /// @param      out_dev             Typeless pointer in device memory storing the output tensor
-    /// @param[in]  in_elementwise_op   The input elementwise operation.
-    /// @param[in]  acc_elementwise_op  The accumulation elementwise operation.
-    ///
-    /// @return     Unique pointer to the Argument class.
-    ///
+    //
+    // @brief      Makes a pointer to Argument class.
+    //
+    // @param[in]  inLengths           Input tensor extent(s) from high to low dimension
+    // @param[in]  inStrides           Input tensor stride(s) from high to low dimension
+    // @param[in]  reduceDims          The dimension(s) the normalization operation is applied
+    // @param[in]  alpha               Typeless pointer in host memory storing the alpha scaling
+    //                                 value as type AccDataType
+    // @param[in]  beta                Typeless pointer in host memory storing the beta scaling
+    //                                 value as type AccDataType
+    // @param[in]  in_dev              Typeless const pointer in device memory storing the input
+    //                                 tensor
+    // @param      out_dev             Typeless pointer in device memory storing the output tensor
+    // @param[in]  in_elementwise_op   The input elementwise operation.
+    // @param[in]  acc_elementwise_op  The accumulation elementwise operation.
+    //
+    // @return     Unique pointer to the Argument class.
+    //
     std::unique_ptr<BaseArgument> MakeArgumentPointer(const std::vector<index_t> inLengths,
                                                       const std::vector<index_t> inStrides,
                                                       const std::vector<int> reduceDims,
diff --git a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
index 82d1dc179d..7bb47e9d3c 100644
--- a/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
+++ b/include/ck/tensor_operation/gpu/grid/block_to_ctile_map.hpp
@@ -271,8 +271,7 @@ struct BlockToCTileMap_KSplit_M00_N0_M01Adapt
     {
     }
 
-    __host__ __device__ constexpr index_t
-    CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
+    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
     {
         const auto M0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I0), MPerBlock);
         const auto N0 = math::integer_divide_ceil(c_grid_desc_m_n.GetLength(I1), NPerBlock);
@@ -625,35 +624,23 @@ struct OffsettedBlockToCTileMap
     index_t block_start_;
 };
 
-///
-/// @brief      Simple tile mapping which creates 3D grid of block of threads.
-///
-/// @paragraph  Description
-///             This Block-to-C-tile-map creates a 3D grid (n_blocks, m_blocks, z_blocks) of thread
-///             blocks. The first 2D are regular 2D tiles created by division of output GEMM
-///             dimenions by corresponding tile size. The third dimension (Z) is a k-split
-///             dimension, which denotes the number of blocks we use to divide work on GEMM K
-///             dimension onto.
-///
-/// @tparam     MPerBlock  Output block tile size in M dimension.
-/// @tparam     NPerBlock  Output block tile size in N dimension.
-///
+/**
+ * @brief      Simple tile mapping which creates 3D grid of block of threads.
+ *
+ * @paragraph  Description
+ *             This Block-to-C-tile-map creates a 3D grid (n_blocks, m_blocks, z_blocks) of thread
+ *             blocks. The first 2D are regular 2D tiles created by division of output GEMM
+ *             dimenions by corresponding tile size. The third dimension (Z) is a k-split dimension,
+ *             which denotes the number of blocks we use to divide work on GEMM K dimension onto.
+ *
+ * @tparam     MPerBlock  Output block tile size in M dimension.
+ * @tparam     NPerBlock  Output block tile size in N dimension.
+ */
 template <index_t MPerBlock, index_t NPerBlock>
 struct BlockToCTileMap_3DGrid_KSplit
 {
-    __host__ __device__ BlockToCTileMap_3DGrid_KSplit() = default;
 
-    ///
-    /// @brief      Constructs a new instance.
-    ///
-    /// @param[in]  top_idx  Swallow blockIdx.
-    ///
-    /// @tparam     TopIdx   The type of block index.
-    ///
-    template <typename TopIdx>
-    __host__ __device__ BlockToCTileMap_3DGrid_KSplit([[maybe_unused]] TopIdx top_idx)
-    {
-    }
+    __host__ __device__ BlockToCTileMap_3DGrid_KSplit() = default;
 
     __host__ __device__ constexpr auto
     CalculateGridSize(index_t M, index_t N, index_t k_split) const
@@ -665,7 +652,8 @@ struct BlockToCTileMap_3DGrid_KSplit
         return std::make_tuple(N0, M0, k_split);
     }
 
-    __device__ constexpr auto CalculateBottomIndex() const
+    template <typename TopIdx>
+    __device__ constexpr auto CalculateBottomIndex(const TopIdx&) const
     {
         return make_tuple(blockIdx.z, blockIdx.y, blockIdx.x);
     }
@@ -684,53 +672,6 @@ struct BlockToCTileMap_3DGrid_KSplit
     }
 };
 
-///
-/// @brief      Block to CTile Map which foster external mechanism for setting up local block id.
-///
-///             In example this type can be easily used to implement tile looping work distribution
-///             scheme.
-///
-/// @tparam     UnderlyingBlockToCTileMap  The type of the local tile mapp.
-///
-template <typename UnderlyingBlockToCTileMap>
-struct LocalBlockToCTileMap
-{
-    using underlying_type = UnderlyingBlockToCTileMap;
-
-    __host__ __device__ LocalBlockToCTileMap(UnderlyingBlockToCTileMap block_to_ctile_map,
-                                             index_t local_id)
-        : block_to_ctile_map_{block_to_ctile_map}, local_block_id_{local_id}
-    {
-    }
-
-    __host__ __device__ constexpr auto CalculateBottomIndex() const
-    {
-        return block_to_ctile_map_.CalculateBottomIndex(make_multi_index(local_block_id_));
-    }
-
-    template <typename CTileIdx, typename CTileDim>
-    __host__ __device__ bool ValidCTileIndex(const CTileIdx& c_tile_idx,
-                                             const CTileDim& c_tile_dim) const
-    {
-        return block_to_ctile_map_.ValidCTileIndex(c_tile_idx, c_tile_dim);
-    }
-
-    template <typename CGridDesc_M_N>
-    __host__ bool CheckValidity(const CGridDesc_M_N& c_grid_desc_m_n) const
-    {
-        return block_to_ctile_map_.CheckValidity(c_grid_desc_m_n);
-    }
-
-    template <typename CGridDesc_M_N>
-    __host__ constexpr index_t CalculateGridSize(const CGridDesc_M_N& c_grid_desc_m_n) const
-    {
-        return block_to_ctile_map_.CalculateGridSize(c_grid_desc_m_n);
-    }
-
-    UnderlyingBlockToCTileMap block_to_ctile_map_;
-    index_t local_block_id_;
-};
-
 enum StreamKReductionStrategy
 {
     Atomic = 0, // sk block use atomic to do reduction
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
index b34b07d53a..f760feb2ed 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_selector.hpp
@@ -4,8 +4,6 @@
 #pragma once
 
 #include <iostream>
-#include <ostream>
-#include <string>
 
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v1.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_pipeline_v2.hpp"
@@ -44,20 +42,4 @@ constexpr auto GridwiseGemmPipeline_Selector()
     }
 }
 
-inline std::string getPipelineVersionString(const PipelineVersion& pv)
-{
-    switch(pv)
-    {
-    case PipelineVersion::v1: return "PipelineVersion::v1";
-    case PipelineVersion::v2: return "PipelineVersion::v2";
-    default: return "Unrecognized pipeline version!";
-    }
-}
-
 } // namespace ck
-
-inline std::ostream& operator<<(std::ostream& os, const ck::PipelineVersion pv)
-{
-    os << ck::getPipelineVersionString(pv);
-    return os;
-}
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index 43a7768451..f9744d7167 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -27,7 +27,8 @@ __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg)
+        kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg,
+                                             const Block2CTileMap& b2c_map)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
@@ -35,12 +36,11 @@ __global__ void
 
     __shared__ uint8_t p_shared[shared_size];
 
-    Block2CTileMap b2c_map{get_block_1d_id()};
-
     GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
         karg, static_cast<void*>(p_shared), b2c_map);
 #else
     ignore = karg;
+    ignore = b2c_map;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -541,6 +541,15 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             make_tuple(Sequence<0, 1>{}, Sequence<2, 3>{}));
     }
 
+    // return block_id to C matrix tile idx (m0, n0) mapping
+    template <typename CGridDesc>
+    __host__ __device__ static constexpr auto MakeCBlockClusterAdaptor(
+        const CGridDesc& c_m_n_grid_desc, index_t /* M01 */, index_t /* N01 */, index_t KBatch)
+    {
+        return BlockToCTileMap_KSplit_M00_N0_M01Adapt<MPerBlock, NPerBlock, CGridDesc>(
+            c_m_n_grid_desc, 8, KBatch);
+    }
+
     __host__ __device__ static constexpr auto
     GetCBlockDescriptor_MBlock_MPerBlock_NBlock_NPerBlock()
     {
@@ -566,28 +575,18 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
     template <bool HasMainKBlockLoop,
               InMemoryDataOperationEnum CGlobalMemoryDataOperation,
               typename Block2CTileMap>
-    __device__ static void Run(const FloatA* p_a_grid,
-                               const FloatB* p_b_grid,
-                               FloatC* p_c_grid,
-                               index_t M,
-                               index_t N,
-                               index_t K,
-                               index_t StrideA,
-                               index_t StrideB,
-                               index_t StrideC,
-                               index_t MPadded,
-                               index_t NPadded,
-                               index_t KPadded,
-                               index_t K0,
-                               index_t k_batch,
+    __device__ static void Run(const Argument& karg,
                                void* __restrict__ p_shared_block,
                                const Block2CTileMap& block_2_ctile_map)
     {
-        const auto a_b_k0_m_k1_grid_desc =
-            MakeAGridDescriptor_KBatch_K0_M_K1(M, MPadded, K, StrideA, k_batch, K0, KPadded);
-        const auto b_b_k0_n_k1_grid_desc =
-            MakeBGridDescriptor_KBatch_K0_N_K1(K, NPadded, N, StrideB, k_batch, K0, KPadded);
-        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(M, N, StrideC);
+        const FloatA* p_a_grid           = karg.p_a_grid;
+        const FloatB* p_b_grid           = karg.p_b_grid;
+        FloatC* p_c_grid                 = karg.p_c_grid;
+        const auto a_b_k0_m_k1_grid_desc = MakeAGridDescriptor_KBatch_K0_M_K1(
+            karg.M, karg.MPadded, karg.K, karg.StrideA, karg.k_batch, karg.K0, karg.KPadded);
+        const auto b_b_k0_n_k1_grid_desc = MakeBGridDescriptor_KBatch_K0_N_K1(
+            karg.K, karg.NPadded, karg.N, karg.StrideB, karg.k_batch, karg.K0, karg.KPadded);
+        const auto c_grid_desc_m_n = MakeCGridDescriptor_M_N(karg.M, karg.N, karg.StrideC);
 
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
@@ -603,7 +602,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
             p_c_grid, c_grid_desc_mblock_mperblock_nblock_nperblock.GetElementSpaceSize());
 
         // divide block work by [KBatch, M, N]
-        const auto block_work_idx = block_2_ctile_map.CalculateBottomIndex();
+        const auto block_work_idx =
+            block_2_ctile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
 
         if(!block_2_ctile_map.ValidCTileIndex(
                block_work_idx,
@@ -1010,34 +1010,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
         }
     }
 
-    template <bool HasMainKBlockLoop,
-              InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-              typename Block2CTileMap>
-    __device__ static void Run(const Argument& karg,
-                               void* __restrict__ p_shared_block,
-                               const Block2CTileMap& block_2_ctile_map)
-    {
-        Run<HasMainKBlockLoop, CGlobalMemoryDataOperation, Block2CTileMap>(karg.p_a_grid,
-                                                                           karg.p_b_grid,
-                                                                           karg.p_c_grid,
-                                                                           karg.M,
-                                                                           karg.N,
-                                                                           karg.K,
-                                                                           karg.StrideA,
-                                                                           karg.StrideB,
-                                                                           karg.StrideC,
-                                                                           karg.MPadded,
-                                                                           karg.NPadded,
-                                                                           karg.KPadded,
-                                                                           karg.K0,
-                                                                           karg.k_batch,
-                                                                           p_shared_block,
-                                                                           block_2_ctile_map);
-    }
-
-    static constexpr auto GetMPerBlock() { return MPerBlock; }
-    static constexpr auto GetNPerBlock() { return NPerBlock; }
-
     static std::string GetTypeString()
     {
         auto str = std::stringstream();
diff --git a/include/ck/utility/sequence.hpp b/include/ck/utility/sequence.hpp
index 63422a10e9..d6bfb2eba1 100644
--- a/include/ck/utility/sequence.hpp
+++ b/include/ck/utility/sequence.hpp
@@ -897,14 +897,3 @@ template <index_t NSize, index_t I>
 using uniform_sequence_gen_t = typename uniform_sequence_gen<NSize, I>::type;
 
 } // namespace ck
-
-template <ck::index_t... Is>
-std::ostream& operator<<(std::ostream& os, const ck::Sequence<Is...>)
-{
-    using S = ck::Sequence<Is...>;
-    os << "{";
-    ck::static_for<0, S::Size() - ck::Number<1>{}, 1>{}(
-        [&](auto i) { os << S::At(i).value << ", "; });
-    os << S::At(S::Size() - ck::Number<1>{}).value << "}";
-    return os;
-}
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
index 2466092124..8f4182a231 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
@@ -14,27 +14,27 @@ namespace ck {
 namespace tensor_operation {
 namespace host {
 
-///
-/// @brief      Reference implementation for forward convolution.
-///
-/// @paragraph
-///             Tensor descriptor in GNCHW/GKCXY/GNKHW dimensional order
-///             Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
-///             as long as dimensions in tensor descriptor is in GNCHW order
-///
-/// @tparam     InDataType               Input tensor data type.
-/// @tparam     WeiDataType              Weights tensor data type.
-/// @tparam     OutDataType              Output tensor data type.
-/// @tparam     InElementwiseOperation   Functor for input tensor elementwise
-///                                      operation.
-/// @tparam     WeiElementwiseOperation  Functor for weights tensor elementwise
-///                                      operation.
-/// @tparam     NDimSpatial  Number of spatial dimensions.
-///
-/// input descriptor in [G, N, C, Do, Ho, Wo] order
-/// weight descriptor in [G, K, C, Z, Y, X] order
-/// output descriptor in [G, N, K, Di, Hi, Wi] order
-/// phyiscal layout is irrelavent
+//
+// @brief      Reference implementation for forward convolution.
+//
+// @paragraph
+//             Tensor descriptor in GNCHW/GKCXY/GNKHW dimensional order
+//             Supports both GNCHW/NGCHW as well as GNHWC/NHWGC physical layout
+//             as long as dimensions in tensor descriptor is in GNCHW order
+//
+// @tparam     InDataType               Input tensor data type.
+// @tparam     WeiDataType              Weights tensor data type.
+// @tparam     OutDataType              Output tensor data type.
+// @tparam     InElementwiseOperation   Functor for input tensor elementwise
+//                                      operation.
+// @tparam     WeiElementwiseOperation  Functor for weights tensor elementwise
+//                                      operation.
+// @tparam     NDimSpatial  Number of spatial dimensions.
+//
+// input descriptor in [G, N, C, Do, Ho, Wo] order
+// weight descriptor in [G, K, C, Z, Y, X] order
+// output descriptor in [G, N, K, Di, Hi, Wi] order
+// phyiscal layout is irrelavent
 template <ck::index_t NDimSpatial,
           typename InDataType,
           typename WeiDataType,
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
index f1bed87ad1..c829e8863d 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -33,26 +33,26 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 // a[m, k] * b[k, n] = e[m, n]
 using device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
-    //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-    //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-    //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-    //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
     // clang-format on
     >;
 
@@ -60,20 +60,20 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 
 using device_grouped_gemm_xdl_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
     // clang-format off
-    //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-    //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-    //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-    //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
-    DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
+        //###################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //###################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //###################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //###################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemm_Xdl<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
index a2762b45c2..8642562fa3 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -33,41 +33,27 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 // a[m, k] * b[k, n] = e[m, n]
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
-    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     PipelineVersion|
-    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                    |
-    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                    |
-    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                    |
-    // Pipline v1
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-
-    // Pipline v2
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // Currently AK1 must equal BK1 !
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
index fc028aa30b..90223fd9bd 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_instance.cpp
@@ -31,41 +31,57 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_kn_mn_irregular_tile_instances = std::tuple<
     // clang-format off
-    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|     PipelineVersion|
-    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                    |
-    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                    |
-    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                    |
-    // Pipline v1
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
-    
-    // Pipline v2
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        // Currently AK1 must equal BK1 !
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 16,16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8>,
+     // DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8>,
+        
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v1>,
+
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Row, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8, PipelineVersion::v2>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
index 729111d054..aa6365cd98 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -32,31 +32,23 @@ static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecializa
 // a[m, k] * b[n, k] = e[m, n]
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // clang-format off
-    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         PipelineVersion|
-    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                        |
-    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                        |
-    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                        |
-    // Pipeline v1
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v1>,
-    // Pipeline v2
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v2>
-
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>
     // clang-format on
     >;
 
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
index 14c82de360..f4460b360b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm/device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_instance.cpp
@@ -31,31 +31,30 @@ static constexpr auto GemmMNKPadding = ck::tensor_operation::device::GemmSpecial
 
 using device_grouped_gemm_xdl_splitk_f16_f16_f16_mk_nk_mn_irregular_tile_instances = std::tuple<
     // clang-format off
-    //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|         PipelineVersion|
-    //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|                        |
-    //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                        |
-    //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                        |
-    // Pipeline v1
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v1>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v1>,
-    
-    // Pipeline v2
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    64,   8,   8,   32,   32,    2,    4,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    64,   8,   8,   32,   32,    1,    2,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
-  //DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    64,   8,   8,   32,   32,    1,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    64,   8,   8,   32,   32,    2,    1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,               8, ck::PipelineVersion::v2>,
-    DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    64,   8,   8,   32,   32,    1,    2,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 8,  8, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8, ck::PipelineVersion::v2>
+        //################################|      A|      B|          Ds|      E| AData| BData| AccData| CShuffle|      DsData| EData|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //################################| Layout| Layout|      Layout| Layout|  Type|  Type|    Type| DataType|        Type|  Type| Elementwise| Elementwise| Elementwise| Spacialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
+        //################################|       |       |            |       |      |      |        |         |            |      |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
+        //################################|       |       |            |       |      |      |        |         |            |      |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   192,    64,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 48, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              4>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,    
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   192,    32,    32,   8,   8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   192,    32,   8,   8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,   256,    32,   8,   8,   32,   32,    1,    4,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    32,    64,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    32,    32,   8,   8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,   128,    64,    64,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>,
+        DeviceGroupedGemmXdlSplitKCShuffle<    Row,    Col, Empty_Tuple,    Row,   F16,   F16,     F32,      F16, Empty_Tuple,   F16, PassThrough, PassThrough, PassThrough, GemmMNKPadding,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,         1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8>
     // clang-format on
     >;
 
diff --git a/script/cmake-ck-dev.sh b/script/cmake-ck-dev.sh
index 7f812792aa..da9e92b862 100755
--- a/script/cmake-ck-dev.sh
+++ b/script/cmake-ck-dev.sh
@@ -12,9 +12,7 @@ cmake
 -save-temps=$PWD"                                                                                 \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=ON                                                                                   \
--D GPU_TARGETS="gfx90a"                                                             \
+-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}
-
-#-D GPU_TARGETS="gfx908;gfx90a;gfx940"                                                             \
diff --git a/test/grouped_gemm/test_grouped_gemm_ut_cases.inc b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
index 0ab7c3245e..d94d140d97 100644
--- a/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
+++ b/test/grouped_gemm/test_grouped_gemm_ut_cases.inc
@@ -4,7 +4,7 @@ TEST_P(RRR_F16_F16_F16, TinyCases)
 {
     const std::vector<int> Ms{0, 1};
     constexpr int N = 768;
-    constexpr int K = 1088;
+    constexpr int K = 544;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -17,9 +17,9 @@ TEST_P(RRR_F16_F16_F16, TinyCases)
 
 TEST_P(RRR_F16_F16_F16, SmallCases)
 {
-    const std::vector<int> Ms{2, 3, 4, 5};
+    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
     constexpr int N = 768;
-    constexpr int K = 1088;
+    constexpr int K = 544;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -34,7 +34,7 @@ TEST_P(RRR_F16_F16_F16, MidCases)
 {
     const std::vector<int> Ms{167, 183, 177, 153, 139, 204};
     constexpr int N = 768;
-    constexpr int K = 1088;
+    constexpr int K = 544;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -49,7 +49,7 @@ TEST_P(RRR_F16_F16_F16, Regular)
 {
     const std::vector<int> Ms{64, 128, 256};
     constexpr int N = 768;
-    constexpr int K = 640;
+    constexpr int K = 320;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -79,7 +79,7 @@ TEST_P(RCR_F16_F16_F16, TinyCases)
 {
     const std::vector<int> Ms{0, 1};
     constexpr int N = 768;
-    constexpr int K = 1088;
+    constexpr int K = 544;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -91,9 +91,9 @@ TEST_P(RCR_F16_F16_F16, TinyCases)
 
 TEST_P(RCR_F16_F16_F16, SmallCases)
 {
-    const std::vector<int> Ms{2, 3, 4, 5};
+    const std::vector<int> Ms{2, 1, 3, 4, 5, 0};
     constexpr int N = 768;
-    constexpr int K = 1088;
+    constexpr int K = 544;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -123,7 +123,7 @@ TEST_P(RCR_F16_F16_F16, Regular)
 {
     const std::vector<int> Ms{32, 64, 128, 256};
     constexpr int N = 768;
-    constexpr int K = 640;
+    constexpr int K = 320;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);
@@ -151,9 +151,9 @@ TEST_P(RCR_F16_F16_F16, MNKPadded)
 
 TEST_P(RRR_F16_F16_F16_LargeK, TestLargeKBatch)
 {
-    const std::vector<int> Ms{127, 150, 188, 210};
+    const std::vector<int> Ms{188, 210};
     constexpr int N = 768;
-    constexpr int K = 8192;
+    constexpr int K = 4096;
 
     const std::vector<int> Ns(Ms.size(), N);
     const std::vector<int> Ks(Ms.size(), K);

From f3b02ecfd21cecd1e808a168efcc4e56e3715078 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 11 Oct 2023 20:29:01 -0500
Subject: [PATCH 168/194] simplified buffer_load/store (#971)

* simplified buffer_load/store

* add bfp8/fp8

* fixed

* fixed buffer_load

* fixed buffer_store

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 include/ck/utility/amd_buffer_addressing.hpp | 848 +++++--------------
 1 file changed, 215 insertions(+), 633 deletions(-)

diff --git a/include/ck/utility/amd_buffer_addressing.hpp b/include/ck/utility/amd_buffer_addressing.hpp
index d8094d84d2..027aa72f18 100644
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -299,6 +299,109 @@ enum struct AmdBufferCoherenceEnum
     GLC_SLC          = 3,
 };
 
+template <index_t N, AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
+__device__ typename vector_type<int8_t, N>::type
+amd_buffer_load_impl_raw(int32x4_t src_wave_buffer_resource,
+                         index_t src_thread_addr_offset,
+                         index_t src_wave_addr_offset)
+{
+    static_assert(N == 1 || N == 2 || N == 4 || N == 8 || N == 16 || N == 32 || N == 64,
+                  "wrong! not implemented");
+
+    if constexpr(N == 1)
+    {
+        return llvm_amdgcn_raw_buffer_load_i8(src_wave_buffer_resource,
+                                              src_thread_addr_offset,
+                                              src_wave_addr_offset,
+                                              static_cast<index_t>(coherence));
+    }
+    else if constexpr(N == 2)
+    {
+
+        int16_t tmp = llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
+
+        return bit_cast<int8x2_t>(tmp);
+    }
+    else if constexpr(N == 4)
+    {
+        int32_t tmp = llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
+                                                      src_thread_addr_offset,
+                                                      src_wave_addr_offset,
+                                                      static_cast<index_t>(coherence));
+
+        return bit_cast<int8x4_t>(tmp);
+    }
+    else if constexpr(N == 8)
+    {
+        int32x2_t tmp = llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
+                                                          src_thread_addr_offset,
+                                                          src_wave_addr_offset,
+                                                          static_cast<index_t>(coherence));
+
+        return bit_cast<int8x8_t>(tmp);
+    }
+    else if constexpr(N == 16)
+    {
+        int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                          src_thread_addr_offset,
+                                                          src_wave_addr_offset,
+                                                          static_cast<index_t>(coherence));
+        return bit_cast<int8x16_t>(tmp);
+    }
+    else if constexpr(N == 32)
+    {
+        int32x4_t tmp0 = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                           src_thread_addr_offset,
+                                                           src_wave_addr_offset,
+                                                           static_cast<index_t>(coherence));
+        int32x4_t tmp1 =
+            llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                              src_thread_addr_offset,
+                                              src_wave_addr_offset + 4 * sizeof(int32_t),
+                                              static_cast<index_t>(coherence));
+        vector_type<int32_t, 8> tmp;
+
+        tmp.AsType<int32x4_t>()(Number<0>{}) = tmp0;
+        tmp.AsType<int32x4_t>()(Number<1>{}) = tmp1;
+
+        return bit_cast<int8x32_t>(tmp);
+    }
+    else if constexpr(N == 64)
+    {
+        int32x4_t tmp0 = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                                           src_thread_addr_offset,
+                                                           src_wave_addr_offset,
+                                                           static_cast<index_t>(coherence));
+        int32x4_t tmp1 =
+            llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                              src_thread_addr_offset,
+                                              src_wave_addr_offset + 4 * sizeof(int32_t),
+                                              static_cast<index_t>(coherence));
+        int32x4_t tmp2 =
+            llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                              src_thread_addr_offset,
+                                              src_wave_addr_offset + 8 * sizeof(int32_t),
+                                              static_cast<index_t>(coherence));
+        int32x4_t tmp3 =
+            llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
+                                              src_thread_addr_offset,
+                                              src_wave_addr_offset + 12 * sizeof(int32_t),
+                                              static_cast<index_t>(coherence));
+
+        vector_type<int32_t, 16> tmp;
+
+        tmp.AsType<int32x4_t>()(Number<0>{}) = tmp0;
+        tmp.AsType<int32x4_t>()(Number<1>{}) = tmp1;
+        tmp.AsType<int32x4_t>()(Number<2>{}) = tmp2;
+        tmp.AsType<int32x4_t>()(Number<3>{}) = tmp3;
+
+        return bit_cast<int8x64_t>(tmp);
+    }
+}
+
 template <typename T,
           index_t N,
           AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
@@ -307,315 +410,116 @@ __device__ typename vector_type<T, N>::type amd_buffer_load_impl(int32x4_t src_w
                                                                  index_t src_wave_addr_offset)
 {
     static_assert(
-        (is_same<T, double>::value && (N == 1 || N == 2 || N == 4)) ||
-            (is_same<T, float>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
-            (is_same<T, half_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
-            (is_same<T, bhalf_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
-            (is_same<T, int32_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
+        (is_same<T, double>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
+            (is_same<T, float>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, half_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, bhalf_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, int32_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, f8_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, bf8_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
             (is_same<T, int8_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)),
         "wrong! not implemented");
 
-    if constexpr(is_same<T, double>::value)
+    using r_t     = typename vector_type<T, N>::type;
+    auto raw_data = amd_buffer_load_impl_raw<sizeof(T) * N, coherence>(
+        src_wave_buffer_resource, src_thread_addr_offset, src_wave_addr_offset);
+    return bit_cast<r_t>(raw_data);
+}
+
+template <index_t N, AmdBufferCoherenceEnum coherence = AmdBufferCoherenceEnum::DefaultCoherence>
+__device__ void
+amd_buffer_store_impl_raw(const typename vector_type<int8_t, N>::type src_thread_data,
+                          int32x4_t dst_wave_buffer_resource,
+                          index_t dst_thread_addr_offset,
+                          index_t dst_wave_addr_offset)
+{
+    static_assert(N == 1 || N == 2 || N == 4 || N == 8 || N == 16 || N == 32 || N == 64,
+                  "wrong! not implemented");
+
+    if constexpr(N == 1)
     {
-        // use fp32 load to mimic fp64 load
-        if constexpr(N == 1)
-        {
-            const float2_t tmp =
-                llvm_amdgcn_raw_buffer_load_fp32x2(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset,
-                                                   static_cast<index_t>(coherence));
-
-            return bit_cast<double>(tmp);
-        }
-        else if constexpr(N == 2)
-        {
-            const float4_t tmp =
-                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset,
-                                                   static_cast<index_t>(coherence));
-
-            return bit_cast<double2_t>(tmp);
-        }
-        else if constexpr(N == 4)
-        {
-            const float4_t f32_0 =
-                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset,
-                                                   static_cast<index_t>(coherence));
-
-            const float4_t f32_1 =
-                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset + 4 * sizeof(float),
-                                                   static_cast<index_t>(coherence));
-            vector_type<double, 4> tmp;
-
-            tmp.AsType<double2_t>()(Number<0>{}) = bit_cast<double2_t>(f32_0);
-            tmp.AsType<double2_t>()(Number<1>{}) = bit_cast<double2_t>(f32_1);
-
-            return tmp.AsType<double4_t>()(Number<0>{});
-        }
+        llvm_amdgcn_raw_buffer_store_i8(src_thread_data,
+                                        dst_wave_buffer_resource,
+                                        dst_thread_addr_offset,
+                                        dst_wave_addr_offset,
+                                        static_cast<index_t>(coherence));
     }
-    else if constexpr(is_same<T, float>::value)
+    else if constexpr(N == 2)
     {
-        if constexpr(N == 1)
-        {
-            return llvm_amdgcn_raw_buffer_load_fp32(src_wave_buffer_resource,
-                                                    src_thread_addr_offset,
-                                                    src_wave_addr_offset,
-                                                    static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            return llvm_amdgcn_raw_buffer_load_fp32x2(src_wave_buffer_resource,
-                                                      src_thread_addr_offset,
-                                                      src_wave_addr_offset,
-                                                      static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            return llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                      src_thread_addr_offset,
-                                                      src_wave_addr_offset,
-                                                      static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-            vector_type<float, 8> tmp;
 
-            tmp.AsType<float4_t>()(Number<0>{}) =
-                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset,
-                                                   static_cast<index_t>(coherence));
-
-            tmp.AsType<float4_t>()(Number<1>{}) =
-                llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset + 4 * sizeof(float),
-                                                   static_cast<index_t>(coherence));
-
-            return tmp.AsType<float8_t>()(Number<0>{});
-        }
+        llvm_amdgcn_raw_buffer_store_i16(bit_cast<int16_t>(src_thread_data),
+                                         dst_wave_buffer_resource,
+                                         dst_thread_addr_offset,
+                                         dst_wave_addr_offset,
+                                         static_cast<index_t>(coherence));
     }
-    else if constexpr(is_same<T, half_t>::value)
+    else if constexpr(N == 4)
     {
-        if constexpr(N == 1)
-        {
-            return llvm_amdgcn_raw_buffer_load_fp16(src_wave_buffer_resource,
-                                                    src_thread_addr_offset,
-                                                    src_wave_addr_offset,
-                                                    static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            return llvm_amdgcn_raw_buffer_load_fp16x2(src_wave_buffer_resource,
-                                                      src_thread_addr_offset,
-                                                      src_wave_addr_offset,
-                                                      static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            return llvm_amdgcn_raw_buffer_load_fp16x4(src_wave_buffer_resource,
-                                                      src_thread_addr_offset,
-                                                      src_wave_addr_offset,
-                                                      static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-            // use fp32 load to mimic fp16 load
-            float4_t tmp = llvm_amdgcn_raw_buffer_load_fp32x4(src_wave_buffer_resource,
-                                                              src_thread_addr_offset,
-                                                              src_wave_addr_offset,
-                                                              static_cast<index_t>(coherence));
-
-            return bit_cast<half8_t>(tmp);
-        }
+        llvm_amdgcn_raw_buffer_store_i32(bit_cast<int32_t>(src_thread_data),
+                                         dst_wave_buffer_resource,
+                                         dst_thread_addr_offset,
+                                         dst_wave_addr_offset,
+                                         static_cast<index_t>(coherence));
     }
-    else if constexpr(is_same<T, bhalf_t>::value)
+    else if constexpr(N == 8)
     {
-        if constexpr(N == 1)
-        {
-            return llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset,
-                                                   static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            return llvm_amdgcn_raw_buffer_load_i16x2(src_wave_buffer_resource,
-                                                     src_thread_addr_offset,
-                                                     src_wave_addr_offset,
-                                                     static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            return llvm_amdgcn_raw_buffer_load_i16x4(src_wave_buffer_resource,
-                                                     src_thread_addr_offset,
-                                                     src_wave_addr_offset,
-                                                     static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
-                                                              src_thread_addr_offset,
-                                                              src_wave_addr_offset,
-                                                              static_cast<index_t>(coherence));
-
-            return bit_cast<bhalf8_t>(tmp);
-        }
+        llvm_amdgcn_raw_buffer_store_i32x2(bit_cast<int32x2_t>(src_thread_data),
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset,
+                                           static_cast<index_t>(coherence));
     }
-    else if constexpr(is_same<T, int32_t>::value)
+    else if constexpr(N == 16)
     {
-        if constexpr(N == 1)
-        {
-            return llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
-                                                   src_thread_addr_offset,
-                                                   src_wave_addr_offset,
-                                                   static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            return llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
-                                                     src_thread_addr_offset,
-                                                     src_wave_addr_offset,
-                                                     static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            return llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
-                                                     src_thread_addr_offset,
-                                                     src_wave_addr_offset,
-                                                     static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-            vector_type<int32_t, 8> tmp;
-
-            tmp.AsType<int32x4_t>()(Number<0>{}) =
-                llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
-                                                  src_thread_addr_offset,
-                                                  src_wave_addr_offset,
-                                                  static_cast<index_t>(coherence));
-
-            tmp.AsType<int32x4_t>()(Number<1>{}) =
-                llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
-                                                  src_thread_addr_offset,
-                                                  src_wave_addr_offset + 4 * sizeof(int32_t),
-                                                  static_cast<index_t>(coherence));
-            return tmp.AsType<int32x8_t>()(Number<0>{});
-        }
+        llvm_amdgcn_raw_buffer_store_i32x4(bit_cast<int32x4_t>(src_thread_data),
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset,
+                                           static_cast<index_t>(coherence));
     }
-    else if constexpr(is_same<T, int8_t>::value)
+    else if constexpr(N == 32)
     {
-        if constexpr(N == 1)
-        {
-            return llvm_amdgcn_raw_buffer_load_i8(src_wave_buffer_resource,
-                                                  src_thread_addr_offset,
-                                                  src_wave_addr_offset,
-                                                  static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-#if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            return llvm_amdgcn_raw_buffer_load_i8x2(src_wave_buffer_resource,
-                                                    src_thread_addr_offset,
-                                                    src_wave_addr_offset,
-                                                    static_cast<index_t>(coherence));
-#else
-            int16_t tmp = llvm_amdgcn_raw_buffer_load_i16(src_wave_buffer_resource,
-                                                          src_thread_addr_offset,
-                                                          src_wave_addr_offset,
-                                                          static_cast<index_t>(coherence));
+        vector_type<int32_t, 8> tmp{bit_cast<int32x8_t>(src_thread_data)};
 
-            return bit_cast<int8x2_t>(tmp);
-#endif
-        }
-        else if constexpr(N == 4)
-        {
-#if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            return llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                    src_thread_addr_offset,
-                                                    src_wave_addr_offset,
-                                                    static_cast<index_t>(coherence));
-#else
-            int32_t tmp = llvm_amdgcn_raw_buffer_load_i32(src_wave_buffer_resource,
-                                                          src_thread_addr_offset,
-                                                          src_wave_addr_offset,
-                                                          static_cast<index_t>(coherence));
+        llvm_amdgcn_raw_buffer_store_i32x4(tmp.template AsType<int32x4_t>()[Number<0>{}],
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset,
+                                           static_cast<index_t>(coherence));
 
-            return bit_cast<int8x4_t>(tmp);
-#endif
-        }
-        else if constexpr(N == 8)
-        {
-#if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            vector_type<int8_t, 8> tmp;
+        llvm_amdgcn_raw_buffer_store_i32x4(tmp.template AsType<int32x4_t>()[Number<1>{}],
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset + sizeof(int32_t) * 4,
+                                           static_cast<index_t>(coherence));
+    }
+    else if constexpr(N == 64)
+    {
+        vector_type<int32_t, 16> tmp{bit_cast<int32x16_t>(src_thread_data)};
 
-            tmp.AsType<int8x4_t>()(Number<0>{}) =
-                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                 src_thread_addr_offset,
-                                                 src_wave_addr_offset,
-                                                 static_cast<index_t>(coherence));
+        llvm_amdgcn_raw_buffer_store_i32x4(tmp.template AsType<int32x4_t>()[Number<0>{}],
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset,
+                                           static_cast<index_t>(coherence));
 
-            tmp.AsType<int8x4_t>()(Number<1>{}) =
-                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                 src_thread_addr_offset,
-                                                 src_wave_addr_offset + 4 * sizeof(int8_t),
-                                                 static_cast<index_t>(coherence));
+        llvm_amdgcn_raw_buffer_store_i32x4(tmp.template AsType<int32x4_t>()[Number<1>{}],
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset + sizeof(int32_t) * 4,
+                                           static_cast<index_t>(coherence));
 
-            return tmp.AsType<int8x8_t>()(Number<0>{});
-#else
-            int32x2_t tmp = llvm_amdgcn_raw_buffer_load_i32x2(src_wave_buffer_resource,
-                                                              src_thread_addr_offset,
-                                                              src_wave_addr_offset,
-                                                              static_cast<index_t>(coherence));
+        llvm_amdgcn_raw_buffer_store_i32x4(tmp.template AsType<int32x4_t>()[Number<2>{}],
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset + sizeof(int32_t) * 8,
+                                           static_cast<index_t>(coherence));
 
-            return bit_cast<int8x8_t>(tmp);
-#endif
-        }
-        else if constexpr(N == 16)
-        {
-#if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            vector_type<int8_t, 16> tmp;
-
-            tmp.AsType<int8x4_t>()(Number<0>{}) =
-                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                 src_thread_addr_offset,
-                                                 src_wave_addr_offset,
-                                                 static_cast<index_t>(coherence));
-
-            tmp.AsType<int8x4_t>()(Number<1>{}) =
-                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                 src_thread_addr_offset,
-                                                 src_wave_addr_offset + 4 * sizeof(int8_t),
-                                                 static_cast<index_t>(coherence));
-
-            tmp.AsType<int8x4_t>()(Number<2>{}) =
-                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                 src_thread_addr_offset,
-                                                 src_wave_addr_offset + 8 * sizeof(int8_t),
-                                                 static_cast<index_t>(coherence));
-
-            tmp.AsType<int8x4_t>()(Number<3>{}) =
-                llvm_amdgcn_raw_buffer_load_i8x4(src_wave_buffer_resource,
-                                                 src_thread_addr_offset,
-                                                 src_wave_addr_offset + 12 * sizeof(int8_t),
-                                                 static_cast<index_t>(coherence));
-
-            return tmp.AsType<int8x16_t>()(Number<0>{});
-#else
-            int32x4_t tmp = llvm_amdgcn_raw_buffer_load_i32x4(src_wave_buffer_resource,
-                                                              src_thread_addr_offset,
-                                                              src_wave_addr_offset,
-                                                              static_cast<index_t>(coherence));
-
-            return bit_cast<int8x16_t>(tmp);
-#endif
-        }
+        llvm_amdgcn_raw_buffer_store_i32x4(tmp.template AsType<int32x4_t>()[Number<3>{}],
+                                           dst_wave_buffer_resource,
+                                           dst_thread_addr_offset,
+                                           dst_wave_addr_offset + sizeof(int32_t) * 12,
+                                           static_cast<index_t>(coherence));
     }
 }
 
@@ -628,255 +532,22 @@ __device__ void amd_buffer_store_impl(const typename vector_type<T, N>::type src
                                       index_t dst_wave_addr_offset)
 {
     static_assert(
-        (is_same<T, double>::value && (N == 1 || N == 2)) ||
-            (is_same<T, float>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
-            (is_same<T, half_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
-            (is_same<T, bhalf_t>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
-            (is_same<T, int32_t>::value && (N == 1 || N == 2 || N == 4)) ||
+        (is_same<T, double>::value && (N == 1 || N == 2 || N == 4 || N == 8)) ||
+            (is_same<T, float>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, half_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, bhalf_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, int32_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, f8_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
+            (is_same<T, bf8_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)) ||
             (is_same<T, int8_t>::value && (N == 1 || N == 2 || N == 4 || N == 8 || N == 16)),
         "wrong! not implemented");
 
-    if constexpr(is_same<T, double>::value)
-    {
-        // use fp32 store to mimic fp64 store
-        if constexpr(N == 1)
-        {
-            llvm_amdgcn_raw_buffer_store_fp32x2(bit_cast<float2_t>(src_thread_data),
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            llvm_amdgcn_raw_buffer_store_fp32x4(bit_cast<float4_t>(src_thread_data),
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-        }
-    }
-    else if constexpr(is_same<T, float>::value)
-    {
-        if constexpr(N == 1)
-        {
-            llvm_amdgcn_raw_buffer_store_fp32(src_thread_data,
-                                              dst_wave_buffer_resource,
-                                              dst_thread_addr_offset,
-                                              dst_wave_addr_offset,
-                                              static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            llvm_amdgcn_raw_buffer_store_fp32x2(src_thread_data,
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            llvm_amdgcn_raw_buffer_store_fp32x4(src_thread_data,
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-            vector_type<float, 8> tmp{src_thread_data};
-            llvm_amdgcn_raw_buffer_store_fp32x4(tmp.AsType<float4_t>()[Number<0>{}],
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-            llvm_amdgcn_raw_buffer_store_fp32x4(tmp.AsType<float4_t>()[Number<1>{}],
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset + 4 * sizeof(float),
-                                                static_cast<index_t>(coherence));
-        }
-    }
-    else if constexpr(is_same<T, half_t>::value)
-    {
-        if constexpr(N == 1)
-        {
-            llvm_amdgcn_raw_buffer_store_fp16(src_thread_data,
-                                              dst_wave_buffer_resource,
-                                              dst_thread_addr_offset,
-                                              dst_wave_addr_offset,
-                                              static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            llvm_amdgcn_raw_buffer_store_fp16x2(src_thread_data,
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            llvm_amdgcn_raw_buffer_store_fp16x4(src_thread_data,
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-#if 0
-            vector_type<half_t, 8> tmp{src_thread_data};
+    using r_t = typename vector_type<int8_t, sizeof(T) * N>::type;
 
-            llvm_amdgcn_raw_buffer_store_fp16x4(tmp.AsType<half4_t>()[Number<0>{}],
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-
-            llvm_amdgcn_raw_buffer_store_fp16x4(tmp.AsType<half4_t>()[Number<1>{}],
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset + 4 * sizeof(half_t),
-                                                static_cast<index_t>(coherence));
-#else
-            llvm_amdgcn_raw_buffer_store_fp32x4(bit_cast<float4_t>(src_thread_data),
-                                                dst_wave_buffer_resource,
-                                                dst_thread_addr_offset,
-                                                dst_wave_addr_offset,
-                                                static_cast<index_t>(coherence));
-#endif
-        }
-    }
-    else if constexpr(is_same<T, bhalf_t>::value)
-    {
-        if constexpr(N == 1)
-        {
-            llvm_amdgcn_raw_buffer_store_i16(src_thread_data,
-                                             dst_wave_buffer_resource,
-                                             dst_thread_addr_offset,
-                                             dst_wave_addr_offset,
-                                             static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            llvm_amdgcn_raw_buffer_store_i16x2(src_thread_data,
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            llvm_amdgcn_raw_buffer_store_i16x4(src_thread_data,
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 8)
-        {
-            vector_type<bhalf_t, 8> tmp{src_thread_data};
-
-            llvm_amdgcn_raw_buffer_store_i16x4(tmp.AsType<bhalf4_t>()[Number<0>{}],
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-
-            llvm_amdgcn_raw_buffer_store_i16x4(tmp.AsType<bhalf4_t>()[Number<1>{}],
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset + 4 * sizeof(bhalf_t),
-                                               static_cast<index_t>(coherence));
-        }
-    }
-    else if constexpr(is_same<T, int32_t>::value)
-    {
-        if constexpr(N == 1)
-        {
-            llvm_amdgcn_raw_buffer_store_i32(src_thread_data,
-                                             dst_wave_buffer_resource,
-                                             dst_thread_addr_offset,
-                                             dst_wave_addr_offset,
-                                             static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-            llvm_amdgcn_raw_buffer_store_i32x2(src_thread_data,
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 4)
-        {
-            llvm_amdgcn_raw_buffer_store_i32x4(src_thread_data,
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-        }
-    }
-    else if constexpr(is_same<T, int8_t>::value)
-    {
-        if constexpr(N == 1)
-        {
-            llvm_amdgcn_raw_buffer_store_i8(src_thread_data,
-                                            dst_wave_buffer_resource,
-                                            dst_thread_addr_offset,
-                                            dst_wave_addr_offset,
-                                            static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 2)
-        {
-#if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            llvm_amdgcn_raw_buffer_store_i8x2(src_thread_data,
-                                              dst_wave_buffer_resource,
-                                              dst_thread_addr_offset,
-                                              dst_wave_addr_offset,
-                                              static_cast<index_t>(coherence));
-#else
-            llvm_amdgcn_raw_buffer_store_i16(bit_cast<int16_t>(src_thread_data),
-                                             dst_wave_buffer_resource,
-                                             dst_thread_addr_offset,
-                                             dst_wave_addr_offset,
-                                             static_cast<index_t>(coherence));
-#endif
-        }
-        else if constexpr(N == 4)
-        {
-#if !CK_WORKAROUND_SWDEV_XXXXXX_INT8_BUFFER_LOAD_STORE_ISSUE
-            llvm_amdgcn_raw_buffer_store_i8x4(src_thread_data,
-                                              dst_wave_buffer_resource,
-                                              dst_thread_addr_offset,
-                                              dst_wave_addr_offset,
-                                              static_cast<index_t>(coherence));
-#else
-            llvm_amdgcn_raw_buffer_store_i32(bit_cast<int32_t>(src_thread_data),
-                                             dst_wave_buffer_resource,
-                                             dst_thread_addr_offset,
-                                             dst_wave_addr_offset,
-                                             static_cast<index_t>(coherence));
-#endif
-        }
-        else if constexpr(N == 8)
-        {
-            llvm_amdgcn_raw_buffer_store_i32x2(bit_cast<int32x2_t>(src_thread_data),
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-        }
-        else if constexpr(N == 16)
-        {
-            llvm_amdgcn_raw_buffer_store_i32x4(bit_cast<int32x4_t>(src_thread_data),
-                                               dst_wave_buffer_resource,
-                                               dst_thread_addr_offset,
-                                               dst_wave_addr_offset,
-                                               static_cast<index_t>(coherence));
-        }
-    }
+    amd_buffer_store_impl_raw<sizeof(T) * N, coherence>(bit_cast<r_t>(src_thread_data),
+                                                        dst_wave_buffer_resource,
+                                                        dst_thread_addr_offset,
+                                                        dst_wave_addr_offset);
 }
 
 template <typename T, index_t N>
@@ -1127,54 +798,14 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 
 #if CK_EXPERIMENTAL_USE_BUFFER_LOAD_OOB_CHECK_OFFSET_TRICK
     uint32_t src_addr_shift = src_thread_element_valid ? 0 : 0x80000000;
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-    if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
-        if constexpr(is_same<scalar_t, f8_t>::value)
-#endif
-#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-            if constexpr(is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-            {
-                auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
-                    src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
-                return bit_cast<vector_t>(tmp);
-            }
-            else
-            {
-#endif
-                return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
-                    src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-            }
-#endif
+    return amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+        src_wave_buffer_resource, src_addr_shift + src_thread_addr_offset, 0);
+
 #else
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-    if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
-        if constexpr(is_same<scalar_t, f8_t>::value)
-#endif
-#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-            if constexpr(is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-            {
-                auto tmp = amd_buffer_load_impl<int8_t, vector_size, coherence>(
-                    src_wave_buffer_resource, src_thread_addr_offset, 0);
-                return src_thread_element_valid ? bit_cast<vector_t>(tmp) : vector_t(0);
-            }
-            else
-            {
-#endif
-                vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
-                    src_wave_buffer_resource, src_thread_addr_offset, 0);
-                return src_thread_element_valid ? tmp : vector_t(0);
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-            }
-#endif
+
+    vector_t tmp = amd_buffer_load_impl<scalar_t, vector_size, coherence>(
+        src_wave_buffer_resource, src_thread_addr_offset, 0);
+    return src_thread_element_valid ? tmp : vector_t(0);
 #endif
 }
 
@@ -1232,62 +863,13 @@ __device__ void amd_buffer_store(const typename vector_type_maker<T, N>::type::t
 
 #if CK_EXPERIMENTAL_USE_BUFFER_STORE_OOB_CHECK_OFFSET_TRICK
     uint32_t dst_addr_shift = dst_thread_element_valid ? 0 : 0x80000000;
-
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-    if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
-        if constexpr(is_same<scalar_t, f8_t>::value)
-#endif
-#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-            if constexpr(is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-            {
-                auto tmp = bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
-                    src_thread_data);
-                amd_buffer_store_impl<int8_t, vector_size, coherence>(
-                    tmp, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
-            }
-            else
-            {
-#endif
-                amd_buffer_store_impl<scalar_t, vector_size, coherence>(src_thread_data,
-                                                                        dst_wave_buffer_resource,
-                                                                        dst_addr_shift +
-                                                                            dst_thread_addr_offset,
-                                                                        0);
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-            }
-#endif
+    amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+        src_thread_data, dst_wave_buffer_resource, dst_addr_shift + dst_thread_addr_offset, 0);
 #else
     if(dst_thread_element_valid)
     {
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-        if constexpr(is_same<scalar_t, f8_t>::value || is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 && !defined CK_ENABLE_BF8
-            if constexpr(is_same<scalar_t, f8_t>::value)
-#endif
-#if !defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-                if constexpr(is_same<scalar_t, bf8_t>::value)
-#endif
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-                {
-                    auto tmp =
-                        bit_cast<typename vector_type_maker<int8_t, vector_size>::type::type>(
-                            src_thread_data);
-                    amd_buffer_store_impl<int8_t, vector_size, coherence>(
-                        tmp, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
-                }
-                else
-                {
-#endif
-                    amd_buffer_store_impl<scalar_t, vector_size, coherence>(
-                        src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-                }
-#endif
+        amd_buffer_store_impl<scalar_t, vector_size, coherence>(
+            src_thread_data, dst_wave_buffer_resource, dst_thread_addr_offset, 0);
     }
 #endif
 }

From a3c80265185ae1a8489d605675a71af206c5b5eb Mon Sep 17 00:00:00 2001
From: "dependabot[bot]" <49699333+dependabot[bot]@users.noreply.github.com>
Date: Thu, 12 Oct 2023 12:43:54 -0600
Subject: [PATCH 169/194] Bump gitpython from 3.1.31 to 3.1.35 in /docs/sphinx
 (#898)

Bumps [gitpython](https://github.com/gitpython-developers/GitPython) from 3.1.31 to 3.1.35.
- [Release notes](https://github.com/gitpython-developers/GitPython/releases)
- [Changelog](https://github.com/gitpython-developers/GitPython/blob/main/CHANGES)
- [Commits](https://github.com/gitpython-developers/GitPython/compare/3.1.31...3.1.35)

---
updated-dependencies:
- dependency-name: gitpython
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
---
 docs/sphinx/requirements.txt | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/sphinx/requirements.txt b/docs/sphinx/requirements.txt
index d5f67eeb58..1a648a7493 100644
--- a/docs/sphinx/requirements.txt
+++ b/docs/sphinx/requirements.txt
@@ -42,7 +42,7 @@ fastjsonschema==2.18.0
     # via rocm-docs-core
 gitdb==4.0.10
     # via gitpython
-gitpython==3.1.31
+gitpython==3.1.35
     # via rocm-docs-core
 idna==3.4
     # via requests
@@ -103,7 +103,7 @@ requests==2.28.2
     # via
     #   pygithub
     #   sphinx
-rocm-docs-core>=0.20.0
+rocm-docs-core==0.24.0
     # via -r requirements.in
 six==1.16.0
     # via

From 2ce9b56c64e157c5828684230b3629f4c37d65da Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Fri, 13 Oct 2023 15:11:43 -0500
Subject: [PATCH 170/194] add vector_type support into thread_copy_v3r1 (#969)

* add vector_type support into thread_copy_v3r1

* remove unncessary type_convert

* fixed datatype

* fixed dataType

* changed API with is_packx_invocable

* changed example

* add missing cmake file

* fixed ci

* fixed cmake

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/60_gemm_multi_ABD/CMakeLists.txt      |  10 +
 .../gemm_multi_ABD_xdl_fp16.cpp               | 363 ++++++++++++++++++
 ...ridwise_gemm_multiple_abd_xdl_cshuffle.hpp |   5 +-
 .../threadwise_tensor_slice_transfer_v3r1.hpp |  77 ++--
 .../threadwise_tensor_slice_transfer_v7r2.hpp | 138 +++----
 include/ck/utility/is_detected.hpp            |   9 +
 6 files changed, 491 insertions(+), 111 deletions(-)
 create mode 100644 example/60_gemm_multi_ABD/CMakeLists.txt
 create mode 100644 example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp

diff --git a/example/60_gemm_multi_ABD/CMakeLists.txt b/example/60_gemm_multi_ABD/CMakeLists.txt
new file mode 100644
index 0000000000..6dbada7ea6
--- /dev/null
+++ b/example/60_gemm_multi_ABD/CMakeLists.txt
@@ -0,0 +1,10 @@
+if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
+    list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+    set(target 0)
+    foreach(gpu IN LISTS GPU_TARGETS)
+        if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+            add_example_executable(example_gemm_multi_ABD_xdl_fp16 gemm_multi_ABD_xdl_fp16.cpp)
+            set(target 1)
+        endif()
+    endforeach()
+endif()
diff --git a/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
new file mode 100644
index 0000000000..13ba639814
--- /dev/null
+++ b/example/60_gemm_multi_ABD/gemm_multi_ABD_xdl_fp16.cpp
@@ -0,0 +1,363 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
+#include "ck/library/utility/check_err.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using ADataType        = F16;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F16;
+using EDataType        = F16;
+
+using ALayout = Row;
+using BLayout = Col;
+using DLayout = Row;
+using ELayout = Row;
+
+struct AddScale
+{
+    static constexpr auto I0 = ck::Number<0>{};
+    static constexpr auto I1 = ck::Number<1>{};
+    static constexpr auto I2 = ck::Number<2>{};
+    static constexpr auto I3 = ck::Number<3>{};
+
+    __host__ __device__ constexpr void
+    operator()(ck::half4_t& a, const ck::half4_t& a0, const ck::half4_t& a1) const
+    {
+        const auto a0_v_t = ck::vector_type<ck::half_t, 4>{a0};
+        const auto a1_v_t = ck::vector_type<ck::half_t, 4>{a1};
+
+        auto r_v_t = ck::vector_type<ck::half_t, 4>{};
+
+        r_v_t.AsType<ck::half_t>()(I0) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I0] + a1_v_t.AsType<ck::half_t>()[I0]);
+        r_v_t.AsType<ck::half_t>()(I1) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I1] + a1_v_t.AsType<ck::half_t>()[I1]);
+        r_v_t.AsType<ck::half_t>()(I2) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I2] + a1_v_t.AsType<ck::half_t>()[I2]);
+        r_v_t.AsType<ck::half_t>()(I3) =
+            scale * (a0_v_t.AsType<ck::half_t>()[I3] + a1_v_t.AsType<ck::half_t>()[I3]);
+
+        a = r_v_t.AsType<ck::half4_t>()[I0];
+    }
+
+    __host__ __device__ constexpr void
+    operator()(ck::half_t& a, const ck::half_t& a0, const ck::half_t& a1) const
+    {
+        a = scale * (a0 + a1);
+    }
+
+    // this attribute controls the copy_function applying element_wise_op with
+    // pack4_data
+    constexpr const static bool is_pack4_invocable = true;
+
+    float scale = 1.0;
+};
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
+        ck::half_t& e, const float& c, const ck::half_t& d) const
+    {
+        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
+    };
+
+    float alpha_;
+    float beta_;
+};
+
+using AElementOp   = AddScale;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultipleABD_Xdl_CShuffle<
+    ck::Tuple<ALayout, ALayout>,
+    ck::Tuple<BLayout>,
+    ck::Tuple<DLayout>,
+    ELayout,
+    ck::Tuple<ADataType, ADataType>,
+    ck::Tuple<BDataType>,
+    AccDataType,
+    CShuffleDataType,
+    ck::Tuple<DDataType>,
+    EDataType,
+    AElementOp,
+    BElementOp,
+    CDEElementOp,
+    GemmSpec,
+    1,
+    256,
+    256,
+    128,
+    32,
+    8,
+    8,
+    32,
+    32,
+    4,
+    2,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    1,
+    1,
+    S<1, 32, 1, 8>,
+    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    // GEMM shape
+    ck::index_t M = 3840;
+    ck::index_t N = 4096;
+    ck::index_t K = 4096;
+
+    ck::index_t StrideA = 4096;
+    ck::index_t StrideB = 4096;
+    ck::index_t StrideD = 4096;
+    ck::index_t StrideE = 4096;
+
+    float alpha = 1.0f;
+    float beta  = 1.0f;
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else if(argc == 6)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        alpha = std::stof(argv[4]);
+        beta  = std::stof(argv[5]);
+    }
+    else if(argc == 13)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+
+        M = std::stoi(argv[4]);
+        N = std::stoi(argv[5]);
+        K = std::stoi(argv[6]);
+
+        StrideA = std::stoi(argv[7]);
+        StrideB = std::stoi(argv[8]);
+        StrideD = std::stoi(argv[9]);
+        StrideE = std::stoi(argv[10]);
+
+        alpha = std::stof(argv[11]);
+        beta  = std::stof(argv[12]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        printf("arg4 to 9: M (256x), N(128x), K(32x), StrideA, StrideB, StrideD, StrideE, alpha, "
+               "beta\n");
+        exit(0);
+    }
+
+    auto f_host_tensor_descriptor =
+        [](std::size_t row, std::size_t col, std::size_t stride, auto layout) {
+            using namespace ck::literals;
+
+            if(std::is_same<decltype(layout), ck::tensor_layout::gemm::RowMajor>::value)
+            {
+                return HostTensorDescriptor({row, col}, {stride, 1_uz});
+            }
+            else
+            {
+                return HostTensorDescriptor({row, col}, {1_uz, stride});
+            }
+        };
+
+    Tensor<ADataType> a0_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<ADataType> a1_m_k(f_host_tensor_descriptor(M, K, StrideA, ALayout{}));
+    Tensor<BDataType> b_k_n(f_host_tensor_descriptor(K, N, StrideB, BLayout{}));
+    Tensor<DDataType> d_m_n(f_host_tensor_descriptor(M, N, StrideD, DLayout{}));
+    Tensor<EDataType> e_m_n_host_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+    Tensor<EDataType> e_m_n_device_result(f_host_tensor_descriptor(M, N, StrideE, ELayout{}));
+
+    std::cout << "a0_m_k: " << a0_m_k.mDesc << std::endl;
+    std::cout << "a1_m_k: " << a1_m_k.mDesc << std::endl;
+    std::cout << "b_k_n: " << b_k_n.mDesc << std::endl;
+    std::cout << "d_m_n: " << d_m_n.mDesc << std::endl;
+    std::cout << "e_m_n: " << e_m_n_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_2<ADataType>{-5, 5});
+        b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_2<DDataType>{-5, 5});
+        break;
+    default:
+        a0_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        a1_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_m_n.GenerateTensorValue(GeneratorTensor_3<DDataType>{-0.5, 0.5});
+    }
+
+    DeviceMem a0_device_buf(sizeof(ADataType) * a0_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem a1_device_buf(sizeof(ADataType) * a1_m_k.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_k_n.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_m_n.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_m_n_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_m_k.mData.data());
+    a1_device_buf.ToDevice(a1_m_k.mData.data());
+    b_device_buf.ToDevice(b_k_n.mData.data());
+    d_device_buf.ToDevice(d_m_n.mData.data());
+    e_device_buf.ToDevice(e_m_n_device_result.mData.data());
+
+    auto a_element_op   = AElementOp{0.2};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument =
+        device_op.MakeArgument(std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
+                                                          a1_device_buf.GetDeviceBuffer()},
+                               std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
+                               std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+                               e_device_buf.GetDeviceBuffer(),
+                               M,
+                               N,
+                               K,
+                               std::array<ck::index_t, 2>{StrideA, StrideA},
+                               std::array<ck::index_t, 1>{StrideB},
+                               std::array<ck::index_t, 1>{StrideD},
+                               StrideE,
+                               a_element_op,
+                               b_element_op,
+                               cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_gemm with the specified compilation parameters does "
+            "not support this GEMM problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N;
+
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
+
+    e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+    if(do_verification)
+    {
+        Tensor<CShuffleDataType> c_m_n({M, N});
+
+        Tensor<ADataType> a_m_k({M, K});
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int k = 0; k < K; ++k)
+            {
+                a_element_op(a_m_k(m, k), a0_m_k(m, k), a1_m_k(m, k));
+            }
+        }
+
+        using ReferenceGemmInstance = ck::tensor_operation::host::ReferenceGemm<ADataType,
+                                                                                BDataType,
+                                                                                CShuffleDataType,
+                                                                                AccDataType,
+                                                                                PassThrough,
+                                                                                BElementOp,
+                                                                                PassThrough>;
+        auto ref_gemm               = ReferenceGemmInstance{};
+        auto ref_invoker            = ref_gemm.MakeInvoker();
+
+        auto ref_argument =
+            ref_gemm.MakeArgument(a_m_k, b_k_n, c_m_n, PassThrough{}, b_element_op, PassThrough{});
+
+        ref_invoker.Run(ref_argument);
+
+        for(int m = 0; m < M; ++m)
+        {
+            for(int n = 0; n < N; ++n)
+            {
+                cde_element_op(e_m_n_host_result(m, n), c_m_n(m, n), d_m_n(m, n));
+            }
+        }
+
+        e_device_buf.FromDevice(e_m_n_device_result.mData.data());
+
+        return ck::utils::check_err(e_m_n_device_result, e_m_n_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
index 10a0842dac..89424468c6 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp
@@ -428,7 +428,7 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
             [&](auto i) {
                 using ALayout = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
 
-                return MakeAGridDescriptor_M_N<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
+                return MakeAGridDescriptor_M_K<ALayout, GemmSpec>(MRaws[i], KRaws[i], AsStride[i]);
             },
             Number<NumATensor>{});
     }
@@ -656,7 +656,8 @@ struct GridwiseGemmMultipleABD_xdl_cshuffle
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ComputeDataType,
+            ComputeDataType, // ComputeDataType for A
+            ComputeDataType, // ComputeDataType for B
             AccDataType,
             decltype(a_block_desc_ak0_m_ak1),
             decltype(b_block_desc_bk0_n_bk1),
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
index 32ea8ae39c..0b2300fbe5 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v3r1.hpp
@@ -9,6 +9,7 @@
 #include "ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp"
 #include "ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer.hpp"
 #include "ck/tensor/static_tensor.hpp"
+#include "ck/utility/is_detected.hpp"
 
 namespace ck {
 
@@ -211,10 +212,44 @@ struct ThreadwiseTensorSliceTransfer_v3r1
             auto src_vector_container = src_vector_type{
                 src_buf.template Get<src_vector_t>(src_coord_.GetOffset(), is_src_valid)};
 
+            using dst_vector_type = vector_type_maker_t<DstData, SrcScalarPerVector>;
+            using dst_vector_t    = typename dst_vector_type::type;
+            dst_vector_type op_r_v;
+
+            constexpr auto get_elem_op_vec_len = []() {
+                if constexpr(is_detected<is_pack8_invocable_t, decltype(src_element_op_)>::value)
+                {
+                    if constexpr(decltype(src_element_op_)::is_pack8_invocable)
+                        return math::min(8, SrcScalarPerVector);
+                }
+                if constexpr(is_detected<is_pack4_invocable_t, decltype(src_element_op_)>::value)
+                {
+                    if constexpr(decltype(src_element_op_)::is_pack4_invocable)
+                        return math::min(4, SrcScalarPerVector);
+                }
+                if constexpr(is_detected<is_pack2_invocable_t, decltype(src_element_op_)>::value)
+                {
+                    if constexpr(decltype(src_element_op_)::is_pack2_invocable)
+                        return math::min(2, SrcScalarPerVector);
+                }
+                return 1;
+            };
+
+            constexpr index_t elem_op_vec_len = get_elem_op_vec_len();
+
+            using src_elem_op_vec_t = typename vector_type<SrcData, elem_op_vec_len>::type;
+            using dst_elem_op_vec_t = typename vector_type<DstData, elem_op_vec_len>::type;
+
+            static_for<0, SrcScalarPerVector / elem_op_vec_len, 1>{}([&](auto idx) {
+                // apply the src elementwise op and convert to DstData under the hood if needed
+                src_element_op_(op_r_v.template AsType<dst_elem_op_vec_t>()(idx),
+                                src_vector_container.template AsType<src_elem_op_vec_t>()[idx]);
+            });
+
             // copy data from src_vector_container into src_thread_scratch_
             src_thread_scratch_tuple_(thread_scratch_id)
-                .template SetAsType<src_vector_t>(
-                    src_data_idx_seq, src_vector_container.template AsType<src_vector_t>()[I0]);
+                .template SetAsType<dst_vector_t>(src_data_idx_seq,
+                                                  op_r_v.template AsType<dst_vector_t>()[I0]);
 
             constexpr auto move_on_dim = [&]() constexpr
             {
@@ -267,19 +302,15 @@ struct ThreadwiseTensorSliceTransfer_v3r1
     {
 #if !CK_EXPERIMENTAL_USE_IN_REGISTER_SUB_DWORD_TRANSPOSE
         static_ford<SliceLengths>{}([&](auto idx) {
-            // convert from SrcData to DstData here
-            dst_thread_scratch_(idx) =
-                type_convert<DstData>(src_thread_scratch_tuple_[thread_scratch_id][idx]);
+            dst_thread_scratch_(idx) = src_thread_scratch_tuple_[thread_scratch_id][idx];
         });
 #else
         // sub-dword transpose between src_thread_scratch_ and dst_thread_scratch_
         // TODO make this logic more generic for more sub-dword datatype
         if constexpr(SrcVectorDim != DstVectorDim &&
-                     ((is_same<half_t, remove_cvref_t<SrcData>>::value &&
-                       is_same<half_t, remove_cvref_t<DstData>>::value &&
+                     ((is_same<half_t, remove_cvref_t<DstData>>::value &&
                        SrcScalarPerVector % 2 == 0 && DstScalarPerVector % 2 == 0) ||
-                      (is_same<int8_t, remove_cvref_t<SrcData>>::value &&
-                       is_same<int8_t, remove_cvref_t<DstData>>::value &&
+                      (is_same<int8_t, remove_cvref_t<DstData>>::value &&
                        SrcScalarPerVector % 4 == 0 && DstScalarPerVector % 4 == 0)))
         {
             // each transpose does
@@ -313,7 +344,7 @@ struct ThreadwiseTensorSliceTransfer_v3r1
                 constexpr auto data_idx_seq = generate_sequence_v2(
                     [&](auto i) { return Number<data_idx[i]>{}; }, Number<nDim>{});
 
-                using src_vector_t = vector_type_maker_t<SrcData, SrcScalarPerVector>;
+                using src_vector_t = vector_type_maker_t<DstData, SrcScalarPerVector>;
                 using dst_vector_t = vector_type_maker_t<DstData, DstScalarPerVector>;
 
                 // get DstScalarPerVector # of read-only references to src vectors from
@@ -336,17 +367,16 @@ struct ThreadwiseTensorSliceTransfer_v3r1
                     Number<num_dst_vector>{});
 
                 // do data transpose
-                transpose_vectors<SrcData, DstScalarPerVector, SrcScalarPerVector>{}(
+                transpose_vectors<DstData, DstScalarPerVector, SrcScalarPerVector>{}(
                     src_vector_refs, dst_vector_refs);
             });
         }
-
-        static_ford<SliceLengths>{}([&](auto idx) {
-            // apply the src elementwise op and convert to DstData under the hood if needed
-            DstData dst_v;
-            src_element_op_(dst_v, src_thread_scratch_tuple_[thread_scratch_id][idx]);
-            dst_thread_scratch_(idx) = dst_v;
-        });
+        else
+        {
+            static_ford<SliceLengths>{}([&](auto idx) {
+                dst_thread_scratch_(idx) = src_thread_scratch_tuple_[thread_scratch_id][idx];
+            });
+        }
 #endif
     }
 
@@ -761,11 +791,12 @@ struct ThreadwiseTensorSliceTransfer_v3r1
     static constexpr auto src_thread_scratch_desc_ = decltype(GetSrcThreadScratchDescriptor()){};
     static constexpr auto dst_thread_scratch_desc_ = decltype(GetDstThreadScratchDescriptor()){};
 
-    using SrcThreadScratch = StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
-                                                             SrcData,
-                                                             SrcScalarPerVector,
-                                                             decltype(src_thread_scratch_desc_),
-                                                             true>;
+    using SrcThreadScratch =
+        StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
+                                        DstData, // apply data_convert with SrcThreadScratch
+                                        SrcScalarPerVector,
+                                        decltype(src_thread_scratch_desc_),
+                                        true>;
 
     using DstThreadScratch = StaticTensorTupleOfVectorBuffer<AddressSpaceEnum::Vgpr,
                                                              DstData,
diff --git a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp
index 003a2b96d9..299a4b9e7d 100644
--- a/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp
+++ b/include/ck/tensor_operation/gpu/thread/threadwise_tensor_slice_transfer_v7r2.hpp
@@ -132,9 +132,6 @@ struct ThreadwiseTensorSliceTransfer_v7r2
             Number<num>{});
     }
 
-    template <typename T>
-    using has_vec_len = decltype(std::declval<T&>().vec_len);
-
     // SrcDescs: Tuple<const SrcDesc0&, const SrcDesc1&, ...>
     // SrcBuffers: Tuple<const SrcBuffer0&, const SrcBuffer1&, ...>
     template <typename SrcBuffers,
@@ -159,94 +156,63 @@ struct ThreadwiseTensorSliceTransfer_v7r2
                                                            is_src_valid);
             });
 
-            if constexpr(is_detected<has_vec_len, decltype(element_op_)>::value)
-            {
-                constexpr auto elem_op_vec_len = decltype(element_op_)::vec_len;
+            constexpr auto get_elem_op_vec_len = []() {
+                if constexpr(is_detected<is_pack8_invocable_t, decltype(element_op_)>::value)
+                {
+                    if constexpr(decltype(element_op_)::is_pack8_invocable)
+                        return math::min(8, SrcScalarPerVector);
+                }
+                if constexpr(is_detected<is_pack4_invocable_t, decltype(element_op_)>::value)
+                {
+                    if constexpr(decltype(element_op_)::is_pack4_invocable)
+                        return math::min(4, SrcScalarPerVector);
+                }
+                if constexpr(is_detected<is_pack2_invocable_t, decltype(element_op_)>::value)
+                {
+                    if constexpr(decltype(element_op_)::is_pack2_invocable)
+                        return math::min(2, SrcScalarPerVector);
+                }
+                return 1;
+            };
 
-                static_assert(is_same<remove_cvref_t<decltype(elem_op_vec_len)>, index_t>::value,
-                              "vec_len in element_op_ type is not index_t");
+            constexpr index_t elem_op_vec_len = get_elem_op_vec_len();
 
-                static_assert(elem_op_vec_len == 1 || elem_op_vec_len == 2 ||
-                                  elem_op_vec_len == 4 || elem_op_vec_len == 8,
-                              "vec_len in element_op_ must be 1, 2, 4, 8");
+            // apply pointwise function
+            static_for<0, SrcScalarPerVector / elem_op_vec_len, 1>{}([&](auto i) {
+                // get reference to src data
+                const auto src_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto iSrc) -> const auto& {
+                        using SrcData = remove_cvref_t<tuple_element_t<iSrc.value, SrcDatas>>;
 
-                static_assert(SrcScalarPerVector % elem_op_vec_len == 0,
-                              "vec_len in element_op_ cannot be divided by SrcScalarPerVector!");
+                        using elem_op_vec_t = typename vector_type<SrcData, elem_op_vec_len>::type;
+
+                        return src_vectors[iSrc].template AsType<elem_op_vec_t>()[i];
+                    },
+                    Number<nSrc>{});
+
+                // get reference to dst data
+                auto dst_data_refs = generate_tie(
+                    // return type should be lvalue
+                    [&](auto iDst) -> auto& {
+                        using DstData = remove_cvref_t<tuple_element_t<iDst.value, DstDatas>>;
+
+                        using elem_op_vec_t = typename vector_type<DstData, elem_op_vec_len>::type;
+
+                        return dst_vectors(iDst).template AsType<elem_op_vec_t>()(i);
+                    },
+                    Number<nDst>{});
 
                 // apply pointwise function
-                static_for<0, SrcScalarPerVector / elem_op_vec_len, 1>{}([&](auto i) {
-                    // get reference to src data
-                    const auto src_data_refs = generate_tie(
-                        // return type should be lvalue
-                        [&](auto iSrc) -> const auto& {
-                            using SrcData = remove_cvref_t<tuple_element_t<iSrc.value, SrcDatas>>;
-
-                            using elem_op_vec_t =
-                                typename vector_type<SrcData, elem_op_vec_len>::type;
-
-                            return src_vectors[iSrc].template AsType<elem_op_vec_t>()[i];
-                        },
-                        Number<nSrc>{});
-
-                    // get reference to dst data
-                    auto dst_data_refs = generate_tie(
-                        // return type should be lvalue
-                        [&](auto iDst) -> auto& {
-                            using DstData = remove_cvref_t<tuple_element_t<iDst.value, DstDatas>>;
-
-                            using elem_op_vec_t =
-                                typename vector_type<DstData, elem_op_vec_len>::type;
-
-                            return dst_vectors(iDst).template AsType<elem_op_vec_t>()(i);
-                        },
-                        Number<nDst>{});
-
-                    // apply pointwise function
-                    // pointwise function signature:
-                    // element_op_(dst_data_refs[I0],
-                    //             dst_data_refs[I1],
-                    //             ...,
-                    //             src_data_refs[I0],
-                    //             src_data_refs[I1],
-                    //             ...)
-                    unpack2(element_op_, dst_data_refs, src_data_refs);
-                });
-            }
-            else
-            {
-                // apply pointwise function
-                static_for<0, SrcScalarPerVector, 1>{}([&](auto i) {
-                    // get reference to src data
-                    const auto src_data_refs = generate_tie(
-                        // return type should be lvalue
-                        [&](auto iSrc) -> const auto& {
-                            using SrcData = remove_cvref_t<tuple_element_t<iSrc.value, SrcDatas>>;
-
-                            return src_vectors[iSrc].template AsType<SrcData>()[i];
-                        },
-                        Number<nSrc>{});
-
-                    // get reference to dst data
-                    auto dst_data_refs = generate_tie(
-                        // return type should be lvalue
-                        [&](auto iDst) -> auto& {
-                            using DstData = remove_cvref_t<tuple_element_t<iDst.value, DstDatas>>;
-
-                            return dst_vectors(iDst).template AsType<DstData>()(i);
-                        },
-                        Number<nDst>{});
-
-                    // apply pointwise function
-                    // pointwise function signature:
-                    // element_op_(dst_data_refs[I0],
-                    //             dst_data_refs[I1],
-                    //             ...,
-                    //             src_data_refs[I0],
-                    //             src_data_refs[I1],
-                    //             ...)
-                    unpack2(element_op_, dst_data_refs, src_data_refs);
-                });
-            }
+                // pointwise function signature:
+                // element_op_(dst_data_refs[I0],
+                //             dst_data_refs[I1],
+                //             ...,
+                //             src_data_refs[I0],
+                //             src_data_refs[I1],
+                //             ...)
+                unpack2(element_op_, dst_data_refs, src_data_refs);
+            });
 
             dst_vectors_tuple_(iAccess) = dst_vectors;
 
diff --git a/include/ck/utility/is_detected.hpp b/include/ck/utility/is_detected.hpp
index 31a4bf24d6..7a324a6c45 100644
--- a/include/ck/utility/is_detected.hpp
+++ b/include/ck/utility/is_detected.hpp
@@ -31,4 +31,13 @@ struct nonesuch
 template <template <class...> class Op, class... Args>
 using is_detected = typename detail::detector<nonesuch, void, Op, Args...>::value_t;
 
+template <typename T>
+using is_pack2_invocable_t = decltype(std::declval<T&>().is_pack2_invocable);
+
+template <typename T>
+using is_pack4_invocable_t = decltype(std::declval<T&>().is_pack4_invocable);
+
+template <typename T>
+using is_pack8_invocable_t = decltype(std::declval<T&>().is_pack8_invocable);
+
 } // namespace ck

From fa753f27ba774e0b831433dc6e3d2576ce8616ce Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Fri, 13 Oct 2023 16:27:11 -0500
Subject: [PATCH 171/194] Add splitk gemm fp16 @ fp16 with fp8 compute
 instances (#983)

* Add ComputeType

* Update for compatibility

* Add instances

* Update profiler api
---
 .../gpu/device/device_gemm_splitk.hpp         |  9 +-
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp |  3 +-
 .../gpu/gemm_splitk.hpp                       | 56 ++++++++++++-
 .../gpu/gemm_splitk/CMakeLists.txt            |  6 +-
 ...f16_f16_f16_comp_fp8_km_kn_mn_instance.cpp | 62 ++++++++++++++
 ...f16_f16_f16_comp_fp8_km_nk_mn_instance.cpp | 62 ++++++++++++++
 ...f16_f16_f16_comp_fp8_mk_kn_mn_instance.cpp | 83 +++++++++++++++++++
 ...f16_f16_f16_comp_fp8_mk_nk_mn_instance.cpp | 79 ++++++++++++++++++
 .../profiler/profile_gemm_splitk_impl.hpp     |  9 +-
 profiler/src/profile_gemm_splitk.cpp          | 60 +++++++++-----
 10 files changed, 398 insertions(+), 31 deletions(-)
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_nk_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_kn_mn_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_nk_mn_instance.cpp

diff --git a/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp b/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
index 6407aa7e09..dfc27e726e 100644
--- a/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_gemm_splitk.hpp
@@ -20,7 +20,8 @@ template <typename ALayout,
           typename CDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CElementwiseOperation>
+          typename CElementwiseOperation,
+          typename ComputeType = CDataType>
 struct DeviceGemmSplitK : public BaseOperator
 {
     virtual std::unique_ptr<BaseArgument> MakeArgumentPointer(const void* p_a,
@@ -48,7 +49,8 @@ template <typename ALayout,
           typename CDataType,
           typename AElementwiseOperation,
           typename BElementwiseOperation,
-          typename CElementwiseOperation>
+          typename CElementwiseOperation,
+          typename ComputeType = CDataType>
 using DeviceGemmSplitKPtr = std::unique_ptr<DeviceGemmSplitK<ALayout,
                                                              BLayout,
                                                              CLayout,
@@ -57,7 +59,8 @@ using DeviceGemmSplitKPtr = std::unique_ptr<DeviceGemmSplitK<ALayout,
                                                              CDataType,
                                                              AElementwiseOperation,
                                                              BElementwiseOperation,
-                                                             CElementwiseOperation>>;
+                                                             CElementwiseOperation,
+                                                             ComputeType>>;
 
 } // namespace device
 } // namespace tensor_operation
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index e7365daed2..d20b008b88 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -69,7 +69,8 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                              CDataType,
                                                              AElementwiseOperation,
                                                              BElementwiseOperation,
-                                                             CElementwiseOperation>
+                                                             CElementwiseOperation,
+                                                             ComputeType>
 {
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index 4cba0875d9..e14df2d750 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -98,6 +98,26 @@ void add_device_gemm_xdl_splitk_f16_f8_f16_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemmSplitK<Row, Col, Row, F16, F8, F16, PassThrough, PassThrough, PassThrough>>>&
         instances);
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances);
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances);
 #endif
 
 template <typename ADataType,
@@ -105,7 +125,8 @@ template <typename ADataType,
           typename CDataType,
           typename ALayout,
           typename BLayout,
-          typename CLayout>
+          typename CLayout,
+          typename ComputeType>
 struct DeviceOperationInstanceFactory<
     ck::tensor_operation::device::DeviceGemmSplitK<ALayout,
                                                    BLayout,
@@ -115,7 +136,8 @@ struct DeviceOperationInstanceFactory<
                                                    CDataType,
                                                    ck::tensor_operation::element_wise::PassThrough,
                                                    ck::tensor_operation::element_wise::PassThrough,
-                                                   ck::tensor_operation::element_wise::PassThrough>>
+                                                   ck::tensor_operation::element_wise::PassThrough,
+                                                   ComputeType>>
 {
     using DeviceOp = DeviceGemmSplitK<ALayout,
                                       BLayout,
@@ -125,7 +147,8 @@ struct DeviceOperationInstanceFactory<
                                       CDataType,
                                       ck::tensor_operation::element_wise::PassThrough,
                                       ck::tensor_operation::element_wise::PassThrough,
-                                      ck::tensor_operation::element_wise::PassThrough>;
+                                      ck::tensor_operation::element_wise::PassThrough,
+                                      ComputeType>;
 
     static auto GetInstances()
     {
@@ -158,7 +181,7 @@ struct DeviceOperationInstanceFactory<
 #endif
 #ifdef CK_ENABLE_FP16
         else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
-                          is_same_v<CDataType, half_t>)
+                          is_same_v<CDataType, half_t> && is_same_v<ComputeType, half_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -231,6 +254,31 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f16_f8_f16_km_nk_mn_instances(op_ptrs);
             }
         }
+        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                          is_same_v<CDataType, half_t> && is_same_v<ComputeType, f8_t>)
+        {
+            if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
+                         is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Row> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_kn_mn_instances(op_ptrs);
+            }
+            else if constexpr(is_same_v<ALayout, Col> && is_same_v<BLayout, Col> &&
+                              is_same_v<CLayout, Row>)
+            {
+                add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_nk_mn_instances(op_ptrs);
+            }
+        }
+
 #endif
         return op_ptrs;
     }
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
index 36c0b50e8e..ec4c27598f 100644
--- a/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/CMakeLists.txt
@@ -15,6 +15,10 @@ list(APPEND GEMM_SPLITK_INSTANCES device_gemm_xdl_splitk_f32_f32_f32_mk_kn_mn_in
                                   device_gemm_xdl_splitk_f16_fp8_f16_mk_kn_mn_instance.cpp
                                   device_gemm_xdl_splitk_f16_fp8_f16_mk_nk_mn_instance.cpp
                                   device_gemm_xdl_splitk_f16_fp8_f16_km_kn_mn_instance.cpp
-                                  device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp)
+                                  device_gemm_xdl_splitk_f16_fp8_f16_km_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_nk_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_kn_mn_instance.cpp
+                                  device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_nk_mn_instance.cpp)
 
 add_instance_library(device_gemm_splitk_instance ${GEMM_SPLITK_INSTANCES})
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_kn_mn_instance.cpp
new file mode 100644
index 0000000000..f27e3197bf
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_kn_mn_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_kn_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,     F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,     Row,     Row, PassThrough, PassThrough, PassThrough,    GemmDefault,  256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,     F8>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_nk_mn_instance.cpp
new file mode 100644
index 0000000000..421f90e8c5
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_km_nk_mn_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_nk_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              4,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              2,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Col,      Col,    Row, PassThrough, PassThrough, PassThrough,   GemmDefault,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,              2,              1,              8,      true,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Col, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f16_f16_comp_f8_km_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_kn_mn_instance.cpp
new file mode 100644
index 0000000000..814a9aabf2
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_kn_mn_instance.cpp
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+// static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
+static constexpr auto GemmMNPadding = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+
+using device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_generic_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               2,    F8>
+    // clang-format on
+    >;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 48, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   192,    64,     4,  8,   32,   32,    3,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   192,     4,  8,   32,   32,    1,    3,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 24, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   192,    32,     4,  8,   32,   32,    3,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,    64,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              2,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    64,    32,     4,  8,   32,   32,    1,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              4,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,      Row,    Row, PassThrough, PassThrough, PassThrough, GemmMNPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             2,              1,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_generic_instances{});
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_kn_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_nk_mn_instance.cpp
new file mode 100644
index 0000000000..05803cfbec
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/gemm_splitk/device_gemm_xdl_splitk_f16_f16_f16_comp_fp8_mk_nk_mn_instance.cpp
@@ -0,0 +1,79 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp"
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F8  = ck::f8_t;
+using F16 = ck::half_t;
+using F32 = float;
+
+using Row = ck::tensor_layout::gemm::RowMajor;
+using Col = ck::tensor_layout::gemm::ColumnMajor;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_generic_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              1,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              1,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               2,    F8>
+    // clang-format on
+    >;
+
+// Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+using device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_instances = std::tuple<
+    // clang-format off
+        //#########################|AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C|          GEMM| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle|     CBlockTransferClusterLengths|  CBlockTransfer| Compute|
+        //#########################| Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|Specialization|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave| _MBlock_MXdlPerWave_MWaveMPerXdl| ScalarPerVector|    Type|
+        //#########################|     |      |      |        |        |        |        |   Operation|   Operation|   Operation|              |      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle| _NBlock_NXdlPerWave_NWaveNPerXdl|   _NWaveNPerXdl|        |
+        //#########################|     |      |      |        |        |        |        |            |            |            |              |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                                 |                |        |
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   256,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   128,   256,     4,  8,   32,   32,    2,    4,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,   128,   128,     4,  8,   32,   32,    4,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   128,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,   128,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,    64,   128,     4,  8,   32,   32,    2,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    64,    64,     4,  8,   32,   32,    2,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,   128,    64,     4,  8,   32,   32,    2,    1,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   256,    64,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 64, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 64, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,   128,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 32, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,   128,    32,   128,     4,  8,   32,   32,    1,    2,  S<1, 4, 32, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 32, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 8>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    64,    32,     4,  8,   32,   32,    2,    1,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F8>,
+        DeviceGemmXdlSplitKCShuffle<  F16,   F16,   F16,     F32,     Row,     Col,     Row, PassThrough, PassThrough, PassThrough,    MNKPadding,    64,    32,    64,     4,  8,   32,   32,    1,    2,  S<1, 4, 16, 1>,  S<0, 2, 1, 3>,  S<0, 2, 1, 3>,              3,              8,              8,      true,  S<1, 4, 16, 1>,  S<0, 1, 3, 2>,  S<0, 1, 3, 2>,             3,              8,              8,      true,           1,           1,                   S<1, 16, 1, 4>,               8,    F8>
+    // clang-format on
+    >;
+
+void add_device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_instances(
+    std::vector<std::unique_ptr<
+        DeviceGemmSplitK<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough, F8>>>&
+        instances)
+{
+    add_device_operation_instances(
+        instances, device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_generic_instances{});
+    add_device_operation_instances(instances,
+                                   device_gemm_xdl_splitk_f16_f16_f16_comp_f8_mk_nk_mn_instances{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/include/profiler/profile_gemm_splitk_impl.hpp b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
index 495513f665..329a7c07ca 100644
--- a/profiler/include/profiler/profile_gemm_splitk_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_splitk_impl.hpp
@@ -30,7 +30,8 @@ template <typename ADataType,
           typename CDataType,
           typename ALayout,
           typename BLayout,
-          typename CLayout>
+          typename CLayout,
+          typename ComputeType = CDataType>
 bool profile_gemm_splitk_impl(int do_verification,
                               int init_method,
                               bool do_log,
@@ -103,7 +104,8 @@ bool profile_gemm_splitk_impl(int do_verification,
                                                                     CDataType,
                                                                     AElementOp,
                                                                     BElementOp,
-                                                                    CElementOp>;
+                                                                    CElementOp,
+                                                                    ComputeType>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
@@ -120,7 +122,8 @@ bool profile_gemm_splitk_impl(int do_verification,
                                                                                 AccDataType,
                                                                                 AElementOp,
                                                                                 BElementOp,
-                                                                                CElementOp>;
+                                                                                CElementOp,
+                                                                                ComputeType>;
 
         auto ref_gemm    = ReferenceGemmInstance{};
         auto ref_invoker = ref_gemm.MakeInvoker();
diff --git a/profiler/src/profile_gemm_splitk.cpp b/profiler/src/profile_gemm_splitk.cpp
index 9c805fc1d1..d8fa3d872a 100644
--- a/profiler/src/profile_gemm_splitk.cpp
+++ b/profiler/src/profile_gemm_splitk.cpp
@@ -25,6 +25,7 @@ enum struct GemmDataType
     INT8_INT8_INT8, // 3
     F8_F16_F16,     // 4
     F16_F8_F16,     // 5
+    F16_F16_F16_F8, // 6
 };
 
 #define OP_NAME "gemm_splitk"
@@ -35,7 +36,8 @@ int profile_gemm_splitk(int argc, char* argv[])
     if(argc != 15)
     {
         printf("arg1: tensor operation (" OP_NAME ": " OP_DESC ")\n");
-        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8; 4: f8@f16; 5: f16@f8)\n");
+        printf("arg2: data type (0: fp32; 1: fp16; 2: bf16; 3: int8; 4: f8@f16; 5: f16@f8; 6: f16, "
+               "comp f8)\n");
         printf("arg3: matrix layout (0: A[m, k] * B[k, n] = C[m, n];\n");
         printf("                     1: A[m, k] * B[n, k] = C[m, n];\n");
         printf("                     2: A[k, m] * B[k, n] = C[m, n];\n");
@@ -80,7 +82,8 @@ int profile_gemm_splitk(int argc, char* argv[])
                        auto c_type,
                        auto a_layout,
                        auto b_layout,
-                       auto c_layout) {
+                       auto c_layout,
+                       auto compute_type) {
         using ADataType   = decltype(a_type);
         using BDataType   = decltype(b_type);
         using AccDataType = decltype(acc_type);
@@ -90,6 +93,8 @@ int profile_gemm_splitk(int argc, char* argv[])
         using BLayout = decltype(b_layout);
         using CLayout = decltype(c_layout);
 
+        using ComputeType = decltype(compute_type);
+
         const int DefaultStrideA = ck::is_same_v<ALayout, Row> ? K : M;
         const int DefaultStrideB = ck::is_same_v<BLayout, Row> ? N : K;
         const int DefaultStrideC = ck::is_same_v<CLayout, Row> ? N : M;
@@ -100,7 +105,8 @@ int profile_gemm_splitk(int argc, char* argv[])
                                                            CDataType,
                                                            ALayout,
                                                            BLayout,
-                                                           CLayout>(
+                                                           CLayout,
+                                                           ComputeType>(
             do_verification,
             init_method,
             do_log,
@@ -118,68 +124,84 @@ int profile_gemm_splitk(int argc, char* argv[])
 
     if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_KN_MN)
     {
-        return profile(F32{}, F32{}, F32{}, F32{}, Row{}, Row{}, Row{});
+        return profile(F32{}, F32{}, F32{}, F32{}, Row{}, Row{}, Row{}, F32{});
     }
     else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::MK_NK_MN)
     {
-        return profile(F32{}, F32{}, F32{}, F32{}, Row{}, Col{}, Row{});
+        return profile(F32{}, F32{}, F32{}, F32{}, Row{}, Col{}, Row{}, F32{});
     }
     else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_KN_MN)
     {
-        return profile(F32{}, F32{}, F32{}, F32{}, Col{}, Row{}, Row{});
+        return profile(F32{}, F32{}, F32{}, F32{}, Col{}, Row{}, Row{}, F32{});
     }
     else if(data_type == GemmDataType::F32_F32_F32 && layout == GemmMatrixLayout::KM_NK_MN)
     {
-        return profile(F32{}, F32{}, F32{}, F32{}, Col{}, Col{}, Row{});
+        return profile(F32{}, F32{}, F32{}, F32{}, Col{}, Col{}, Row{}, F32{});
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
-        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{});
+        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
-        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{});
+        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
     {
-        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{});
+        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
     {
-        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{});
+        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{}, F16{});
     }
 #if defined CK_ENABLE_FP8
     else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
-        return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{});
+        return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
-        return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{});
+        return profile(F8{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::KM_KN_MN)
     {
-        return profile(F8{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{});
+        return profile(F8{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F8_F16_F16 && layout == GemmMatrixLayout::KM_NK_MN)
     {
-        return profile(F8{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{});
+        return profile(F8{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::MK_KN_MN)
     {
-        return profile(F16{}, F8{}, F32{}, F16{}, Row{}, Row{}, Row{});
+        return profile(F16{}, F8{}, F32{}, F16{}, Row{}, Row{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::MK_NK_MN)
     {
-        return profile(F16{}, F8{}, F32{}, F16{}, Row{}, Col{}, Row{});
+        return profile(F16{}, F8{}, F32{}, F16{}, Row{}, Col{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::KM_KN_MN)
     {
-        return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Row{}, Row{});
+        return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Row{}, Row{}, F16{});
     }
     else if(data_type == GemmDataType::F16_F8_F16 && layout == GemmMatrixLayout::KM_NK_MN)
     {
-        return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Col{}, Row{});
+        return profile(F16{}, F8{}, F32{}, F16{}, Col{}, Col{}, Row{}, F16{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16_F8 && layout == GemmMatrixLayout::MK_KN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Row{}, Row{}, F8{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16_F8 && layout == GemmMatrixLayout::MK_NK_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Row{}, Col{}, Row{}, F8{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16_F8 && layout == GemmMatrixLayout::KM_KN_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Row{}, Row{}, F8{});
+    }
+    else if(data_type == GemmDataType::F16_F16_F16_F8 && layout == GemmMatrixLayout::KM_NK_MN)
+    {
+        return profile(F16{}, F16{}, F32{}, F16{}, Col{}, Col{}, Row{}, F8{});
     }
 #endif
     else

From 707ad0025f3fea189f8861ff741861c45aac786f Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 16 Oct 2023 08:40:26 -0700
Subject: [PATCH 172/194] Add hipTensor build and test to CK CI. (#990)

* add a hipTensor test to CI

* use jenkins git plugin

* change hipTensor folder location in CI

* change the git method for hipTensor

* run tests usign ctest

* check the hipTensor contents

* only build hipTensor on MI100/200

* pull hipTensor as zip archive

* fix jenkins syntax

* add path to the CK installation

* combine build commands into one shell

* change jenkins syntax for CK installer path

* try different syntax

* allow unzip overwrite

* fix jenkins file syntax

* remove any old versions of hipTensor before building

* add option to select hipTensor branch for testing
---
 Dockerfile  |  1 +
 Jenkinsfile | 29 +++++++++++++++++++++++++++++
 2 files changed, 30 insertions(+)

diff --git a/Dockerfile b/Dockerfile
index 75ca0c7f9f..eb3305a42e 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -49,6 +49,7 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     vim \
     nano \
     zlib1g-dev \
+    zip \
     openssh-server \
     clang-format-12 \
     kmod && \
diff --git a/Jenkinsfile b/Jenkinsfile
index 60a53c54fb..38bd3d7df4 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -526,6 +526,26 @@ def Build_CK(Map conf=[:]){
                            stash "ckprofiler_0.2.0_amd64.deb"
                         }
                     }
+                    if (params.hipTensor_test && navi_node == 0 ){
+                        //build and test hipTensor
+                        sh """#!/bin/bash
+                            rm -rf "${params.hipTensor_branch}".zip
+                            rm -rf hipTensor-"${params.hipTensor_branch}"
+                            wget https://github.com/ROCmSoftwarePlatform/hipTensor/archive/refs/heads/"${params.hipTensor_branch}".zip
+                            unzip -o "${params.hipTensor_branch}".zip
+                        """
+                        dir("hipTensor-${params.hipTensor_branch}"){
+                            sh """#!/bin/bash
+                                mkdir -p build
+                                ls -ltr
+                                CC=hipcc CXX=hipcc cmake -Bbuild . -D CMAKE_PREFIX_PATH="/opt/rocm;${env.WORKSPACE}/install"
+                                cmake --build build -- -j
+                            """
+                        }
+                        dir("hipTensor-${params.hipTensor_branch}/build"){
+                            sh 'ctest'
+                        }
+                    }
                 }
             }
         }
@@ -654,6 +674,15 @@ pipeline {
             name: "DL_KERNELS",
             defaultValue: false,
             description: "Select whether to build DL kernels (default: OFF)")
+        booleanParam(
+            name: "hipTensor_test",
+            defaultValue: true,
+            description: "Use the CK build to verify hipTensor build and tests (default: ON)")
+        string(
+            name: 'hipTensor_branch',
+            defaultValue: 'mainline',
+            description: 'Specify which branch of hipTensor to use (default: mainline)')
+
     }
     environment{
         dbuser = "${dbuser}"

From 39430bfdeb7247f0fdf4ba07559619ab1ab9a415 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Mon, 16 Oct 2023 17:42:59 -0500
Subject: [PATCH 173/194] workaround with float (#992)

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 include/ck/utility/type_convert.hpp | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 8b70a6bfb4..ccbd5db644 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -146,7 +146,7 @@ inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return type_convert<f8_t>(type_convert<float>(x));
-#else
+#elif 0
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -154,6 +154,8 @@ inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#else
+    return type_convert<f8_t>(type_convert<float>(x));
 #endif
 }
 
@@ -164,9 +166,11 @@ inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // use native conversion to float and convert to fp16
     return type_convert<half_t>(type_convert<float>(x));
-#else
+#elif 0
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<f8_t, half_t, negative_zero_nan>(x);
+#else
+    return type_convert<half_t>(type_convert<float>(x));
 #endif
 }
 #endif
@@ -222,7 +226,7 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return type_convert<bf8_t>(type_convert<float>(x));
-#else
+#elif 0
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -230,6 +234,8 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
+#else
+    return type_convert<bf8_t>(type_convert<float>(x));
 #endif
 }
 
@@ -240,9 +246,11 @@ inline __host__ __device__ half_t type_convert<half_t, bf8_t>(bf8_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // use native conversion to float and convert to fp16
     return type_convert<half_t>(type_convert<float>(x));
-#else
+#elif 0
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<bf8_t, half_t, negative_zero_nan>(x);
+#else
+    return type_convert<half_t>(type_convert<float>(x));
 #endif
 }
 #endif
@@ -354,7 +362,7 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
 #else
-    return type_convert<f8_t>(type_convert<float>(x));
+    return f8_convert_sr<f8_t>(type_convert<float>(x));
 #endif
 }
 #endif
@@ -406,7 +414,7 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
 #else
-    return type_convert<bf8_t>(type_convert<float>(x));
+    return f8_convert_sr<bf8_t>(type_convert<float>(x));
 #endif
 }
 #endif

From 16d7c4d2f710b06f07469da3efbf39f87c28b746 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 17 Oct 2023 10:32:26 +0200
Subject: [PATCH 174/194] Add grouped conv bwd weight wmma (#985)

* Add grouped conv bwd weight wmma

* Update README, changelog, profiler

* Minor fixes

* Fix grouped conv bwd wei dl kernel

* Minor fixes

* Minor stylistic fixes
---
 CHANGELOG.md                                  |   2 +-
 .../20_grouped_conv_bwd_weight/CMakeLists.txt |  13 +-
 example/20_grouped_conv_bwd_weight/common.hpp |  40 +-
 .../grouped_conv_bwd_weight_dl_fp16.cpp       |  21 +-
 .../grouped_conv_bwd_weight_wmma_fp16.cpp     |  88 ++
 .../grouped_conv_bwd_weight_xdl_bf16.cpp      |  21 +-
 .../grouped_conv_bwd_weight_xdl_fp16.cpp      |  21 +-
 ..._conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp |  21 +-
 .../run_grouped_conv_bwd_weight_example.inc   |  22 +-
 ...conv_bwd_data_multiple_d_wmma_cshuffle.hpp |   2 +-
 .../device_grouped_conv_bwd_weight_dl.hpp     |  31 +-
 ..._grouped_conv_bwd_weight_wmma_cshuffle.hpp | 877 ++++++++++++++++++
 ...e_grouped_conv_bwd_weight_xdl_cshuffle.hpp |  35 +-
 ...uped_conv_fwd_multiple_d_wmma_cshuffle.hpp |   2 +-
 .../device/impl/device_grouped_conv_utils.hpp |   9 +-
 ...gridwise_gemm_multiple_d_wmma_cshuffle.hpp |  41 +-
 ...ce_grouped_conv_bwd_weight_dl_instance.hpp |   2 -
 ..._grouped_conv_bwd_weight_wmma_instance.hpp | 119 +++
 .../grouped_convolution_backward_weight.hpp   | 126 +++
 .../grouped_conv1d_bwd_weight/CMakeLists.txt  |  18 +-
 ...weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp |   0
 ..._weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp |   0
 ..._weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp |   0
 ...weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp |   0
 ..._weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp |   0
 ..._weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp |   0
 ...eight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp |   0
 ...weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp |   0
 ...weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp |   0
 .../grouped_conv2d_bwd_weight/CMakeLists.txt  |  24 +-
 ...ght_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |   0
 ...ight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   0
 ...ight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   0
 ...ght_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |   0
 ...ight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |   0
 ...ight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |   0
 ...ht_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp |   0
 ...ght_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp |   0
 ...ght_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp |   0
 ...ht_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp |   0
 ...ght_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp |   0
 ...ght_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp |   0
 .../grouped_conv3d_bwd_weight/CMakeLists.txt  |  36 +-
 ..._dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |   0
 ...t_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |   0
 ...t_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |   0
 ..._dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |   0
 ...t_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |   0
 ...t_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |   0
 ...hwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp |  35 +
 ...wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |  35 +
 ...dhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp |  35 +
 ..._wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp |  35 +
 ...wgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp |  35 +
 ...wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |  35 +
 ...hwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp |  35 +
 ..._wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp |  35 +
 ...xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp |   0
 ..._xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp |   0
 ..._xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp |   0
 ...xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp |   0
 ...kzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp |   0
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp |   0
 ..._xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp |   0
 profiler/README.md                            |   6 +-
 .../src/profile_grouped_conv_bwd_weight.cpp   |  22 +-
 test/grouped_convnd_bwd_weight/CMakeLists.txt |  17 +-
 .../test_grouped_convnd_bwd_weight.cpp        |  46 +-
 ...ouped_convnd_bwd_weight_interface_wmma.cpp | 191 ++++
 ...ouped_convnd_bwd_weight_interface_xdl.cpp} |   0
 70 files changed, 1960 insertions(+), 173 deletions(-)
 create mode 100644 example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_wmma_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp
 create mode 100644 library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => dl}/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => dl}/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => dl}/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => dl}/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => dl}/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => dl}/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => xdl}/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => xdl}/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/{ => xdl}/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => dl}/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => dl}/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => dl}/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => dl}/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => dl}/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => dl}/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => xdl}/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => xdl}/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => xdl}/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => xdl}/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => xdl}/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/{ => xdl}/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => dl}/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => dl}/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => dl}/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => dl}/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => dl}/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => dl}/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp (100%)
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp (100%)
 rename library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/{ => xdl}/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp (100%)
 create mode 100644 test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface_wmma.cpp
 rename test/grouped_convnd_bwd_weight/{test_grouped_convnd_bwd_weight_interface.cpp => test_grouped_convnd_bwd_weight_interface_xdl.cpp} (100%)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 10d92d8e3e..f88df081da 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -14,7 +14,7 @@ None
 ### Additions
 - Added an image to a column kernel (#867)
 - Added a column to an image kernel (#930)
-- Support for 3D grouped convolution forward on RDNA 3 GPUs (#935)
+- Support for 3D grouped convolution on RDNA 3 GPUs (#935, #950, #985)
 - Grouped convolution support for small K and C (#822 #879 #897)
 - Support for NHWGC (2D and 3D) grouped convolution backward weight (#769 #804)
 - Support for bf16/f32/f16 and NHWGC (2D and 3d) grouped convolution backward data (#757 #799)
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index c83b384a01..1ecbab5825 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -1,7 +1,8 @@
-list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
+ if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
    add_custom_target(example_grouped_conv_bwd_weight)
    add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
    if(result EQUAL 0)
@@ -18,6 +19,14 @@ foreach(gpu IN LISTS GPU_TARGETS)
     endif()
    endif()
    set(target 1)
+  endif()
+  if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
+   add_custom_target(example_grouped_conv_bwd_weight)
+   add_example_executable(example_grouped_conv_bwd_weight_wmma_fp16 grouped_conv_bwd_weight_wmma_fp16.cpp)
+   if(result EQUAL 0)
+    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_wmma_fp16)
+   endif()
+   set(target 1)
  endif()
 endforeach()
 
diff --git a/example/20_grouped_conv_bwd_weight/common.hpp b/example/20_grouped_conv_bwd_weight/common.hpp
index 0e7eb00e4b..1d2206ff72 100644
--- a/example/20_grouped_conv_bwd_weight/common.hpp
+++ b/example/20_grouped_conv_bwd_weight/common.hpp
@@ -46,25 +46,21 @@ struct CommonLayoutSetting
     using OutputLayout = OutputLay;
 };
 
-template <ck::index_t NDimSpatial>
-struct CommonLayoutSettingSelector;
-
 namespace ctl = ck::tensor_layout::convolution;
-
-template <>
-struct CommonLayoutSettingSelector<1> final : CommonLayoutSetting<ctl::GNWC, ctl::GKXC, ctl::GNWK>
-{
-};
-
-template <>
-struct CommonLayoutSettingSelector<2> final
-    : CommonLayoutSetting<ctl::GNHWC, ctl::GKYXC, ctl::GNHWK>
-{
-};
-
-template <>
-struct CommonLayoutSettingSelector<3> final
-    : CommonLayoutSetting<ctl::GNDHWC, ctl::GKZYXC, ctl::GNDHWK>
+template <ck::index_t NDimSpatial>
+struct CommonLayoutSettingSelector
+    : CommonLayoutSetting<ck::tuple_element_t<NDimSpatial - 1,
+                                              ck::Tuple<ck::tensor_layout::convolution::GNWC,
+                                                        ck::tensor_layout::convolution::GNHWC,
+                                                        ck::tensor_layout::convolution::GNDHWC>>,
+                          ck::tuple_element_t<NDimSpatial - 1,
+                                              ck::Tuple<ck::tensor_layout::convolution::GKXC,
+                                                        ck::tensor_layout::convolution::GKYXC,
+                                                        ck::tensor_layout::convolution::GKZYXC>>,
+                          ck::tuple_element_t<NDimSpatial - 1,
+                                              ck::Tuple<ck::tensor_layout::convolution::GNWK,
+                                                        ck::tensor_layout::convolution::GNHWK,
+                                                        ck::tensor_layout::convolution::GNDHWK>>>
 {
 };
 
@@ -84,10 +80,10 @@ struct ExecutionConfig final
     bool time_kernel     = false;
 };
 
-#define DefaultConvParam                                                      \
-    ck::utils::conv::ConvParam                                                \
-    {                                                                         \
-        2, 4, 1, 128, 256, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, { 1, 1 } \
+#define DefaultConvParam                                                                         \
+    ck::utils::conv::ConvParam                                                                   \
+    {                                                                                            \
+        3, 4, 1, 128, 256, {3, 3, 3}, {14, 14, 14}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, { 1, 1, 1 } \
     }
 
 inline void print_help_msg()
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
index dccde28aa0..44ba3d8e02 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_dl_fp16.cpp
@@ -76,4 +76,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
 
 #include "run_grouped_conv_bwd_weight_example.inc"
 
-int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
+
+    if(!parse_cmd_args(argc, argv, config, conv_param))
+    {
+        return 1;
+    }
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
+    case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
+    case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
+    default: break;
+    }
+
+    return 1;
+}
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_wmma_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_wmma_fp16.cpp
new file mode 100644
index 0000000000..2c2950b17e
--- /dev/null
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_wmma_fp16.cpp
@@ -0,0 +1,88 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "common.hpp"
+
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp"
+
+using InDataType  = F16;
+using WeiDataType = F16;
+using OutDataType = F16;
+using AccDataType = F32;
+
+using InElementOp  = PassThrough;
+using WeiElementOp = PassThrough;
+using OutElementOp = PassThrough;
+
+template <ck::index_t NDimSpatial>
+using DeviceConvBwdWeightInstance =
+    ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Wmma_CShuffle<
+        NDimSpatial,
+        ck::tensor_layout::convolution::GNDHWC,
+        ck::tensor_layout::convolution::GKZYXC,
+        ck::tensor_layout::convolution::GNDHWK,
+        InDataType,           // InDataType
+        WeiDataType,          // WeiDataType
+        OutDataType,          // OutDataType
+        AccDataType,          // AccDataType
+        InElementOp,          // InElementwiseOperation
+        WeiElementOp,         // WeiElementwiseOperation
+        OutElementOp,         // OutElementwiseOperation
+        ConvBwdWeightDefault, // ConvolutionBackwardWeightSpecialization
+        256,                  // BlockSize
+        128,                  // MPerBlock
+        128,                  // NPerBlock
+        4,                    // K0PerBlock
+        8,                    // K1
+        16,                   // MPerWMMA
+        16,                   // NPerWMMA
+        4,                    // MRepeat
+        2,                    // NRepeat
+        S<4, 64, 1>,          // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<0, 2, 1>,           // ABlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1>,           // ABlockTransferSrcAccessOrder
+        1,                    // ABlockTransferSrcVectorDim
+        1,                    // ABlockTransferSrcScalarPerVector
+        8,                    // ABlockTransferDstScalarPerVector_AK1
+        true,                 // ABlockLdsExtraM
+        S<4, 64, 1>,          // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<0, 2, 1>,           // BBlockTransferThreadClusterArrangeOrder
+        S<0, 2, 1>,           // BBlockTransferSrcAccessOrder
+        1,                    // BBlockTransferSrcVectorDim
+        1,                    // BBlockTransferSrcScalarPerVector
+        8,                    // BBlockTransferDstScalarPerVector_BK1
+        true,                 // BBlockLdsExtraN
+        4,
+        2,
+        S<1, 32, 1, 8>,
+        1>;
+
+template <ck::index_t NDimSpatial>
+using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWeight<NDimSpatial,
+                                                                                     InDataType,
+                                                                                     WeiDataType,
+                                                                                     OutDataType,
+                                                                                     InElementOp,
+                                                                                     WeiElementOp,
+                                                                                     OutElementOp>;
+
+#include "run_grouped_conv_bwd_weight_example.inc"
+
+int main(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
+
+    if(!parse_cmd_args(argc, argv, config, conv_param))
+    {
+        return 1;
+    }
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
+    default: break;
+    }
+
+    return 1;
+}
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
index 467209ef28..80b9724930 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_bf16.cpp
@@ -78,4 +78,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
 
 #include "run_grouped_conv_bwd_weight_example.inc"
 
-int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
+
+    if(!parse_cmd_args(argc, argv, config, conv_param))
+    {
+        return 1;
+    }
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
+    case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
+    case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
+    default: break;
+    }
+
+    return 1;
+}
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
index 8b751b0d98..5c1c97d615 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16.cpp
@@ -77,4 +77,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
 
 #include "run_grouped_conv_bwd_weight_example.inc"
 
-int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
+
+    if(!parse_cmd_args(argc, argv, config, conv_param))
+    {
+        return 1;
+    }
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
+    case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
+    case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
+    default: break;
+    }
+
+    return 1;
+}
diff --git a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
index c678867e74..0d0d58eb2a 100644
--- a/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
+++ b/example/20_grouped_conv_bwd_weight/grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp
@@ -83,4 +83,23 @@ using HostConvBwdWeightInstance = ck::tensor_operation::host::ReferenceConvBwdWe
 
 #include "run_grouped_conv_bwd_weight_example.inc"
 
-int main(int argc, char* argv[]) { return !run_grouped_conv_bwd_weight_example(argc, argv); }
+int main(int argc, char* argv[])
+{
+    ExecutionConfig config;
+    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
+
+    if(!parse_cmd_args(argc, argv, config, conv_param))
+    {
+        return 1;
+    }
+
+    switch(conv_param.num_dim_spatial_)
+    {
+    case 1: return !run_grouped_conv_bwd_weight<1>(config, conv_param);
+    case 2: return !run_grouped_conv_bwd_weight<2>(config, conv_param);
+    case 3: return !run_grouped_conv_bwd_weight<3>(config, conv_param);
+    default: break;
+    }
+
+    return 1;
+}
diff --git a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
index 755d6addd8..73d15f3ea8 100644
--- a/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
+++ b/example/20_grouped_conv_bwd_weight/run_grouped_conv_bwd_weight_example.inc
@@ -5,7 +5,7 @@ template <ck::index_t NDimSpatial>
 bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
                                  const ck::utils::conv::ConvParam& conv_param)
 {
-    // Dl op doesn't support split_k > 1
+    // Dl and WMMA ops don't support split_k > 1
     constexpr ck::index_t split_k = 1;
 
     const auto in_g_n_c_wis_desc =
@@ -143,23 +143,3 @@ bool run_grouped_conv_bwd_weight(const ExecutionConfig& config,
 
     return true;
 }
-
-bool run_grouped_conv_bwd_weight_example(int argc, char* argv[])
-{
-    ExecutionConfig config;
-    ck::utils::conv::ConvParam conv_param = DefaultConvParam;
-
-    if(!parse_cmd_args(argc, argv, config, conv_param))
-    {
-        return false;
-    }
-
-    switch(conv_param.num_dim_spatial_)
-    {
-    case 1: return run_grouped_conv_bwd_weight<1>(config, conv_param);
-    case 2: return run_grouped_conv_bwd_weight<2>(config, conv_param);
-    case 3: return run_grouped_conv_bwd_weight<3>(config, conv_param);
-    }
-
-    return false;
-}
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp
index 35581de72b..e2f9f24bf5 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp
@@ -565,7 +565,7 @@ struct DeviceGroupedConvBwdDataMultipleD_Wmma_CShuffle
                 auto launch_kernel = [&](auto has_main_k_block_loop) {
                     constexpr bool has_main_loop = has_main_k_block_loop.value;
 
-                    const auto kernel = kernel_grouped_conv_fwd_multiple_d_wmma_cshuffle<
+                    const auto kernel = kernel_grouped_conv_multiple_d_wmma_cshuffle<
                         GridwiseGemm,
                         ADataType,
                         BDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
index ef57e6e4d2..e2c9dca904 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
@@ -12,6 +12,7 @@
 #include "ck/tensor_description/tensor_descriptor_helper.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_dl_v1r3.hpp"
 #include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
@@ -22,32 +23,6 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-namespace {
-
-struct ComputePtrOffsetOfStridedBatch
-{
-    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideA_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideB_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetCPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideC_);
-    }
-
-    index_t BatchStrideA_;
-    index_t BatchStrideB_;
-    index_t BatchStrideC_;
-};
-
-} // namespace
-
 template <typename GridwiseGemm,
           typename FloatAB,
           typename FloatC,
@@ -952,7 +927,7 @@ struct DeviceGroupedConvBwdWeight_Dl : public DeviceGroupedConvBwdWeight<NDimSpa
         Block2CTileMap block_2_ctile_map_;
 
         // for computing batch offset
-        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
+        ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_;
 
         // element-wise op
         OutElementwiseOperation a_element_op_;
@@ -1024,7 +999,7 @@ struct DeviceGroupedConvBwdWeight_Dl : public DeviceGroupedConvBwdWeight<NDimSpa
                     remove_reference_t<DeviceOp::BGridDesc_B_K0_N0_N1_K1>,
                     remove_reference_t<DeviceOp::CGridDesc_M0_M10_M11_N0_N10_N11>,
                     remove_reference_t<DeviceOp::Block2CTileMap>,
-                    ComputePtrOffsetOfStridedBatch,
+                    ComputePtrOffsetOfStridedBatch<I0>,
                     has_main_loop,
                     has_double_loop>;
 
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp
new file mode 100644
index 0000000000..e8a721eb36
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp
@@ -0,0 +1,877 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <numeric>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+template <index_t NDimSpatial,
+          typename InLayout,
+          typename WeiLayout,
+          typename OutLayout,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename AccDataType,
+          typename InElementwiseOperation,
+          typename WeiElementwiseOperation,
+          typename OutElementwiseOperation,
+          ConvolutionBackwardWeightSpecialization ConvBackwardWeightSpecialization,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t K0PerBlock,
+          index_t K1,
+          index_t MPerWMMA,
+          index_t NPerWMMA,
+          index_t MRepeat,
+          index_t NRepeat,
+          typename ABlockTransferThreadClusterLengths_K0_M_K1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_K1,
+          bool ABlockLdsAddExtraM,
+          typename BBlockTransferThreadClusterLengths_K0_N_K1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_K1,
+          bool BBlockLdsAddExtraN,
+          index_t CShuffleMRepeatPerShuffle,
+          index_t CShuffleNRepeatPerShuffle,
+          typename CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CShuffleBlockTransferScalarPerVector_NPerBlock,
+          index_t NumGemmKPrefetchStage                        = 1,
+          LoopScheduler LoopSched                              = make_default_loop_scheduler(),
+          ck::PipelineVersion PipelineVer                      = ck::PipelineVersion::v1,
+          typename ck::enable_if<NDimSpatial == 3, bool>::type = false>
+struct DeviceGroupedConvBwdWeight_Wmma_CShuffle
+    : public DeviceGroupedConvBwdWeight<NDimSpatial,
+                                        InLayout,
+                                        WeiLayout,
+                                        OutLayout,
+                                        InDataType,
+                                        WeiDataType,
+                                        OutDataType,
+                                        InElementwiseOperation,
+                                        WeiElementwiseOperation,
+                                        OutElementwiseOperation>
+{
+    using DeviceOp = DeviceGroupedConvBwdWeight_Wmma_CShuffle;
+
+    using ADataType = OutDataType;
+    using BDataType = InDataType;
+    using CDataType = WeiDataType;
+
+    using AElementwiseOperation = OutElementwiseOperation;
+    using BElementwiseOperation = InElementwiseOperation;
+    using CElementwiseOperation = WeiElementwiseOperation;
+
+    // TODO make A/B datatype different
+    using ABDataType = InDataType;
+
+    // 3d
+    static constexpr bool is_NDHWGK_GKZYXC_NDHWGC =
+        is_same_v<InLayout, tensor_layout::convolution::NDHWGC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::NDHWGK>;
+    static constexpr bool is_GNDHWK_GKZYXC_GNDHWC =
+        is_same_v<InLayout, tensor_layout::convolution::GNDHWC> &&
+        is_same_v<WeiLayout, tensor_layout::convolution::GKZYXC> &&
+        is_same_v<OutLayout, tensor_layout::convolution::GNDHWK>;
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+    static constexpr auto I4 = Number<4>{};
+    static constexpr auto I5 = Number<5>{};
+
+    static constexpr auto GemmK1Number = Number<K1>{};
+    static constexpr index_t KPerBlock = K0PerBlock * GemmK1Number;
+
+    template <index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    constexpr static auto
+    make_out_grid_desc(const index_t N,
+                       const index_t Do,
+                       const index_t Ho,
+                       const index_t Wo,
+                       const index_t K,
+                       const std::array<index_t, NDimSpatial + 3>& output_strides)
+    {
+        const index_t WoStride = output_strides[5];
+        const auto KStride     = Number<1>{};
+        return make_naive_tensor_descriptor(make_tuple(N * Do * Ho * Wo, K),
+                                            make_tuple(WoStride, KStride));
+    }
+
+    template <index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    constexpr static auto
+    make_in_grid_desc(const index_t N,
+                      const index_t Di,
+                      const index_t Hi,
+                      const index_t Wi,
+                      const index_t C,
+                      const std::array<index_t, NDimSpatial + 3>& input_strides)
+    {
+        const index_t NStride  = input_strides[1];
+        const index_t DiStride = input_strides[3];
+        const index_t HiStride = input_strides[4];
+        const index_t WiStride = input_strides[5];
+        const auto CStride     = input_strides[2];
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            return make_naive_tensor_descriptor(make_tuple(N * Di * Hi * Wi, C),
+                                                make_tuple(WiStride, CStride));
+        }
+        else
+        {
+            return make_naive_tensor_descriptor(
+                make_tuple(N, Di, Hi, Wi, C),
+                make_tuple(NStride, DiStride, HiStride, WiStride, CStride));
+        }
+    }
+
+    template <index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    constexpr static auto
+    make_wei_grid_desc(const index_t K,
+                       const index_t Z,
+                       const index_t Y,
+                       const index_t X,
+                       const index_t C,
+                       const std::array<index_t, NDimSpatial + 3>& weights_strides)
+    {
+        const auto CStride = Number<1>{};
+        const auto KStride = weights_strides[1];
+        return make_naive_tensor_descriptor(make_tuple(K, Z * Y * X * C),
+                                            make_tuple(KStride, CStride));
+    }
+
+    template <index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    static auto MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N(
+        const index_t N,
+        const index_t K,
+        const index_t C,
+        const std::array<index_t, NDimSpatial>& input_spatial_lengths,
+        const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
+        const std::array<index_t, NDimSpatial>& output_spatial_lengths,
+        const std::array<index_t, NDimSpatial + 3>& input_strides,
+        const std::array<index_t, NDimSpatial + 3>& weights_strides,
+        const std::array<index_t, NDimSpatial + 3>& output_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+        const std::array<index_t, NDimSpatial>& input_left_pads,
+        const std::array<index_t, NDimSpatial>& input_right_pads)
+    {
+        using namespace ck;
+
+        const index_t Di = input_spatial_lengths[0];
+        const index_t Hi = input_spatial_lengths[1];
+        const index_t Wi = input_spatial_lengths[2];
+
+        const index_t Do = output_spatial_lengths[0];
+        const index_t Ho = output_spatial_lengths[1];
+        const index_t Wo = output_spatial_lengths[2];
+
+        const index_t Z = filter_spatial_lengths[0];
+        const index_t Y = filter_spatial_lengths[1];
+        const index_t X = filter_spatial_lengths[2];
+
+        const index_t ConvStrideD = conv_filter_strides[0];
+        const index_t ConvStrideH = conv_filter_strides[1];
+        const index_t ConvStrideW = conv_filter_strides[2];
+
+        const index_t ConvDilationD = conv_filter_dilations[0];
+        const index_t ConvDilationH = conv_filter_dilations[1];
+        const index_t ConvDilationW = conv_filter_dilations[2];
+
+        const index_t InLeftPadD = input_left_pads[0];
+        const index_t InLeftPadH = input_left_pads[1];
+        const index_t InLeftPadW = input_left_pads[2];
+
+        const index_t InRightPadD = input_right_pads[0];
+        const index_t InRightPadH = input_right_pads[1];
+        const index_t InRightPadW = input_right_pads[2];
+
+        const index_t GemmKTotal = N * Do * Ho * Wo;
+        const index_t GemmM      = K;
+        const index_t GemmN      = C * Z * X * Y;
+
+        const auto PadGemmM = (MPerBlock - GemmM % MPerBlock) % MPerBlock;
+        const auto PadGemmN = (NPerBlock - GemmN % NPerBlock) % NPerBlock;
+
+        const index_t GemmK0 =
+            math::integer_divide_ceil(GemmKTotal, GemmK1Number * K0PerBlock) * K0PerBlock;
+        const index_t GemmKPad = GemmK0 * GemmK1Number;
+
+        const auto out_grid_desc = make_out_grid_desc<NDim>(N, Do, Ho, Wo, K, output_strides);
+        const auto in_grid_desc  = make_in_grid_desc<NDim>(N, Di, Hi, Wi, C, input_strides);
+        const auto wei_grid_desc = make_wei_grid_desc<NDim>(K, Z, Y, X, C, weights_strides);
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            // B: input tensor
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                              wei_grid_desc);
+        }
+        else
+        {
+            // A: output tensor
+            const auto out_gemmkpad_gemmm_grid_desc = transform_tensor_descriptor(
+                out_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc = transform_tensor_descriptor(
+                out_gemmkpad_gemmm_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmM)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            // B: input tensor
+            const auto in_n_dip_hip_wip_c_grid_desc = transform_tensor_descriptor(
+                in_grid_desc,
+                make_tuple(make_pass_through_transform(N),
+                           make_pad_transform(Di, InLeftPadD, InRightPadD),
+                           make_pad_transform(Hi, InLeftPadH, InRightPadH),
+                           make_pad_transform(Wi, InLeftPadW, InRightPadW),
+                           make_pass_through_transform(C)),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}));
+
+            const auto in_n_z_do_y_ho_x_wo_c_grid_desc = transform_tensor_descriptor(
+                in_n_dip_hip_wip_c_grid_desc,
+                make_tuple(
+                    make_pass_through_transform(N),
+                    make_embed_transform(make_tuple(Z, Do), make_tuple(ConvDilationD, ConvStrideD)),
+                    make_embed_transform(make_tuple(Y, Ho), make_tuple(ConvDilationH, ConvStrideH)),
+                    make_embed_transform(make_tuple(X, Wo), make_tuple(ConvDilationW, ConvStrideW)),
+                    make_pass_through_transform(C)),
+                make_tuple(
+                    Sequence<0>{}, Sequence<1>{}, Sequence<2>{}, Sequence<3>{}, Sequence<4>{}),
+                make_tuple(Sequence<0>{},
+                           Sequence<1, 2>{},
+                           Sequence<3, 4>{},
+                           Sequence<5, 6>{},
+                           Sequence<7>{}));
+
+            const auto in_gemmktotal_gemmn_grid_desc = transform_tensor_descriptor(
+                in_n_z_do_y_ho_x_wo_c_grid_desc,
+                make_tuple(make_merge_transform(make_tuple(Z, Y, X, C)),
+                           make_merge_transform(make_tuple(N, Do, Ho, Wo))),
+                make_tuple(Sequence<1, 3, 5, 7>{}, Sequence<0, 2, 4, 6>{}),
+                make_tuple(Sequence<1>{}, Sequence<0>{}));
+
+            const auto in_gemmkpad_gemmn_grid_desc = transform_tensor_descriptor(
+                in_gemmktotal_gemmn_grid_desc,
+                make_tuple(make_right_pad_transform(GemmKTotal, GemmKPad - GemmKTotal),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc = transform_tensor_descriptor(
+                in_gemmkpad_gemmn_grid_desc,
+                make_tuple(make_unmerge_transform(make_tuple(GemmK0, GemmK1Number)),
+                           make_pass_through_transform(GemmN)),
+                make_tuple(Sequence<0>{}, Sequence<1>{}),
+                make_tuple(Sequence<0, 2>{}, Sequence<1>{}));
+
+            // Pad
+            const auto out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    out_gemmkbatch_gemmk0_gemmm_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmM, PadGemmM),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+            const auto in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc =
+                transform_tensor_descriptor(
+                    in_gemmkbatch_gemmk0_gemmn_gemmk1_grid_desc,
+                    make_tuple(make_pass_through_transform(GemmK0),
+                               make_right_pad_transform(GemmN, PadGemmN),
+                               make_pass_through_transform(GemmK1Number)),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}),
+                    make_tuple(Sequence<0>{}, Sequence<1>{}, Sequence<2>{}));
+
+            const auto wei_gemmm_gemmn_pad_grid_desc =
+                transform_tensor_descriptor(wei_grid_desc,
+                                            make_tuple(make_right_pad_transform(GemmM, PadGemmM),
+                                                       make_right_pad_transform(GemmN, PadGemmN)),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}),
+                                            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+            return make_tuple(out_gemmkbatch_gemmk0_gemmm_gemmk1_pad_grid_desc,
+                              in_gemmkbatch_gemmk0_gemmn_gemmk1_pad_grid_desc,
+                              wei_gemmm_gemmn_pad_grid_desc);
+        }
+    }
+
+    template <index_t NDim, typename ck::enable_if<NDim == 3, bool>::type = false>
+    static auto GetABCGridDesc()
+    {
+        const index_t dim = 1;
+        const std::array<index_t, NDimSpatial> lengths{1, 1, 1};
+        const std::array<index_t, NDimSpatial + 3> strides{1, 1, 1, 1, 1, 1};
+        const std::array<index_t, NDimSpatial> params{1, 1, 1};
+        return MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<3>(dim,
+                                                                  dim,
+                                                                  dim,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  lengths,
+                                                                  strides,
+                                                                  strides,
+                                                                  strides,
+                                                                  params,
+                                                                  params,
+                                                                  params,
+                                                                  params);
+    }
+
+    using ABCGridDescs = decltype(GetABCGridDesc<NDimSpatial>());
+
+    using AGridDesc_K0_M_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I0])>;
+    using BGridDesc_K0_N_K1 = remove_cvref_t<decltype(ABCGridDescs{}[I1])>;
+    using CGridDesc_M_N     = remove_cvref_t<decltype(ABCGridDescs{}[I2])>;
+
+    using GridwiseGemm = GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle<
+        // DataType Family
+        ADataType,
+        BDataType,
+        AccDataType,
+        CDataType,
+        Tuple<>,
+        CDataType,
+        // InMemory Data Descriptor
+        AGridDesc_K0_M_K1,
+        BGridDesc_K0_N_K1,
+        Tuple<>,
+        CGridDesc_M_N,
+        // ElementwiseOp Family
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        // Tiling Family
+        MPerBlock,
+        NPerBlock,
+        K0PerBlock,
+        MPerWMMA,
+        NPerWMMA,
+        K1,
+        MRepeat,
+        NRepeat,
+        // ThreadCluster Family
+        BlockSize,
+        ABlockTransferThreadClusterLengths_K0_M_K1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_K1,
+        false,
+        ABlockLdsAddExtraM,
+        BBlockTransferThreadClusterLengths_K0_N_K1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_K1,
+        false,
+        BBlockLdsAddExtraN,
+        CShuffleMRepeatPerShuffle,
+        CShuffleNRepeatPerShuffle,
+        CShuffleBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CShuffleBlockTransferScalarPerVector_NPerBlock,
+        NumGemmKPrefetchStage,
+        LoopSched,
+        PipelineVer>;
+
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(Tuple<>{}));
+
+    using CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            CGridDesc_M_N{}));
+
+    using Block2CTileMap = decltype(GridwiseGemm::MakeDefaultBlock2CTileMap(
+        CGridDesc_M_N{}, I1 /* M01 */, I1 /* N01 */));
+
+    struct Argument : public BaseArgument
+    {
+        Argument(const InDataType* p_in_grid,
+                 WeiDataType* p_wei_grid,
+                 const OutDataType* p_out_grid,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads,
+                 InElementwiseOperation in_element_op,
+                 WeiElementwiseOperation wei_element_op,
+                 OutElementwiseOperation out_element_op,
+                 index_t split_k)
+            : p_a_grid_{p_out_grid},
+              p_b_grid_{p_in_grid},
+              p_c_grid_{p_wei_grid},
+              a_grid_desc_kbatch_k0_m_k1_{},
+              b_grid_desc_kbatch_k0_n_k1_{},
+              c_grid_desc_m_n_{},
+              c_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              block_2_ctile_map_{},
+              compute_ptr_offset_of_batch_{},
+              a_element_op_{out_element_op},
+              b_element_op_{in_element_op},
+              c_element_op_{wei_element_op},
+              Conv_G_{a_g_n_c_wis_lengths[0]},
+              Conv_N_{a_g_n_c_wis_lengths[1]},
+              Conv_K_{b_g_k_c_xs_lengths[1]},
+              Conv_C_{a_g_n_c_wis_lengths[2]},
+              input_spatial_lengths_{},
+              filter_spatial_lengths_{},
+              output_spatial_lengths_{},
+              conv_filter_strides_{conv_filter_strides},
+              input_left_pads_{input_left_pads},
+              input_right_pads_{input_right_pads},
+              k_batch_{split_k}
+        {
+            constexpr index_t spatial_offset = 3;
+            std::copy(begin(a_g_n_c_wis_lengths) + spatial_offset,
+                      end(a_g_n_c_wis_lengths),
+                      begin(input_spatial_lengths_));
+            std::copy(begin(b_g_k_c_xs_lengths) + spatial_offset,
+                      end(b_g_k_c_xs_lengths),
+                      begin(filter_spatial_lengths_));
+            std::copy(begin(e_g_n_k_wos_lengths) + spatial_offset,
+                      end(e_g_n_k_wos_lengths),
+                      begin(output_spatial_lengths_));
+
+            const auto descs =
+                DeviceOp::MakeABCGridDescriptor_A_K0_M_K1_B_K0_N_K1_C_M_N<NDimSpatial>(
+                    Conv_N_,
+                    Conv_K_,
+                    Conv_C_,
+                    input_spatial_lengths_,
+                    filter_spatial_lengths_,
+                    output_spatial_lengths_,
+                    a_g_n_c_wis_strides,
+                    b_g_k_c_xs_strides,
+                    e_g_n_k_wos_strides,
+                    conv_filter_strides,
+                    conv_filter_dilations,
+                    input_left_pads,
+                    input_right_pads);
+
+            a_grid_desc_kbatch_k0_m_k1_ = descs[I0];
+            b_grid_desc_kbatch_k0_n_k1_ = descs[I1];
+            c_grid_desc_m_n_            = descs[I2];
+
+            block_2_ctile_map_ = GridwiseGemm::MakeDefaultBlock2CTileMap(
+                c_grid_desc_m_n_, I1 /* M01 */, I1 /* N01 */);
+
+            // A/B/C Batch Stride
+            compute_ptr_offset_of_batch_.BatchStrideA_ = e_g_n_k_wos_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideB_ = a_g_n_c_wis_strides[0];
+            compute_ptr_offset_of_batch_.BatchStrideE_ =
+                Conv_K_ * Conv_C_ *
+                std::accumulate(begin(filter_spatial_lengths_),
+                                end(filter_spatial_lengths_),
+                                index_t{1},
+                                std::multiplies<>{});
+
+            if(GridwiseGemm::CheckValidity(a_grid_desc_kbatch_k0_m_k1_,
+                                           b_grid_desc_kbatch_k0_n_k1_,
+                                           c_grid_desc_m_n_,
+                                           block_2_ctile_map_))
+            {
+                c_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        c_grid_desc_m_n_);
+            }
+        }
+
+        const ADataType* p_a_grid_;
+        const BDataType* p_b_grid_;
+        CDataType* p_c_grid_;
+        AGridDesc_K0_M_K1 a_grid_desc_kbatch_k0_m_k1_;
+        BGridDesc_K0_N_K1 b_grid_desc_kbatch_k0_n_k1_;
+        CGridDesc_M_N c_grid_desc_m_n_;
+        CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock c_grid_desc_mblock_mperblock_nblock_nperblock_;
+
+        Block2CTileMap block_2_ctile_map_;
+
+        // for computing batch offset
+        ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_;
+
+        OutElementwiseOperation a_element_op_;
+        InElementwiseOperation b_element_op_;
+        WeiElementwiseOperation c_element_op_;
+
+        // for checking IsSupportedArgument()
+        const index_t Conv_G_;
+        const index_t Conv_N_;
+        const index_t Conv_K_;
+        const index_t Conv_C_;
+        std::array<index_t, NDimSpatial> input_spatial_lengths_;
+        std::array<index_t, NDimSpatial> filter_spatial_lengths_;
+        std::array<index_t, NDimSpatial> output_spatial_lengths_;
+        const std::array<index_t, NDimSpatial>& conv_filter_strides_;
+        const std::array<index_t, NDimSpatial>& input_left_pads_;
+        const std::array<index_t, NDimSpatial>& input_right_pads_;
+        const index_t k_batch_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        void Print(const Argument& arg)
+        {
+            std::cout << "arg.a_grid_desc_kbatch_k0_m_k1_{"
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I0) << ", "
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1) << ", "
+                      << arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I2) << "}" << std::endl;
+
+            std::cout << "arg.b_grid_desc_kbatch_k0_n_k1_{"
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I0) << ", "
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I1) << ", "
+                      << arg.b_grid_desc_kbatch_k0_n_k1_.GetLength(I2) << "}" << std::endl;
+
+            std::cout << "arg.c_grid_desc_m_n_{" << arg.c_grid_desc_m_n_.GetLength(I0) << ", "
+                      << arg.c_grid_desc_m_n_.GetLength(I1) << "}" << std::endl;
+        }
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(stream_config.log_level_ > 0)
+            {
+                Print(arg);
+            }
+
+            if(!GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
+                                            arg.b_grid_desc_kbatch_k0_n_k1_,
+                                            arg.c_grid_desc_m_n_,
+                                            arg.block_2_ctile_map_))
+            {
+                throw std::runtime_error(
+                    "wrong! GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle has invalid "
+                    "setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_ctile_map_.CalculateGridSize(arg.c_grid_desc_m_n_) * arg.Conv_G_;
+
+            const auto K0 = arg.a_grid_desc_kbatch_k0_m_k1_.GetLength(I1);
+
+            const bool has_main_k0_block_loop = GridwiseGemm::CalculateHasMainKBlockLoop(K0);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_grouped_conv_multiple_d_wmma_cshuffle<
+                    GridwiseGemm,
+                    ADataType,
+                    BDataType,
+                    typename GridwiseGemm::DsGridPointer,
+                    CDataType,
+                    OutElementwiseOperation,
+                    InElementwiseOperation,
+                    WeiElementwiseOperation,
+                    AGridDesc_K0_M_K1,
+                    BGridDesc_K0_N_K1,
+                    DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    remove_reference_t<typename GridwiseGemm::DefaultBlock2CTileMap>,
+                    ComputePtrOffsetOfStridedBatch<I0>,
+                    has_main_loop>;
+
+                using EmptyTuple = Tuple<>;
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_a_grid_,
+                                              arg.p_b_grid_,
+                                              EmptyTuple{}, // Ds
+                                              arg.p_c_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.c_element_op_,
+                                              arg.Conv_G_,
+                                              arg.a_grid_desc_kbatch_k0_m_k1_,
+                                              arg.b_grid_desc_kbatch_k0_n_k1_,
+                                              DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock{},
+                                              arg.c_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.block_2_ctile_map_,
+                                              arg.compute_ptr_offset_of_batch_);
+            };
+
+            if(has_main_k0_block_loop)
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static constexpr bool IsValidCompilationParameter()
+    {
+        // TODO: properly implement this check
+        return true;
+    }
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        // check device
+        if(get_device_name() == "gfx1100" || get_device_name() == "gfx1101" ||
+           get_device_name() == "gfx1102")
+        {
+            if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, int32_t>))
+            {
+                return false;
+            }
+        }
+        else
+        {
+            return false;
+        }
+
+        // TODO: Add support for split_k > 1
+        if(arg.k_batch_ != 1)
+        {
+            return false;
+        }
+
+        if constexpr(!(is_NDHWGK_GKZYXC_NDHWGC || is_GNDHWK_GKZYXC_GNDHWC))
+        {
+            return false;
+        }
+
+        if constexpr(ConvBackwardWeightSpecialization ==
+                     ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0)
+        {
+            // check if it's a 1x1 convolution with stride=1 and no padding
+            for(int i = 0; i < NDimSpatial; i++)
+            {
+                if(!(arg.filter_spatial_lengths_[i] == 1 && arg.conv_filter_strides_[i] == 1 &&
+                     arg.input_left_pads_[i] == 0 && arg.input_right_pads_[i] == 0))
+                {
+                    return false;
+                }
+            }
+        }
+
+        // vector load A/B matrix from global memory
+        if(!(ABlockTransferSrcVectorDim == 1 && BBlockTransferSrcVectorDim == 1 &&
+             arg.Conv_K_ % ABlockTransferSrcScalarPerVector == 0 &&
+             arg.Conv_C_ % BBlockTransferSrcScalarPerVector == 0))
+        {
+            return false;
+        }
+
+        // vector store C matrix into global memory
+        if(!(arg.Conv_C_ % CShuffleBlockTransferScalarPerVector_NPerBlock == 0))
+        {
+            return false;
+        }
+
+        // Gridwise GEMM size
+        return GridwiseGemm::CheckValidity(arg.a_grid_desc_kbatch_k0_m_k1_,
+                                           arg.b_grid_desc_kbatch_k0_n_k1_,
+                                           arg.c_grid_desc_m_n_,
+                                           arg.block_2_ctile_map_);
+    }
+
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto
+    MakeArgument(const InDataType* p_in_grid,
+                 WeiDataType* p_wei_grid,
+                 const OutDataType* p_out_grid,
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                 const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                 const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                 const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                 const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                 const std::array<index_t, NDimSpatial>& input_left_pads,
+                 const std::array<index_t, NDimSpatial>& input_right_pads,
+                 InElementwiseOperation in_element_op,
+                 WeiElementwiseOperation wei_element_op,
+                 OutElementwiseOperation out_element_op,
+                 const index_t split_k)
+    {
+        return Argument{p_in_grid,
+                        p_wei_grid,
+                        p_out_grid,
+                        a_g_n_c_wis_lengths, // input
+                        a_g_n_c_wis_strides,
+                        b_g_k_c_xs_lengths, // weight
+                        b_g_k_c_xs_strides,
+                        e_g_n_k_wos_lengths, // output
+                        e_g_n_k_wos_strides,
+                        conv_filter_strides,
+                        conv_filter_dilations,
+                        input_left_pads,
+                        input_right_pads,
+                        in_element_op,
+                        wei_element_op,
+                        out_element_op,
+                        split_k};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(const void* p_in_grid,
+                        void* p_wei_grid,
+                        const void* p_out_grid,
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_lengths, // input
+                        const std::array<index_t, NDimSpatial + 3>& a_g_n_c_wis_strides,
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_lengths, // weight
+                        const std::array<index_t, NDimSpatial + 3>& b_g_k_c_xs_strides,
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_lengths, // output
+                        const std::array<index_t, NDimSpatial + 3>& e_g_n_k_wos_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_strides,
+                        const std::array<index_t, NDimSpatial>& conv_filter_dilations,
+                        const std::array<index_t, NDimSpatial>& input_left_pads,
+                        const std::array<index_t, NDimSpatial>& input_right_pads,
+                        InElementwiseOperation in_element_op,
+                        WeiElementwiseOperation wei_element_op,
+                        OutElementwiseOperation out_element_op,
+                        const index_t split_k) override
+    {
+        return std::make_unique<Argument>(static_cast<const InDataType*>(p_in_grid),
+                                          static_cast<WeiDataType*>(p_wei_grid),
+                                          static_cast<const OutDataType*>(p_out_grid),
+                                          a_g_n_c_wis_lengths, // input
+                                          a_g_n_c_wis_strides,
+                                          b_g_k_c_xs_lengths, // weight
+                                          b_g_k_c_xs_strides,
+                                          e_g_n_k_wos_lengths, // output
+                                          e_g_n_k_wos_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          in_element_op,
+                                          wei_element_op,
+                                          out_element_op,
+                                          split_k);
+    }
+
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceGroupedConvBwdWeight_Wmma_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << K0PerBlock << ", "
+            << getConvBackwardWeightSpecializationString(ConvBackwardWeightSpecialization) << ", "
+            << K1 << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << ABlockTransferDstScalarPerVector_K1 << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferDstScalarPerVector_K1 << ", "
+            << CShuffleMRepeatPerShuffle << ", "
+            << CShuffleNRepeatPerShuffle << ", "
+            << CShuffleBlockTransferScalarPerVector_NPerBlock
+            << ">";
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
index d4bba3bf64..cbb8643627 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_xdl_cshuffle.hpp
@@ -14,6 +14,7 @@
 #include "ck/tensor_operation/gpu/device/device_grouped_conv_bwd_weight.hpp"
 #include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
 #include "ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/device_prop.hpp"
 #include "ck/host_utility/kernel_launch.hpp"
 
@@ -21,32 +22,6 @@ namespace ck {
 namespace tensor_operation {
 namespace device {
 
-namespace {
-
-struct ComputePtrOffsetOfStridedBatch
-{
-    __host__ __device__ constexpr long_index_t GetAPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideA_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetBPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideB_);
-    }
-
-    __host__ __device__ constexpr long_index_t GetCPtrOffset(index_t g_idx) const
-    {
-        return g_idx * static_cast<long_index_t>(BatchStrideC_);
-    }
-
-    index_t BatchStrideA_;
-    index_t BatchStrideB_;
-    index_t BatchStrideC_;
-};
-
-} // namespace
-
 template <typename GridwiseGemm,
           typename FloatA,
           typename FloatB,
@@ -1222,7 +1197,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
         Block2CTileMap block_2_ctile_map_;
 
         // for computing batch offset
-        ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch_;
+        ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_;
 
         index_t M01_;
         index_t N01_;
@@ -1301,7 +1276,7 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
                     remove_reference_t<DeviceOp::BGridDesc_K0_N_K1>,
                     remove_reference_t<DeviceOp::CGridDesc_MBlock_MPerBlock_NBlock_NPerBlock>,
                     remove_reference_t<DeviceOp::Block2CTileMap>,
-                    ComputePtrOffsetOfStridedBatch,
+                    ComputePtrOffsetOfStridedBatch<I0>,
                     has_main_loop>;
 
                 return launch_and_time_kernel(stream_config,
@@ -1348,6 +1323,10 @@ struct DeviceGroupedConvBwdWeight_Xdl_CShuffle
 
     static bool IsSupportedArgument(const Argument& arg)
     {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
         if constexpr(NDimSpatial == 1)
         {
             if constexpr(!is_GNWK_GKXC_GNWC)
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
index e4e793eff8..3e303f5c5e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_wmma_cshuffle.hpp
@@ -471,7 +471,7 @@ struct DeviceGroupedConvFwdMultipleD_Wmma_CShuffle
             auto launch_kernel = [&](auto has_main_k_block_loop) {
                 constexpr bool has_main_loop = has_main_k_block_loop.value;
 
-                const auto kernel = kernel_grouped_conv_fwd_multiple_d_wmma_cshuffle<
+                const auto kernel = kernel_grouped_conv_multiple_d_wmma_cshuffle<
                     GridwiseOp,
                     ADataType,
                     BDataType,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
index 948232ed6f..e19a00299e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp
@@ -43,7 +43,13 @@ struct ComputePtrOffsetOfStridedBatch
         return ds_offset;
     }
 
-    __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+    [[maybe_unused]] __host__ __device__ constexpr long_index_t GetEPtrOffset(index_t g_idx) const
+    {
+        return g_idx * static_cast<long_index_t>(BatchStrideE_);
+    }
+
+    // alias for kernels without multiple D
+    [[maybe_unused]] __host__ __device__ constexpr long_index_t GetCPtrOffset(index_t g_idx) const
     {
         return g_idx * static_cast<long_index_t>(BatchStrideE_);
     }
@@ -52,6 +58,7 @@ struct ComputePtrOffsetOfStridedBatch
     index_t BatchStrideB_;
     Array<ck::index_t, NumDTensor> BatchStrideDs_;
     index_t BatchStrideE_;
+    index_t& BatchStrideC_ = BatchStrideE_; // alias for kernels without multiple D
 };
 
 } // namespace device
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
index 98ade85a30..53b2169bc6 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_d_wmma_cshuffle.hpp
@@ -36,7 +36,7 @@ __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
-        kernel_grouped_conv_fwd_multiple_d_wmma_cshuffle(
+        kernel_grouped_conv_multiple_d_wmma_cshuffle(
             const ADataType* __restrict__ p_a_grid,
             const BDataType* __restrict__ p_b_grid,
             DsPointer p_ds_grid,
@@ -452,11 +452,11 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
     }
 
     // block_id to matrix tile idx (m0, n0) mapping are controlled by {M01, N01}
+    // CheckValidity for kernels without multi D
     template <typename Block2CTileMap>
     __host__ __device__ static constexpr bool
     CheckValidity(const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
                   const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
-                  const DsGridDesc_M_N& ds_grid_desc_m_n,
                   const EGridDesc_M_N& e_grid_desc_m_n,
                   const Block2CTileMap& block_2_ctile_map)
     {
@@ -471,18 +471,6 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
         const auto N  = b_grid_desc_k0_n_k1.GetLength(I1);
         const auto K0 = a_grid_desc_k0_m_k1.GetLength(I0);
 
-        bool valid = true;
-
-        static_for<0, NumDTensor, 1>{}([&](auto i) {
-            valid = valid && (M == ds_grid_desc_m_n[i].GetLength(I0) &&
-                              N == ds_grid_desc_m_n[i].GetLength(I1));
-        });
-
-        if(!valid)
-        {
-            return false;
-        }
-
         if(!(M == e_grid_desc_m_n.GetLength(I0) && N == e_grid_desc_m_n.GetLength(I1) &&
              K0 == b_grid_desc_k0_n_k1.GetLength(I0) && K1 == a_grid_desc_k0_m_k1.GetLength(I2) &&
              K1 == b_grid_desc_k0_n_k1.GetLength(I2)))
@@ -517,6 +505,31 @@ struct GridwiseGemmMultipleD_k0mk1_k0nk1_mn_wmma_cshuffle
         return true;
     }
 
+    template <typename Block2CTileMap>
+    __host__ __device__ static constexpr bool
+    CheckValidity(const AGridDesc_K0_M_K1& a_grid_desc_k0_m_k1,
+                  const BGridDesc_K0_N_K1& b_grid_desc_k0_n_k1,
+                  const DsGridDesc_M_N& ds_grid_desc_m_n,
+                  const EGridDesc_M_N& e_grid_desc_m_n,
+                  const Block2CTileMap& block_2_ctile_map)
+    {
+        const auto M = a_grid_desc_k0_m_k1.GetLength(I1);
+        const auto N = b_grid_desc_k0_n_k1.GetLength(I1);
+        bool valid   = true;
+        static_for<0, NumDTensor, 1>{}([&](auto i) {
+            valid = valid && (M == ds_grid_desc_m_n[i].GetLength(I0) &&
+                              N == ds_grid_desc_m_n[i].GetLength(I1));
+        });
+
+        if(!valid)
+        {
+            return false;
+        }
+
+        return CheckValidity(
+            a_grid_desc_k0_m_k1, b_grid_desc_k0_n_k1, e_grid_desc_m_n, block_2_ctile_map);
+    }
+
     __host__ __device__ static constexpr bool CalculateHasMainKBlockLoop(index_t K)
     {
         const index_t num_loop = K / (K0PerBlock * K1);
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
index 29fe298e42..c3e333e720 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_dl_instance.hpp
@@ -6,8 +6,6 @@
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
-
 namespace ck {
 namespace tensor_operation {
 namespace device {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp
new file mode 100644
index 0000000000..40c4d558b8
--- /dev/null
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp
@@ -0,0 +1,119 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using F16 = ck::half_t;
+using F32 = float;
+using I8  = int8_t;
+using I32 = int32_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using namespace ck::tensor_layout::convolution;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvBwdWeightDefault =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Default;
+
+static constexpr auto ConvBwdWeightFilter1x1Stride1Pad0 =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_wmma_f16_instances =
+    std::tuple<
+        // clang-format off
+        //#####################################|    NumDim|       A|       B|       C| AData| BData|  CData| AccData|            A|           B|            C|    ConvForward| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|       CShuffle|       CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|   Spatial|  Layout|  Layout|  Layout|  Type|  Type|   Type|    Type|  Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeatPerWave| NRepeatPerWave|            _MBlock_MPerBlock| ScalarPerVector|
+        //#####################################|          |        |        |        |      |      |       |        |    Operation|   Operation|    Operation|               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |     PerShuffle|     PerShuffle|            _NBlock_NPerBlock|      _NPerBlock|
+        //#####################################|          |        |        |        |      |      |       |        |             |            |             |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |               |               |                             |                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    64,    64,     4,  8,    16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          256,   128,   256,     8,  8,    16,   16,       2,       8,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          256,   256,   128,     8,  8,    16,   16,       8,       2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,               2>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          256,   256,    64,     8,  8,    16,   16,       4,       2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,               4>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          256,    64,   256,     8,  8,    16,   16,       2,       4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 16>,              4>,
+        // blocksize=128
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    64,   128,     8,  8,    16,   16,       2,       4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   128,    64,     8,  8,    16,   16,       4,       2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   128,   128,     8,  8,    16,   16,       4,       4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    32,   256,     8,  8,    16,   16,       1,       8,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   256,    32,     8,  8,    16,   16,       8,       1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        // blocksize=64
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           64,    64,    32,     8,  8,    16,   16,       4,       1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           64,    32,    64,     8,  8,    16,   16,       1,       4,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           64,    64,    64,     8,  8,    16,   16,       2,       4,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           64,   128,    32,     8,  8,    16,   16,       8,       1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           64,    32,   128,     8,  8,    16,   16,       1,       8,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        // blocksize=32
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    16,    32,     8,  8,    16,   16,       1,       2,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    16,    64,     8,  8,    16,   16,       1,       4,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    32,    64,     8,  8,    16,   16,       2,       4,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    32,    32,     8,  8,    16,   16,       2,       2,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    64,    32,     8,  8,    16,   16,       4,       2,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    64,    16,     8,  8,    16,   16,       4,       1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    32,    16,     8,  8,    16,   16,       2,       1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+        // clang-format on
+        >;
+
+template <index_t NDSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename CLayout,
+          ConvolutionBackwardWeightSpecialization ConvSpec>
+using device_grouped_conv_bwd_weight_wmma_i8_instances =
+    std::tuple<
+        // clang-format off
+        //#####################################|    NumDim|       A|       B|       C| AData| BData|  CData| AccData|            A|           B|            C|    ConvForward| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|       CShuffle|       CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //#####################################|   Spatial|  Layout|  Layout|  Layout|  Type|  Type|   Type|    Type|  Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeatPerWave| NRepeatPerWave|            _MBlock_MPerBlock| ScalarPerVector|
+        //#####################################|          |        |        |        |      |      |       |        |    Operation|   Operation|    Operation|               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |     PerShuffle|     PerShuffle|            _NBlock_NPerBlock|      _NPerBlock|
+        //#####################################|          |        |        |        |      |      |       |        |             |            |             |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |               |               |                             |                |
+        // generic instance
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout,  I8,    I8,  I8,  I32,   PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    64,    64,     4,  8,     16,   16,       2,       2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               1,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,             1,             8,         1,           1,           1,               S<1, 32, 1, 4>,               1>,
+        // blocksize=256
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          256,    64,   256,     8,   8,    16,   16,       2,       4,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               2,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              8,         1,           1,           1,               S<1, 32, 1, 8>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          256,   256,    64,     8,   8,    16,   16,       4,       2,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            2,              8,         1,           1,           1,               S<1, 64, 1, 4>,               8>,
+        // blocksize=128
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   128,   256,     8,   8,    16,   16,       4,       8,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    64,   256,     8,   8,    16,   16,       2,       8,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               4,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    32,   256,     8,   8,    16,   16,       1,       8,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               2,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,    64,   128,     8,   8,    16,   16,       2,       4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               4,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   128,    64,     8,   8,    16,   16,       4,       2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            4,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   256,    32,     8,   8,    16,   16,       8,       1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            2,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   256,    64,     8,   8,    16,   16,       8,       2,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            4,              8,         1,           1,           1,               S<1, 16, 1, 8>,               2>,      
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   256,   128,     8,   8,    16,   16,       8,       4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>, 
+        // blocksize=64
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           64,    32,   128,     8,   8,    16,   16,       1,       8,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               4,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           64,    64,   128,     8,   8,    16,   16,       2,       8,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 32, 1, 2>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           64,   128,    64,     8,   8,    16,   16,       8,       2,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           64,   128,    32,     8,   8,    16,   16,       8,       1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,      S<8, 8, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            4,              8,         1,           1,           1,               S<1, 16, 1, 4>,               8>,
+        // blocksize=32
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    16,    64,     8,   8,    16,   16,       1,       4,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               4,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    64,    64,     8,   8,    16,   16,       4,       4,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,           16,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    32,    32,     8,   8,    16,   16,       2,       2,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            8,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>,  
+        DeviceGroupedConvBwdWeight_Wmma_CShuffle<NDSpatial, ALayout, BLayout, CLayout, I8,    I8,  I8,  I32,    PassThrough, PassThrough, PassThrough,       ConvSpec,           32,    64,    16,     8,   8,    16,   16,       4,       1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              16,              8,         1,      S<8, 4, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,            4,              8,         1,           1,           1,               S<1, 16, 1, 2>,               8>
+        // clang-format on
+        >;
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index 9f536f210d..9310e9e57d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -163,6 +163,30 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances(
@@ -177,6 +201,31 @@ void add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instances
                                                            PassThrough,
                                                            PassThrough>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
 #ifdef CK_ENABLE_BF16
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
     std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
@@ -202,6 +251,30 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances
                                                            PassThrough,
                                                            PassThrough,
                                                            PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
 #endif
 #ifdef CK_ENABLE_FP32
 void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instances(
@@ -231,6 +304,31 @@ void add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_
                                                            BF8,
                                                            F8>>>& instances);
 #endif
+#ifdef CK_ENABLE_INT8
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances);
+#endif
 
 #ifdef DL_KERNELS
 // dl
@@ -694,6 +792,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instances(
                         op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
@@ -708,6 +810,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                     add_device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
                         op_ptrs);
                 }
+#endif
+#ifdef CK_ENABLE_INT8
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                                  is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
 #endif
             }
             else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
@@ -737,6 +849,10 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
                         op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+                        op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
@@ -752,6 +868,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                         op_ptrs);
                 }
 #endif
+#ifdef CK_ENABLE_INT8
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                                  is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
+                        op_ptrs);
+                    add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+                        op_ptrs);
+                }
+#endif
 #if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                                   is_same_v<OutDataType, half_t> &&
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt
index 660f2544ac..cfd829f87e 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/CMakeLists.txt
@@ -1,16 +1,16 @@
 set(GROUPED_CONV1D_BWD_WEIGHT
-    device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
-    device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
-    device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp)
+    xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
+    xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
+    xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp)
 
 if(DL_KERNELS)
     list(APPEND GROUPED_CONV1D_BWD_WEIGHT
-        device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
-        device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
-        device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
-        device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
-        device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
-        device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp)
+        dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
+        dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
+        dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
+        dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
+        dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
+        dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp)
 endif()
 
 add_instance_library(device_grouped_conv1d_bwd_weight_instance ${GROUPED_CONV1D_BWD_WEIGHT})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/dl/device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv1d_bwd_weight/xdl/device_grouped_conv1d_bwd_weight_xdl_gnwc_gkxc_gnwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
index 578e9029e9..8a896b06c7 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/CMakeLists.txt
@@ -1,19 +1,19 @@
 set(GROUPED_CONV2D_BWD_WEIGHT
-    device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
-    device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-    device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
-    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
-    device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp)
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+    xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp)
 
 if(DL_KERNELS)
     list(APPEND GROUPED_CONV2D_BWD_WEIGHT
-        device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
-        device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
-        device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
-        device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
-        device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
-        device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp)
+        dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
+        dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
+        dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
+        dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
+        dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
+        dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp)
 endif()
 
 add_instance_library(device_grouped_conv2d_bwd_weight_instance ${GROUPED_CONV2D_BWD_WEIGHT})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/dl/device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_gnhwc_gkyxc_gnhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv2d_bwd_weight/xdl/device_grouped_conv2d_bwd_weight_xdl_nhwgc_gkyxc_nhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
index 65ea645a39..a2c4b1f80b 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
@@ -1,20 +1,30 @@
 set(GROUPED_CONV3D_BWD_WEIGHT
-    device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
-    device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp)
+    xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+    xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+    xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
+    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp)
 
 if(DL_KERNELS)
     list(APPEND GROUPED_CONV3D_BWD_WEIGHT
-        device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
-        device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
-        device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
-        device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
-        device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-        device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp)
+        dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+        dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
+        dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
+        dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+        dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
+        dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp)
 endif()
 
+list(APPEND GROUPED_CONV3D_BWD_WEIGHT
+    wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
+    wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp)
+
 add_instance_library(device_grouped_conv3d_bwd_weight_instance ${GROUPED_CONV3D_BWD_WEIGHT})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/dl/device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..26171031aa
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_f16_instances<3,
+                                                          GNDHWC,
+                                                          GKZYXC,
+                                                          GNDHWK,
+                                                          ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
new file mode 100644
index 0000000000..9399655700
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_f16_instances<3,
+                                                          GNDHWC,
+                                                          GKZYXC,
+                                                          GNDHWK,
+                                                          ConvBwdWeightDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..3187173677
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_i8_instances<3,
+                                                         GNDHWC,
+                                                         GKZYXC,
+                                                         GNDHWK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
new file mode 100644
index 0000000000..19451b855e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           GNDHWC,
+                                                           GKZYXC,
+                                                           GNDHWK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_i8_instances<3,
+                                                         GNDHWC,
+                                                         GKZYXC,
+                                                         GNDHWK,
+                                                         ConvBwdWeightDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..71b4365d62
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_f16_instances<3,
+                                                          NDHWGC,
+                                                          GKZYXC,
+                                                          NDHWGK,
+                                                          ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
new file mode 100644
index 0000000000..06134d1f20
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           F16,
+                                                           F16,
+                                                           F16,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_f16_instances<3,
+                                                          NDHWGC,
+                                                          GKZYXC,
+                                                          NDHWGK,
+                                                          ConvBwdWeightDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
new file mode 100644
index 0000000000..39e2cc8c28
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_i8_instances<3,
+                                                         NDHWGC,
+                                                         GKZYXC,
+                                                         NDHWGK,
+                                                         ConvBwdWeightFilter1x1Stride1Pad0>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
new file mode 100644
index 0000000000..0d56619aeb
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+#include "ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_wmma_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+void add_device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instances(
+    std::vector<std::unique_ptr<DeviceGroupedConvBwdWeight<3,
+                                                           NDHWGC,
+                                                           GKZYXC,
+                                                           NDHWGK,
+                                                           int8_t,
+                                                           int8_t,
+                                                           int8_t,
+                                                           PassThrough,
+                                                           PassThrough,
+                                                           PassThrough>>>& instances)
+{
+    add_device_operation_instances(
+        instances,
+        device_grouped_conv_bwd_weight_wmma_i8_instances<3,
+                                                         NDHWGC,
+                                                         GKZYXC,
+                                                         NDHWGK,
+                                                         ConvBwdWeightDefault>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
similarity index 100%
rename from library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
rename to library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
diff --git a/profiler/README.md b/profiler/README.md
index d081fc33e9..a2d370b210 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -147,7 +147,9 @@ GB/s: 127.947
 # arg1: tensor operation (grouped_conv_bwd_weight: Grouped Convolution Backward Weight)
 # arg2: data type (0: Input fp32, Weight fp32, Output fp32
 #                  1: Input fp16, Weight fp16, Output fp16
-#                  2: Input bf16, Weight fp32, Output bf16)
+#                  2: Input bf16, Weight fp32, Output bf16
+#                  3: Input fp16, Weight fp16, Output fp16, Gemm bf8@fp8
+#                  4: Input int8, Weight int8, Output int8)
 # arg3: tensor layout (0: Input[G, N, C, Hi, Wi], Weight[G, K, C, Y, X], Output[G, N, K, Ho, Wo]
 #                      1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, N, Ho, Wo, K]
 #                      2: Input[N, Hi, Wi, G, C], Weight[G, K, Y, X, C], Output[N, Ho, Wo, G, K]
@@ -167,7 +169,7 @@ GB/s: 127.947
 # SplitK
 
  ################                   op   datatype  layout  verify  init  log  time  Ndims  G   N   K   C  Y  X  Hi  Wi  Sy  Sx  Dy  Dx  LeftPy  LeftPx  RightPy  RightPx  SplitK
-./bin/ckProfiler grouped_conv_bwd_weight          1       0       1     1    0     1      2 32 256 256 512  3  3  28  28   1   1   1   1       1       0        0        0       1
+./bin/ckProfiler grouped_conv_bwd_weight         1       1      0     1    0     1      2 32 256 256 512  3  3  28  28   1   1   1   1       1       0        0        0       1
 
  ```
 
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index 9a9b5a0f74..bd1016f286 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -20,10 +20,11 @@ enum struct ConvLayout
 
 enum struct ConvDataType
 {
-    F32_F32_F32,       // 0
-    F16_F16_F16,       // 1
-    BF16_F32_BF16,     // 2
-    F16_F16_F16_BF8_F8 // 3
+    F32_F32_F32,        // 0
+    F16_F16_F16,        // 1
+    BF16_F32_BF16,      // 2
+    F16_F16_F16_BF8_F8, // 3
+    I8_I8_I8            // 4
 };
 
 #define OP_NAME "grouped_conv_bwd_weight"
@@ -35,7 +36,8 @@ static void print_helper_msg()
               << "arg2: data type (0: Input fp32, Weight fp32, Output fp32\n"
               << "                 1: Input fp16, Weight fp16, Output fp16\n"
               << "                 2: Input bf16, Weight fp32, Output bf16\n"
-              << "                 3: Input fp16, Weight fp16, Output fp16, Gemm bf8@fp8)\n"
+              << "                 3: Input fp16, Weight fp16, Output fp16, Gemm bf8@fp8\n"
+              << "                 4: Input int8, Weight int8, Output int8)\n"
               << "arg3: tensor layout (0: Input[G, N, C, Hi, Wi], Weight[G, K, C, Y, X], Output[G, "
                  "N, K, Ho, Wo]\n"
               << "                     1: Input[G, N, Hi, Wi, C], Weight[G, K, Y, X, C], Output[G, "
@@ -196,6 +198,11 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
             // fp32 atomic add is used for weight tensor in bf16 kernel
             return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
+        else if(data_type == ConvDataType::I8_I8_I8)
+        {
+            return profile(
+                I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, int8_t{}, int8_t{}, int8_t{}, int8_t{}, int8_t{});
+        }
     }
     else if(num_dim_spatial == 3 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
     {
@@ -216,6 +223,11 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
         {
             return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{}, BF8{}, F8{});
         }
+        else if(data_type == ConvDataType::I8_I8_I8)
+        {
+            return profile(
+                I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, int8_t{}, int8_t{}, int8_t{}, int8_t{}, int8_t{});
+        }
     }
 
     std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/test/grouped_convnd_bwd_weight/CMakeLists.txt b/test/grouped_convnd_bwd_weight/CMakeLists.txt
index f1a3f04b77..b167943c97 100644
--- a/test/grouped_convnd_bwd_weight/CMakeLists.txt
+++ b/test/grouped_convnd_bwd_weight/CMakeLists.txt
@@ -1,11 +1,20 @@
-list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
+list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
+
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
+ if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
    add_gtest_executable(test_grouped_convnd_bwd_weight test_grouped_convnd_bwd_weight.cpp)
    target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
-   add_gtest_executable(test_grouped_convnd_bwd_weight_interface test_grouped_convnd_bwd_weight_interface.cpp)
-   target_link_libraries(test_grouped_convnd_bwd_weight_interface PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+   add_gtest_executable(test_grouped_convnd_bwd_weight_interface test_grouped_convnd_bwd_weight_interface_xdl.cpp)
+   target_link_libraries(test_grouped_convnd_bwd_weight_interface PRIVATE utility)
+   set(target 1)
+ endif()
+ if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
+   add_gtest_executable(test_grouped_convnd_bwd_weight test_grouped_convnd_bwd_weight.cpp)
+   target_link_libraries(test_grouped_convnd_bwd_weight PRIVATE utility device_grouped_conv1d_bwd_weight_instance device_grouped_conv2d_bwd_weight_instance device_grouped_conv3d_bwd_weight_instance)
+   add_gtest_executable(test_grouped_convnd_bwd_weight_interface test_grouped_convnd_bwd_weight_interface_wmma.cpp)
+   target_link_libraries(test_grouped_convnd_bwd_weight_interface PRIVATE utility)
    set(target 1)
  endif()
 endforeach()
\ No newline at end of file
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
index 2132020320..856f9fd15c 100644
--- a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight.cpp
@@ -11,6 +11,7 @@
 
 #include "ck/utility/common_header.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/host_utility/device_prop.hpp"
 
 #include "profiler/profile_grouped_conv_bwd_weight_impl.hpp"
 
@@ -33,8 +34,9 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
 
     bool skip_case(const ck::utils::conv::ConvParam& params, const ck::index_t split_k)
     {
-        // Odd K or C values are supported only by DL kernel (only applies to fp16)
-        // DL kernel currently supports only `split_k=1`
+        // Odd K or C values are supported only by DL and WMMA
+        // kernels (only applies to fp16)
+        // DL and WMMA kernels currently support only `split_k=1`
         if constexpr(std::is_same_v<InDataType, ck::half_t>)
         {
             if(split_k != 1 && (params.K_ % 2 != 0 || params.C_ % 2 != 0))
@@ -53,6 +55,42 @@ class TestGroupedConvndBwdWeight : public ::testing::Test
             }
         }
 
+        const bool is_navi3x = ck::get_device_name() == "gfx1100" ||
+                               ck::get_device_name() == "gfx1101" ||
+                               ck::get_device_name() == "gfx1102";
+        if(is_navi3x)
+        {
+            // on navi3x only support for 3d is implemented
+            if constexpr(NDimSpatial{} != 3)
+            {
+                return true;
+            }
+            // on navi3x only support for i8 and fp16 is implemented
+            if constexpr(!((std::is_same_v<InDataType, int8_t> &&
+                            std::is_same_v<WeiDataType, int8_t> &&
+                            std::is_same_v<OutDataType, int8_t>) ||
+                           (std::is_same_v<InDataType, ck::half_t> &&
+                            std::is_same_v<WeiDataType, ck::half_t> &&
+                            std::is_same_v<OutDataType, ck::half_t>)))
+            {
+                return true;
+            }
+            // WMMA kernel is only supported for split_k=1
+            if(split_k != 1)
+            {
+                return true;
+            }
+        }
+        else
+        {
+            // support for i8 is only implemented on navi3x
+            if constexpr(std::is_same_v<InDataType, int8_t> &&
+                         std::is_same_v<WeiDataType, int8_t> && std::is_same_v<OutDataType, int8_t>)
+            {
+                return true;
+            }
+        }
+
         return false;
     }
 
@@ -120,9 +158,11 @@ using KernelTypes3d = ::testing::Types<
     std::tuple<float, float, float, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
     std::tuple<ck::half_t, ck::half_t, ck::half_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
     std::tuple<ck::bhalf_t, float, ck::bhalf_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
+    std::tuple<int8_t, int8_t, int8_t, GNDHWC, GKZYXC, GNDHWK, ck::Number<3>>,
     std::tuple<float, float, float, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
     std::tuple<ck::half_t, ck::half_t, ck::half_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
-    std::tuple<ck::bhalf_t, float, ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>>;
+    std::tuple<ck::bhalf_t, float, ck::bhalf_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>,
+    std::tuple<int8_t, int8_t, int8_t, NDHWGC, GKZYXC, NDHWGK, ck::Number<3>>>;
 
 TYPED_TEST_SUITE(TestGroupedConvndBwdWeight1d, KernelTypes1d);
 TYPED_TEST_SUITE(TestGroupedConvndBwdWeight2d, KernelTypes2d);
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface_wmma.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface_wmma.cpp
new file mode 100644
index 0000000000..1dcb8f866d
--- /dev/null
+++ b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface_wmma.cpp
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <cstdlib>
+#include <iostream>
+#include <initializer_list>
+#include <tuple>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/convolution_backward_weight_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_wmma_cshuffle.hpp"
+
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+#include <gtest/gtest.h>
+
+using F16         = ck::half_t;
+using F32         = float;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+using ConvolutionBackwardWeightSpecialization =
+    ck::tensor_operation::device::ConvolutionBackwardWeightSpecialization;
+
+static constexpr auto ConvBwdWeightDefault = ConvolutionBackwardWeightSpecialization::Default;
+static constexpr auto Filter1x1Stride1Pad0 =
+    ConvolutionBackwardWeightSpecialization::Filter1x1Stride1Pad0;
+
+template <typename Tuple, ConvolutionBackwardWeightSpecialization ConvSpec>
+class TestGroupedConvndBwdWeight : public ::testing::Test
+{
+    protected:
+    using OutLayout                          = std::tuple_element_t<0, Tuple>;
+    using WeiLayout                          = std::tuple_element_t<1, Tuple>;
+    using InLayout                           = std::tuple_element_t<2, Tuple>;
+    static constexpr ck::index_t NDimSpatial = std::tuple_element_t<3, Tuple>{};
+
+    // clang-format off
+    using GroupedConvBwdWeightDeviceInstance = ck::tensor_operation::device::DeviceGroupedConvBwdWeight_Wmma_CShuffle
+        //|    NumDim|       A|       B|       C| AData| BData|  CData| AccData|            A|           B|            C|    ConvForward| Block|  MPer|  NPer|  KPer| K1|  MPer| NPer| MRepeat| NRepeat|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|       CShuffle|       CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
+        //|   Spatial|  Layout|  Layout|  Layout|  Type|  Type|   Type|    Type|  Elementwise| Elementwise|  Elementwise| Specialization|  Size| Block| Block| Block|   |  WMMA| WMMA|        |        |   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MRepeatPerWave| NRepeatPerWave|            _MBlock_MPerBlock| ScalarPerVector|
+        //|          |        |        |        |      |      |       |        |    Operation|   Operation|    Operation|               |      |      |      |      |   |      |     |        |        | Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |     PerShuffle|     PerShuffle|            _NBlock_NPerBlock|      _NPerBlock|
+        //|          |        |        |        |      |      |       |        |             |            |             |               |      |      |      |      |   |      |     |        |        |                |               |               |               |               |               |          |                |               |               |              |               |               |          |               |               |                             |                |
+    <NDimSpatial, InLayout, WeiLayout, OutLayout,  F16,   F16,  F16,  F32,  PassThrough, PassThrough, PassThrough,       ConvSpec,          128,   128,   128,     8,  8,    16,   16,       4,       4,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,               8>;
+    // clang-format on
+
+    ck::utils::conv::ConvParam conv_param;
+
+    template <ck::index_t SplitK>
+    bool Run()
+    {
+
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        std::array<ck::index_t, NDimSpatial + 3> input_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> filter_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> output_lengths{};
+        std::array<ck::index_t, NDimSpatial + 3> input_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> weights_strides{};
+        std::array<ck::index_t, NDimSpatial + 3> output_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+        std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+        std::array<ck::index_t, NDimSpatial> input_left_pads{};
+        std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+        auto range_copy = [](const auto& from, auto to) { std::copy(begin(from), end(from), to); };
+
+        range_copy(in_g_n_c_wis_desc.GetLengths(), begin(input_lengths));
+        range_copy(in_g_n_c_wis_desc.GetStrides(), begin(input_strides));
+        range_copy(wei_g_k_c_xs_desc.GetLengths(), begin(filter_lengths));
+        range_copy(wei_g_k_c_xs_desc.GetStrides(), begin(weights_strides));
+        range_copy(out_g_n_k_wos_desc.GetLengths(), begin(output_lengths));
+        range_copy(out_g_n_k_wos_desc.GetStrides(), begin(output_strides));
+        range_copy(conv_param.conv_filter_strides_, begin(conv_filter_strides));
+        range_copy(conv_param.conv_filter_dilations_, begin(conv_filter_dilations));
+        range_copy(conv_param.input_left_pads_, begin(input_left_pads));
+        range_copy(conv_param.input_right_pads_, begin(input_right_pads));
+
+        auto conv = GroupedConvBwdWeightDeviceInstance{};
+
+        auto argument = conv.MakeArgument(nullptr,
+                                          nullptr,
+                                          nullptr,
+                                          input_lengths,
+                                          input_strides,
+                                          filter_lengths,
+                                          weights_strides,
+                                          output_lengths,
+                                          output_strides,
+                                          conv_filter_strides,
+                                          conv_filter_dilations,
+                                          input_left_pads,
+                                          input_right_pads,
+                                          PassThrough{},
+                                          PassThrough{},
+                                          PassThrough{},
+                                          SplitK);
+        return conv.IsSupportedArgument(argument);
+    }
+};
+
+using namespace ck::tensor_layout::convolution;
+
+using KernelTypes3d = ::testing::Types<std::tuple<GNDHWK, GKZYXC, GNDHWC, ck::Number<3>>,
+                                       std::tuple<NDHWGK, GKZYXC, NDHWGC, ck::Number<3>>>;
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightFilter1x13d
+    : public TestGroupedConvndBwdWeight<Tuple, Filter1x1Stride1Pad0>
+{
+};
+
+template <typename Tuple>
+class TestGroupedConvndBwdWeightDefault3d
+    : public TestGroupedConvndBwdWeight<Tuple, ConvBwdWeightDefault>
+{
+};
+
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightFilter1x13d, KernelTypes3d);
+TYPED_TEST_SUITE(TestGroupedConvndBwdWeightDefault3d, KernelTypes3d);
+
+TYPED_TEST(TestGroupedConvndBwdWeightFilter1x13d, SpecializationCheck)
+{
+    // Check filter 3x3x3 instead of 1x1x1
+    this->conv_param = {
+        3, 2, 4, 192, 192, {3, 3, 3}, {28, 28, 28}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    bool is_supported = this->template Run<1>();
+    EXPECT_FALSE(is_supported);
+
+    // Check strides 2x2x2 instead of 1x1x1
+    this->conv_param = {
+        3, 2, 4, 192, 192, {1, 1, 1}, {28, 28, 28}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    is_supported = this->template Run<1>();
+    EXPECT_FALSE(is_supported);
+
+    // Check with pad
+    this->conv_param = {
+        3, 2, 4, 192, 192, {1, 1, 1}, {28, 28, 28}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}};
+    is_supported = this->template Run<1>();
+    EXPECT_FALSE(is_supported);
+
+    // Supported version
+    this->conv_param = {
+        3, 2, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    is_supported = this->template Run<1>();
+    EXPECT_TRUE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeightDefault3d, VectorLoadCheck)
+{
+    // vector load for A
+    this->conv_param = {
+        3, 2, 128, 129, 256, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    bool is_supported = this->template Run<1>();
+    EXPECT_FALSE(is_supported);
+    // vector load for B, E, Ds
+    this->conv_param = {
+        3, 2, 128, 128, 257, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    is_supported = this->template Run<1>();
+    EXPECT_FALSE(is_supported);
+}
+
+TYPED_TEST(TestGroupedConvndBwdWeightDefault3d, SplitKCheck)
+{
+    // SplitK=1
+    this->conv_param = {
+        3, 2, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    bool is_supported = this->template Run<1>();
+    EXPECT_TRUE(is_supported);
+    // SplitK=2
+    this->conv_param = {
+        3, 2, 128, 128, 256, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}};
+    is_supported = this->template Run<2>();
+    EXPECT_FALSE(is_supported);
+}
diff --git a/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp b/test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface_xdl.cpp
similarity index 100%
rename from test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface.cpp
rename to test/grouped_convnd_bwd_weight/test_grouped_convnd_bwd_weight_interface_xdl.cpp

From bf0addb5753fb44e39e33e0edfa2158d6f1ffce7 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 17 Oct 2023 09:24:02 -0500
Subject: [PATCH 175/194] added ab_elementwise_op support into splitK Gemm
 (#956)

* add ab_elementwise

* fixed ci

* fixed a merge issue

* fixed pr comments

* fixed a conflict

* remove 61_example

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 .../impl/device_gemm_xdl_splitk_c_shuffle.hpp | 102 +++++++++++++++---
 .../gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp  |  27 +++--
 2 files changed, 104 insertions(+), 25 deletions(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
index d20b008b88..b0193857c8 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl_splitk_c_shuffle.hpp
@@ -127,7 +127,50 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
         PipelineVer,
         ComputeType>;
 
-    using Argument              = typename GridwiseGemm::Argument;
+    struct Argument : public GridwiseGemm::Argument
+    {
+        Argument(const ADataType* p_a_grid_,
+                 const BDataType* p_b_grid_,
+                 CDataType* p_c_grid_,
+                 index_t M_,
+                 index_t N_,
+                 index_t K_,
+                 index_t StrideA_,
+                 index_t StrideB_,
+                 index_t StrideC_,
+                 index_t MPadded_,
+                 index_t NPadded_,
+                 index_t KPadded_,
+                 index_t K0_,
+                 index_t k_batch_,
+                 AElementwiseOperation a_element_op_,
+                 BElementwiseOperation b_element_op_,
+                 CElementwiseOperation c_element_op_)
+            : GridwiseGemm::Argument(p_a_grid_,
+                                     p_b_grid_,
+                                     p_c_grid_,
+                                     M_,
+                                     N_,
+                                     K_,
+                                     StrideA_,
+                                     StrideB_,
+                                     StrideC_,
+                                     MPadded_,
+                                     NPadded_,
+                                     KPadded_,
+                                     K0_,
+                                     k_batch_),
+              a_element_op(a_element_op_),
+              b_element_op(b_element_op_),
+              c_element_op(c_element_op_)
+        {
+        }
+
+        AElementwiseOperation a_element_op;
+        BElementwiseOperation b_element_op;
+        CElementwiseOperation c_element_op;
+    };
+
     using DefaultBlock2CTileMap = typename GridwiseGemm::DefaultBlock2CTileMap;
 
     // Invoker
@@ -168,8 +211,17 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                      karg.M * karg.N * sizeof(CDataType),
                                                      stream_config.stream_id_));
 
-                ave_time = launch_and_time_kernel(
-                    stream_config, kernel, dim3(gdx, gdy, gdz), dim3(BlockSize), 0, karg, b2c_map);
+                ave_time =
+                    launch_and_time_kernel(stream_config,
+                                           kernel,
+                                           dim3(gdx, gdy, gdz),
+                                           dim3(BlockSize),
+                                           0,
+                                           static_cast<typename GridwiseGemm::Argument>(karg),
+                                           b2c_map,
+                                           karg.a_element_op,
+                                           karg.b_element_op,
+                                           karg.c_element_op);
             };
 
             if(has_main_k0_block_loop)
@@ -180,7 +232,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              true,
                                                              InMemoryDataOperationEnum::Set,
-                                                             DefaultBlock2CTileMap>;
+                                                             DefaultBlock2CTileMap,
+                                                             AElementwiseOperation,
+                                                             BElementwiseOperation,
+                                                             CElementwiseOperation>;
 
                     Run(kernel);
                 }
@@ -190,7 +245,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              true,
                                                              InMemoryDataOperationEnum::AtomicAdd,
-                                                             DefaultBlock2CTileMap>;
+                                                             DefaultBlock2CTileMap,
+                                                             AElementwiseOperation,
+                                                             BElementwiseOperation,
+                                                             CElementwiseOperation>;
 
                     Run(kernel);
                 }
@@ -203,7 +261,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              false,
                                                              InMemoryDataOperationEnum::Set,
-                                                             DefaultBlock2CTileMap>;
+                                                             DefaultBlock2CTileMap,
+                                                             AElementwiseOperation,
+                                                             BElementwiseOperation,
+                                                             CElementwiseOperation>;
 
                     Run(kernel);
                 }
@@ -213,7 +274,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         kernel_gemm_xdlops_v2r4r2_simplified<GridwiseGemm,
                                                              false,
                                                              InMemoryDataOperationEnum::AtomicAdd,
-                                                             DefaultBlock2CTileMap>;
+                                                             DefaultBlock2CTileMap,
+                                                             AElementwiseOperation,
+                                                             BElementwiseOperation,
+                                                             CElementwiseOperation>;
 
                     Run(kernel);
                 }
@@ -261,12 +325,12 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                              index_t StrideA,
                              index_t StrideB,
                              index_t StrideC,
-                             AElementwiseOperation,
-                             BElementwiseOperation,
-                             CElementwiseOperation,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CElementwiseOperation c_element_op,
                              index_t KBatch)
     {
-        return Argument{p_a,
+        return Argument(p_a,
                         p_b,
                         p_c,
                         M,
@@ -279,7 +343,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                         GridwiseGemm::CalculateNPadded(N),
                         GridwiseGemm::CalculateKPadded(K, KBatch),
                         GridwiseGemm::CalculateK0(K, KBatch),
-                        KBatch};
+                        KBatch,
+                        a_element_op,
+                        b_element_op,
+                        c_element_op);
     }
 
     static auto MakeInvoker() { return Invoker{}; }
@@ -294,9 +361,9 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                                       index_t StrideA,
                                                       index_t StrideB,
                                                       index_t StrideC,
-                                                      AElementwiseOperation,
-                                                      BElementwiseOperation,
-                                                      CElementwiseOperation,
+                                                      AElementwiseOperation a_element_op,
+                                                      BElementwiseOperation b_element_op,
+                                                      CElementwiseOperation c_element_op,
                                                       ck::index_t KBatch = 1) override
     {
         return std::make_unique<Argument>(static_cast<const ADataType*>(p_a),
@@ -312,7 +379,10 @@ struct DeviceGemmXdlSplitKCShuffle : public DeviceGemmSplitK<ALayout,
                                           GridwiseGemm::CalculateNPadded(N),
                                           GridwiseGemm::CalculateKPadded(K, KBatch),
                                           GridwiseGemm::CalculateK0(K, KBatch),
-                                          KBatch);
+                                          KBatch,
+                                          a_element_op,
+                                          b_element_op,
+                                          c_element_op);
     }
 
     // polymorphic
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
index f9744d7167..d969b3f657 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r4r2.hpp
@@ -22,13 +22,19 @@ namespace ck {
 template <typename GridwiseGemm,
           bool HasMainKBlockLoop,
           InMemoryDataOperationEnum CGlobalMemoryDataOperation,
-          typename Block2CTileMap>
+          typename Block2CTileMap,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CElementwiseOperation>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
     __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
 #endif
         kernel_gemm_xdlops_v2r4r2_simplified(typename GridwiseGemm::Argument karg,
-                                             const Block2CTileMap& b2c_map)
+                                             const Block2CTileMap& b2c_map,
+                                             const AElementwiseOperation a_element_op,
+                                             const BElementwiseOperation b_element_op,
+                                             const CElementwiseOperation c_element_op)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
     defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
@@ -37,10 +43,13 @@ __global__ void
     __shared__ uint8_t p_shared[shared_size];
 
     GridwiseGemm::template Run<HasMainKBlockLoop, CGlobalMemoryDataOperation>(
-        karg, static_cast<void*>(p_shared), b2c_map);
+        karg, static_cast<void*>(p_shared), b2c_map, a_element_op, b_element_op, c_element_op);
 #else
     ignore = karg;
     ignore = b2c_map;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = c_element_op;
 #endif // end of if (defined(__gfx908__) || defined(__gfx90a__))
 }
 
@@ -577,7 +586,10 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
               typename Block2CTileMap>
     __device__ static void Run(const Argument& karg,
                                void* __restrict__ p_shared_block,
-                               const Block2CTileMap& block_2_ctile_map)
+                               const Block2CTileMap& block_2_ctile_map,
+                               const AElementwiseOperation a_element_op = AElementwiseOperation{},
+                               const BElementwiseOperation b_element_op = BElementwiseOperation{},
+                               const CElementwiseOperation c_element_op = CElementwiseOperation{})
     {
         const FloatA* p_a_grid           = karg.p_a_grid;
         const FloatB* p_b_grid           = karg.p_b_grid;
@@ -590,9 +602,6 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
         const auto c_grid_desc_mblock_mperblock_nblock_nperblock =
             MakeCGridDesc_MBlock_MPerBlock_NBlock_NPerBlock(c_grid_desc_m_n);
-        const AElementwiseOperation a_element_op = AElementwiseOperation{};
-        const BElementwiseOperation b_element_op = BElementwiseOperation{};
-        const CElementwiseOperation c_element_op = CElementwiseOperation{};
 
         const auto a_grid_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_a_grid, a_b_k0_m_k1_grid_desc.GetElementSpaceSize());
@@ -761,8 +770,8 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_v2r4r2
 
         auto blockwise_gemm = BlockwiseGemmXdlops_k0mk1_k0nk1_m0n0m1n1m2m3m4n2_Selector<
             BlockSize,
-            ComputeType,
-            ComputeType,
+            ComputeType, // ComputeType A
+            ComputeType, // ComputeType B
             FloatAcc,
             decltype(a_k0_m_k1_block_desc),
             decltype(b_k0_n_k1_block_desc),

From 1cc36ba5fb805aa99cbaba38b5607dda85f0f99d Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Tue, 17 Oct 2023 20:17:58 -0500
Subject: [PATCH 176/194] Add contraction_multi_abd (#972)

* add gridwise_multi_abd

* move element_op into RunRead

* merge element_wise op with data read

* add multiABD example

* allow packed elementwise_op

* changed example

* clean

* clean

* add is_detected

* fix

* minor fix

* add scaleAdd_vec4 example

* init commit for contraction_multi_ABD

* add examples

* add examples of multiA and broadcast

* update example

* fixed comments

* Update cmake-ck-dev.sh

* Update cmake-ck-dev.sh

* Add comments into the example

* Update CMakeLists.txt

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 example/60_gemm_multi_ABD/CMakeLists.txt      |  18 +-
 .../61_contraction_multi_ABD/CMakeLists.txt   |   8 +
 .../contraction_multi_ABD_xdl_fp16.cpp        | 328 +++++++
 .../device_contraction_multiple_abd.hpp       |  61 ++
 ..._contraction_multiple_abd_xdl_cshuffle.hpp | 846 ++++++++++++++++++
 5 files changed, 1251 insertions(+), 10 deletions(-)
 create mode 100644 example/61_contraction_multi_ABD/CMakeLists.txt
 create mode 100644 example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
 create mode 100644 include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
 create mode 100644 include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp

diff --git a/example/60_gemm_multi_ABD/CMakeLists.txt b/example/60_gemm_multi_ABD/CMakeLists.txt
index 6dbada7ea6..57bc0b33ef 100644
--- a/example/60_gemm_multi_ABD/CMakeLists.txt
+++ b/example/60_gemm_multi_ABD/CMakeLists.txt
@@ -1,10 +1,8 @@
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-    list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
-    set(target 0)
-    foreach(gpu IN LISTS GPU_TARGETS)
-        if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-            add_example_executable(example_gemm_multi_ABD_xdl_fp16 gemm_multi_ABD_xdl_fp16.cpp)
-            set(target 1)
-        endif()
-    endforeach()
-endif()
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+    if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+        add_example_executable(example_gemm_multi_ABD_xdl_fp16 gemm_multi_ABD_xdl_fp16.cpp)
+        set(target 1)
+    endif()
+endforeach()
diff --git a/example/61_contraction_multi_ABD/CMakeLists.txt b/example/61_contraction_multi_ABD/CMakeLists.txt
new file mode 100644
index 0000000000..42500b64e6
--- /dev/null
+++ b/example/61_contraction_multi_ABD/CMakeLists.txt
@@ -0,0 +1,8 @@
+list(APPEND gpu_list2 gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+     add_example_executable(example_contraction_multi_ABD_xdl_fp16 contraction_multi_ABD_xdl_fp16.cpp)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
new file mode 100644
index 0000000000..6634dacd25
--- /dev/null
+++ b/example/61_contraction_multi_ABD/contraction_multi_ABD_xdl_fp16.cpp
@@ -0,0 +1,328 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iostream>
+#include <numeric>
+#include <initializer_list>
+#include <cstdlib>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/literals.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_contraction.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/numeric.hpp"
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using F16 = ck::half_t;
+using F32 = float;
+
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+using A0DataType       = F16;
+using A1DataType       = F32;
+using BDataType        = F16;
+using AccDataType      = F32;
+using CShuffleDataType = F32;
+using DDataType        = F16;
+using EDataType        = F16;
+
+static constexpr ck::index_t NumDimM = 2;
+static constexpr ck::index_t NumDimN = 2;
+static constexpr ck::index_t NumDimK = 2;
+
+struct AlphaBetaAdd
+{
+    AlphaBetaAdd(float alpha, float beta) : alpha_(alpha), beta_(beta){};
+
+    template <typename E, typename C, typename D>
+    __host__ __device__ constexpr void operator()(E& e, const C& c, const D& d) const;
+
+    template <>
+    __host__ __device__ constexpr void operator()<ck::half_t, float, ck::half_t>(
+        ck::half_t& e, const float& c, const ck::half_t& d) const
+    {
+        e = ck::type_convert<ck::half_t>(alpha_ * c + beta_ * ck::type_convert<float>(d));
+    };
+
+    float alpha_;
+    float beta_;
+};
+
+struct Multiply
+{
+    __host__ __device__ constexpr void
+    operator()(ck::half_t& a, const ck::half_t& a0, const float& a1) const
+    {
+        a = ck::type_convert<ck::half_t>(ck::type_convert<float>(a0) * a1);
+    }
+};
+
+using AElementOp   = Multiply;
+using BElementOp   = PassThrough;
+using CDEElementOp = AlphaBetaAdd;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+using DeviceOpInstance = ck::tensor_operation::device::DeviceContractionMultipleABD_Xdl_CShuffle<
+    NumDimM,
+    NumDimN,
+    NumDimK,
+    ck::Tuple<A0DataType, A1DataType>,
+    ck::Tuple<BDataType>,
+    AccDataType,
+    CShuffleDataType,
+    ck::Tuple<DDataType>,
+    EDataType,
+    AElementOp,
+    BElementOp,
+    CDEElementOp,
+    GemmSpec,
+    1,
+    256,
+    256,
+    128,
+    32,
+    8,
+    8,
+    32,
+    32,
+    4,
+    2,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    S<4, 64, 1>,
+    S<1, 0, 2>,
+    S<1, 0, 2>,
+    2,
+    8,
+    8,
+    1,
+    1,
+    1,
+    S<1, 32, 1, 8>,
+    8>;
+
+int main(int argc, char* argv[])
+{
+    bool do_verification = true;
+    int init_method      = 1;
+    bool time_kernel     = false;
+
+    float alpha = 1.0f;
+    float beta  = 1.0f;
+
+    // A0[M0, M1, K0, K1]
+    std::vector<ck::index_t> a0_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a0_ms_ks_strides{128 * 32 * 64, 32 * 64, 64, 1};
+    // A1[M1, K1] -> A1[M0, M1, K0, K1]
+    std::vector<ck::index_t> a1_ms_ks_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> a1_ms_ks_strides{0, 64, 0, 1};
+    // B[N0, N1, K0, K1]
+    std::vector<ck::index_t> b_ns_ks_lengths{32, 64, 32, 64};
+    std::vector<ck::index_t> b_ns_ks_strides{64 * 32 * 64, 32 * 64, 64, 1};
+    // D[M0, M1, N0, N1]
+    std::vector<ck::index_t> d_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> d_ms_ns_strides{128 * 32 * 64, 32 * 64, 64, 1};
+    // E[M0, M1, N0, N1]
+    std::vector<ck::index_t> e_ms_ns_lengths{30, 128, 32, 64};
+    std::vector<ck::index_t> e_ms_ns_strides{128 * 32 * 64, 32 * 64, 64, 1};
+
+    if(argc == 1)
+    {
+        // use default case
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        printf("arg1: verification (0=no, 1=yes)\n");
+        printf("arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n");
+        printf("arg3: time kernel (0=no, 1=yes)\n");
+        exit(0);
+    }
+
+    Tensor<A0DataType> a0_ms_ks(a0_ms_ks_lengths, a0_ms_ks_strides);
+    Tensor<A1DataType> a1_ms_ks(a1_ms_ks_lengths, a1_ms_ks_strides);
+    Tensor<BDataType> b_ns_ks(b_ns_ks_lengths, b_ns_ks_strides);
+    Tensor<EDataType> d_ms_ns(d_ms_ns_lengths, d_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+    Tensor<EDataType> e_ms_ns_device_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+    std::cout << "a0_ms_ks: " << a0_ms_ks.mDesc << std::endl;
+    std::cout << "a1_ms_ks: " << a1_ms_ks.mDesc << std::endl;
+    std::cout << "b_ns_ks: " << b_ns_ks.mDesc << std::endl;
+    std::cout << "d_ms_ns: " << d_ms_ns.mDesc << std::endl;
+    std::cout << "e_ms_ns: " << e_ms_ns_host_result.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        a0_ms_ks.GenerateTensorValue(GeneratorTensor_2<A0DataType>{-5, 5});
+        a1_ms_ks.GenerateTensorValue(GeneratorTensor_2<A1DataType>{-5, 5});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
+        break;
+    default:
+        a0_ms_ks.GenerateTensorValue(GeneratorTensor_3<A0DataType>{0.0, 1.0});
+        a1_ms_ks.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0.0, 1.0});
+        b_ns_ks.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        d_ms_ns.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        break;
+    }
+
+    DeviceMem a0_device_buf(sizeof(A0DataType) * a0_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem a1_device_buf(sizeof(A1DataType) * a1_ms_ks.mDesc.GetElementSpaceSize());
+    DeviceMem b_device_buf(sizeof(BDataType) * b_ns_ks.mDesc.GetElementSpaceSize());
+    DeviceMem d_device_buf(sizeof(DDataType) * d_ms_ns.mDesc.GetElementSpaceSize());
+    DeviceMem e_device_buf(sizeof(EDataType) * e_ms_ns_device_result.mDesc.GetElementSpaceSize());
+
+    a0_device_buf.ToDevice(a0_ms_ks.mData.data());
+    a1_device_buf.ToDevice(a1_ms_ks.mData.data());
+    b_device_buf.ToDevice(b_ns_ks.mData.data());
+    d_device_buf.ToDevice(d_ms_ns.mData.data());
+
+    // set zero
+    e_device_buf.SetZero();
+
+    auto a_element_op   = AElementOp{};
+    auto b_element_op   = BElementOp{};
+    auto cde_element_op = CDEElementOp{alpha, beta};
+
+    // do GEMM
+    auto device_op = DeviceOpInstance{};
+    auto invoker   = device_op.MakeInvoker();
+    auto argument  = device_op.MakeArgument(
+        std::array<const void*, 2>{a0_device_buf.GetDeviceBuffer(),
+                                   a1_device_buf.GetDeviceBuffer()},
+        std::array<const void*, 1>{b_device_buf.GetDeviceBuffer()},
+        std::array<const void*, 1>{d_device_buf.GetDeviceBuffer()},
+        e_device_buf.GetDeviceBuffer(),
+        std::array<std::vector<ck::index_t>, 2>{a0_ms_ks_lengths, a1_ms_ks_lengths},
+        std::array<std::vector<ck::index_t>, 2>{a0_ms_ks_strides, a1_ms_ks_strides},
+        std::array<std::vector<ck::index_t>, 1>{b_ns_ks_lengths},
+        std::array<std::vector<ck::index_t>, 1>{b_ns_ks_strides},
+        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
+        std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
+        e_ms_ns_lengths,
+        e_ms_ns_strides,
+        a_element_op,
+        b_element_op,
+        cde_element_op);
+
+    if(!device_op.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_contraction with the specified compilation parameters does "
+            "not support this problem");
+    }
+
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    if(time_kernel)
+    {
+        ck::index_t M =
+            ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
+
+        ck::index_t N = ck::accumulate_n<ck::index_t>(
+            e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
+
+        ck::index_t K = ck::accumulate_n<ck::index_t>(
+            a0_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
+
+        std::size_t flop = std::size_t(2) * M * N * K;
+        std::size_t num_btype =
+            sizeof(A0DataType) * M * K + sizeof(BDataType) * K * N + +sizeof(EDataType) * M * N;
+
+        float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+
+        float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
+                  << " GB/s" << std::endl;
+    }
+
+    if(do_verification)
+    {
+
+        Tensor<CShuffleDataType> c_ms_ns_host_result(e_ms_ns_lengths, e_ms_ns_strides);
+
+        Tensor<A0DataType> a_ms_ks(a0_ms_ks_lengths, a0_ms_ks_strides);
+
+        for(size_t m0 = 0; m0 < a_ms_ks.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < a_ms_ks.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t k0 = 0; k0 < a_ms_ks.mDesc.GetLengths()[2]; ++k0)
+                {
+                    for(size_t k1 = 0; k1 < a_ms_ks.mDesc.GetLengths()[3]; ++k1)
+                    {
+                        a_element_op(a_ms_ks(m0, m1, k0, k1),
+                                     a0_ms_ks(m0, m1, k0, k1),
+                                     a1_ms_ks(m0, m1, k0, k1));
+                    }
+                }
+            }
+        }
+
+        using ReferenceOpInstance =
+            ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
+                                                                      NumDimN,
+                                                                      NumDimK,
+                                                                      A0DataType,
+                                                                      BDataType,
+                                                                      CShuffleDataType,
+                                                                      AccDataType,
+                                                                      PassThrough,
+                                                                      BElementOp>;
+
+        auto ref_op      = ReferenceOpInstance{};
+        auto ref_invoker = ref_op.MakeInvoker();
+
+        Tensor<float> empty_tensor(std::vector<ck::index_t>{}, std::vector<ck::index_t>{});
+        auto ref_argument =
+            ref_op.MakeArgument(a_ms_ks, b_ns_ks, c_ms_ns_host_result, PassThrough{}, b_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        for(size_t m0 = 0; m0 < e_ms_ns_host_result.mDesc.GetLengths()[0]; ++m0)
+        {
+            for(size_t m1 = 0; m1 < e_ms_ns_host_result.mDesc.GetLengths()[1]; ++m1)
+            {
+                for(size_t n0 = 0; n0 < e_ms_ns_host_result.mDesc.GetLengths()[2]; ++n0)
+                {
+                    for(size_t n1 = 0; n1 < e_ms_ns_host_result.mDesc.GetLengths()[3]; ++n1)
+                    {
+                        cde_element_op(e_ms_ns_host_result(m0, m1, n0, n1),
+                                       c_ms_ns_host_result(m0, m1, n0, n1),
+                                       d_ms_ns(m0, m1, n0, n1));
+                    }
+                }
+            }
+        }
+
+        e_device_buf.FromDevice(e_ms_ns_device_result.mData.data());
+
+        return ck::utils::check_err(e_ms_ns_device_result, e_ms_ns_host_result) ? 0 : 1;
+    }
+
+    return 0;
+}
diff --git a/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
new file mode 100644
index 0000000000..3116d40474
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <array>
+
+#include "ck/tensor_operation/gpu/device/device_base.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A0[M0, M1, ... K0, K1, ...], ...
+//   input : B0[N0, N1, ... K0, K1, ...], ...
+//   input : D0[M0, M1, ... N0, N1, ...], D1[M0, M1, ... N0, N1, ...], ...
+//   output : E[M0, M1, ... N0, N1, ...]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          typename AsDataType,
+          typename BsDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation>
+struct DeviceContractionMultipleABD : public BaseOperator
+{
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    virtual std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
+                        std::array<const void*, NumBTensor> p_bs,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
+                        const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
+                        const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
+                        const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
+                        const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
+                        const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
+                        const std::vector<index_t>& e_ms_ns_length,
+                        const std::vector<index_t>& e_ms_ns_stride,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) = 0;
+
+    virtual std::unique_ptr<BaseInvoker> MakeInvokerPointer() = 0;
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
new file mode 100644
index 0000000000..a4c6767434
--- /dev/null
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
@@ -0,0 +1,846 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <iostream>
+#include <sstream>
+
+#include "ck/utility/common_header.hpp"
+#include "ck/tensor_description/tensor_descriptor.hpp"
+#include "ck/tensor_description/tensor_descriptor_helper.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_contraction_multiple_abd.hpp"
+#include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
+#include "ck/tensor_operation/gpu/device/matrix_padder.hpp"
+#include "ck/tensor_operation/gpu/grid/gridwise_gemm_multiple_abd_xdl_cshuffle.hpp"
+#include "ck/host_utility/device_prop.hpp"
+#include "ck/host_utility/kernel_launch.hpp"
+
+namespace ck {
+
+template <typename GridwiseGemm,
+          typename AsPointer,
+          typename BsPointer,
+          typename DsPointer,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          typename AsGridDesc_AK0_M_AK1,
+          typename BsGridDesc_BK0_N_BK1,
+          typename DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock,
+          typename Block2ETileMap,
+          bool HasMainKBlockLoop>
+__global__ void
+#if CK_USE_LAUNCH_BOUNDS
+    __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
+#endif
+        kernel_contraction_multiple_abd_xdl_cshuffle(
+            AsPointer p_as_grid,
+            BsPointer p_bs_grid,
+            DsPointer p_ds_grid,
+            EDataType* __restrict__ p_e_grid,
+            const AElementwiseOperation a_element_op,
+            const BElementwiseOperation b_element_op,
+            const CDEElementwiseOperation cde_element_op,
+            const AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1,
+            const BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1,
+            const DsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                ds_grid_desc_mblock_mperblock_nblock_nperblock,
+            const EGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock
+                e_grid_desc_mblock_mperblock_nblock_nperblock,
+            const Block2ETileMap block_2_etile_map)
+{
+#if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx908__) || defined(__gfx90a__) || \
+    defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__))
+    __shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
+
+    GridwiseGemm::template Run<HasMainKBlockLoop>(p_as_grid,
+                                                  p_bs_grid,
+                                                  p_ds_grid,
+                                                  p_e_grid,
+                                                  p_shared,
+                                                  a_element_op,
+                                                  b_element_op,
+                                                  cde_element_op,
+                                                  as_grid_desc_ak0_m_ak1,
+                                                  bs_grid_desc_bk0_n_bk1,
+                                                  ds_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  e_grid_desc_mblock_mperblock_nblock_nperblock,
+                                                  block_2_etile_map);
+#else
+    ignore = p_as_grid;
+    ignore = p_bs_grid;
+    ignore = p_ds_grid;
+    ignore = p_e_grid;
+    ignore = a_element_op;
+    ignore = b_element_op;
+    ignore = cde_element_op;
+    ignore = as_grid_desc_ak0_m_ak1;
+    ignore = bs_grid_desc_bk0_n_bk1;
+    ignore = ds_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = e_grid_desc_mblock_mperblock_nblock_nperblock;
+    ignore = block_2_etile_map;
+#endif
+}
+
+} // namespace ck
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+
+// GEMM:
+//   input : A[M, K]
+//   input : B[N, K]
+//   input : D0[M, N], D1[M, N], ...
+//   output : E[M, N]
+//   C = a_op(A) * b_op(B)
+//   E = cde_op(C, D0, D1, ...)
+// Assume:
+//   D0, D1, ... and E have the same layout
+template <index_t NumDimM,
+          index_t NumDimN,
+          index_t NumDimK,
+          typename AsDataType,
+          typename BsDataType,
+          typename AccDataType,
+          typename CShuffleDataType,
+          typename DsDataType,
+          typename EDataType,
+          typename AElementwiseOperation,
+          typename BElementwiseOperation,
+          typename CDEElementwiseOperation,
+          GemmSpecialization GemmSpec,
+          index_t NumGemmKPrefetchStage,
+          index_t BlockSize,
+          index_t MPerBlock,
+          index_t NPerBlock,
+          index_t KPerBlock,
+          index_t AK1,
+          index_t BK1,
+          index_t MPerXDL,
+          index_t NPerXDL,
+          index_t MXdlPerWave,
+          index_t NXdlPerWave,
+          typename ABlockTransferThreadClusterLengths_AK0_M_AK1,
+          typename ABlockTransferThreadClusterArrangeOrder,
+          typename ABlockTransferSrcAccessOrder,
+          index_t ABlockTransferSrcVectorDim,
+          index_t ABlockTransferSrcScalarPerVector,
+          index_t ABlockTransferDstScalarPerVector_AK1,
+          index_t ABlockLdsExtraM,
+          typename BBlockTransferThreadClusterLengths_BK0_N_BK1,
+          typename BBlockTransferThreadClusterArrangeOrder,
+          typename BBlockTransferSrcAccessOrder,
+          index_t BBlockTransferSrcVectorDim,
+          index_t BBlockTransferSrcScalarPerVector,
+          index_t BBlockTransferDstScalarPerVector_BK1,
+          index_t BBlockLdsExtraN,
+          index_t CShuffleMXdlPerWavePerShuffle,
+          index_t CShuffleNXdlPerWavePerShuffle,
+          typename CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+          index_t CDEBlockTransferScalarPerVector_NPerBlock,
+          LoopScheduler LoopSched     = make_default_loop_scheduler(),
+          PipelineVersion PipelineVer = PipelineVersion::v1>
+struct DeviceContractionMultipleABD_Xdl_CShuffle
+    : public DeviceContractionMultipleABD<NumDimM,
+                                          NumDimN,
+                                          NumDimK,
+                                          AsDataType,
+                                          BsDataType,
+                                          DsDataType,
+                                          EDataType,
+                                          AElementwiseOperation,
+                                          BElementwiseOperation,
+                                          CDEElementwiseOperation>
+{
+    using DeviceOp = DeviceContractionMultipleABD_Xdl_CShuffle;
+
+    static constexpr index_t NumATensor = AsDataType::Size();
+    static constexpr index_t NumBTensor = BsDataType::Size();
+    static constexpr index_t NumDTensor = DsDataType::Size();
+
+    static constexpr auto I0 = Number<0>{};
+    static constexpr auto I1 = Number<1>{};
+    static constexpr auto I2 = Number<2>{};
+    static constexpr auto I3 = Number<3>{};
+
+    using ComputeDataType = EDataType;
+
+    // GridwiseGemm
+    using GridwiseGemm = GridwiseGemmMultipleABD_xdl_cshuffle<
+        AsDataType,
+        BsDataType,
+        ComputeDataType,
+        AccDataType,
+        CShuffleDataType,
+        DsDataType,
+        EDataType,
+        AElementwiseOperation,
+        BElementwiseOperation,
+        CDEElementwiseOperation,
+        InMemoryDataOperationEnum::Set,
+        NumGemmKPrefetchStage,
+        BlockSize,
+        MPerBlock,
+        NPerBlock,
+        KPerBlock,
+        AK1,
+        BK1,
+        MPerXDL,
+        NPerXDL,
+        MXdlPerWave,
+        NXdlPerWave,
+        ABlockTransferThreadClusterLengths_AK0_M_AK1,
+        ABlockTransferThreadClusterArrangeOrder,
+        ABlockTransferSrcAccessOrder,
+        ABlockTransferSrcVectorDim,
+        ABlockTransferSrcScalarPerVector,
+        ABlockTransferDstScalarPerVector_AK1,
+        false,
+        ABlockLdsExtraM,
+        BBlockTransferThreadClusterLengths_BK0_N_BK1,
+        BBlockTransferThreadClusterArrangeOrder,
+        BBlockTransferSrcAccessOrder,
+        BBlockTransferSrcVectorDim,
+        BBlockTransferSrcScalarPerVector,
+        BBlockTransferDstScalarPerVector_BK1,
+        false,
+        BBlockLdsExtraN,
+        CShuffleMXdlPerWavePerShuffle,
+        CShuffleNXdlPerWavePerShuffle,
+        CDEBlockTransferClusterLengths_MBlock_MPerBlock_NBlock_NPerBlock,
+        CDEBlockTransferScalarPerVector_NPerBlock,
+        LoopSched,
+        PipelineVer>;
+
+    static constexpr auto matrix_padder =
+        ck::tensor_operation::device::MatrixPadder<GemmSpec, index_t, index_t, index_t>{
+            MPerBlock, NPerBlock, KPerBlock};
+
+    static auto MakeAGridDescriptor_M_K(const std::vector<index_t>& a_ms_ks_lengths_,
+                                        const std::vector<index_t>& a_ms_ks_strides_)
+    {
+        assert(a_ms_ks_lengths_.size() == NumDimM + NumDimK &&
+               a_ms_ks_strides_.size() == NumDimM + NumDimK);
+
+        const auto to_tuple = [&](auto& vec, auto num) {
+            return generate_tuple([&](auto i) { return vec[i]; }, num);
+        };
+
+        const auto a_ms_ks_lengths = to_tuple(a_ms_ks_lengths_, Number<NumDimM + NumDimK>{});
+        const auto a_ms_ks_strides = to_tuple(a_ms_ks_strides_, Number<NumDimM + NumDimK>{});
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
+
+        // dimension Ids for K0, K1, ...
+        constexpr auto kDimIds =
+            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimK, 1>::type{};
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(a_ms_ks_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto kLengths = get_container_subset(a_ms_ks_lengths, kDimIds);
+
+        // naive tensor A[M0, M1, M2, ..., K0, K1, K2...]
+        const auto a_grid_desc_ms_ks =
+            make_naive_tensor_descriptor(a_ms_ks_lengths, a_ms_ks_strides);
+
+        // transformed tensor A[MRaw = M0 * M1 * M2 * ... , KRaw = K0 * K1 * K2 * ...]
+        const auto a_grid_desc_mraw_kraw = transform_tensor_descriptor(
+            a_grid_desc_ms_ks,
+            make_tuple(make_merge_transform(mLengths), make_merge_transform(kLengths)),
+            make_tuple(mDimIds, kDimIds),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return matrix_padder.PadADescriptor_M_K(a_grid_desc_mraw_kraw);
+    }
+
+    __host__ __device__ static auto
+    MakeAsGridDescriptor_M_K(const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_lengths,
+                             const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeAGridDescriptor_M_K(as_ms_ks_lengths[i], as_ms_ks_strides[i]);
+            },
+            Number<NumATensor>{});
+    }
+
+    // Assume: B[N0, N1, N2, ..., K0, K1, K2, ...]
+    static auto MakeBGridDescriptor_N_K(const std::vector<index_t>& b_ns_ks_lengths_,
+                                        const std::vector<index_t>& b_ns_ks_strides_)
+    {
+        assert(b_ns_ks_lengths_.size() == NumDimN + NumDimK &&
+               b_ns_ks_strides_.size() == NumDimN + NumDimK);
+
+        const auto to_tuple = [&](auto& vec, auto num) {
+            return generate_tuple([&](auto i) { return vec[i]; }, num);
+        };
+
+        const auto b_ns_ks_lengths = to_tuple(b_ns_ks_lengths_, Number<NumDimN + NumDimK>{});
+        const auto b_ns_ks_strides = to_tuple(b_ns_ks_strides_, Number<NumDimN + NumDimK>{});
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds = typename arithmetic_sequence_gen<0, NumDimN, 1>::type{};
+
+        // dimension Ids for K0, K1, ...
+        constexpr auto kDimIds =
+            typename arithmetic_sequence_gen<NumDimN, NumDimN + NumDimK, 1>::type{};
+
+        // lengths for K0, K1, ...
+        const auto kLengths = get_container_subset(b_ns_ks_lengths, kDimIds);
+
+        // lengths for N0, N1, ...
+        const auto nLengths = get_container_subset(b_ns_ks_lengths, nDimIds);
+
+        // naive tensor B[N0, N1, N2, ..., K0, K1, K2, ...]
+        const auto b_grid_desc_ns_ks =
+            make_naive_tensor_descriptor(b_ns_ks_lengths, b_ns_ks_strides);
+
+        // transformed tensor B[NRaw = N0 * N1 * N2 * ..., KRaw = K0 * K1 * K2 * ...]
+        const auto b_grid_desc_nraw_kraw = transform_tensor_descriptor(
+            b_grid_desc_ns_ks,
+            make_tuple(make_merge_transform(nLengths), make_merge_transform(kLengths)),
+            make_tuple(nDimIds, kDimIds),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return matrix_padder.PadBDescriptor_N_K(b_grid_desc_nraw_kraw);
+    }
+
+    __host__ __device__ static auto
+    MakeBsGridDescriptor_N_K(const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_lengths,
+                             const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeBGridDescriptor_N_K(bs_ns_ks_lengths[i], bs_ns_ks_strides[i]);
+            },
+            Number<NumBTensor>{});
+    }
+
+    // assume E[M0, M1, M2, ..., N0, N1, N2...]
+    static auto MakeEGridDescriptor_M_N(const std::vector<index_t>& e_ms_ns_lengths_,
+                                        const std::vector<index_t>& e_ms_ns_strides_)
+    {
+        assert(e_ms_ns_lengths_.size() == NumDimM + NumDimN &&
+               e_ms_ns_strides_.size() == NumDimM + NumDimN);
+
+        const auto to_tuple = [&](auto& vec, auto num) {
+            return generate_tuple([&](auto i) { return vec[i]; }, num);
+        };
+
+        const auto e_ms_ns_lengths = to_tuple(e_ms_ns_lengths_, Number<NumDimM + NumDimN>{});
+        const auto e_ms_ns_strides = to_tuple(e_ms_ns_strides_, Number<NumDimM + NumDimN>{});
+
+        // dimension Ids for M0, M1, ...
+        constexpr auto mDimIds = typename arithmetic_sequence_gen<0, NumDimM, 1>::type{};
+
+        // dimension Ids for N0, N1, ...
+        constexpr auto nDimIds =
+            typename arithmetic_sequence_gen<NumDimM, NumDimM + NumDimN, 1>::type{};
+
+        // lengths for M0, M1, ...
+        const auto mLengths = get_container_subset(e_ms_ns_lengths, mDimIds);
+
+        // lengths for K0, K1, ...
+        const auto nLengths = get_container_subset(e_ms_ns_lengths, nDimIds);
+
+        // naive tensor E[M0, M1, M2, ..., N0, N1, N2...]
+        const auto e_grid_desc_ms_ns =
+            make_naive_tensor_descriptor(e_ms_ns_lengths, e_ms_ns_strides);
+
+        // transformed tensor E[MRaw = M0 * M1 * M2 * ... , NRaw = N0 * N1 * N2 * ...]
+        const auto e_grid_desc_mraw_nraw = transform_tensor_descriptor(
+            e_grid_desc_ms_ns,
+            make_tuple(make_merge_transform(mLengths), make_merge_transform(nLengths)),
+            make_tuple(mDimIds, nDimIds),
+            make_tuple(Sequence<0>{}, Sequence<1>{}));
+
+        return matrix_padder.PadCDescriptor_M_N(e_grid_desc_mraw_nraw);
+    }
+
+    static auto
+    MakeDsGridDescriptor_M_N(const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_lengths,
+                             const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_strides)
+    {
+        return generate_tuple(
+            [&](auto i) {
+                return MakeEGridDescriptor_M_N(ds_ms_ns_lengths[i], ds_ms_ns_strides[i]);
+            },
+            Number<NumDTensor>{});
+    }
+
+    // desc for problem definition
+    using AsGridDesc_M_K = remove_cvref_t<decltype(MakeAsGridDescriptor_M_K({}, {}))>;
+    using BsGridDesc_N_K = remove_cvref_t<decltype(MakeBsGridDescriptor_N_K({}, {}))>;
+    using DsGridDesc_M_N = remove_cvref_t<decltype(MakeDsGridDescriptor_M_N({}, {}))>;
+    using EGridDesc_M_N  = remove_cvref_t<decltype(MakeEGridDescriptor_M_N({}, {}))>;
+
+    // desc for blockwise copy
+    using AsGridDesc_AK0_M_AK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(AsGridDesc_M_K{}))>;
+    using BsGridDesc_BK0_N_BK1 =
+        remove_cvref_t<decltype(GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(BsGridDesc_N_K{}))>;
+    using DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock = remove_cvref_t<
+        decltype(GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            DsGridDesc_M_N{}))>;
+    using EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock =
+        remove_cvref_t<decltype(GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+            EGridDesc_M_N{}))>;
+
+    // block-to-e-tile map
+    using Block2ETileMap =
+        remove_cvref_t<decltype(GridwiseGemm::MakeBlock2ETileMap(EGridDesc_M_N{}))>;
+
+    // Argument
+    struct Argument : public BaseArgument
+    {
+        Argument(std::array<const void*, NumATensor> p_as_grid,
+                 std::array<const void*, NumBTensor> p_bs_grid,
+                 std::array<const void*, NumDTensor> p_ds_grid,
+                 void* p_e_grid,
+                 const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
+                 const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
+                 const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
+                 const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
+                 const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
+                 const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
+                 const std::vector<index_t>& e_ms_ns_length,
+                 const std::vector<index_t>& e_ms_ns_stride,
+                 AElementwiseOperation a_element_op,
+                 BElementwiseOperation b_element_op,
+                 CDEElementwiseOperation cde_element_op)
+            : p_as_grid_{},
+              p_bs_grid_{},
+              p_ds_grid_{},
+              p_e_grid_{static_cast<EDataType*>(p_e_grid)},
+              as_grid_desc_m_k_{},
+              bs_grid_desc_n_k_{},
+              ds_grid_desc_m_n_{},
+              e_grid_desc_m_n_{MakeEGridDescriptor_M_N(e_ms_ns_length, e_ms_ns_stride)},
+              as_grid_desc_ak0_m_ak1_{},
+              bs_grid_desc_bk0_n_bk1_{},
+              ds_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              e_grid_desc_mblock_mperblock_nblock_nperblock_{},
+              block_2_etile_map_{GridwiseGemm::MakeBlock2ETileMap(e_grid_desc_m_n_)},
+              a_element_op_{a_element_op},
+              b_element_op_{b_element_op},
+              cde_element_op_{cde_element_op}
+        {
+            // populate pointer, desc for As
+            static_for<0, NumATensor, 1>{}([&](auto i) {
+                // using ALayout   = remove_cvref_t<tuple_element_t<i.value, AsLayout>>;
+                using ADataType = remove_cvref_t<tuple_element_t<i.value, AsDataType>>;
+
+                // A pointer
+                p_as_grid_(i) = static_cast<const ADataType*>(p_as_grid[i]);
+
+                // A desc
+                as_grid_desc_m_k_(i) =
+                    MakeAGridDescriptor_M_K(a_ms_ks_lengths[i], a_ms_ks_strides[i]);
+            });
+
+            // populate pointer, desc for Bs
+            static_for<0, NumBTensor, 1>{}([&](auto i) {
+                // using BLayout   = remove_cvref_t<tuple_element_t<i.value, BsLayout>>;
+                using BDataType = remove_cvref_t<tuple_element_t<i.value, BsDataType>>;
+
+                // B pointer
+                p_bs_grid_(i) = static_cast<const BDataType*>(p_bs_grid[i]);
+
+                // B desc
+                bs_grid_desc_n_k_(i) =
+                    MakeBGridDescriptor_N_K(b_ns_ks_lengths[i], b_ns_ks_strides[i]);
+            });
+
+            // populate pointer, desc for Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                // using DLayout   = remove_cvref_t<tuple_element_t<i.value, DsLayout>>;
+                using DDataType = remove_cvref_t<tuple_element_t<i.value, DsDataType>>;
+
+                // D pointer
+                p_ds_grid_(i) = static_cast<const DDataType*>(p_ds_grid[i]);
+
+                // D desc
+                ds_grid_desc_m_n_(i) =
+                    MakeEGridDescriptor_M_N(d_ms_ns_lengths[i], d_ms_ns_strides[i]);
+            });
+
+            // populate desc for Ds/E
+            if(GridwiseGemm::CheckValidity(as_grid_desc_m_k_,
+                                           bs_grid_desc_n_k_,
+                                           ds_grid_desc_m_n_,
+                                           e_grid_desc_m_n_,
+                                           block_2_etile_map_))
+            {
+                as_grid_desc_ak0_m_ak1_ =
+                    GridwiseGemm::MakeAsGridDescriptor_AK0_M_AK1(as_grid_desc_m_k_);
+
+                bs_grid_desc_bk0_n_bk1_ =
+                    GridwiseGemm::MakeBsGridDescriptor_BK0_N_BK1(bs_grid_desc_n_k_);
+
+                ds_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeDsGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        ds_grid_desc_m_n_);
+
+                e_grid_desc_mblock_mperblock_nblock_nperblock_ =
+                    GridwiseGemm::MakeEGridDescriptor_MBlock_MPerBlock_NBlock_NPerBlock(
+                        e_grid_desc_m_n_);
+            }
+
+            // for sanity check of vector memory access
+            for(index_t i = 0; i < NumATensor; ++i)
+            {
+                a_mz_stride_[i] = a_ms_ks_strides[i][NumDimM - 1];
+                a_kz_stride_[i] = a_ms_ks_strides[i][NumDimM + NumDimK - 1];
+            }
+
+            for(index_t i = 0; i < NumBTensor; ++i)
+            {
+                b_nz_stride_[i] = b_ns_ks_strides[i][NumDimN - 1];
+                b_kz_stride_[i] = b_ns_ks_strides[i][NumDimN + NumDimK - 1];
+            }
+
+            for(index_t i = 0; i < NumDTensor; ++i)
+            {
+                ds_nz_stride_[i] = d_ms_ns_strides[i][NumDimM + NumDimN - 1];
+            }
+
+            e_nz_stride_ = e_ms_ns_stride[NumDimM + NumDimN - 1];
+        }
+
+        // pointers
+        typename GridwiseGemm::AsGridPointer p_as_grid_;
+        typename GridwiseGemm::BsGridPointer p_bs_grid_;
+        typename GridwiseGemm::DsGridPointer p_ds_grid_;
+        EDataType* p_e_grid_;
+
+        // tensor descriptors for problem definiton
+        AsGridDesc_M_K as_grid_desc_m_k_;
+        BsGridDesc_N_K bs_grid_desc_n_k_;
+        DsGridDesc_M_N ds_grid_desc_m_n_;
+        EGridDesc_M_N e_grid_desc_m_n_;
+
+        // tensor descriptors for block/thread-wise copy
+        AsGridDesc_AK0_M_AK1 as_grid_desc_ak0_m_ak1_;
+        BsGridDesc_BK0_N_BK1 bs_grid_desc_bk0_n_bk1_;
+        DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock
+            ds_grid_desc_mblock_mperblock_nblock_nperblock_;
+        EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock e_grid_desc_mblock_mperblock_nblock_nperblock_;
+
+        // block-to-e-tile map
+        Block2ETileMap block_2_etile_map_;
+
+        // element-wise op
+        AElementwiseOperation a_element_op_;
+        BElementwiseOperation b_element_op_;
+        CDEElementwiseOperation cde_element_op_;
+
+        // Strides for the last M/N/K dimensions of A/B/Ds/E
+        //   for sanity check of vector load/store
+        std::array<index_t, NumATensor> a_mz_stride_;
+        std::array<index_t, NumATensor> a_kz_stride_;
+
+        std::array<index_t, NumBTensor> b_nz_stride_;
+        std::array<index_t, NumBTensor> b_kz_stride_;
+
+        std::array<index_t, NumDTensor> ds_nz_stride_;
+        index_t e_nz_stride_;
+    };
+
+    // Invoker
+    struct Invoker : public BaseInvoker
+    {
+        using Argument = DeviceOp::Argument;
+
+        float Run(const Argument& arg, const StreamConfig& stream_config = StreamConfig{})
+        {
+            if(!GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
+                                            arg.bs_grid_desc_n_k_,
+                                            arg.ds_grid_desc_m_n_,
+                                            arg.e_grid_desc_m_n_,
+                                            arg.block_2_etile_map_))
+            {
+                throw std::runtime_error("wrong! GridwiseGemm has invalid setting");
+            }
+
+            const index_t grid_size =
+                arg.block_2_etile_map_.CalculateGridSize(arg.e_grid_desc_m_n_);
+
+            auto launch_kernel = [&](auto has_main_k_block_loop) {
+                constexpr bool has_main_loop = has_main_k_block_loop.value;
+
+                const auto kernel = kernel_contraction_multiple_abd_xdl_cshuffle<
+                    GridwiseGemm,
+                    typename GridwiseGemm::AsGridPointer,
+                    typename GridwiseGemm::BsGridPointer,
+                    typename GridwiseGemm::DsGridPointer,
+                    EDataType,
+                    AElementwiseOperation,
+                    BElementwiseOperation,
+                    CDEElementwiseOperation,
+                    DeviceOp::AsGridDesc_AK0_M_AK1,
+                    DeviceOp::BsGridDesc_BK0_N_BK1,
+                    DeviceOp::DsGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    DeviceOp::EGridDesc_MBlock_MPerBlock_NBlock_NPerBlock,
+                    DeviceOp::Block2ETileMap,
+                    has_main_loop>;
+
+                return launch_and_time_kernel(stream_config,
+                                              kernel,
+                                              dim3(grid_size),
+                                              dim3(BlockSize),
+                                              0,
+                                              arg.p_as_grid_,
+                                              arg.p_bs_grid_,
+                                              arg.p_ds_grid_,
+                                              arg.p_e_grid_,
+                                              arg.a_element_op_,
+                                              arg.b_element_op_,
+                                              arg.cde_element_op_,
+                                              arg.as_grid_desc_ak0_m_ak1_,
+                                              arg.bs_grid_desc_bk0_n_bk1_,
+                                              arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.e_grid_desc_mblock_mperblock_nblock_nperblock_,
+                                              arg.block_2_etile_map_);
+            };
+
+            const auto K = arg.as_grid_desc_m_k_[I0].GetLength(I1);
+
+            if(GridwiseGemm::CalculateHasMainKBlockLoop(K))
+            {
+                return launch_kernel(integral_constant<bool, true>{});
+            }
+            else
+            {
+                return launch_kernel(integral_constant<bool, false>{});
+            }
+        }
+
+        // polymorphic
+        float Run(const BaseArgument* p_arg,
+                  const StreamConfig& stream_config = StreamConfig{}) override
+        {
+            return Run(*dynamic_cast<const Argument*>(p_arg), stream_config);
+        }
+    };
+
+    static bool IsSupportedArgument(const Argument& arg)
+    {
+        if(!ck::is_xdl_supported())
+        {
+            return false;
+        }
+
+        // check vector load/store
+        {
+            bool all_valid = true;
+
+            static_for<0, NumATensor, 1>{}([&](auto i) {
+                // vector memory access of A: could be on M or AK1 dimension
+                if constexpr(ABlockTransferSrcVectorDim == 1)
+                {
+                    if(!(arg.a_mz_stride_[i] == 1 && arg.as_grid_desc_ak0_m_ak1_[i].GetLength(I1) %
+                                                             ABlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+                else
+                {
+                    if(!(arg.a_kz_stride_[i] == 1 && arg.as_grid_desc_ak0_m_ak1_[i].GetLength(I2) %
+                                                             ABlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+            });
+
+            // vector memory access of B: could be on N or BK1 dimension
+            static_for<0, NumBTensor, 1>{}([&](auto i) {
+                if constexpr(BBlockTransferSrcVectorDim == 1)
+                {
+                    if(!(arg.b_nz_stride_[i] == 1 && arg.bs_grid_desc_bk0_n_bk1_[i].GetLength(I1) %
+                                                             BBlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+                else
+                {
+                    if(!(arg.b_kz_stride_[i] == 1 && arg.bs_grid_desc_bk0_n_bk1_[i].GetLength(I2) %
+                                                             BBlockTransferSrcScalarPerVector ==
+                                                         0))
+                    {
+                        all_valid = false;
+                    }
+                }
+            });
+
+            // check vector load of Ds
+            static_for<0, NumDTensor, 1>{}([&](auto i) {
+                if(!(arg.ds_nz_stride_[i] == 1 &&
+                     arg.ds_grid_desc_mblock_mperblock_nblock_nperblock_[i].GetLength(I3) %
+                             CDEBlockTransferScalarPerVector_NPerBlock ==
+                         0))
+                {
+                    all_valid = false;
+                }
+            });
+
+            // vector memory access of E: always on NPerBlock dimension
+            if(!(arg.e_nz_stride_ == 1 &&
+                 arg.e_grid_desc_mblock_mperblock_nblock_nperblock_.GetLength(I3) %
+                         CDEBlockTransferScalarPerVector_NPerBlock ==
+                     0))
+            {
+                all_valid = false;
+            }
+
+            if(!all_valid)
+            {
+                return false;
+            }
+        }
+
+        return GridwiseGemm::CheckValidity(arg.as_grid_desc_m_k_,
+                                           arg.bs_grid_desc_n_k_,
+                                           arg.ds_grid_desc_m_n_,
+                                           arg.e_grid_desc_m_n_,
+                                           arg.block_2_etile_map_);
+    }
+
+    // polymorphic
+    bool IsSupportedArgument(const BaseArgument* p_arg) override
+    {
+        return IsSupportedArgument(*dynamic_cast<const Argument*>(p_arg));
+    }
+
+    static auto MakeArgument(std::array<const void*, NumATensor> p_as,
+                             std::array<const void*, NumBTensor> p_bs,
+                             std::array<const void*, NumDTensor> p_ds,
+                             void* p_e,
+                             const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_lengths,
+                             const std::array<std::vector<index_t>, NumATensor>& a_ms_ks_strides,
+                             const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_lengths,
+                             const std::array<std::vector<index_t>, NumBTensor>& b_ns_ks_strides,
+                             const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_lengths,
+                             const std::array<std::vector<index_t>, NumDTensor>& d_ms_ns_strides,
+                             const std::vector<index_t>& e_ms_ns_length,
+                             const std::vector<index_t>& e_ms_ns_stride,
+                             AElementwiseOperation a_element_op,
+                             BElementwiseOperation b_element_op,
+                             CDEElementwiseOperation cde_element_op)
+    {
+        return Argument{p_as,
+                        p_bs,
+                        p_ds,
+                        p_e,
+                        a_ms_ks_lengths,
+                        a_ms_ks_strides,
+                        b_ns_ks_lengths,
+                        b_ns_ks_strides,
+                        d_ms_ns_lengths,
+                        d_ms_ns_strides,
+                        e_ms_ns_length,
+                        e_ms_ns_stride,
+                        a_element_op,
+                        b_element_op,
+                        cde_element_op};
+    }
+
+    static auto MakeInvoker() { return Invoker{}; }
+
+    // polymorphic
+    std::unique_ptr<BaseArgument>
+    MakeArgumentPointer(std::array<const void*, NumATensor> p_as,
+                        std::array<const void*, NumBTensor> p_bs,
+                        std::array<const void*, NumDTensor> p_ds,
+                        void* p_e,
+                        const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_lengths,
+                        const std::array<std::vector<index_t>, NumATensor>& as_ms_ks_strides,
+                        const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_lengths,
+                        const std::array<std::vector<index_t>, NumBTensor>& bs_ns_ks_strides,
+                        const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_lengths,
+                        const std::array<std::vector<index_t>, NumDTensor>& ds_ms_ns_strides,
+                        const std::vector<index_t>& e_ms_ns_length,
+                        const std::vector<index_t>& e_ms_ns_stride,
+                        AElementwiseOperation a_element_op,
+                        BElementwiseOperation b_element_op,
+                        CDEElementwiseOperation cde_element_op) override
+    {
+        return std::make_unique<Argument>(p_as,
+                                          p_bs,
+                                          p_ds,
+                                          p_e,
+                                          as_ms_ks_lengths,
+                                          as_ms_ks_strides,
+                                          bs_ns_ks_lengths,
+                                          bs_ns_ks_strides,
+                                          ds_ms_ns_lengths,
+                                          ds_ms_ns_strides,
+                                          e_ms_ns_length,
+                                          e_ms_ns_stride,
+                                          a_element_op,
+                                          b_element_op,
+                                          cde_element_op);
+    }
+
+    // polymorphic
+    std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
+    {
+        return std::make_unique<Invoker>(Invoker{});
+    }
+
+    // polymorphic
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        std::map<LoopScheduler, std::string> LoopSchedToString{
+            {LoopScheduler::Default, "Default"}, {LoopScheduler::Interwave, "Interwave"}};
+
+        std::map<PipelineVersion, std::string> PipelineVersionToString{{PipelineVersion::v1, "v1"},
+                                                                       {PipelineVersion::v2, "v2"}};
+
+        // clang-format off
+        str << "DeviceContractionMultipleABD_Xdl_CShuffle"
+            << "<"
+            << BlockSize << ", "
+            << MPerBlock << ", "
+            << NPerBlock << ", "
+            << KPerBlock << ", "
+            << AK1 << ", "
+            << BK1 << ", "
+            << MPerXDL << ", "
+            << NPerXDL << ", "
+            << MXdlPerWave << ", "
+            << NXdlPerWave << ", "
+            << ABlockTransferSrcScalarPerVector << ", "
+            << BBlockTransferSrcScalarPerVector << ", "
+            << CShuffleMXdlPerWavePerShuffle << ", "
+            << CShuffleNXdlPerWavePerShuffle << ", "
+            << getGemmSpecializationString(GemmSpec)
+            << ">"
+            << " LoopScheduler: "
+            << LoopSchedToString[LoopSched] << ", "
+            << "PipelineVersion: "
+            << PipelineVersionToString[PipelineVer];
+        // clang-format on
+
+        return str.str();
+    }
+};
+
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

From bf435140dc3d28c67552bd314554bea93db12708 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 18 Oct 2023 11:14:14 -0500
Subject: [PATCH 177/194] Clean DTYPES conditions in CMake (#974)

* Add a condition to build fp8 instances

* simplified buffer_load/store

* add bfp8/fp8

* fixed

* remove all f8/bf8 condition include folder

* fixed cmake conditions

* fixed DTYPES=fp16/bfp16

* fix

* fixed buffer_load

* fixed buffer_store

* fix

* clean example cmake files

* fixed ci

* fixed cit

---------

Co-authored-by: Rostyslav Geyyer <rosty.geyyer@amd.com>
Co-authored-by: Jing Zhang <jizha@amd.com>
---
 CMakeLists.txt                                |   6 +-
 example/01_gemm/CMakeLists.txt                |  68 +++-----
 .../04_gemm_add_add_fastgelu/CMakeLists.txt   |  44 +++--
 .../CMakeLists.txt                            |  47 +++---
 example/15_grouped_gemm/CMakeLists.txt        |  50 +++---
 .../CMakeLists.txt                            |  94 +++++------
 .../20_grouped_conv_bwd_weight/CMakeLists.txt |  50 +++---
 example/22_cgemm/CMakeLists.txt               |  24 ++-
 example/24_batched_gemm/CMakeLists.txt        |  27 ++-
 .../CMakeLists.txt                            |  64 ++++----
 .../CMakeLists.txt                            |  46 +++---
 example/35_splitK_gemm/CMakeLists.txt         |  35 ++--
 .../CMakeLists.txt                            |  39 +++--
 example/39_permute/CMakeLists.txt             |  14 +-
 example/52_im2col_col2im/CMakeLists.txt       |  19 ++-
 example/CMakeLists.txt                        |   6 +
 .../element/unary_element_wise_operation.hpp  |   6 -
 .../tensor_operation/gpu/warp/xdlops_gemm.hpp |  32 +---
 include/ck/utility/amd_xdlops.hpp             |  15 +-
 include/ck/utility/data_type.hpp              |  30 +---
 include/ck/utility/f8_utils.hpp               |   4 -
 include/ck/utility/type_convert.hpp           |   8 -
 .../device_operation_instance_factory.hpp     |   8 +-
 .../gpu/convolution_backward_data.hpp         |  24 ++-
 .../gpu/convolution_forward.hpp               |  15 +-
 .../gpu/gemm_splitk.hpp                       |  14 +-
 .../grouped_convolution_backward_weight.hpp   |  79 ++++-----
 .../include/ck/library/utility/check_err.hpp  |   4 -
 .../gpu/CMakeLists.txt                        | 155 +++++++++---------
 .../grouped_conv3d_bwd_data/CMakeLists.txt    |  13 +-
 .../gpu/grouped_conv3d_fwd/CMakeLists.txt     |  19 +--
 .../gpu/grouped_gemm_fixed_nk/CMakeLists.txt  |  20 +--
 profiler/src/CMakeLists.txt                   |  18 +-
 .../src/profile_grouped_conv_bwd_weight.cpp   |  38 ++---
 test/CMakeLists.txt                           | 116 ++++++-------
 test/contraction/CMakeLists.txt               |  16 +-
 test/conv_tensor_rearrange/CMakeLists.txt     |   1 +
 37 files changed, 558 insertions(+), 710 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index b09da41a83..1d4ac3f14f 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -32,12 +32,10 @@ if (DTYPES)
     if (DTYPES MATCHES "fp8")
         add_definitions(-DCK_ENABLE_FP8)
         set(CK_ENABLE_FP8 "ON")
-        add_compile_options(-Wno-bit-int-extension)
     endif()
     if (DTYPES MATCHES "bf8")
         add_definitions(-DCK_ENABLE_BF8)
         set(CK_ENABLE_BF8 "ON")
-        add_compile_options(-Wno-bit-int-extension)
     endif()
     if (DTYPES MATCHES "fp16")
         add_definitions(-DCK_ENABLE_FP16)
@@ -59,9 +57,11 @@ if (DTYPES)
 else()
     add_definitions(-DCK_ENABLE_INT8 -DCK_ENABLE_FP8 -DCK_ENABLE_BF8 -DCK_ENABLE_FP16 -DCK_ENABLE_FP32 -DCK_ENABLE_FP64 -DCK_ENABLE_BF16)
     set(CK_ENABLE_ALL_DTYPES "ON")
-    add_compile_options(-Wno-bit-int-extension) # enable fp8 and bf8
 endif()
 
+#for f8/bf8_t type
+add_compile_options(-Wno-bit-int-extension)
+
 if(DL_KERNELS)
     add_definitions(-DDL_KERNELS)
     set(CK_ENABLE_DL_KERNELS "ON")
diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt
index 703d0e3834..42d7311d3d 100644
--- a/example/01_gemm/CMakeLists.txt
+++ b/example/01_gemm/CMakeLists.txt
@@ -1,82 +1,60 @@
 add_custom_target(example_gemm_dl)
 
 add_example_executable(example_gemm_dl_fp32 gemm_dl_fp32.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_dl example_gemm_dl_fp32)
-endif()
+add_example_dependencies(example_gemm_dl example_gemm_dl_fp32)
+
 add_example_executable(example_gemm_dl_fp16 gemm_dl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_dl example_gemm_dl_fp16)
-endif()
+add_example_dependencies(example_gemm_dl example_gemm_dl_fp16)
+
 add_example_executable(example_gemm_dpp_fp16 gemm_dpp_fp16.cpp)
+
 add_example_executable(example_gemm_dl_int8 gemm_dl_int8.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_dl example_gemm_dl_int8)
-endif()
+add_example_dependencies(example_gemm_dl example_gemm_dl_int8)
 if(USE_BITINT_EXTENSION_INT4)
     add_example_executable(example_gemm_dl_int4 gemm_dl_int4.cpp)
-    add_dependencies(example_gemm_dl example_gemm_dl_int4)
+    add_example_dependencies(example_gemm_dl example_gemm_dl_int4)
 endif(USE_BITINT_EXTENSION_INT4)
 
 add_custom_target(example_gemm_xdl)
 add_example_executable(example_gemm_xdl_fp16 gemm_xdl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16)
+
 add_example_executable(example_gemm_xdl_wavelet_fp16 gemm_xdl_wavelet_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_wavelet_fp16)
+
 add_example_executable(example_gemm_xdl_skip_b_lds_fp16 gemm_xdl_skip_b_lds_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_skip_b_lds_fp16)
 if(GPU_TARGETS MATCHES "gfx1100" OR GPU_TARGETS MATCHES "gfx1101" OR GPU_TARGETS MATCHES "gfx1102")
     add_custom_target(example_gemm_wmma)
     add_example_executable(example_gemm_wmma_fp16 gemm_wmma_fp16.cpp)
-    if(result EQUAL 0)
-      add_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
-    endif()
+    add_example_dependencies(example_gemm_wmma example_gemm_wmma_fp16)
 endif()
 
 add_example_executable(example_gemm_xdl_bf16 gemm_xdl_bf16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16)
 
-  add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
-endif()
+add_example_executable(example_gemm_xdl_bf16_rtn gemm_xdl_bf16_rtn.cpp)
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_rtn)
 
 add_example_executable(example_gemm_xdl_int8 gemm_xdl_int8.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_int8)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_int8)
 
 if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_int4)
+    add_example_executable(example_gemm_xdl_int4 gemm_xdl_int4.cpp)
+    add_example_dependencies(example_gemm_xdl example_gemm_xdl_int4)
 endif(USE_BITINT_EXTENSION_INT4)
 
 # FIXME: re-enable this exampe as test when SWDEV-335738 is fixed
 add_example_executable_no_testing(example_gemm_xdl_fp64 gemm_xdl_fp64.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp64)
 
 add_example_executable(example_gemm_xdl_streamk gemm_xdl_streamk.cpp)
 
-
 add_example_executable(example_gemm_xdl_fp8 gemm_xdl_fp8.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8)
 
 add_example_executable(example_gemm_xdl_fp8_bf8 gemm_xdl_fp8_bf8.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp8_bf8)
 
 add_example_executable(example_gemm_xdl_fp16_fp8 gemm_xdl_fp16_fp8.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
-endif()
+add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8)
diff --git a/example/04_gemm_add_add_fastgelu/CMakeLists.txt b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
index 9fe833dda0..3486b15571 100644
--- a/example/04_gemm_add_add_fastgelu/CMakeLists.txt
+++ b/example/04_gemm_add_add_fastgelu/CMakeLists.txt
@@ -1,28 +1,24 @@
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
-    add_custom_target(example_gemm_add_add_fastgelu_xdl)
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
-    if(result EQUAL 0)
-      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
+    if(gpu IN_LIST gpu_list AND target EQUAL 0)
+        add_custom_target(example_gemm_add_add_fastgelu_xdl)
+        add_example_executable(example_gemm_add_add_fastgelu_xdl_bf16 gemm_add_add_fastgelu_xdl_bf16.cpp)
+        add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_bf16)
+
+        add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
+        add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
+
+        add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
+        add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
+
+        if(USE_BITINT_EXTENSION_INT4)
+            add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
+            add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
+        endif(USE_BITINT_EXTENSION_INT4)
+
+        add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
+        add_example_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
+        set(target 1)
     endif()
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp16 gemm_add_add_fastgelu_xdl_fp16.cpp)
-    if(result EQUAL 0)
-      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp16)
-    endif()
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_fp32 gemm_add_add_fastgelu_xdl_fp32.cpp)
-    if(result EQUAL 0)
-      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_fp32)
-    endif()
-    if(USE_BITINT_EXTENSION_INT4)
-       add_example_executable(example_gemm_add_add_fastgelu_xdl_int4 gemm_add_add_fastgelu_xdl_int4.cpp)
-       add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int4)
-    endif(USE_BITINT_EXTENSION_INT4)
-    add_example_executable(example_gemm_add_add_fastgelu_xdl_int8 gemm_add_add_fastgelu_xdl_int8.cpp)
-    if(result EQUAL 0)
-      add_dependencies(example_gemm_add_add_fastgelu_xdl example_gemm_add_add_fastgelu_xdl_int8)
-    endif()
-   set(target 1)
- endif()
-endforeach()
\ No newline at end of file
+endforeach()
diff --git a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
index 150d146e31..222a3b7c0b 100644
--- a/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
+++ b/example/10_convnd_fwd_multiple_d_multiple_reduce/CMakeLists.txt
@@ -1,28 +1,25 @@
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
-  add_custom_target(example_convnd_fwd_reduce_xdl)
-  add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
-  endif()
-  add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
-  endif()
-  add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
-  endif()
-  add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
-  endif()
-  if(USE_BITINT_EXTENSION_INT4)
-      add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
-      add_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
-  endif(USE_BITINT_EXTENSION_INT4)
-  set(target 1)
- endif()
-endforeach()
\ No newline at end of file
+    if(gpu IN_LIST gpu_list AND target EQUAL 0)
+        add_custom_target(example_convnd_fwd_reduce_xdl)
+
+        add_example_executable(example_convnd_fwd_max_xdl_int8 convnd_fwd_max_xdl_int8.cpp)
+        add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int8)
+
+        add_example_executable_no_testing(example_convnd_fwd_max_xdl_bf16 convnd_fwd_max_xdl_bf16.cpp)
+        add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_bf16)
+
+        add_example_executable_no_testing(example_convnd_fwd_max_xdl_fp16 convnd_fwd_max_xdl_fp16.cpp)
+        add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp16)
+
+        add_example_executable(example_convnd_fwd_max_xdl_fp32 convnd_fwd_max_xdl_fp32.cpp)
+        add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_fp32)
+
+        if(USE_BITINT_EXTENSION_INT4)
+            add_example_executable(example_convnd_fwd_max_xdl_int4 convnd_fwd_max_xdl_int4.cpp)
+            add_example_dependencies(example_convnd_fwd_reduce_xdl example_convnd_fwd_max_xdl_int4)
+        endif(USE_BITINT_EXTENSION_INT4)
+        set(target 1)
+    endif()
+endforeach()
diff --git a/example/15_grouped_gemm/CMakeLists.txt b/example/15_grouped_gemm/CMakeLists.txt
index f2c76b76f2..84040fcf5c 100644
--- a/example/15_grouped_gemm/CMakeLists.txt
+++ b/example/15_grouped_gemm/CMakeLists.txt
@@ -1,44 +1,32 @@
 add_custom_target(example_grouped_gemm_xdl)
 add_example_executable(example_grouped_gemm_xdl_fp32 grouped_gemm_xdl_fp32.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp32)
+
 add_example_executable(example_grouped_gemm_xdl_fp16 grouped_gemm_xdl_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fp16)
+
 add_example_executable(example_grouped_gemm_multiple_d_dl_fp16 grouped_gemm_multiple_d_dl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_multiple_d_dl_fp16)
+
 add_example_executable(example_grouped_gemm_xdl_splitk_fp16 grouped_gemm_xdl_splitk_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_splitk_fp16)
+
 add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp16 grouped_gemm_xdl_fixed_nk_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp16)
+
 add_example_executable(example_grouped_gemm_xdl_fixed_nk_bias_fp16 grouped_gemm_xdl_fixed_nk_bias_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_bias_fp16)
+
 add_example_executable(example_grouped_gemm_xdl_bf16 grouped_gemm_xdl_bf16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_bf16)
+
 add_example_executable(example_grouped_gemm_xdl_int8 grouped_gemm_xdl_int8.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int8)
+
 add_example_executable(example_grouped_gemm_xdl_fixed_nk_fp8 grouped_gemm_xdl_fixed_nk_fp8.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
-endif()
+add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_fixed_nk_fp8)
 
 if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
-  endif()
+    add_example_executable(example_grouped_gemm_xdl_int4 grouped_gemm_xdl_int4.cpp)
+    add_example_dependencies(example_grouped_gemm_xdl example_grouped_gemm_xdl_int4)
 endif()
diff --git a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
index 67cf5666d7..5955e1d6cb 100644
--- a/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
+++ b/example/16_gemm_multi_d_multi_reduces/CMakeLists.txt
@@ -1,62 +1,48 @@
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(example_gemm_reduce_xdl)
-   add_custom_target(example_gemm_reduce_xdl_max)
-   add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
-   add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
-   add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
-   endif()
-   add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
-   endif()
-   add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
-   endif()
+    if(gpu IN_LIST gpu_list AND target EQUAL 0)
+        add_custom_target(example_gemm_reduce_xdl)
+        add_custom_target(example_gemm_reduce_xdl_max)
+        add_custom_target(example_gemm_reduce_xdl_mean_meansquare)
+        add_custom_target(example_gemm_add_add_mean_meansquare_xdl)
 
-   add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
-   endif()
-   add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
-   endif()
+        add_example_executable(example_gemm_max_xdl_fp16 gemm_max_xdl_fp16.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp16)
 
-   add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
-   endif()
-   add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
-   endif()
+        add_example_executable(example_gemm_add_add_mean_meansquare_xdl_fp16 gemm_add_add_mean_meansquare_xdl_fp16.cpp)
+        add_example_dependencies(example_gemm_add_add_mean_meansquare_xdl example_gemm_add_add_mean_meansquare_xdl_fp16)
 
-   add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
-   endif()
-   add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
-   endif()
-   
-   add_dependencies(example_gemm_reduce_xdl
-                 example_gemm_reduce_xdl_mean_meansquare
-                 example_gemm_reduce_xdl_max
-                 example_gemm_add_add_mean_meansquare_xdl)
+        add_example_executable(example_gemm_mean_meansquare_xdl_fp16 gemm_mean_meansquare_xdl_fp16.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp16)
 
-   if(USE_BITINT_EXTENSION_INT4)
-      add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
-      if(result EQUAL 0)
-        add_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
-      endif()
-   endif()
-   set(target 1)
- endif()
+        add_example_executable(example_gemm_max_xdl_int8 gemm_max_xdl_int8.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int8)
+
+        add_example_executable(example_gemm_add_addsquare_xdl_int8 gemm_add_addsquare_xdl_int8.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_add_addsquare_xdl_int8)
+
+        add_example_executable(example_gemm_max_xdl_fp32 gemm_max_xdl_fp32.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_fp32)
+
+        add_example_executable(example_gemm_mean_meansquare_xdl_fp32 gemm_mean_meansquare_xdl_fp32.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_fp32)
+
+        add_example_executable(example_gemm_max_xdl_bf16 gemm_max_xdl_bf16.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_bf16)
+
+        add_example_executable(example_gemm_mean_meansquare_xdl_bf16 gemm_mean_meansquare_xdl_bf16.cpp)
+        add_example_dependencies(example_gemm_reduce_xdl_mean_meansquare example_gemm_mean_meansquare_xdl_bf16)
+
+        add_example_dependencies(example_gemm_reduce_xdl
+            example_gemm_reduce_xdl_mean_meansquare
+            example_gemm_reduce_xdl_max
+            example_gemm_add_add_mean_meansquare_xdl)
+
+        if(USE_BITINT_EXTENSION_INT4)
+            add_example_executable(example_gemm_max_xdl_int4 gemm_max_xdl_int4.cpp)
+            add_example_dependencies(example_gemm_reduce_xdl_max example_gemm_max_xdl_int4)
+        endif()
+        set(target 1)
+    endif()
 endforeach()
diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index 1ecbab5825..2b0c4a28ce 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -2,36 +2,30 @@ list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
 list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
-   add_custom_target(example_grouped_conv_bwd_weight)
-   add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
-   endif()
-   add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
-   endif()
-   if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-    add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
-    if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
+    if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
+        add_custom_target(example_grouped_conv_bwd_weight)
+        add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16 grouped_conv_bwd_weight_xdl_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16)
+
+        add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
+
+        if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
+            add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
+            add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
+        endif()
+        set(target 1)
+    endif()
+
+    if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
+        add_custom_target(example_grouped_conv_bwd_weight)
+        add_example_executable(example_grouped_conv_bwd_weight_wmma_fp16 grouped_conv_bwd_weight_wmma_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_wmma_fp16)
+        set(target 1)
     endif()
-   endif()
-   set(target 1)
-  endif()
-  if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
-   add_custom_target(example_grouped_conv_bwd_weight)
-   add_example_executable(example_grouped_conv_bwd_weight_wmma_fp16 grouped_conv_bwd_weight_wmma_fp16.cpp)
-   if(result EQUAL 0)
-    add_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_wmma_fp16)
-   endif()
-   set(target 1)
- endif()
 endforeach()
 
 add_custom_target(example_grouped_conv_bwd_weight_dl)
+
 add_example_executable(example_grouped_conv_bwd_weight_dl_fp16 grouped_conv_bwd_weight_dl_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
-endif()
+add_example_dependencies(example_grouped_conv_bwd_weight_dl example_grouped_conv_bwd_weight_dl_fp16)
diff --git a/example/22_cgemm/CMakeLists.txt b/example/22_cgemm/CMakeLists.txt
index f401f7187f..44585b11d0 100644
--- a/example/22_cgemm/CMakeLists.txt
+++ b/example/22_cgemm/CMakeLists.txt
@@ -1,22 +1,18 @@
 add_custom_target(example_cgemm_xdl)
 
 add_example_executable(example_cgemm_xdl_bf16 cgemm_xdl_bf16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
-endif()
+add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_bf16)
+
 add_example_executable(example_cgemm_xdl_fp16 cgemm_xdl_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
-endif()
+add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_fp16)
+
 add_example_executable(example_cgemm_xdl_fp32 cgemm_xdl_fp32.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
-endif()
+add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_fp32)
+
 add_example_executable(example_cgemm_xdl_int8 cgemm_xdl_int8.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
-endif()
+add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_int8)
+
 if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp)
-  add_dependencies(example_cgemm_xdl example_cgemm_xdl_int4)
+    add_example_executable(example_cgemm_xdl_int4 cgemm_xdl_int4.cpp)
+    add_example_dependencies(example_cgemm_xdl example_cgemm_xdl_int4)
 endif()
diff --git a/example/24_batched_gemm/CMakeLists.txt b/example/24_batched_gemm/CMakeLists.txt
index c80540de7d..4cb45be7c9 100644
--- a/example/24_batched_gemm/CMakeLists.txt
+++ b/example/24_batched_gemm/CMakeLists.txt
@@ -1,23 +1,18 @@
 add_custom_target(example_batched_gemm_xdl)
+
 add_example_executable(example_batched_gemm_xdl_fp32 batched_gemm_xdl_fp32.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32)
-endif()
+add_example_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp32)
+
 add_example_executable(example_batched_gemm_xdl_fp16 batched_gemm_xdl_fp16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16)
-endif()
+add_example_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_fp16)
+
 add_example_executable(example_batched_gemm_xdl_bf16 batched_gemm_xdl_bf16.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bf16)
-endif()
+add_example_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_bf16)
+
 add_example_executable(example_batched_gemm_xdl_int8 batched_gemm_xdl_int8.cpp)
-if(result EQUAL 0)
-  add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8)
-endif()
+add_example_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int8)
+
 if(USE_BITINT_EXTENSION_INT4)
-  add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
-  endif()
+    add_example_executable(example_batched_gemm_xdl_int4 batched_gemm_xdl_int4.cpp)
+    add_example_dependencies(example_batched_gemm_xdl example_batched_gemm_xdl_int4)
 endif()
diff --git a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
index 4ab5bd75f2..3a8c2ef52f 100644
--- a/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
+++ b/example/30_grouped_conv_fwd_multiple_d/CMakeLists.txt
@@ -3,44 +3,38 @@ list(APPEND gpu_list2 gfx1100 gfx1101 gfx1102)
 
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
-   add_custom_target(example_grouped_conv_fwd_multiple_d)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
-   endif()
-   add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
-   endif()
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
-   endif()
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
-   endif()
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
-   endif()
-   if(USE_BITINT_EXTENSION_INT4)
-      add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
-      if(result EQUAL 0)
-         add_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
-      endif()
-   endif() # USE_BITINT_EXTENSION_INT4
+    if(gpu IN_LIST gpu_list1 AND target EQUAL 0)
+        add_custom_target(example_grouped_conv_fwd_multiple_d)
 
-   set(target 1)
- endif()
+        add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp16 grouped_conv_fwd_bias_relu_add_xdl_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp16)
+
+        add_example_executable(example_grouped_conv_fwd_xdl_fp16 grouped_conv_fwd_xdl_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_xdl_fp16)
+
+        add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_fp32 grouped_conv_fwd_bias_relu_add_xdl_fp32.cpp)
+        add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_fp32)
+
+        add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_bf16 grouped_conv_fwd_bias_relu_add_xdl_bf16.cpp)
+        add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_bf16)
+
+        add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int8 grouped_conv_fwd_bias_relu_add_xdl_int8.cpp)
+        add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int8)
+
+        if(USE_BITINT_EXTENSION_INT4)
+            add_example_executable(example_grouped_conv_fwd_bias_relu_add_xdl_int4 grouped_conv_fwd_bias_relu_add_xdl_int4.cpp)
+            add_example_dependencies(example_grouped_conv_fwd_multiple_d example_grouped_conv_fwd_bias_relu_add_xdl_int4)
+        endif() # USE_BITINT_EXTENSION_INT4
+
+        set(target 1)
+    endif()
 endforeach()
 
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
-   add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
-   set(target 1)
- endif()
+    if(gpu IN_LIST gpu_list2 AND target EQUAL 0)
+        add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_fp16 grouped_conv_fwd_bias_relu_add_wmma_fp16.cpp)
+        add_example_executable(example_grouped_conv_fwd_bias_relu_add_wmma_int8 grouped_conv_fwd_bias_relu_add_wmma_int8.cpp)
+        set(target 1)
+    endif()
 endforeach()
diff --git a/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt b/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
index 25eb44ae59..2a24abf094 100644
--- a/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
+++ b/example/32_batched_gemm_scale_softmax_gemm/CMakeLists.txt
@@ -1,31 +1,23 @@
 add_custom_target(example_gemm_scale_softmax_gemm)
 
 add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_fp16 batched_gemm_scale_softmax_gemm_xdl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
-endif()
-add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
-endif()
-add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
-endif()
-add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
-endif()
-add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
-endif()
-add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
-endif()
-add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
-endif()
+add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_fp16)
+
+add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
+add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_fp16)
+
+add_example_executable(example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_scale_softmax_gemm_permute_xdl_fp16.cpp)
+add_example_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_scale_softmax_gemm_permute_xdl_fp16)
+
+add_example_executable(example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
+
+add_example_executable(example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16 grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16.cpp)
+add_example_dependencies(example_gemm_scale_softmax_gemm example_grouped_gemm_lower_triangle_scale_softmax_gemm_permute_xdl_fp16)
+
+add_example_executable(example_batched_gemm_scale_softmax_gemm_xdl_bf16 batched_gemm_scale_softmax_gemm_xdl_bf16.cpp)
+add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_xdl_bf16)
+
+add_example_executable(example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16 batched_gemm_scale_softmax_gemm_permute_xdl_bf16.cpp)
+add_example_dependencies(example_gemm_scale_softmax_gemm example_batched_gemm_scale_softmax_gemm_permute_xdl_bf16)
 
diff --git a/example/35_splitK_gemm/CMakeLists.txt b/example/35_splitK_gemm/CMakeLists.txt
index 8970a57648..eff6b6f3fa 100644
--- a/example/35_splitK_gemm/CMakeLists.txt
+++ b/example/35_splitK_gemm/CMakeLists.txt
@@ -4,28 +4,23 @@ foreach(gpu IN LISTS GPU_TARGETS)
  if(gpu IN_LIST gpu_list AND target EQUAL 0)
    add_custom_target(example_splitK_gemm_xdl)
 
-  add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
-   endif()
-  add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
-   endif()
-  add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16)
-   endif()
-  add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
-  if(result EQUAL 0)
-    add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
-   endif()
+   add_example_executable(example_splitK_gemm_xdl_fp32 splitK_gemm_xdl_fp32.cpp)
+   add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp32)
+
+   add_example_executable(example_splitK_gemm_xdl_fp16 splitK_gemm_xdl_fp16.cpp)
+   add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_fp16)
+
+   add_example_executable(example_splitK_gemm_xdl_bf16 splitK_gemm_xdl_bf16.cpp)
+   add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_bf16)
+
+   add_example_executable(example_splitK_gemm_xdl_int8 splitK_gemm_xdl_int8.cpp)
+   add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int8)
+
    if(USE_BITINT_EXTENSION_INT4)
-      add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
-      if(result EQUAL 0)
-        add_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
-      endif()
+       add_example_executable(example_splitK_gemm_xdl_int4 splitK_gemm_xdl_int4.cpp)
+       add_example_dependencies(example_splitK_gemm_xdl example_splitK_gemm_xdl_int4)
    endif()
+
    set(target 1)
  endif()
 endforeach()
diff --git a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
index 96d1a6c3c0..1ae179e950 100644
--- a/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
+++ b/example/38_grouped_conv_bwd_data_multiple_d/CMakeLists.txt
@@ -2,27 +2,26 @@ list(APPEND gpu_list_xdl gfx908 gfx90a gfx940 gfx941 gfx942)
 list(APPEND gpu_list_wmma gfx1100 gfx1101 gfx1102)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
-   add_custom_target(example_grouped_conv_bwd_data)
-   add_example_executable(example_grouped_conv_bwd_data_xdl_fp16 grouped_conv_bwd_data_xdl_fp16.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_xdl_fp16)
-   endif()
-   add_example_executable(example_grouped_conv_bwd_data_bias_relu_xdl_fp16 grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp)
-   if(result EQUAL 0)
-     add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_xdl_fp16)
-   endif()
-   set(target 1)
- endif()
+    if(gpu IN_LIST gpu_list_xdl AND target EQUAL 0)
+        add_custom_target(example_grouped_conv_bwd_data)
+
+        add_example_executable(example_grouped_conv_bwd_data_xdl_fp16 grouped_conv_bwd_data_xdl_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_xdl_fp16)
+
+        add_example_executable(example_grouped_conv_bwd_data_bias_relu_xdl_fp16 grouped_conv_bwd_data_bias_relu_xdl_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_bias_relu_xdl_fp16)
+
+        set(target 1)
+    endif()
 endforeach()
 
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
-   add_custom_target(example_grouped_conv_bwd_data)
-   add_example_executable(example_grouped_conv_bwd_data_wmma_fp16 grouped_conv_bwd_data_wmma_fp16.cpp)
-   if(result EQUAL 0)
-      add_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_wmma_fp16)
-   endif()
-   set(target 1)
- endif()
+    if(gpu IN_LIST gpu_list_wmma AND target EQUAL 0)
+        add_custom_target(example_grouped_conv_bwd_data)
+
+        add_example_executable(example_grouped_conv_bwd_data_wmma_fp16 grouped_conv_bwd_data_wmma_fp16.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_data example_grouped_conv_bwd_data_wmma_fp16)
+
+        set(target 1)
+    endif()
 endforeach()
diff --git a/example/39_permute/CMakeLists.txt b/example/39_permute/CMakeLists.txt
index bcf47b4926..8b850c89a9 100644
--- a/example/39_permute/CMakeLists.txt
+++ b/example/39_permute/CMakeLists.txt
@@ -1,14 +1,10 @@
 add_custom_target(example_permute)
 
 add_example_executable(example_permute_1xHxW_fp16 permute_1xHxW_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_permute example_permute_1xHxW_fp16)
-endif()
+add_example_dependencies(example_permute example_permute_1xHxW_fp16)
+
 add_example_executable(example_permute_NxHxW_fp16 permute_NxHxW_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_permute example_permute_NxHxW_fp16)
-endif()
+add_example_dependencies(example_permute example_permute_NxHxW_fp16)
+
 add_example_executable(example_permute_HxWx4_fp16 permute_HxWx4_fp16.cpp)
-if(result EQUAL 0)
-    add_dependencies(example_permute example_permute_HxWx4_fp16)
-endif()
+add_example_dependencies(example_permute example_permute_HxWx4_fp16)
diff --git a/example/52_im2col_col2im/CMakeLists.txt b/example/52_im2col_col2im/CMakeLists.txt
index a2dec9e805..4dc6c8b4e0 100644
--- a/example/52_im2col_col2im/CMakeLists.txt
+++ b/example/52_im2col_col2im/CMakeLists.txt
@@ -1,12 +1,15 @@
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
-   add_custom_target(example_im2col_col2im)
-   add_example_executable(example_image_to_column_f32 image_to_column_f32.cpp)
-   add_dependencies(example_im2col_col2im example_image_to_column_f32)
-   add_example_executable(example_column_to_image_f32 column_to_image_f32.cpp)
-   add_dependencies(example_im2col_col2im example_column_to_image_f32)
-   set(target 1)
- endif()
+    if(gpu IN_LIST gpu_list AND target EQUAL 0)
+        add_custom_target(example_im2col_col2im)
+
+        add_example_executable(example_image_to_column_f32 image_to_column_f32.cpp)
+        add_example_dependencies(example_im2col_col2im example_image_to_column_f32)
+
+        add_example_executable(example_column_to_image_f32 column_to_image_f32.cpp)
+        add_example_dependencies(example_im2col_col2im example_column_to_image_f32)
+
+        set(target 1)
+    endif()
 endforeach()
diff --git a/example/CMakeLists.txt b/example/CMakeLists.txt
index 7f8704f281..5a53982bb2 100644
--- a/example/CMakeLists.txt
+++ b/example/CMakeLists.txt
@@ -62,6 +62,12 @@ function(add_example_executable EXAMPLE_NAME FILE_NAME)
     set(result ${result} PARENT_SCOPE)
 endfunction(add_example_executable EXAMPLE_NAME)
 
+function(add_example_dependencies EXAMPLE_NAME FILE_NAME)
+    if(result EQUAL 0)
+        add_dependencies(${EXAMPLE_NAME} ${FILE_NAME})
+    endif()
+endfunction(add_example_dependencies EXAMPLE_NAME)
+
 function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
     message("adding example ${EXAMPLE_NAME}")
     set(result 1)
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 05cd5c5b8d..23fdd23e10 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -113,7 +113,6 @@ struct PassThrough
     }
 #endif
 
-#if defined CK_ENABLE_FP8
     template <>
     __host__ __device__ void operator()<f8_t, f8_t>(f8_t& y, const f8_t& x) const
     {
@@ -143,9 +142,7 @@ struct PassThrough
     {
         y = type_convert<f8_t>(x);
     }
-#endif
 
-#if defined CK_ENABLE_BF8
     template <>
     __host__ __device__ void operator()<bf8_t, bf8_t>(bf8_t& y, const bf8_t& x) const
     {
@@ -175,7 +172,6 @@ struct PassThrough
     {
         y = ck::type_convert<bf8_t>(x);
     }
-#endif
 };
 
 struct UnaryConvert
@@ -204,7 +200,6 @@ struct ConvertBF16RTN
     }
 };
 
-#if defined CK_ENABLE_FP8
 struct ConvertF8SR
 {
     // convert to fp8 using stochastic rounding (SR)
@@ -221,7 +216,6 @@ struct ConvertF8SR
         y = f8_convert_sr<Y>(x);
     }
 };
-#endif
 
 struct Scale
 {
diff --git a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
index c8e56fbc56..835075b7f2 100644
--- a/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
+++ b/include/ck/tensor_operation/gpu/warp/xdlops_gemm.hpp
@@ -462,7 +462,6 @@ struct mfma_type<MfmaInstr::mfma_f64_16x16x4f64>
     }
 };
 
-#if defined CK_ENABLE_FP8
 template <>
 struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8f8>
 {
@@ -506,9 +505,7 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8f8>
         intrin_mfma_f32_16x16x32f8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
     }
 };
-#endif
 
-#if defined CK_ENABLE_BF8
 template <>
 struct mfma_type<MfmaInstr::mfma_f32_32x32x16bf8bf8>
 {
@@ -552,9 +549,7 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32bf8bf8>
         intrin_mfma_f32_16x16x32bf8bf8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
     }
 };
-#endif
 
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
 template <>
 struct mfma_type<MfmaInstr::mfma_f32_32x32x16f8bf8>
 {
@@ -598,9 +593,7 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32f8bf8>
         intrin_mfma_f32_16x16x32f8bf8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
     }
 };
-#endif
 
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
 template <>
 struct mfma_type<MfmaInstr::mfma_f32_32x32x16bf8f8>
 {
@@ -644,7 +637,6 @@ struct mfma_type<MfmaInstr::mfma_f32_16x16x32bf8f8>
         intrin_mfma_f32_16x16x32bf8f8<MPerXdlops, NPerXdlops>::Run(a, b, reg_c);
     }
 };
-#endif
 
 template <typename base_type,
           index_t MPerXdlops,
@@ -792,7 +784,6 @@ struct MfmaSelector
     }
 #endif
 
-#if defined CK_ENABLE_FP8
     template <>
     static constexpr auto GetMfma<f8_t, 32, 32>()
     {
@@ -804,9 +795,7 @@ struct MfmaSelector
     {
         return MfmaInstr::mfma_f32_16x16x32f8f8;
     }
-#endif
 
-#if defined CK_ENABLE_BF8
     template <>
     static constexpr auto GetMfma<bf8_t, 32, 32>()
     {
@@ -818,9 +807,7 @@ struct MfmaSelector
     {
         return MfmaInstr::mfma_f32_16x16x32bf8bf8;
     }
-#endif
 
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
     template <>
     static constexpr auto GetMfma<f8_t, 32, 32, bf8_t>()
     {
@@ -832,9 +819,7 @@ struct MfmaSelector
     {
         return MfmaInstr::mfma_f32_16x16x32f8bf8;
     }
-#endif
 
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
     template <>
     static constexpr auto GetMfma<bf8_t, 32, 32, f8_t>()
     {
@@ -846,7 +831,6 @@ struct MfmaSelector
     {
         return MfmaInstr::mfma_f32_16x16x32bf8f8;
     }
-#endif
 
     static constexpr auto selected_mfma =
         mfma_type<GetMfma<base_type, MPerXdlops, NPerXdlops, additional_type>()>{};
@@ -1051,18 +1035,10 @@ struct XdlopsGemm
         static_assert(
             is_same<base_type, double>::value || is_same<base_type, float>::value ||
                 is_same<base_type, half_t>::value || is_same<base_type, bhalf_t>::value ||
-                is_same<base_type, int8_t>::value
-#if defined CK_ENABLE_FP8
-                || is_same<base_type, f8_t>::value
-#endif
-#if defined CK_ENABLE_BF8
-                || is_same<base_type, bf8_t>::value
-#endif
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-                || (is_same<base_type, f8_t>::value && is_same<additional_type, bf8_t>::value) ||
-                (is_same<base_type, bf8_t>::value && is_same<additional_type, f8_t>::value)
-#endif
-                ,
+                is_same<base_type, int8_t>::value || is_same<base_type, f8_t>::value ||
+                is_same<base_type, bf8_t>::value ||
+                (is_same<base_type, f8_t>::value && is_same<additional_type, bf8_t>::value) ||
+                (is_same<base_type, bf8_t>::value && is_same<additional_type, f8_t>::value),
             "base base_type must be double, float, half, bfloat16, int8_t, f8_t or bf8_t!");
 
         static_for<0, KPack / mfma_instr.k_per_blk, 1>{}([&](auto k) {
diff --git a/include/ck/utility/amd_xdlops.hpp b/include/ck/utility/amd_xdlops.hpp
index bf6241e46e..afc066405e 100644
--- a/include/ck/utility/amd_xdlops.hpp
+++ b/include/ck/utility/amd_xdlops.hpp
@@ -1,10 +1,7 @@
 // SPDX-License-Identifier: MIT
 // Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
-#ifndef CK_AMD_XDLOPS_HPP
-#define CK_AMD_XDLOPS_HPP
-
-#include "data_type.hpp"
+#pragma once
 
 namespace ck {
 
@@ -355,7 +352,6 @@ struct intrin_mfma_f64_16x16x4f64<16, 16>
     }
 };
 
-#if defined CK_ENABLE_FP8
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f32_32x32x16f8f8;
 
@@ -418,9 +414,7 @@ struct intrin_mfma_f32_16x16x32f8f8<16, 16>
 #endif
     }
 };
-#endif
 
-#if defined CK_ENABLE_BF8
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f32_32x32x16bf8bf8;
 
@@ -483,9 +477,7 @@ struct intrin_mfma_f32_16x16x32bf8bf8<16, 16>
 #endif
     }
 };
-#endif
 
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f32_32x32x16f8bf8;
 
@@ -548,9 +540,7 @@ struct intrin_mfma_f32_16x16x32f8bf8<16, 16>
 #endif
     }
 };
-#endif
 
-#if defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
 template <index_t MPerWave, index_t NPerWave>
 struct intrin_mfma_f32_32x32x16bf8f8;
 
@@ -613,6 +603,5 @@ struct intrin_mfma_f32_16x16x32bf8f8<16, 16>
 #endif
     }
 };
-#endif
+
 } // namespace ck
-#endif
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index 89100577aa..ceaca27a45 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -9,15 +9,9 @@ namespace ck {
 
 using bhalf_t = ushort;
 using half_t  = _Float16;
-#ifdef CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
-using int4_t = _BitInt(4);
-#endif
-#if defined CK_ENABLE_FP8
-using f8_t = _BitInt(8);
-#endif
-#if defined CK_ENABLE_BF8
-using bf8_t = unsigned _BitInt(8);
-#endif
+using int4_t  = _BitInt(4);
+using f8_t    = _BitInt(8);
+using bf8_t   = unsigned _BitInt(8);
 
 // vector_type
 template <typename T, index_t N>
@@ -148,23 +142,19 @@ struct scalar_type<int4_t>
 };
 #endif
 
-#if defined CK_ENABLE_FP8
 template <>
 struct scalar_type<f8_t>
 {
     using type                           = f8_t;
     static constexpr index_t vector_size = 1;
 };
-#endif
 
-#if defined CK_ENABLE_BF8
 template <>
 struct scalar_type<bf8_t>
 {
     using type                           = bf8_t;
     static constexpr index_t vector_size = 1;
 };
-#endif
 
 template <typename T>
 struct vector_type<T, 1>
@@ -968,24 +958,20 @@ using int8x32_t = typename vector_type<int8_t, 32>::type;
 using int8x64_t = typename vector_type<int8_t, 64>::type;
 
 // f8
-#if defined CK_ENABLE_FP8
 using f8x2_t  = typename vector_type<f8_t, 2>::type;
 using f8x4_t  = typename vector_type<f8_t, 4>::type;
 using f8x8_t  = typename vector_type<f8_t, 8>::type;
 using f8x16_t = typename vector_type<f8_t, 16>::type;
 using f8x32_t = typename vector_type<f8_t, 32>::type;
 using f8x64_t = typename vector_type<f8_t, 64>::type;
-#endif
 
 // bf8
-#if defined CK_ENABLE_BF8
 using bf8x2_t  = typename vector_type<bf8_t, 2>::type;
 using bf8x4_t  = typename vector_type<bf8_t, 4>::type;
 using bf8x8_t  = typename vector_type<bf8_t, 8>::type;
 using bf8x16_t = typename vector_type<bf8_t, 16>::type;
 using bf8x32_t = typename vector_type<bf8_t, 32>::type;
 using bf8x64_t = typename vector_type<bf8_t, 64>::type;
-#endif
 
 template <typename T>
 struct NumericLimits
@@ -1033,7 +1019,6 @@ struct NumericLimits<int4_t>
 };
 #endif // CK_EXPERIMENTAL_BIT_INT_EXTENSION_INT4
 
-#if defined CK_ENABLE_FP8
 template <>
 struct NumericLimits<f8_t>
 {
@@ -1056,9 +1041,7 @@ struct NumericLimits<f8_t>
 
     __host__ __device__ static constexpr f8_t QuietNaN() { return f8_t(binary_qnan); }
 };
-#endif
 
-#if defined CK_ENABLE_BF8
 template <>
 struct NumericLimits<bf8_t>
 {
@@ -1081,7 +1064,6 @@ struct NumericLimits<bf8_t>
 
     __host__ __device__ static constexpr bf8_t QuietNaN() { return bf8_t(binary_qnan); }
 };
-#endif
 
 template <typename T>
 struct NumericUtils
@@ -1120,22 +1102,18 @@ struct NumericUtils<half_t>
     using bitwise_type                  = uint16_t;
 };
 
-#if defined CK_ENABLE_FP8
 template <>
 struct NumericUtils<f8_t>
 {
     static constexpr int exp  = 4;
     static constexpr int mant = 3;
 };
-#endif
 
-#if defined CK_ENABLE_BF8
 template <>
 struct NumericUtils<bf8_t>
 {
     static constexpr int exp  = 5;
     static constexpr int mant = 2;
 };
-#endif
-
+//
 } // namespace ck
diff --git a/include/ck/utility/f8_utils.hpp b/include/ck/utility/f8_utils.hpp
index 217b339b66..b63c82fe9a 100644
--- a/include/ck/utility/f8_utils.hpp
+++ b/include/ck/utility/f8_utils.hpp
@@ -6,8 +6,6 @@
 #include "ck/utility/data_type.hpp"
 
 // these conversions are disabled if native conversions available
-#if !defined(__gfx940__) && !defined(__gfx941__) && !defined(__gfx942__)
-#if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
 namespace ck {
 
 // fp8 rounding modes
@@ -244,5 +242,3 @@ __host__ __device__ Y cast_from_f8(X x)
 }
 
 } // namespace ck::utils
-#endif // #if defined CK_ENABLE_FP8 || defined CK_ENABLE_BF8
-#endif // #if !defined(__gfx940__) && !defined(__gfx941__) && !defined(__gfx942__)
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index ccbd5db644..aba8baa593 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -95,7 +95,6 @@ inline __host__ __device__ constexpr bhalf_t type_convert<bhalf_t, int8_t>(int8_
     return type_convert<bhalf_t>(x_fp32);
 }
 
-#if defined CK_ENABLE_FP8
 // convert fp32 to fp8
 template <>
 inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
@@ -173,9 +172,7 @@ inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
     return type_convert<half_t>(type_convert<float>(x));
 #endif
 }
-#endif
 
-#if defined CK_ENABLE_BF8
 // convert fp32 to bf8
 template <>
 inline __host__ __device__ bf8_t type_convert<bf8_t, float>(float x)
@@ -253,7 +250,6 @@ inline __host__ __device__ half_t type_convert<half_t, bf8_t>(bf8_t x)
     return type_convert<half_t>(type_convert<float>(x));
 #endif
 }
-#endif
 
 // Declare a template function for bf16 conversion using RTN
 template <typename Y, typename X>
@@ -316,7 +312,6 @@ inline __host__ __device__ constexpr bhalf_t bf16_convert_rtn<bhalf_t, half_t>(h
 template <typename Y, typename X>
 __host__ __device__ constexpr Y f8_convert_sr(X x);
 
-#if defined CK_ENABLE_FP8
 // convert fp32 to fp8 with stochastic rounding
 template <>
 inline __host__ __device__ f8_t f8_convert_sr<f8_t, float>(float x)
@@ -365,9 +360,7 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
     return f8_convert_sr<f8_t>(type_convert<float>(x));
 #endif
 }
-#endif
 
-#if defined CK_ENABLE_BF8
 // convert fp32 to bf8 with stochastic rounding
 template <>
 inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, float>(float x)
@@ -417,6 +410,5 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
     return f8_convert_sr<bf8_t>(type_convert<float>(x));
 #endif
 }
-#endif
 
 } // namespace ck
diff --git a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
index ea11fd2e1a..ba21e7a251 100644
--- a/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
+++ b/library/include/ck/library/tensor_operation_instance/device_operation_instance_factory.hpp
@@ -20,12 +20,8 @@ using F16  = ck::half_t;
 using BF16 = ck::bhalf_t;
 using I8   = int8_t;
 using I32  = int32_t;
-#if defined CK_ENABLE_FP8
-using F8 = ck::f8_t;
-#endif
-#if defined CK_ENABLE_BF8
-using BF8 = ck::bf8_t;
-#endif
+using F8   = ck::f8_t;
+using BF8  = ck::bf8_t;
 
 using Empty_Tuple = ck::Tuple<>;
 
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
index 571ff0b672..56ab6be0c3 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_backward_data.hpp
@@ -240,11 +240,13 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
         if constexpr(NumDimSpatial == 1 && is_same_v<InLayout, NWC> && is_same_v<WeiLayout, KXC> &&
                      is_same_v<OutLayout, NWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_conv1d_bwd_data_xdl_nwc_kxc_nwk_f32_instances(op_ptrs);
             }
+#endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
@@ -267,17 +269,23 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWC> &&
-                          is_same_v<WeiLayout, KYXC> && is_same_v<OutLayout, NHWK>)
+        if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWC> &&
+                     is_same_v<WeiLayout, KYXC> && is_same_v<OutLayout, NHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_conv2d_bwd_data_xdl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
-#ifdef DL_KERNELS
-                add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
-#endif
             }
+#endif
+#if defined(DL_KERNELS) && defined(CK_ENABLE_FP32)
+            if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
+                         is_same_v<OutDataType, float>)
+            {
+                add_device_conv2d_bwd_data_dl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
+            }
+#endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
@@ -306,14 +314,16 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceConvBw
             }
 #endif
         }
-        else if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWC> &&
-                          is_same_v<WeiLayout, KZYXC> && is_same_v<OutLayout, NDHWK>)
+        if constexpr(NumDimSpatial == 3 && is_same_v<InLayout, NDHWC> &&
+                     is_same_v<WeiLayout, KZYXC> && is_same_v<OutLayout, NDHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_conv3d_bwd_data_xdl_ndhwc_kzyxc_ndhwk_f32_instances(op_ptrs);
             }
+#endif
 #ifdef CK_ENABLE_FP16
             if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
                          is_same_v<OutDataType, half_t>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
index ad2da3364f..5e594366e6 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/convolution_forward.hpp
@@ -98,30 +98,31 @@ struct DeviceOperationInstanceFactory<
         if constexpr(NumDimSpatial == 2 && is_same_v<InLayout, NHWC> &&
                      is_same_v<WeiLayout, KYXC> && is_same_v<OutLayout, NHWK>)
         {
+#ifdef CK_ENABLE_FP32
             if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
                          is_same_v<OutDataType, float>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f32_instances(op_ptrs);
             }
+#endif
 #ifdef CK_ENABLE_FP16
-            else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                              is_same_v<OutDataType, half_t>)
+            if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                         is_same_v<OutDataType, half_t>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
                 add_device_conv2d_fwd_xdl_c_shuffle_nhwc_kyxc_nhwk_f16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_BF16
-            else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                              is_same_v<WeiDataType, ck::bhalf_t> &&
-                              is_same_v<OutDataType, ck::bhalf_t>)
+            if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                         is_same_v<WeiDataType, ck::bhalf_t> && is_same_v<OutDataType, ck::bhalf_t>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_bf16_instances(op_ptrs);
             }
 #endif
 #ifdef CK_ENABLE_INT8
-            else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
-                              is_same_v<OutDataType, int8_t>)
+            if constexpr(is_same_v<InDataType, int8_t> && is_same_v<WeiDataType, int8_t> &&
+                         is_same_v<OutDataType, int8_t>)
             {
                 add_device_conv2d_fwd_xdl_nhwc_kyxc_nhwk_int8_instances(op_ptrs);
             }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
index e14df2d750..8ad6ddca9d 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/gemm_splitk.hpp
@@ -155,7 +155,7 @@ struct DeviceOperationInstanceFactory<
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 #ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<ADataType, float> && is_same_v<BDataType, float> &&
-                     is_same_v<CDataType, float>)
+                     is_same_v<CDataType, float> && is_same_v<ComputeType, float>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -180,8 +180,8 @@ struct DeviceOperationInstanceFactory<
         }
 #endif
 #ifdef CK_ENABLE_FP16
-        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
-                          is_same_v<CDataType, half_t> && is_same_v<ComputeType, half_t>)
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<CDataType, half_t> && is_same_v<ComputeType, half_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -206,8 +206,8 @@ struct DeviceOperationInstanceFactory<
         }
 #endif
 #if(defined(CK_ENABLE_FP16) || defined(CK_ENABLE_FP8))
-        else if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
-                          is_same_v<CDataType, half_t>)
+        if constexpr(is_same_v<ADataType, f8_t> && is_same_v<BDataType, half_t> &&
+                     is_same_v<CDataType, half_t> && is_same_v<ComputeType, half_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
@@ -230,8 +230,8 @@ struct DeviceOperationInstanceFactory<
                 add_device_gemm_xdl_splitk_f8_f16_f16_km_nk_mn_instances(op_ptrs);
             }
         }
-        else if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
-                          is_same_v<CDataType, half_t>)
+        if constexpr(is_same_v<ADataType, half_t> && is_same_v<BDataType, f8_t> &&
+                     is_same_v<CDataType, half_t> && is_same_v<ComputeType, half_t>)
         {
             if constexpr(is_same_v<ALayout, Row> && is_same_v<BLayout, Row> &&
                          is_same_v<CLayout, Row>)
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
index 9310e9e57d..f15008349a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_convolution_backward_weight.hpp
@@ -627,8 +627,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_FP16
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_f16_instances(op_ptrs);
@@ -637,9 +637,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_BF16
-                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                                  is_same_v<WeiDataType, float> &&
-                                  is_same_v<OutDataType, ck::bhalf_t>)
+                if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv1d_bwd_weight_dl_gnwc_gkxc_gnwk_bf16_f32_bf16_instances(
@@ -650,8 +649,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
             }
-            else if constexpr(is_same_v<InLayout, NWGC> && is_same_v<WeiLayout, GKXC> &&
-                              is_same_v<OutLayout, NWGK>)
+            if constexpr(is_same_v<InLayout, NWGC> && is_same_v<WeiLayout, GKXC> &&
+                         is_same_v<OutLayout, NWGK>)
             {
 #ifdef DL_KERNELS
 #ifdef CK_ENABLE_FP32
@@ -662,16 +661,15 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_FP16
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_f16_instances(op_ptrs);
                 }
 #endif
 #ifdef CK_ENABLE_BF16
-                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                                  is_same_v<WeiDataType, float> &&
-                                  is_same_v<OutDataType, ck::bhalf_t>)
+                if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, ck::bhalf_t>)
                 {
                     add_device_grouped_conv1d_bwd_weight_dl_nwgc_gkxc_nwgk_bf16_f32_bf16_instances(
                         op_ptrs);
@@ -680,7 +678,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
             }
         }
-        else if constexpr(NumDimSpatial == 2)
+        if constexpr(NumDimSpatial == 2)
         {
             if constexpr(is_same_v<InLayout, GNHWC> && is_same_v<WeiLayout, GKYXC> &&
                          is_same_v<OutLayout, GNHWK>)
@@ -698,8 +696,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_FP16
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_f16_instances(
@@ -710,9 +708,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_BF16
-                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                                  is_same_v<WeiDataType, float> &&
-                                  is_same_v<OutDataType, ck::bhalf_t>)
+                if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_gnhwc_gkyxc_gnhwk_bf16_f32_bf16_instances(
@@ -723,8 +720,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
             }
-            else if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
-                              is_same_v<OutLayout, NHWGK>)
+            if constexpr(is_same_v<InLayout, NHWGC> && is_same_v<WeiLayout, GKYXC> &&
+                         is_same_v<OutLayout, NHWGK>)
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
@@ -739,8 +736,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_FP16
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_f16_instances(
@@ -751,9 +748,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_BF16
-                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                                  is_same_v<WeiDataType, float> &&
-                                  is_same_v<OutDataType, ck::bhalf_t>)
+                if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv2d_bwd_weight_dl_nhwgc_gkyxc_nhwgk_bf16_f32_bf16_instances(
@@ -765,7 +761,7 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
 #endif
             }
         }
-        else if constexpr(NumDimSpatial == 3)
+        if constexpr(NumDimSpatial == 3)
         {
             if constexpr(is_same_v<InLayout, GNDHWC> && is_same_v<WeiLayout, GKZYXC> &&
                          is_same_v<OutLayout, GNDHWK>)
@@ -783,8 +779,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_FP16
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_f16_instances(
@@ -799,9 +795,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_BF16
-                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                                  is_same_v<WeiDataType, float> &&
-                                  is_same_v<OutDataType, ck::bhalf_t>)
+                if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_gndhwc_gkzyxc_gndhwk_bf16_f32_bf16_instances(
@@ -822,8 +817,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
             }
-            else if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
-                              is_same_v<OutLayout, NDHWGK>)
+            if constexpr(is_same_v<InLayout, NDHWGC> && is_same_v<WeiLayout, GKZYXC> &&
+                         is_same_v<OutLayout, NDHWGK>)
             {
 #ifdef CK_ENABLE_FP32
                 if constexpr(is_same_v<InDataType, float> && is_same_v<WeiDataType, float> &&
@@ -838,10 +833,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_FP16
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t> &&
-                                  is_same_v<ComputeTypeA, half_t> &&
-                                  is_same_v<ComputeTypeB, half_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, half_t> &&
+                             is_same_v<ComputeTypeB, half_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_f16_instances(
@@ -856,9 +850,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #ifdef CK_ENABLE_BF16
-                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
-                                  is_same_v<WeiDataType, float> &&
-                                  is_same_v<OutDataType, ck::bhalf_t>)
+                if constexpr(is_same_v<InDataType, ck::bhalf_t> && is_same_v<WeiDataType, float> &&
+                             is_same_v<OutDataType, ck::bhalf_t>)
                 {
 #ifdef DL_KERNELS
                     add_device_grouped_conv3d_bwd_weight_dl_ndhwgc_gkzyxc_ndhwgk_bf16_f32_bf16_instances(
@@ -879,9 +872,9 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceGroupe
                 }
 #endif
 #if defined CK_ENABLE_FP16 && defined CK_ENABLE_FP8 && defined CK_ENABLE_BF8
-                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
-                                  is_same_v<OutDataType, half_t> &&
-                                  is_same_v<ComputeTypeA, bf8_t> && is_same_v<ComputeTypeB, f8_t>)
+                if constexpr(is_same_v<InDataType, half_t> && is_same_v<WeiDataType, half_t> &&
+                             is_same_v<OutDataType, half_t> && is_same_v<ComputeTypeA, bf8_t> &&
+                             is_same_v<ComputeTypeB, f8_t>)
                 {
                     add_device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_f8_instances(
                         op_ptrs);
diff --git a/library/include/ck/library/utility/check_err.hpp b/library/include/ck/library/utility/check_err.hpp
index c0f9ba2edc..a3df884eee 100644
--- a/library/include/ck/library/utility/check_err.hpp
+++ b/library/include/ck/library/utility/check_err.hpp
@@ -230,7 +230,6 @@ check_err(const Range& out,
     return res;
 }
 
-#if defined CK_ENABLE_FP8
 template <typename Range, typename RefRange>
 std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
                   std::is_same_v<ranges::range_value_t<Range>, f8_t>),
@@ -275,9 +274,7 @@ check_err(const Range& out,
     }
     return res;
 }
-#endif
 
-#if defined CK_ENABLE_BF8
 template <typename Range, typename RefRange>
 std::enable_if_t<(std::is_same_v<ranges::range_value_t<Range>, ranges::range_value_t<RefRange>> &&
                   std::is_same_v<ranges::range_value_t<Range>, bf8_t>),
@@ -322,7 +319,6 @@ check_err(const Range& out,
     }
     return res;
 }
-#endif
 
 } // namespace utils
 } // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
index 9c1f72eca3..e1952a886c 100644
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
@@ -2,44 +2,44 @@ function(add_instance_library INSTANCE_NAME)
     message("adding instance ${INSTANCE_NAME}")
     set(result 1)
     if(DEFINED DTYPES)
-      foreach(source IN LISTS ARGN)
-        set(test 0)
-        foreach(type IN LISTS DTYPES)
+        foreach(source IN LISTS ARGN)
+            set(test 0)
+            foreach(type IN LISTS DTYPES)
                 if(type MATCHES "fp16")
-                        set(type1 "_f16")
+                    set(type1 "_f16")
                 elseif(type MATCHES "fp32")
-                        set(type1 "_f32")
+                    set(type1 "_f32")
                 elseif(type MATCHES "fp8")
-                        set(type1 "_f8")
+                    set(type1 "_f8")
                 elseif(type MATCHES "bf16")
-                        set(type1 "_b16")
+                    set(type1 "_b16")
                 elseif(type MATCHES "fp64")
-                        set(type1 "_f64")
+                    set(type1 "_f64")
                 elseif(type MATCHES "int8")
-                        set(type1 "_i8")
+                    set(type1 "_i8")
                 endif()
                 #make an exception for reduction kernels
-                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}" OR "${source}" MATCHES "device_reduce_instance")
-                        #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                        set(test 0)
-                        break()
+                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}" OR "${source}" MATCHES "device_reduce_instance" OR ${source} MATCHES "device_image_to_column")
+                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                    set(test 0)
+                    break()
                 elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                  NOT(source MATCHES type OR source MATCHES type1))
-                        #if filename contains a type which doesn't match any selected type, mark it for removal
-                        set(test 1)
+                    source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
+                    NOT(source MATCHES type OR source MATCHES type1))
+                    #if filename contains a type which doesn't match any selected type, mark it for removal
+                    set(test 1)
                 endif()
-        endforeach()
-        if(test EQUAL 1)
+            endforeach()
+            if(test EQUAL 1)
                 message("removing instance ${source} ")
                 list(REMOVE_ITEM ARGN "${source}")
-        endif()
-      endforeach()
+            endif()
+        endforeach()
     endif()
     foreach(source IN LISTS ARGN)
         if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
-                message("removing dl instance ${source} ")
-                list(REMOVE_ITEM ARGN "${source}")
+            message("removing dl instance ${source} ")
+            list(REMOVE_ITEM ARGN "${source}")
         endif()
     endforeach()
     #only continue if there are some source files left on the list
@@ -49,8 +49,10 @@ function(add_instance_library INSTANCE_NAME)
         set_target_properties(${INSTANCE_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON)
         clang_tidy_check(${INSTANCE_NAME})
         set(result 0)
+        message("add_instance_library ${INSTANCE_NAME}")
+    else()
+        message("skip_instance_libary ${INSTANCE_NAME}")
     endif()
-    #message("add_instance_library returns ${result}")
     set(result ${result} PARENT_SCOPE)
 endfunction(add_instance_library INSTANCE_NAME)
 
@@ -58,65 +60,70 @@ endfunction(add_instance_library INSTANCE_NAME)
 file(GLOB dir_list LIST_DIRECTORIES true *)
 set(CK_DEVICE_INSTANCES)
 FOREACH(subdir_path ${dir_list})
-set(target_dir)
-IF(IS_DIRECTORY "${subdir_path}")
-    set(cmake_instance)
-    file(READ "${subdir_path}/CMakeLists.txt" cmake_instance)
-    set(add_inst 0)
-    if(("${cmake_instance}" MATCHES "_fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
+    set(target_dir)
+    IF(IS_DIRECTORY "${subdir_path}")
+        set(cmake_instance)
+        file(READ "${subdir_path}/CMakeLists.txt" cmake_instance)
+        set(add_inst 0)
+        if(("${cmake_instance}" MATCHES "_fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
             message("fp8 instance found!")
             set(add_inst 1)
-    endif()
-    if(("${cmake_instance}" MATCHES "_fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
+        endif()
+        if(("${cmake_instance}" MATCHES "_fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
             message("fp16 instance found!")
             set(add_inst 1)
-    endif()
-    if(("${cmake_instance}" MATCHES "_fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
+        endif()
+        if(("${cmake_instance}" MATCHES "_fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
             message("fp32 instance found!")
             set(add_inst 1)
-    endif()
-    if(("${cmake_instance}" MATCHES "_fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
+        endif()
+        if(("${cmake_instance}" MATCHES "_fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
             message("fp64 instance found!")
             set(add_inst 1)
-    endif()
-    if("${cmake_instance}" MATCHES "_bf16" AND DTYPES MATCHES "bf16")
+        endif()
+        if("${cmake_instance}" MATCHES "_bf16" AND DTYPES MATCHES "bf16")
             message("bf16 instance found!")
             set(add_inst 1)
-    endif()
-    if(("${cmake_instance}" MATCHES "_int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
+        endif()
+        if(("${cmake_instance}" MATCHES "_int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
             message("int8 instance found!")
             set(add_inst 1)
-    endif()
-    if(NOT "${cmake_instance}" MATCHES "_fp8" OR
-                   NOT "${cmake_instance}" MATCHES "_f8" OR
-                   NOT "${cmake_instance}" MATCHES "_fp16" OR
-                   NOT "${cmake_instance}" MATCHES "_f16" OR
-                   NOT "${cmake_instance}" MATCHES "_fp32" OR
-                   NOT "${cmake_instance}" MATCHES "_f32" OR
-                   NOT "${cmake_instance}" MATCHES "_fp64" OR
-                   NOT "${cmake_instance}" MATCHES "_f64" OR
-                   NOT "${cmake_instance}" MATCHES "_bf16" OR
-                   NOT "${cmake_instance}" MATCHES "_int8" OR
-                   NOT "${cmake_instance}" MATCHES "_i8" OR
-                   NOT "${cmake_instance}" MATCHES "_int4" OR
-                   NOT DEFINED DTYPES)
-             message("instance should be built for all types!")
-             set(add_inst 1)
-     endif()
-    if("${cmake_instance}" MATCHES "quantization" AND DEFINED DTYPES AND NOT DTYPES MATCHES "int8")
-           message("quantization instances will not be built!")
-           set(add_inst 0)
-    endif()
-    if("${cmake_instance}" MATCHES "ONLY DL_KERNELS" AND NOT DEFINED DL_KERNELS)
-	    message("Found only dl instances, but DL_KERNELS is not set. Skipping.")
+        endif()
+        if(NOT ("${cmake_instance}" MATCHES "_fp8" OR
+                "${cmake_instance}" MATCHES "_f8" OR
+                "${cmake_instance}" MATCHES "_fp16" OR
+                "${cmake_instance}" MATCHES "_f16" OR
+                "${cmake_instance}" MATCHES "_fp32" OR
+                "${cmake_instance}" MATCHES "_f32" OR
+                "${cmake_instance}" MATCHES "_fp64" OR
+                "${cmake_instance}" MATCHES "_f64" OR
+                "${cmake_instance}" MATCHES "_bf16" OR
+                "${cmake_instance}" MATCHES "_int8" OR
+                "${cmake_instance}" MATCHES "_i8" OR
+                "${cmake_instance}" MATCHES "_int4"))
+            message("instance should be built for all types!")
+            set(add_inst 1)
+        endif()
+        if(NOT DEFINED DTYPES)
+            set(add_inst 1)
+        endif()
+        if(("${cmake_instance}" MATCHES "quantization") AND (DEFINED DTYPES) AND (NOT DTYPES MATCHES "int8"))
+            message("quantization instances will not be built!")
             set(add_inst 0)
-    endif()
-    if(add_inst EQUAL 1)
-      get_filename_component(target_dir ${subdir_path} NAME)
-      add_subdirectory(${target_dir})
-      list(APPEND CK_DEVICE_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
-    endif()
-ENDIF()
+        endif()
+        if(("${cmake_instance}" MATCHES "ONLY DL_KERNELS") AND (NOT DEFINED DL_KERNELS))
+            message("Found only dl instances, but DL_KERNELS is not set. Skipping.")
+            set(add_inst 0)
+        endif()
+        if((add_inst EQUAL 1))
+            get_filename_component(target_dir ${subdir_path} NAME)
+            add_subdirectory(${target_dir})
+            list(APPEND CK_DEVICE_INSTANCES $<TARGET_OBJECTS:device_${target_dir}_instance>)
+            message("add_instance_directory ${subdir_path}")
+        else()
+            message("skip_instance_directory ${subdir_path}")
+        endif()
+    ENDIF()
 ENDFOREACH()
 
 add_library(device_operations STATIC ${CK_DEVICE_INSTANCES})
@@ -158,11 +165,11 @@ target_compile_options(device_operations PRIVATE
 
 # install(TARGETS device_operations LIBRARY DESTINATION lib)
 rocm_install(TARGETS device_operations
-        EXPORT device_operationsTargets)
+    EXPORT device_operationsTargets)
 
 rocm_install(DIRECTORY ${DEV_OPS_INC_DIRS} DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/ck)
 rocm_install(EXPORT device_operationsTargets
-        FILE composable_kerneldevice_operationsTargets.cmake
-        NAMESPACE composable_kernel::
-        DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
+    FILE composable_kerneldevice_operationsTargets.cmake
+    NAMESPACE composable_kernel::
+    DESTINATION ${CMAKE_INSTALL_LIBDIR}/cmake/composable_kernel
 )
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
index 0927cf225d..836e671bf2 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_data/CMakeLists.txt
@@ -1,11 +1,10 @@
-add_instance_library(device_grouped_conv3d_bwd_data_instance
+set(GROUPED_CONV3D_BWD_DATA
    xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
    xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-   xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
@@ -13,5 +12,11 @@ add_instance_library(device_grouped_conv3d_bwd_data_instance
    wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
    wmma/device_grouped_conv3d_bwd_data_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
-   wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
-)
+   wmma/device_grouped_conv3d_bwd_data_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp)
+
+if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "bf8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+   list(APPEND GROUPED_CONV3D_BWD_DATA
+      xdl/device_grouped_conv3d_bwd_data_xdl_ndhwgc_gkzyxc_ndhwgk_input_f16_comp_bf8_f8_instance.cpp)
+endif()
+
+add_instance_library(device_grouped_conv3d_bwd_data_instance ${GROUPED_CONV3D_BWD_DATA})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
index c3cc4cb054..bada661028 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_fwd/CMakeLists.txt
@@ -1,33 +1,32 @@
-add_instance_library(device_grouped_conv3d_fwd_instance
+set(GROUPED_CONV3D_FWD
    xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_f32_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_gndhwc_gkzyxc_gndhwk_int8_instance.cpp
-
-
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
    xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_int8_instance.cpp
-   xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp
-
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp
-
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1p0_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1p0_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1p0_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1p0_instance.cpp
-
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_1x1s1p0_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_1x1s1p0_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_1x1s1p0_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_1x1s1p0_instance.cpp
-
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_f16_oddc_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_gndhwc_gkzyxc_gndhwk_i8_oddc_instance.cpp
    wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_f16_oddc_instance.cpp
-   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp
-)
+   wmma/device_grouped_conv3d_fwd_wmma_ndhwgc_gkzyxc_ndhwgk_i8_oddc_instance.cpp)
+
+if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+    list(APPEND GROUPED_CONV3D_FWD
+      xdl/device_grouped_conv3d_fwd_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_fp8_instance.cpp)
+endif()
+
+add_instance_library(device_grouped_conv3d_fwd_instance ${GROUPED_CONV3D_FWD})
diff --git a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
index 51a42c3d8d..3b48954d22 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_gemm_fixed_nk/CMakeLists.txt
@@ -1,18 +1,10 @@
 set(GROUPED_GEMM_FIXED_NK_INSTANCES)
 
-if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
-  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp)
-  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp)
-endif()
-
-if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
-  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp)
-  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp)
-endif()
-
-if((DTYPES MATCHES "int8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
-  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp)
-  list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp)
-endif()
+list(APPEND GROUPED_GEMM_FIXED_NK_INSTANCES device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_kn_mn_instance.cpp
+                                            device_grouped_gemm_xdl_fixed_nk_f16_f16_f16_mk_nk_mn_instance.cpp
+                                            device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_kn_mn_instance.cpp
+                                            device_grouped_gemm_xdl_fixed_nk_f16_fp8_f16_mk_nk_mn_instance.cpp
+                                            device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_kn_mn_instance.cpp
+                                            device_grouped_gemm_xdl_fixed_nk_f16_i8_f16_mk_nk_mn_instance.cpp)
 
 add_instance_library(device_grouped_gemm_fixed_nk_instance ${GROUPED_GEMM_FIXED_NK_INSTANCES})
diff --git a/profiler/src/CMakeLists.txt b/profiler/src/CMakeLists.txt
index 8d9a47cd6b..e1d8afc2a4 100644
--- a/profiler/src/CMakeLists.txt
+++ b/profiler/src/CMakeLists.txt
@@ -25,8 +25,6 @@ set(PROFILER_SOURCES
     profile_batchnorm_fwd.cpp
     profile_batchnorm_bwd.cpp
     profile_batchnorm_infer.cpp
-    profile_contraction_bilinear.cpp
-    profile_contraction_scale.cpp
     profile_grouped_conv_bwd_data.cpp
     profile_conv_tensor_rearrange.cpp
 )
@@ -46,6 +44,11 @@ if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
   list(APPEND PROFILER_SOURCES profile_grouped_gemm_fastgelu.cpp)
 endif()
 
+if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+  list(APPEND PROFILER_SOURCES profile_contraction_bilinear.cpp)
+  list(APPEND PROFILER_SOURCES profile_contraction_scale.cpp)
+endif()
+
 set(PROFILER_EXECUTABLE ckProfiler)
 
 add_executable(${PROFILER_EXECUTABLE} ${PROFILER_SOURCES})
@@ -76,8 +79,6 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_normalization_instan
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_softmax_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_reduce_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batchnorm_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
-target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_pool3d_fwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_avg_pool3d_bwd_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_max_pool_bwd_instance)
@@ -85,9 +86,18 @@ target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv2d_bwd_d
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_grouped_conv3d_bwd_data_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_image_to_column_instance)
 target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_column_to_image_instance)
+
+if(DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64" OR NOT DEFINED DTYPES)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_bilinear_instance)
+  target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_contraction_scale_instance)
+endif()
+
+
+
 if(DL_KERNELS)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_batched_gemm_multi_d_instance)
 endif()
+
 if(DTYPES MATCHES "fp16" OR NOT DEFINED DTYPES)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_fastgelu_instance)
   target_link_libraries(${PROFILER_EXECUTABLE} PRIVATE device_gemm_add_relu_add_layernorm_instance)
diff --git a/profiler/src/profile_grouped_conv_bwd_weight.cpp b/profiler/src/profile_grouped_conv_bwd_weight.cpp
index bd1016f286..6ed7cf5e48 100644
--- a/profiler/src/profile_grouped_conv_bwd_weight.cpp
+++ b/profiler/src/profile_grouped_conv_bwd_weight.cpp
@@ -86,12 +86,8 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
     using F32  = float;
     using F16  = ck::half_t;
     using BF16 = ck::bhalf_t;
-#ifdef CK_ENABLE_FP8
-    using F8 = ck::f8_t;
-#endif
-#ifdef CK_ENABLE_BF8
-    using BF8 = ck::bf8_t;
-#endif
+    using F8   = ck::f8_t;
+    using BF8  = ck::bf8_t;
 
     using namespace ck::tensor_layout::convolution;
 
@@ -141,59 +137,59 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
         {
             return profile(I1, GNWC{}, GKXC{}, GNWK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        if(data_type == ConvDataType::F16_F16_F16)
         {
             return profile(I1, GNWC{}, GKXC{}, GNWK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
-        else if(data_type == ConvDataType::BF16_F32_BF16)
+        if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
             return profile(I1, GNWC{}, GKXC{}, GNWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
-    else if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
+    if(num_dim_spatial == 2 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
             return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        if(data_type == ConvDataType::F16_F16_F16)
         {
             return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
-        else if(data_type == ConvDataType::BF16_F32_BF16)
+        if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
             return profile(I2, GNHWC{}, GKYXC{}, GNHWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
-    else if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    if(num_dim_spatial == 2 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
             return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        if(data_type == ConvDataType::F16_F16_F16)
         {
             return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
-        else if(data_type == ConvDataType::BF16_F32_BF16)
+        if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
             return profile(I2, NHWGC{}, GKYXC{}, NHWGK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
     }
-    else if(num_dim_spatial == 3 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
+    if(num_dim_spatial == 3 && layout == ConvLayout::GNHWC_GKYXC_GNHWK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
             return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        if(data_type == ConvDataType::F16_F16_F16)
         {
             return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
-        else if(data_type == ConvDataType::BF16_F32_BF16)
+        if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
             return profile(I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
@@ -204,22 +200,22 @@ int profile_grouped_conv_bwd_weight(int argc, char* argv[])
                 I3, GNDHWC{}, GKZYXC{}, GNDHWK{}, int8_t{}, int8_t{}, int8_t{}, int8_t{}, int8_t{});
         }
     }
-    else if(num_dim_spatial == 3 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
+    if(num_dim_spatial == 3 && layout == ConvLayout::NHWGC_GKYXC_NHWGK)
     {
         if(data_type == ConvDataType::F32_F32_F32)
         {
             return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F32{}, F32{}, F32{}, F32{}, F32{});
         }
-        else if(data_type == ConvDataType::F16_F16_F16)
+        if(data_type == ConvDataType::F16_F16_F16)
         {
             return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{}, F16{}, F16{});
         }
-        else if(data_type == ConvDataType::BF16_F32_BF16)
+        if(data_type == ConvDataType::BF16_F32_BF16)
         {
             // fp32 atomic add is used for weight tensor in bf16 kernel
             return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, BF16{}, F32{}, BF16{}, BF16{}, BF16{});
         }
-        else if(data_type == ConvDataType::F16_F16_F16_BF8_F8)
+        if(data_type == ConvDataType::F16_F16_F16_BF8_F8)
         {
             return profile(I3, NDHWGC{}, GKZYXC{}, NDHWGK{}, F16{}, F16{}, F16{}, BF8{}, F8{});
         }
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 4b2c7bbf38..265f428b55 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -11,40 +11,40 @@ function(add_test_executable TEST_NAME)
     message("adding test ${TEST_NAME}")
     set(result 1)
     if(DEFINED DTYPES)
-    foreach(source IN LISTS ARGN)
-        set(test 0)
-        foreach(type IN LISTS DTYPES)
-            if(type MATCHES "fp16")
-                set(type1 "_f16")
-            elseif(type MATCHES "fp32")
-                set(type1 "_f32")
-            elseif(type MATCHES "fp8")
-                set(type1 "_f8")
-            elseif(type MATCHES "bf16")
-                set(type1 "_b16")
-            elseif(type MATCHES "fp64")
-                set(type1 "_f64")
-            elseif(type MATCHES "int8")
-                set(type1 "_i8")
-            endif()
-            if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
-                #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                set(test 0)
-                break()
-            elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                NOT(source MATCHES type OR source MATCHES type1))
+        foreach(source IN LISTS ARGN)
+            set(test 0)
+            foreach(type IN LISTS DTYPES)
+                if(type MATCHES "fp16")
+                    set(type1 "_f16")
+                elseif(type MATCHES "fp32")
+                    set(type1 "_f32")
+                elseif(type MATCHES "fp8")
+                    set(type1 "_f8")
+                elseif(type MATCHES "bf16")
+                    set(type1 "_b16")
+                elseif(type MATCHES "fp64")
+                    set(type1 "_f64")
+                elseif(type MATCHES "int8")
+                    set(type1 "_i8")
+                endif()
+                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
+                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                    set(test 0)
+                    break()
+                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                    source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
+                    NOT(source MATCHES type OR source MATCHES type1))
                     #if filename contains a type which doesn't match any selected type, mark it for removal
                     set(test 1)
+                endif()
+            endforeach()
+            if(test EQUAL 1)
+                message("removing test ${source} ")
+                list(REMOVE_ITEM ARGN "${source}")
             endif()
         endforeach()
-        if(test EQUAL 1)
-            message("removing test ${source} ")
-            list(REMOVE_ITEM ARGN "${source}")
-        endif()
-      endforeach()
-      endif()
-      foreach(source IN LISTS ARGN)
+    endif()
+    foreach(source IN LISTS ARGN)
         if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
             message("removing dl test ${source} ")
             list(REMOVE_ITEM ARGN "${source}")
@@ -70,38 +70,38 @@ function(add_gtest_executable TEST_NAME)
     message("adding gtest ${TEST_NAME}")
     set(result 1)
     if(DEFINED DTYPES)
-    foreach(source IN LISTS ARGN)
-        set(test 0)
-        foreach(type IN LISTS DTYPES)
-            if(type MATCHES "fp16")
-                set(type1 "_f16")
-            elseif(type MATCHES "fp32")
-                set(type1 "_f32")
-            elseif(type MATCHES "fp8")
-                set(type1 "_f8")
-            elseif(type MATCHES "bf16")
-                set(type1 "_b16")
-            elseif(type MATCHES "fp64")
-                set(type1 "_f64")
-            elseif(type MATCHES "int8")
-                set(type1 "_i8")
-            endif()
-            if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
-                #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                set(test 0)
-                break()
-            elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                NOT(source MATCHES type OR source MATCHES type1))
+        foreach(source IN LISTS ARGN)
+            set(test 0)
+            foreach(type IN LISTS DTYPES)
+                if(type MATCHES "fp16")
+                    set(type1 "_f16")
+                elseif(type MATCHES "fp32")
+                    set(type1 "_f32")
+                elseif(type MATCHES "fp8")
+                    set(type1 "_f8")
+                elseif(type MATCHES "bf16")
+                    set(type1 "_b16")
+                elseif(type MATCHES "fp64")
+                    set(type1 "_f64")
+                elseif(type MATCHES "int8")
+                    set(type1 "_i8")
+                endif()
+                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
+                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
+                    set(test 0)
+                    break()
+                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
+                    source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
+                    NOT(source MATCHES type OR source MATCHES type1))
                     #if filename contains a type which doesn't match any selected type, mark it for removal
                     set(test 1)
+                endif()
+            endforeach()
+            if(test EQUAL 1)
+                message("removing gtest ${source} ")
+                list(REMOVE_ITEM ARGN "${source}")
             endif()
         endforeach()
-        if(test EQUAL 1)
-            message("removing gtest ${source} ")
-            list(REMOVE_ITEM ARGN "${source}")
-        endif()
-    endforeach()
     endif()
     foreach(source IN LISTS ARGN)
         if(NOT DEFINED DL_KERNELS AND source MATCHES "_dl")
diff --git a/test/contraction/CMakeLists.txt b/test/contraction/CMakeLists.txt
index 1f6e0ed341..a86e72fddb 100644
--- a/test/contraction/CMakeLists.txt
+++ b/test/contraction/CMakeLists.txt
@@ -1,11 +1,13 @@
-add_gtest_executable(test_contraction test_contraction.cpp)
-target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
 list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
 set(target 0)
 foreach(gpu IN LISTS GPU_TARGETS)
- if(gpu IN_LIST gpu_list AND target EQUAL 0)
-    add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
-    target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
-   set(target 1)
- endif()
+    if(gpu IN_LIST gpu_list AND target EQUAL 0)
+        if((DTYPES MATCHES "fp32" OR DTYPES MATCHES "fp64") OR NOT DEFINED DTYPES)
+            add_gtest_executable(test_contraction test_contraction.cpp)
+            target_link_libraries(test_contraction PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+            add_gtest_executable(test_contraction_interface test_contraction_interface.cpp)
+            target_link_libraries(test_contraction_interface PRIVATE utility device_contraction_bilinear_instance device_contraction_scale_instance)
+            set(target 1)
+        endif()
+    endif()
 endforeach()
diff --git a/test/conv_tensor_rearrange/CMakeLists.txt b/test/conv_tensor_rearrange/CMakeLists.txt
index f6ad263242..05ca4a9ffb 100644
--- a/test/conv_tensor_rearrange/CMakeLists.txt
+++ b/test/conv_tensor_rearrange/CMakeLists.txt
@@ -1,4 +1,5 @@
 add_gtest_executable(test_conv_tensor_rearrange test_conv_tensor_rearrange.cpp)
 target_link_libraries(test_conv_tensor_rearrange PRIVATE utility device_image_to_column_instance device_column_to_image_instance)
+
 add_gtest_executable(test_conv_tensor_rearrange_interface test_conv_tensor_rearrange_interface.cpp)
 target_link_libraries(test_conv_tensor_rearrange_interface PRIVATE utility)

From 58338bb203efc9e4fa21c7837bf3531a7bad3694 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Wed, 18 Oct 2023 18:30:13 -0500
Subject: [PATCH 178/194] fixed math-ci error; suspend a warning (#996)

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 cmake/EnableCompilerWarnings.cmake                               | 1 +
 .../device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp | 1 +
 2 files changed, 2 insertions(+)

diff --git a/cmake/EnableCompilerWarnings.cmake b/cmake/EnableCompilerWarnings.cmake
index 66139cc710..87cb8cdf11 100644
--- a/cmake/EnableCompilerWarnings.cmake
+++ b/cmake/EnableCompilerWarnings.cmake
@@ -70,6 +70,7 @@ else()
             -Wno-option-ignored
             -Wsign-compare
             -Wno-extra-semi-stmt
+            -Wno-unused-template
         )
         if (CMAKE_${COMPILER}_COMPILER_ID MATCHES "Clang")
             list(APPEND CMAKE_COMPILER_WARNINGS
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
index a4c6767434..309a32bf2f 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_contraction_multiple_abd_xdl_cshuffle.hpp
@@ -5,6 +5,7 @@
 
 #include <iostream>
 #include <sstream>
+#include <vector>
 
 #include "ck/utility/common_header.hpp"
 #include "ck/tensor_description/tensor_descriptor.hpp"

From 3696fe1c76f50b2d16742632128c41f7869bc119 Mon Sep 17 00:00:00 2001
From: rocking <ChunYu.Lai@amd.com>
Date: Thu, 19 Oct 2023 07:36:29 +0800
Subject: [PATCH 179/194] Layernorm and groupnorm support to save mean and
 inverse std in forward (#929)

* save mean and inverse std in normalization

* Save mean and inverse std in splitK

* Vector save mean and inv std

* Modify instance for save mean and std

* simplify the layernorm example

* Save mean and std in groupnorm example

* Save mean and inv std in ckProfiler and test

* Remove compute data type from base class

* Save mean and inv std in client example

* Add changelog

* clang format

* Fix compile error

* Refine naming

* Avoid error in bf16

* revert changelog
---
 client_example/05_layernorm/layernorm2d.cpp   |  42 ++-
 .../18_groupnorm/groupnorm_swish.cpp          | 122 ++++++---
 ...as_relu_add_layernorm_xdl_welford_fp16.cpp |   5 +-
 example/27_layernorm/layernorm_fp16.cpp       |  19 +-
 .../27_layernorm/layernorm_splitk_fp16.cpp    |  19 +-
 .../27_layernorm/run_layernorm_example.inc    |  82 ++++--
 .../groupnorm_sigmoid_mul_fp16.cpp            |  17 +-
 .../42_groupnorm/groupnorm_splitk_fp16.cpp    |  19 +-
 example/42_groupnorm/groupnorm_swish_fp16.cpp |  19 +-
 .../42_groupnorm/run_groupnorm_example.inc    |  53 +++-
 .../elementwise_layernorm_blockwise.cpp       |  17 +-
 .../gpu/device/device_normalization.hpp       |   8 +-
 .../device/impl/device_normalization_impl.hpp | 121 +++++++--
 .../impl/device_normalization_splitk_impl.hpp | 256 ++++++++++++------
 .../gridwise_normalization_naive_variance.hpp | 117 +++++++-
 .../gridwise_normalization_selector.hpp       |  66 +++--
 .../gridwise_normalization_splitk_2nd.hpp     |  89 +++++-
 ...ridwise_normalization_welford_variance.hpp | 117 +++++++-
 .../cpu/reference_groupnorm.hpp               |  75 +++--
 .../cpu/reference_layernorm.hpp               |  61 +++--
 .../gpu/normalization.hpp                     |  17 +-
 .../gpu/normalization_swish.hpp               |  18 +-
 .../device_groupnorm_f16_instance.cpp         |   2 +-
 ...oupnorm_swish_f16_f32_f32_f16_instance.cpp |   2 +-
 .../device_groupnorm_swish_f16_instance.cpp   |   2 +-
 .../device_layernorm2d_f16_instance.cpp       |   2 +-
 .../device_layernorm4d_f16_instance.cpp       |   2 +-
 .../normalization_instance_common.hpp         | 242 ++++++++---------
 .../profile_elementwise_layernorm_impl.hpp    |  17 +-
 ...ofile_gemm_add_relu_add_layernorm_impl.hpp |   5 +-
 .../profiler/profile_groupnorm_impl.hpp       | 116 ++++++--
 .../profiler/profile_layernorm_impl.hpp       | 116 ++++++--
 profiler/src/profile_groupnorm.cpp            |   4 +-
 profiler/src/profile_layernorm.cpp            |   4 +-
 test/normalization/test_groupnorm_fp16.cpp    |  17 +-
 test/normalization/test_groupnorm_fp32.cpp    |  17 +-
 test/normalization/test_layernorm2d_fp16.cpp  |  15 +-
 test/normalization/test_layernorm2d_fp32.cpp  |  15 +-
 38 files changed, 1393 insertions(+), 544 deletions(-)

diff --git a/client_example/05_layernorm/layernorm2d.cpp b/client_example/05_layernorm/layernorm2d.cpp
index 3ee7cead7b..7a8e5fec99 100644
--- a/client_example/05_layernorm/layernorm2d.cpp
+++ b/client_example/05_layernorm/layernorm2d.cpp
@@ -12,12 +12,14 @@
 
 #include "ck/library/tensor_operation_instance/gpu/normalization.hpp"
 
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+using XDataType              = ck::half_t;
+using GammaDataType          = ck::half_t;
+using BetaDataType           = ck::half_t;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using PassThrough            = ck::tensor_operation::element_wise::PassThrough;
+
+#define SAVE_MEAN_INV_STD
 
 constexpr int Rank         = 2;
 constexpr int NumReduceDim = 1;
@@ -50,12 +52,16 @@ int main(int argc, char* argv[])
     SimpleDeviceMem gamma_device_buf(sizeof(GammaDataType) * N);
     SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * N);
     SimpleDeviceMem y_device_buf(sizeof(YDataType) * xy_size);
+#ifdef SAVE_MEAN_INV_STD
+    SimpleDeviceMem save_mean_device_buf(sizeof(SaveMeanInvStdDataType) * M);
+    SimpleDeviceMem save_inv_std_device_buf(sizeof(SaveMeanInvStdDataType) * M);
+#endif
 
     using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                                        GammaDataType,
                                                                        BetaDataType,
-                                                                       ComputeDataType,
                                                                        YDataType,
+                                                                       SaveMeanInvStdDataType,
                                                                        PassThrough,
                                                                        Rank,
                                                                        NumReduceDim>;
@@ -84,14 +90,21 @@ int main(int argc, char* argv[])
                                                         {0, 1},      // gammaStrides
                                                         {0, 1},      // betaStrides
                                                         {Stride, 1}, // yStrides
+                                                        {1},         // save_mean Strides
+                                                        {1},         // save_inv_std Strides
                                                         {1},         // reduceDims
                                                         1e-4,
                                                         x_device_buf.GetDeviceBuffer(),
                                                         gamma_device_buf.GetDeviceBuffer(),
                                                         beta_device_buf.GetDeviceBuffer(),
                                                         y_device_buf.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+                                                        save_mean_device_buf.GetDeviceBuffer(),
+                                                        save_inv_std_device_buf.GetDeviceBuffer(),
+#else
                                                         nullptr,
                                                         nullptr,
+#endif
                                                         PassThrough{});
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
@@ -109,6 +122,10 @@ int main(int argc, char* argv[])
             std::size_t num_byte = sizeof(XDataType) * M * N + sizeof(GammaDataType) * N +
                                    sizeof(BetaDataType) * N + sizeof(YDataType) * M * N;
 
+#ifdef SAVE_MEAN_INV_STD
+            num_byte += sizeof(SaveMeanInvStdDataType) * M * 2;
+#endif
+
             float gb_per_sec = num_byte / 1.E6 / ave_time;
 
             std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
@@ -140,17 +157,24 @@ int main(int argc, char* argv[])
 
         auto argument_ptr = op_ptr->MakeArgumentPointer({M, N},      // lengths
                                                         {Stride, 1}, // xStrides
-                                                        {1},         // gammaStrides
-                                                        {1},         // betaStrides
+                                                        {0, 1},      // gammaStrides
+                                                        {0, 1},      // betaStrides
                                                         {Stride, 1}, // yStrides
+                                                        {1},         // save_mean Strides
+                                                        {1},         // save_inv_std Strides
                                                         {1},         // reduceDims
                                                         1e-4,
                                                         x_device_buf.GetDeviceBuffer(),
                                                         gamma_device_buf.GetDeviceBuffer(),
                                                         beta_device_buf.GetDeviceBuffer(),
                                                         y_device_buf.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+                                                        save_mean_device_buf.GetDeviceBuffer(),
+                                                        save_inv_std_device_buf.GetDeviceBuffer(),
+#else
                                                         nullptr,
                                                         nullptr,
+#endif
                                                         PassThrough{});
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
diff --git a/client_example/18_groupnorm/groupnorm_swish.cpp b/client_example/18_groupnorm/groupnorm_swish.cpp
index df0a9ceec6..abe7492c65 100644
--- a/client_example/18_groupnorm/groupnorm_swish.cpp
+++ b/client_example/18_groupnorm/groupnorm_swish.cpp
@@ -12,12 +12,14 @@
 
 #include "ck/library/tensor_operation_instance/gpu/normalization_swish.hpp"
 
-using XDataType       = ck::half_t;
-using GammaDataType   = float;
-using BetaDataType    = float;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using Swish           = ck::tensor_operation::element_wise::Swish;
+using XDataType              = ck::half_t;
+using GammaDataType          = float;
+using BetaDataType           = float;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using Swish                  = ck::tensor_operation::element_wise::Swish;
+
+#define SAVE_MEAN_INV_STD
 
 constexpr int Rank         = 5;
 constexpr int NumReduceDim = 3;
@@ -49,19 +51,24 @@ int main(int argc, char* argv[])
     std::size_t xy_size         = N * H * W * G * C;
     std::size_t gamma_beta_size = G * C;
 
-    std::vector<ck::index_t> xy_strides         = {H * W * G * C, W * G * C, G * C, C, 1};
-    std::vector<ck::index_t> gamma_beta_strides = {0, 0, 0, C, 1};
+    std::vector<ck::index_t> xy_strides                = {H * W * G * C, W * G * C, G * C, C, 1};
+    std::vector<ck::index_t> gamma_beta_strides        = {0, 0, 0, C, 1};
+    std::vector<ck::index_t> save_mean_inv_std_strides = {G, 1};
 
     SimpleDeviceMem x_device_buf(sizeof(XDataType) * xy_size);
     SimpleDeviceMem gamma_device_buf(sizeof(GammaDataType) * gamma_beta_size);
     SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * gamma_beta_size);
     SimpleDeviceMem y_device_buf(sizeof(YDataType) * xy_size);
+#ifdef SAVE_MEAN_INV_STD
+    SimpleDeviceMem save_mean_device_buf(sizeof(SaveMeanInvStdDataType) * N * G);
+    SimpleDeviceMem save_inv_std_device_buf(sizeof(SaveMeanInvStdDataType) * N * G);
+#endif
 
     using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                                        GammaDataType,
                                                                        BetaDataType,
-                                                                       ComputeDataType,
                                                                        YDataType,
+                                                                       SaveMeanInvStdDataType,
                                                                        Swish,
                                                                        Rank,
                                                                        NumReduceDim>;
@@ -75,19 +82,26 @@ int main(int argc, char* argv[])
     const auto& generic_op_ptr = op_ptrs[0];
 
     auto generic_argument_ptr =
-        generic_op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
-                                            xy_strides,         // xStrides
-                                            gamma_beta_strides, // gammaStrides
-                                            gamma_beta_strides, // betaStrides
-                                            xy_strides,         // yStrides
-                                            {1, 2, 4},          // reduceDims
+        generic_op_ptr->MakeArgumentPointer({N, H, W, G, C},           // lengths
+                                            xy_strides,                // xStrides
+                                            gamma_beta_strides,        // gammaStrides
+                                            gamma_beta_strides,        // betaStrides
+                                            xy_strides,                // yStrides
+                                            save_mean_inv_std_strides, // save_mean Strides
+                                            save_mean_inv_std_strides, // save_inv_std Strides
+                                            {1, 2, 4},                 // reduceDims
                                             1e-6,
                                             x_device_buf.GetDeviceBuffer(),
                                             gamma_device_buf.GetDeviceBuffer(),
                                             beta_device_buf.GetDeviceBuffer(),
                                             y_device_buf.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+                                            save_mean_device_buf.GetDeviceBuffer(),
+                                            save_inv_std_device_buf.GetDeviceBuffer(),
+#else
                                             nullptr,
                                             nullptr,
+#endif
                                             Swish{});
 
     if(!generic_op_ptr->IsSupportedArgument(generic_argument_ptr.get()))
@@ -107,21 +121,29 @@ int main(int argc, char* argv[])
 
     for(int i = 0; i < op_ptrs.size(); ++i)
     {
-        auto& op_ptr      = op_ptrs[i];
-        auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
-                                                        xy_strides,         // xStrides
-                                                        gamma_beta_strides, // gammaStrides
-                                                        gamma_beta_strides, // betaStrides
-                                                        xy_strides,         // yStrides
-                                                        {1, 2, 4},          // reduceDims
-                                                        1e-6,
-                                                        x_device_buf.GetDeviceBuffer(),
-                                                        gamma_device_buf.GetDeviceBuffer(),
-                                                        beta_device_buf.GetDeviceBuffer(),
-                                                        y_device_buf.GetDeviceBuffer(),
-                                                        nullptr,
-                                                        nullptr,
-                                                        Swish{});
+        auto& op_ptr = op_ptrs[i];
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer({N, H, W, G, C},           // lengths
+                                        xy_strides,                // xStrides
+                                        gamma_beta_strides,        // gammaStrides
+                                        gamma_beta_strides,        // betaStrides
+                                        xy_strides,                // yStrides
+                                        save_mean_inv_std_strides, // save_mean Strides
+                                        save_mean_inv_std_strides, // save_inv_std Strides
+                                        {1, 2, 4},                 // reduceDims
+                                        1e-6,
+                                        x_device_buf.GetDeviceBuffer(),
+                                        gamma_device_buf.GetDeviceBuffer(),
+                                        beta_device_buf.GetDeviceBuffer(),
+                                        y_device_buf.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+                                        save_mean_device_buf.GetDeviceBuffer(),
+                                        save_inv_std_device_buf.GetDeviceBuffer(),
+#else
+                                        nullptr,
+                                        nullptr,
+#endif
+                                        Swish{});
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
@@ -139,6 +161,10 @@ int main(int argc, char* argv[])
                 sizeof(XDataType) * xy_size + sizeof(GammaDataType) * gamma_beta_size +
                 sizeof(BetaDataType) * gamma_beta_size + sizeof(YDataType) * xy_size;
 
+#ifdef SAVE_MEAN_INV_STD
+            num_byte += sizeof(SaveMeanInvStdDataType) * N * G * 2;
+#endif
+
             float gb_per_sec = num_byte / 1.E6 / ave_time;
 
             std::cout << "Perf: " << std::setw(10) << ave_time << " ms, " << gb_per_sec << " GB/s, "
@@ -169,20 +195,28 @@ int main(int argc, char* argv[])
         std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
                   << std::endl;
 
-        auto argument_ptr = op_ptr->MakeArgumentPointer({N, H, W, G, C},    // lengths
-                                                        xy_strides,         // xStrides
-                                                        gamma_beta_strides, // gammaStrides
-                                                        gamma_beta_strides, // betaStrides
-                                                        xy_strides,         // yStrides
-                                                        {1, 2, 4},          // reduceDims
-                                                        1e-6,
-                                                        x_device_buf.GetDeviceBuffer(),
-                                                        gamma_device_buf.GetDeviceBuffer(),
-                                                        beta_device_buf.GetDeviceBuffer(),
-                                                        y_device_buf.GetDeviceBuffer(),
-                                                        nullptr,
-                                                        nullptr,
-                                                        Swish{});
+        auto argument_ptr =
+            op_ptr->MakeArgumentPointer({N, H, W, G, C},           // lengths
+                                        xy_strides,                // xStrides
+                                        gamma_beta_strides,        // gammaStrides
+                                        gamma_beta_strides,        // betaStrides
+                                        xy_strides,                // yStrides
+                                        save_mean_inv_std_strides, // save_mean Strides
+                                        save_mean_inv_std_strides, // save_inv_std Strides
+                                        {1, 2, 4},                 // reduceDims
+                                        1e-6,
+                                        x_device_buf.GetDeviceBuffer(),
+                                        gamma_device_buf.GetDeviceBuffer(),
+                                        beta_device_buf.GetDeviceBuffer(),
+                                        y_device_buf.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+                                        save_mean_device_buf.GetDeviceBuffer(),
+                                        save_inv_std_device_buf.GetDeviceBuffer(),
+#else
+                                        nullptr,
+                                        nullptr,
+#endif
+                                        Swish{});
 
         auto invoker_ptr = op_ptr->MakeInvokerPointer();
 
diff --git a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
index fc58ca19f8..6a92e9a2f5 100644
--- a/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
+++ b/example/21_gemm_layernorm/gemm_bias_relu_add_layernorm_xdl_welford_fp16.cpp
@@ -114,12 +114,15 @@ void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
                                                                               BetaDataType,
                                                                               HDataType,
                                                                               AccDataType,
+                                                                              AccDataType,
                                                                               HElementOp,
                                                                               2,
                                                                               1>;
 
     Tensor<EMeanVarDataType> e_m_n(HostTensorDescriptor{M, N});
     Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
+    Tensor<AccDataType> save_mean({M});
+    Tensor<AccDataType> save_inv_std({M});
 
     auto ref_gemm         = ReferenceGemm{};
     auto ref_gemm_invoker = ref_gemm.MakeInvoker();
@@ -145,7 +148,7 @@ void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
     auto ref_layernorm_invoker = ref_layernorm.MakeInvoker();
 
     auto ref_layernorm_argument = ref_layernorm.MakeArgument(
-        e_m_n, gamma_n, beta_n, h_m_n, h_element_op, {M, N}, {1}, epsilon);
+        e_m_n, gamma_n, beta_n, h_m_n, save_mean, save_inv_std, h_element_op, {M, N}, {1}, epsilon);
     ref_layernorm_invoker.Run(ref_layernorm_argument);
 }
 
diff --git a/example/27_layernorm/layernorm_fp16.cpp b/example/27_layernorm/layernorm_fp16.cpp
index bb8b954f0a..255452e769 100644
--- a/example/27_layernorm/layernorm_fp16.cpp
+++ b/example/27_layernorm/layernorm_fp16.cpp
@@ -3,12 +3,15 @@
 
 #include "common.hpp"
 
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+using XDataType              = ck::half_t;
+using GammaDataType          = ck::half_t;
+using BetaDataType           = ck::half_t;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using ComputeDataType        = float;
+using PassThrough            = ck::tensor_operation::element_wise::PassThrough;
+
+#define SAVE_MEAN_INV_STD
 
 constexpr int Rank         = 2;
 constexpr int NumReduceDim = 1;
@@ -19,6 +22,7 @@ using DeviceInstance =
                                                           BetaDataType,
                                                           ComputeDataType,
                                                           YDataType,
+                                                          SaveMeanInvStdDataType,
                                                           PassThrough,
                                                           Rank,
                                                           NumReduceDim,
@@ -33,7 +37,8 @@ using DeviceInstance =
                                                           8,   // GammaScalarPerVector
                                                           1,   // BetaVecDim (0=M, 1=K)
                                                           8,   // BetaScalarPerVector
-                                                          8>;  // OutScalarPerVector
+                                                          8,   // YScalarPerVector
+                                                          1>;  // SaveMeanInvStdScalarPerVector
 #include "run_layernorm_example.inc"
 
 int main() { return run_groupnorm_example<DeviceInstance>(); }
diff --git a/example/27_layernorm/layernorm_splitk_fp16.cpp b/example/27_layernorm/layernorm_splitk_fp16.cpp
index e0378d028b..e2a85bddc5 100644
--- a/example/27_layernorm/layernorm_splitk_fp16.cpp
+++ b/example/27_layernorm/layernorm_splitk_fp16.cpp
@@ -3,12 +3,15 @@
 
 #include "common.hpp"
 
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using PassThrough     = ck::tensor_operation::element_wise::PassThrough;
+using XDataType              = ck::half_t;
+using GammaDataType          = ck::half_t;
+using BetaDataType           = ck::half_t;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using ComputeDataType        = float;
+using PassThrough            = ck::tensor_operation::element_wise::PassThrough;
+
+#define SAVE_MEAN_INV_STD
 
 constexpr int Rank         = 2;
 constexpr int NumReduceDim = 1;
@@ -19,6 +22,7 @@ using DeviceInstance =
                                                                 BetaDataType,
                                                                 ComputeDataType,
                                                                 YDataType,
+                                                                SaveMeanInvStdDataType,
                                                                 PassThrough,
                                                                 Rank,
                                                                 NumReduceDim,
@@ -33,7 +37,8 @@ using DeviceInstance =
                                                                 8,   // GammaScalarPerVector
                                                                 1,   // BetaVecDim (0=M, 1=K)
                                                                 8,   // BetaScalarPerVector
-                                                                8>;  // YScalarPerVector
+                                                                8,   // YScalarPerVector
+                                                                1>; // SaveMeanInvStdScalarPerVector
 
 #include "run_layernorm_example.inc"
 
diff --git a/example/27_layernorm/run_layernorm_example.inc b/example/27_layernorm/run_layernorm_example.inc
index 95200b540a..c8f599a39c 100644
--- a/example/27_layernorm/run_layernorm_example.inc
+++ b/example/27_layernorm/run_layernorm_example.inc
@@ -10,22 +10,13 @@ int run_groupnorm_example()
 
     ck::index_t M      = 1024;
     ck::index_t N      = 1024;
-    ck::index_t Stride = N;
 
-    auto f_host_tensor_descriptor1d = [](std::size_t len, std::size_t stride) {
-        return HostTensorDescriptor({len}, {stride});
-    };
-
-    auto f_host_tensor_descriptor2d = [](std::size_t row, std::size_t col, std::size_t stride) {
-        using namespace ck::literals;
-
-        return HostTensorDescriptor({row, col}, {stride, 1_uz});
-    };
-
-    Tensor<XDataType> x(f_host_tensor_descriptor2d(M, N, Stride));
-    Tensor<GammaDataType> gamma(f_host_tensor_descriptor1d(N, 1));
-    Tensor<BetaDataType> beta(f_host_tensor_descriptor1d(N, 1));
-    Tensor<YDataType> y(f_host_tensor_descriptor2d(M, N, Stride));
+    Tensor<XDataType> x({M, N});
+    Tensor<GammaDataType> gamma({N});
+    Tensor<BetaDataType> beta({N});
+    Tensor<YDataType> y({M, N});
+    Tensor<SaveMeanInvStdDataType> save_mean({M});
+    Tensor<SaveMeanInvStdDataType> save_inv_std({M});
 
     x.GenerateTensorValue(GeneratorTensor_3<XDataType>{0.0, 1.0});
     gamma.GenerateTensorValue(GeneratorTensor_3<GammaDataType>{0.0, 1.0});
@@ -35,6 +26,11 @@ int run_groupnorm_example()
     DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
     DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
     DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
+#ifdef SAVE_MEAN_INV_STD
+    DeviceMem save_mean_dev(sizeof(SaveMeanInvStdDataType) * save_mean.mDesc.GetElementSpaceSize());
+    DeviceMem save_inv_std_dev(sizeof(SaveMeanInvStdDataType) *
+                               save_inv_std.mDesc.GetElementSpaceSize());
+#endif
 
     x_dev.ToDevice(x.mData.data());
     gamma_dev.ToDevice(gamma.mData.data());
@@ -47,14 +43,23 @@ int run_groupnorm_example()
         {0, 1},
         {0, 1},
         std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
+        std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
+                                    save_mean.mDesc.GetStrides().end()},
+        std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
+                                    save_mean.mDesc.GetStrides().end()},
         {1},
         1e-4,
         x_dev.GetDeviceBuffer(),
         gamma_dev.GetDeviceBuffer(),
         beta_dev.GetDeviceBuffer(),
         y_dev.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+        save_mean_dev.GetDeviceBuffer(),
+        save_inv_std_dev.GetDeviceBuffer(),
+#else
         nullptr,
         nullptr,
+#endif
         PassThrough{});
 
     if(!device_instance.IsSupportedArgument(argument_ptr.get()))
@@ -72,24 +77,45 @@ int run_groupnorm_example()
 
     bool pass = true;
     {
-        Tensor<YDataType> host_y(f_host_tensor_descriptor2d(M, N, Stride));
-        using ReferenceInstance = ck::tensor_operation::host::ReferenceLayernorm<XDataType,
-                                                                                 GammaDataType,
-                                                                                 BetaDataType,
-                                                                                 YDataType,
-                                                                                 ComputeDataType,
-                                                                                 PassThrough,
-                                                                                 Rank,
-                                                                                 NumReduceDim>;
+        Tensor<YDataType> host_y({M, N});
+        Tensor<SaveMeanInvStdDataType> host_save_mean({M});
+        Tensor<SaveMeanInvStdDataType> host_save_inv_std({M});
+
+        using ReferenceInstance =
+            ck::tensor_operation::host::ReferenceLayernorm<XDataType,
+                                                           GammaDataType,
+                                                           BetaDataType,
+                                                           YDataType,
+                                                           SaveMeanInvStdDataType,
+                                                           ComputeDataType,
+                                                           PassThrough,
+                                                           Rank,
+                                                           NumReduceDim>;
 
         ReferenceInstance ref;
-        auto ref_argument =
-            ref.MakeArgument(x, gamma, beta, host_y, PassThrough{}, {M, N}, {1}, 1e-4);
-        auto ref_invoker = ref.MakeInvoker();
+        auto ref_argument = ref.MakeArgument(x,
+                                             gamma,
+                                             beta,
+                                             host_y,
+                                             host_save_mean,
+                                             host_save_inv_std,
+                                             PassThrough{},
+                                             {M, N},
+                                             {1},
+                                             1e-4);
+        auto ref_invoker  = ref.MakeInvoker();
         ref_invoker.Run(ref_argument);
 
         y_dev.FromDevice(y.mData.data());
-        pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
+        pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results (y)", 1e-3, 1e-3);
+#ifdef SAVE_MEAN_INV_STD
+        save_mean_dev.FromDevice(save_mean.mData.data());
+        save_inv_std_dev.FromDevice(save_inv_std.mData.data());
+        pass &= ck::utils::check_err(
+            save_mean, host_save_mean, "Error: Incorrect results (mean)", 1e-3, 1e-3);
+        pass &= ck::utils::check_err(
+            save_inv_std, host_save_inv_std, "Error: Incorrect results (inv_std)", 1e-3, 1e-3);
+#endif
     }
 
     return (pass ? 0 : 1);
diff --git a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
index b36bd761b3..0ede570e62 100644
--- a/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_sigmoid_mul_fp16.cpp
@@ -6,11 +6,14 @@
 constexpr int Rank         = 5;
 constexpr int NumReduceDim = 3;
 
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
+using XDataType              = ck::half_t;
+using GammaDataType          = ck::half_t;
+using BetaDataType           = ck::half_t;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using ComputeDataType        = float;
+
+#define SAVE_MEAN_INV_STD
 
 struct YElementOp
 {
@@ -39,6 +42,7 @@ using DeviceInstance =
                                                           BetaDataType,
                                                           ComputeDataType,
                                                           YDataType,
+                                                          SaveMeanInvStdDataType,
                                                           YElementOp,
                                                           Rank,
                                                           NumReduceDim,
@@ -53,7 +57,8 @@ using DeviceInstance =
                                                           2,    // GammaScalarPerVector
                                                           1,    // BetaVecDim (0=M, 1=K)
                                                           2,    // BetaScalarPerVector
-                                                          2>;   // OutScalarPerVector
+                                                          2,    // YScalarPerVector
+                                                          1>;   // SaveMeanInvStdScalarPerVector
 
 #include "run_groupnorm_example.inc"
 
diff --git a/example/42_groupnorm/groupnorm_splitk_fp16.cpp b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
index 057b240a63..5f56268e02 100644
--- a/example/42_groupnorm/groupnorm_splitk_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_splitk_fp16.cpp
@@ -6,12 +6,15 @@
 constexpr int Rank         = 5;
 constexpr int NumReduceDim = 3;
 
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using YElementOp      = ck::tensor_operation::element_wise::Swish;
+using XDataType              = ck::half_t;
+using GammaDataType          = ck::half_t;
+using BetaDataType           = ck::half_t;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using ComputeDataType        = float;
+using YElementOp             = ck::tensor_operation::element_wise::Swish;
+
+#define SAVE_MEAN_INV_STD
 
 using DeviceInstance =
     ck::tensor_operation::device::DeviceNormalizationSplitKImpl<XDataType,
@@ -19,6 +22,7 @@ using DeviceInstance =
                                                                 BetaDataType,
                                                                 ComputeDataType,
                                                                 YDataType,
+                                                                SaveMeanInvStdDataType,
                                                                 YElementOp,
                                                                 Rank,
                                                                 NumReduceDim,
@@ -33,7 +37,8 @@ using DeviceInstance =
                                                                 2,   // GammaScalarPerVector
                                                                 1,   // BetaVecDim (0=M, 1=K)
                                                                 2,   // BetaScalarPerVector
-                                                                2>;  // OutScalarPerVector
+                                                                2,   // YScalarPerVector
+                                                                1>; // SaveMeanInvStdScalarPerVector
 
 #include "run_groupnorm_example.inc"
 
diff --git a/example/42_groupnorm/groupnorm_swish_fp16.cpp b/example/42_groupnorm/groupnorm_swish_fp16.cpp
index 363f22ed4c..97cd4698aa 100644
--- a/example/42_groupnorm/groupnorm_swish_fp16.cpp
+++ b/example/42_groupnorm/groupnorm_swish_fp16.cpp
@@ -6,12 +6,15 @@
 constexpr int Rank         = 5;
 constexpr int NumReduceDim = 3;
 
-using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
-using YDataType       = ck::half_t;
-using ComputeDataType = float;
-using YElementOp      = ck::tensor_operation::element_wise::Swish;
+using XDataType              = ck::half_t;
+using GammaDataType          = ck::half_t;
+using BetaDataType           = ck::half_t;
+using YDataType              = ck::half_t;
+using SaveMeanInvStdDataType = float;
+using ComputeDataType        = float;
+using YElementOp             = ck::tensor_operation::element_wise::Swish;
+
+#define SAVE_MEAN_INV_STD
 
 using DeviceInstance =
     ck::tensor_operation::device::DeviceNormalizationImpl<XDataType,
@@ -19,6 +22,7 @@ using DeviceInstance =
                                                           BetaDataType,
                                                           ComputeDataType,
                                                           YDataType,
+                                                          SaveMeanInvStdDataType,
                                                           YElementOp,
                                                           Rank,
                                                           NumReduceDim,
@@ -33,7 +37,8 @@ using DeviceInstance =
                                                           2,    // GammaScalarPerVector
                                                           1,    // BetaVecDim (0=M, 1=K)
                                                           2,    // BetaScalarPerVector
-                                                          2>;   // OutScalarPerVector
+                                                          2,    // YScalarPerVector
+                                                          1>;   // SaveMeanInvStdScalarPerVector
 
 #include "run_groupnorm_example.inc"
 
diff --git a/example/42_groupnorm/run_groupnorm_example.inc b/example/42_groupnorm/run_groupnorm_example.inc
index 16065c8d46..da41e90639 100644
--- a/example/42_groupnorm/run_groupnorm_example.inc
+++ b/example/42_groupnorm/run_groupnorm_example.inc
@@ -34,6 +34,8 @@ int run_groupnorm_example(int argc, char* argv[])
     Tensor<YDataType> y({N, H, W, G, C});
     Tensor<GammaDataType> gamma({G, C});
     Tensor<BetaDataType> beta({G, C});
+    Tensor<SaveMeanInvStdDataType> save_mean({N, G});
+    Tensor<SaveMeanInvStdDataType> save_inv_std({N, G});
 
     ck::utils::FillUniformDistribution<XDataType>{0.f, 1.f}(x);
     ck::utils::FillUniformDistribution<GammaDataType>{0.f, 1.f}(gamma);
@@ -43,6 +45,11 @@ int run_groupnorm_example(int argc, char* argv[])
     DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
     DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
     DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
+#ifdef SAVE_MEAN_INV_STD
+    DeviceMem save_mean_dev(sizeof(SaveMeanInvStdDataType) * save_mean.mDesc.GetElementSpaceSize());
+    DeviceMem save_inv_std_dev(sizeof(SaveMeanInvStdDataType) *
+                               save_inv_std.mDesc.GetElementSpaceSize());
+#endif
 
     x_dev.ToDevice(x.mData.data());
     gamma_dev.ToDevice(gamma.mData.data());
@@ -57,14 +64,23 @@ int run_groupnorm_example(int argc, char* argv[])
         {0, 0, 0, C, 1},
         {0, 0, 0, C, 1},
         std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
+        std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
+                                    save_mean.mDesc.GetStrides().end()},
+        std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
+                                    save_mean.mDesc.GetStrides().end()},
         {1, 2, 4}, // reduction dimension: [H, W, C]
         1e-6,
         x_dev.GetDeviceBuffer(),
         gamma_dev.GetDeviceBuffer(),
         beta_dev.GetDeviceBuffer(),
         y_dev.GetDeviceBuffer(),
+#ifdef SAVE_MEAN_INV_STD
+        save_mean_dev.GetDeviceBuffer(),
+        save_inv_std_dev.GetDeviceBuffer(),
+#else
         nullptr,
         nullptr,
+#endif
         y_element_op);
 
     if(!device_instance.IsSupportedArgument(argument_ptr.get()))
@@ -92,21 +108,40 @@ int run_groupnorm_example(int argc, char* argv[])
     bool pass = true;
     {
         Tensor<YDataType> host_y({N, H, W, G, C});
-        using ReferenceInstance = ck::tensor_operation::host::ReferenceGroupnorm<XDataType,
-                                                                                 GammaDataType,
-                                                                                 BetaDataType,
-                                                                                 YDataType,
-                                                                                 ComputeDataType,
-                                                                                 YElementOp>;
+        Tensor<SaveMeanInvStdDataType> host_save_mean(HostTensorDescriptor{N, G});
+        Tensor<SaveMeanInvStdDataType> host_save_inv_std(HostTensorDescriptor{N, G});
+        using ReferenceInstance =
+            ck::tensor_operation::host::ReferenceGroupnorm<XDataType,
+                                                           GammaDataType,
+                                                           BetaDataType,
+                                                           YDataType,
+                                                           SaveMeanInvStdDataType,
+                                                           ComputeDataType,
+                                                           YElementOp>;
 
         ReferenceInstance ref;
-        auto ref_argument =
-            ref.MakeArgument(x, gamma, beta, host_y, y_element_op, {N, H, W, G, C}, 1e-6);
-        auto ref_invoker = ref.MakeInvoker();
+        auto ref_argument = ref.MakeArgument(x,
+                                             gamma,
+                                             beta,
+                                             host_y,
+                                             host_save_mean,
+                                             host_save_inv_std,
+                                             y_element_op,
+                                             {N, H, W, G, C},
+                                             1e-6);
+        auto ref_invoker  = ref.MakeInvoker();
         ref_invoker.Run(ref_argument);
 
         y_dev.FromDevice(y.mData.data());
         pass &= ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
+#ifdef SAVE_MEAN_INV_STD
+        save_mean_dev.FromDevice(save_mean.mData.data());
+        save_inv_std_dev.FromDevice(save_inv_std.mData.data());
+        pass &= ck::utils::check_err(
+            save_mean, host_save_mean, "Error: Incorrect results (mean)", 1e-3, 1e-3);
+        pass &= ck::utils::check_err(
+            save_inv_std, host_save_inv_std, "Error: Incorrect results (inv_std)", 1e-3, 1e-3);
+#endif
     }
 
     return (pass ? 0 : 1);
diff --git a/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp b/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
index 76361f87a5..c02d540983 100644
--- a/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
+++ b/example/45_elementwise_normalization/elementwise_layernorm_blockwise.cpp
@@ -167,20 +167,31 @@ int main()
                            XElementwiseOperation>(x, a, b, mn, XElementwiseOperation{});
 
         Tensor<YDataType> host_y(f_host_tensor_descriptor2d(M, N, Stride));
+        Tensor<AccDataType> host_save_mean({M});
+        Tensor<AccDataType> host_save_inv_std({M});
         using ReferenceInstance =
             ck::tensor_operation::host::ReferenceLayernorm<XDataType,
                                                            GammaDataType,
                                                            BetaDataType,
                                                            YDataType,
                                                            AccDataType,
+                                                           AccDataType,
                                                            YElementwiseOperation,
                                                            Rank,
                                                            NumReduceDim>;
 
         ReferenceInstance ref;
-        auto ref_argument =
-            ref.MakeArgument(x, gamma, beta, host_y, YElementwiseOperation{}, {M, N}, {1}, 1e-4);
-        auto ref_invoker = ref.MakeInvoker();
+        auto ref_argument = ref.MakeArgument(x,
+                                             gamma,
+                                             beta,
+                                             host_y,
+                                             host_save_mean,
+                                             host_save_inv_std,
+                                             YElementwiseOperation{},
+                                             {M, N},
+                                             {1},
+                                             1e-4);
+        auto ref_invoker  = ref.MakeInvoker();
         ref_invoker.Run(ref_argument);
 
         y_dev.FromDevice(y.mData.data());
diff --git a/include/ck/tensor_operation/gpu/device/device_normalization.hpp b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
index 1f178f9fcb..97e83ebab2 100644
--- a/include/ck/tensor_operation/gpu/device/device_normalization.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_normalization.hpp
@@ -14,8 +14,8 @@ namespace device {
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename ComputeDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim>
@@ -27,6 +27,8 @@ struct DeviceNormalization : public BaseOperator
                         const std::vector<index_t> gammaStrides,
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
+                        const std::vector<index_t> saveMeanStrides,
+                        const std::vector<index_t> saveInvStdStrides,
                         const std::vector<index_t> reduceDims,
                         double epsilon,
                         const void* p_x,
@@ -43,16 +45,16 @@ struct DeviceNormalization : public BaseOperator
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename ComputeDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim>
 using DeviceNormalizationPtr = std::unique_ptr<DeviceNormalization<XDataType,
                                                                    GammaDataType,
                                                                    BetaDataType,
-                                                                   ComputeDataType,
                                                                    YDataType,
+                                                                   SaveMeanInvStdDataType,
                                                                    YElementwiseOperation,
                                                                    Rank,
                                                                    NumReduceDim>>;
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
index ea0d805043..1ef3350185 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_impl.hpp
@@ -28,6 +28,7 @@ template <typename XDataType,
           typename BetaDataType,
           typename ComputeDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim,
@@ -43,12 +44,13 @@ template <typename XDataType,
           index_t BetaSrcVectorDim,
           index_t BetaSrcVectorSize,
           index_t YDstVectorSize,
+          index_t SaveMeanInvStdDstVectorSize,
           bool UseWelford = true>
 struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                                             GammaDataType,
                                                             BetaDataType,
-                                                            ComputeDataType,
                                                             YDataType,
+                                                            SaveMeanInvStdDataType,
                                                             YElementwiseOperation,
                                                             Rank,
                                                             NumReduceDim>
@@ -64,18 +66,24 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
          (BetaSrcVectorDim == 1 && KThreadSliceSize % BetaSrcVectorSize == 0)),
         "Invalid thread slice sizes and/or beta vector sizes configuration, please check!");
 
+    static_assert(MThreadSliceSize % SaveMeanInvStdDstVectorSize == 0,
+                  "Invalid thread slice sizes and/or save mean and inverse std vector sizes "
+                  "configuration, please check!");
+
     using PassThrough = tensor_operation::element_wise::PassThrough;
 
+    static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
     static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
     static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
 
+    static constexpr bool reduceAllDim = (NumInvariantDim == 0);
+    static_assert(!reduceAllDim); // TODO
+
     static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
                                     const std::vector<index_t>& inStrides,
                                     int numBlockTileIteration)
     {
-        constexpr index_t NumInvariantDim  = Rank - NumReduceDim;
         static constexpr index_t numSrcDim = Rank;
-        static constexpr bool reduceAllDim = (NumInvariantDim == 0);
 
         const auto tupleSrcLengths = make_tuple_from_array(inLengths, Number<numSrcDim>{});
         const auto tupleSrcStrides = make_tuple_from_array(inStrides, Number<numSrcDim>{});
@@ -133,7 +141,37 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         return (in_grid_desc_m_k_padded);
     };
 
+    static auto MakeSaveMeanInvStdDescriptor_M(const std::vector<index_t>& lengths,
+                                               const std::vector<index_t>& strides)
+    {
+        using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+
+        const auto tupleSrcLengths = make_tuple_from_array_and_index_seq(lengths, InvariantDims{});
+        const auto tupleSrcStrides = make_tuple_from_array_and_index_seq(strides, InvariantDims{});
+
+        const auto desc = make_naive_tensor_descriptor(tupleSrcLengths, tupleSrcStrides);
+
+        const auto grid_desc_m =
+            transform_tensor_descriptor(desc,
+                                        make_tuple(make_merge_transform(tupleSrcLengths)),
+                                        make_tuple(InvariantDims{}),
+                                        make_tuple(Sequence<0>{}));
+
+        const auto invariantLength = grid_desc_m.GetLength(Number<0>{});
+        const auto pad_M =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto grid_desc_m_padded = transform_tensor_descriptor(
+            grid_desc_m,
+            make_tuple(make_right_pad_transform(invariantLength, pad_M)),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0>{}));
+
+        return grid_desc_m_padded;
+    }
+
     using GridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1));
+    using GridDesc_M   = decltype(MakeSaveMeanInvStdDescriptor_M({1}, {1}));
 
     struct Argument : public BaseArgument
     {
@@ -142,17 +180,23 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                  const std::vector<index_t> gammaStrides,
                  const std::vector<index_t> betaStrides,
                  const std::vector<index_t> yStrides,
+                 const std::vector<index_t> saveMeanStrides,
+                 const std::vector<index_t> saveInvStdStrides,
                  const std::vector<index_t> reduceDims,
                  YElementwiseOperation y_elementwise_op,
                  double epsilon,
                  const XDataType* p_x,
                  const GammaDataType* p_gamma,
                  const BetaDataType* p_beta,
-                 YDataType* p_y)
+                 YDataType* p_y,
+                 SaveMeanInvStdDataType* p_saveMean,
+                 SaveMeanInvStdDataType* p_saveInvStd)
             : p_x_(p_x),
               p_gamma_(p_gamma),
               p_beta_(p_beta),
               p_y_(p_y),
+              p_saveMean_(p_saveMean),
+              p_saveInvStd_(p_saveInvStd),
               y_elementwise_op_(y_elementwise_op)
         {
             epsilon_ = static_cast<ComputeDataType>(epsilon);
@@ -162,16 +206,14 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
             yStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(yStrides, reduceDims);
             gammaStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(gammaStrides, reduceDims);
             betaStrides_  = shuffle_tensor_dimensions<Rank, NumReduceDim>(betaStrides, reduceDims);
+            saveMeanStrides_   = saveMeanStrides;
+            saveInvStdStrides_ = saveInvStdStrides;
 
-            long_index_t invariant_length;
-            long_index_t reduce_length;
+            std::tie(MRaw_, KRaw_) = get_2d_lengths<Rank, NumReduceDim>(Lengths_);
 
-            std::tie(invariant_length, reduce_length) =
-                get_2d_lengths<Rank, NumReduceDim>(Lengths_);
+            numBlockTileIteration_ = math::integer_divide_ceil(KRaw_, K_BlockTileSize);
 
-            numBlockTileIteration_ = math::integer_divide_ceil(reduce_length, K_BlockTileSize);
-
-            gridSize_ = math::integer_divide_ceil(invariant_length, M_BlockTileSize);
+            gridSize_ = math::integer_divide_ceil(MRaw_, M_BlockTileSize);
 
             x_grid_desc_m_k_ = MakeSrc2dDescriptor(Lengths_, xStrides_, numBlockTileIteration_);
             gamma_grid_desc_m_k_ =
@@ -179,9 +221,16 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
             beta_grid_desc_m_k_ =
                 MakeSrc2dDescriptor(Lengths_, betaStrides_, numBlockTileIteration_);
             y_grid_desc_m_k_ = MakeSrc2dDescriptor(Lengths_, yStrides_, numBlockTileIteration_);
+            save_mean_grid_desc_m_    = MakeSaveMeanInvStdDescriptor_M(Lengths_, saveMeanStrides);
+            save_inv_std_grid_desc_m_ = MakeSaveMeanInvStdDescriptor_M(Lengths_, saveInvStdStrides);
 
             isSweeponce_ =
                 x_grid_desc_m_k_.GetLength(Number<1>{}) <= KThreadClusterSize * KThreadSliceSize;
+
+            if constexpr(NumInvariantDim == 0)
+                invariant_lowest_length_ = 1;
+            else
+                invariant_lowest_length_ = Lengths_[NumInvariantDim - 1];
         }
 
         ComputeDataType epsilon_;
@@ -190,12 +239,16 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         const GammaDataType* p_gamma_;
         const BetaDataType* p_beta_;
         YDataType* p_y_;
+        SaveMeanInvStdDataType* p_saveMean_;
+        SaveMeanInvStdDataType* p_saveInvStd_;
 
         std::vector<index_t> Lengths_;
         std::vector<index_t> xStrides_;
         std::vector<index_t> gammaStrides_;
         std::vector<index_t> betaStrides_;
         std::vector<index_t> yStrides_;
+        std::vector<index_t> saveMeanStrides_;
+        std::vector<index_t> saveInvStdStrides_;
 
         YElementwiseOperation y_elementwise_op_;
 
@@ -206,7 +259,14 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
         GridDesc_M_K gamma_grid_desc_m_k_;
         GridDesc_M_K beta_grid_desc_m_k_;
         GridDesc_M_K y_grid_desc_m_k_;
+        GridDesc_M save_mean_grid_desc_m_;
+        GridDesc_M save_inv_std_grid_desc_m_;
         bool isSweeponce_;
+
+        index_t MRaw_; // invarient length
+        index_t KRaw_; // reduce length
+
+        index_t invariant_lowest_length_;
     };
 
     struct Invoker : public BaseInvoker
@@ -217,9 +277,11 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                                            GammaDataType,
                                                            BetaDataType,
                                                            YDataType,
+                                                           SaveMeanInvStdDataType,
                                                            ComputeDataType,
                                                            YElementwiseOperation,
                                                            GridDesc_M_K,
+                                                           GridDesc_M,
                                                            BlockSize,
                                                            MThreadClusterSize,
                                                            KThreadClusterSize,
@@ -233,6 +295,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                                            BetaSrcVectorSize,
                                                            XYSrcVectorDim,
                                                            YDstVectorSize,
+                                                           SaveMeanInvStdDstVectorSize,
                                                            UseWelford>(arg.isSweeponce_);
 
             float avg_time = 0;
@@ -245,12 +308,16 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                                                arg.gamma_grid_desc_m_k_,
                                                arg.beta_grid_desc_m_k_,
                                                arg.y_grid_desc_m_k_,
+                                               arg.save_mean_grid_desc_m_,
+                                               arg.save_inv_std_grid_desc_m_,
                                                arg.numBlockTileIteration_,
                                                arg.epsilon_,
                                                arg.p_x_,
                                                arg.p_gamma_,
                                                arg.p_beta_,
                                                arg.p_y_,
+                                               arg.p_saveMean_,
+                                               arg.p_saveInvStd_,
                                                arg.y_elementwise_op_);
 
             return (avg_time);
@@ -267,8 +334,6 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
     {
         const Argument* p_arg_ = dynamic_cast<const Argument*>(p_arg);
 
-        constexpr index_t NumInvariantDim = Rank - NumReduceDim;
-
         if constexpr(XYSrcVectorDim == 0)
         {
             if constexpr(NumInvariantDim == 0)
@@ -277,13 +342,15 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
             }
             else
             {
+                printf("!!!! %d\n", p_arg_->invariant_lowest_length_);
+
                 if(p_arg_->xStrides_[NumInvariantDim - 1] != 1)
                     return false;
 
-                if(p_arg_->invariant_lowest_length % XSrcVectorSize != 0)
+                if(p_arg_->invariant_lowest_length_ % XSrcVectorSize != 0)
                     return false;
 
-                if(p_arg_->invariant_lowest_length % YDstVectorSize != 0)
+                if(p_arg_->invariant_lowest_length_ % YDstVectorSize != 0)
                     return false;
             };
         }
@@ -325,7 +392,7 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
             if(p_arg_->betaStrides_[NumInvariantDim - 1] != 1)
                 return (false);
 
-            if(p_arg_->invariant_lowest_length % BetaSrcVectorSize != 0)
+            if(p_arg_->invariant_lowest_length_ % BetaSrcVectorSize != 0)
                 return (false);
         }
         else // if fastest dim is reduced
@@ -337,6 +404,9 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                 return (false);
         }
 
+        if(p_arg_->invariant_lowest_length_ % SaveMeanInvStdDstVectorSize != 0)
+            return false;
+
         return true;
     };
 
@@ -346,6 +416,8 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                         const std::vector<index_t> gammaStrides,
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
+                        const std::vector<index_t> saveMeanStrides,
+                        const std::vector<index_t> saveInvStdStrides,
                         const std::vector<index_t> reduceDims,
                         double epsilon,
                         const void* p_x,
@@ -353,27 +425,30 @@ struct DeviceNormalizationImpl : public DeviceNormalization<XDataType,
                         const void* p_beta,
                         void* p_y,
                         void* p_saveMean,
-                        void* p_saveInvVar,
+                        void* p_saveInvStd,
                         YElementwiseOperation y_elementwise_op) override
     {
-        // TODO
-        // Optional cache of the intermediate results (mean and InvVariance) during the
-        // forward pass could speedup in the backward
-        ignore = p_saveMean;
-        ignore = p_saveInvVar;
+        if(lengths.size() != Rank || xStrides.size() != Rank || gammaStrides.size() != Rank ||
+           betaStrides.size() != Rank || yStrides.size() != Rank ||
+           saveMeanStrides.size() != NumInvariantDim || saveInvStdStrides.size() != NumInvariantDim)
+            throw std::runtime_error("dimension is incorrect");
 
         return std::make_unique<Argument>(lengths,
                                           xStrides,
                                           gammaStrides,
                                           betaStrides,
                                           yStrides,
+                                          saveMeanStrides,
+                                          saveInvStdStrides,
                                           reduceDims,
                                           y_elementwise_op,
                                           epsilon,
                                           static_cast<const XDataType*>(p_x),
                                           static_cast<const GammaDataType*>(p_gamma),
                                           static_cast<const BetaDataType*>(p_beta),
-                                          static_cast<YDataType*>(p_y));
+                                          static_cast<YDataType*>(p_y),
+                                          static_cast<SaveMeanInvStdDataType*>(p_saveMean),
+                                          static_cast<SaveMeanInvStdDataType*>(p_saveInvStd));
     };
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
index 8b2b3c41bf..e969e493c6 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_normalization_splitk_impl.hpp
@@ -19,7 +19,7 @@
 namespace ck {
 template <typename GridwiseWelford,
           typename XDataType,
-          typename MeanVarDataType,
+          typename WorkspaceMeanVarDataType,
           typename ComputeDataType,
           typename XGridDesc_M_K,
           typename MeanVarGridDesc_M_KBlock>
@@ -28,8 +28,8 @@ kernel_normalizationSplitK1st(const XGridDesc_M_K x_grid_desc_m_k,
                               const MeanVarGridDesc_M_KBlock mean_var_grid_desc_m_kblock,
                               index_t num_k_block_tile_iteration,
                               const XDataType* const __restrict__ p_x_global,
-                              MeanVarDataType* const __restrict__ p_welford_mean,
-                              MeanVarDataType* const __restrict__ p_welford_variance,
+                              WorkspaceMeanVarDataType* const __restrict__ p_welford_mean,
+                              WorkspaceMeanVarDataType* const __restrict__ p_welford_variance,
                               int32_t* const __restrict__ p_welford_count)
 {
     GridwiseWelford::Run(x_grid_desc_m_k,
@@ -42,16 +42,18 @@ kernel_normalizationSplitK1st(const XGridDesc_M_K x_grid_desc_m_k,
 };
 
 template <typename GridwiseWelfordNormalization,
-          typename MeanVarDataType,
+          typename WorkspaceMeanVarDataType,
           typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename ComputeDataType,
           typename YElementwiseOperation,
           typename MeanVarGridDesc_M_KBlock,
           typename CountGridDesc_M_KBlock,
-          typename XYGammaBetaGridDesc_M_K>
+          typename XYGammaBetaGridDesc_M_K,
+          typename SaveMeanInvStdGridDesc_M>
 __global__ void
 kernel_normalizationSplitK2nd(const MeanVarGridDesc_M_KBlock mean_var_grid_desc_m_kblock,
                               const CountGridDesc_M_KBlock count_grid_desc_m_kblock,
@@ -59,17 +61,21 @@ kernel_normalizationSplitK2nd(const MeanVarGridDesc_M_KBlock mean_var_grid_desc_
                               const XYGammaBetaGridDesc_M_K gamma_grid_desc_m_k,
                               const XYGammaBetaGridDesc_M_K beta_grid_desc_m_k,
                               const XYGammaBetaGridDesc_M_K y_grid_desc_m_k,
+                              const SaveMeanInvStdGridDesc_M save_mean_grid_desc_m,
+                              const SaveMeanInvStdGridDesc_M save_inv_std_grid_desc_m,
                               index_t num_k_mean_var_count_iteration,
                               index_t num_k_block_tile_iteration,
                               index_t k_grid_size,
                               ComputeDataType epsilon,
-                              const MeanVarDataType* const p_mean_global,
-                              const MeanVarDataType* const p_variance_global,
+                              const WorkspaceMeanVarDataType* const p_mean_global,
+                              const WorkspaceMeanVarDataType* const p_variance_global,
                               const int32_t* const p_welford_count_global,
                               const XDataType* const __restrict__ p_x_global,
                               const GammaDataType* const __restrict__ p_gamma_global,
                               const BetaDataType* const __restrict__ p_beta_global,
                               YDataType* const __restrict__ p_y_global,
+                              SaveMeanInvStdDataType* const __restrict__ p_save_mean_global,
+                              SaveMeanInvStdDataType* const __restrict__ p_save_inv_std_global,
                               const YElementwiseOperation y_elementwise_op)
 {
     GridwiseWelfordNormalization::Run(mean_var_grid_desc_m_kblock,
@@ -78,6 +84,8 @@ kernel_normalizationSplitK2nd(const MeanVarGridDesc_M_KBlock mean_var_grid_desc_
                                       gamma_grid_desc_m_k,
                                       beta_grid_desc_m_k,
                                       y_grid_desc_m_k,
+                                      save_mean_grid_desc_m,
+                                      save_inv_std_grid_desc_m,
                                       num_k_mean_var_count_iteration,
                                       num_k_block_tile_iteration,
                                       k_grid_size,
@@ -89,6 +97,8 @@ kernel_normalizationSplitK2nd(const MeanVarGridDesc_M_KBlock mean_var_grid_desc_
                                       p_gamma_global,
                                       p_beta_global,
                                       p_y_global,
+                                      p_save_mean_global,
+                                      p_save_inv_std_global,
                                       y_elementwise_op);
 };
 } // namespace ck
@@ -107,6 +117,7 @@ template <typename XDataType,
           typename BetaDataType,
           typename ComputeDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim,
@@ -121,17 +132,18 @@ template <typename XDataType,
           index_t GammaSrcVectorSize,
           index_t BetaSrcVectorDim,
           index_t BetaSrcVectorSize,
-          index_t YDstVectorSize>
+          index_t YDstVectorSize,
+          index_t SaveMeanInvStdDstVectorSize>
 struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                                                                   GammaDataType,
                                                                   BetaDataType,
-                                                                  ComputeDataType,
                                                                   YDataType,
+                                                                  SaveMeanInvStdDataType,
                                                                   YElementwiseOperation,
                                                                   Rank,
                                                                   NumReduceDim>
 {
-    using MeanVarDataType = ComputeDataType;
+    using WorkspaceMeanVarDataType = SaveMeanInvStdDataType;
 
     static_assert(BlockSize == MThreadClusterSize * KThreadClusterSize);
     static_assert(
@@ -144,22 +156,28 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
          (BetaSrcVectorDim == 1 && KThreadSliceSize % BetaSrcVectorSize == 0)),
         "Invalid thread slice sizes and/or beta vector sizes configuration, please check!");
 
+    static_assert(MThreadSliceSize % SaveMeanInvStdDstVectorSize == 0,
+                  "Invalid thread slice sizes and/or save mean and inverse std vector sizes "
+                  "configuration, please check!");
+
     using PassThrough = tensor_operation::element_wise::PassThrough;
 
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
 
+    static constexpr index_t NumInvariantDim = Rank - NumReduceDim;
     static constexpr index_t M_BlockTileSize = MThreadClusterSize * MThreadSliceSize;
     static constexpr index_t K_BlockTileSize = KThreadClusterSize * KThreadSliceSize;
 
+    static constexpr bool reduceAllDim = (NumInvariantDim == 0);
+    static_assert(!reduceAllDim); // TODO
+
     static auto MakeSrc2dDescriptor(const std::vector<index_t>& inLengths,
                                     const std::vector<index_t>& inStrides,
                                     int kBlockSize,
                                     int numBlockTileIteration)
     {
-        constexpr index_t NumInvariantDim  = Rank - NumReduceDim;
         static constexpr index_t numSrcDim = Rank;
-        static constexpr bool reduceAllDim = (NumInvariantDim == 0);
 
         const auto tupleSrcLengths = make_tuple_from_array(inLengths, Number<numSrcDim>{});
         const auto tupleSrcStrides = make_tuple_from_array(inStrides, Number<numSrcDim>{});
@@ -219,7 +237,7 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
     };
 
     template <typename DoPads, index_t MPerTile, index_t KPerTile>
-    static auto MakeMeanVarDescriptor_M_K(index_t M, index_t K)
+    static auto MakeWorkspaceMeanVarDescriptor_M_K(index_t M, index_t K)
     {
         const auto grid_desc_m_k =
             make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(K, I1));
@@ -227,26 +245,57 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
     }
 
     template <typename DoPads, index_t MPerTile, index_t KPerTile>
-    static auto MakeCountDescriptor_M_K(index_t M, index_t K)
+    static auto MakeWorkspaceCountDescriptor_M_K(index_t M, index_t K)
     {
         const auto grid_desc_m_k =
             make_naive_tensor_descriptor(make_tuple(M, K), make_tuple(I0, I1));
         return PadTensorDescriptor(grid_desc_m_k, make_tuple(MPerTile, KPerTile), DoPads{});
     }
 
+    static auto MakeSaveMeanInvStdDescriptor_M(const std::vector<index_t>& lengths,
+                                               const std::vector<index_t>& strides)
+    {
+        using InvariantDims = typename arithmetic_sequence_gen<0, NumInvariantDim, 1>::type;
+
+        const auto tupleSrcLengths = make_tuple_from_array_and_index_seq(lengths, InvariantDims{});
+        const auto tupleSrcStrides = make_tuple_from_array_and_index_seq(strides, InvariantDims{});
+
+        const auto desc = make_naive_tensor_descriptor(tupleSrcLengths, tupleSrcStrides);
+
+        const auto grid_desc_m =
+            transform_tensor_descriptor(desc,
+                                        make_tuple(make_merge_transform(tupleSrcLengths)),
+                                        make_tuple(InvariantDims{}),
+                                        make_tuple(Sequence<0>{}));
+
+        const auto invariantLength = grid_desc_m.GetLength(Number<0>{});
+        const auto pad_M =
+            math::integer_least_multiple(invariantLength, M_BlockTileSize) - invariantLength;
+
+        auto grid_desc_m_padded = transform_tensor_descriptor(
+            grid_desc_m,
+            make_tuple(make_right_pad_transform(invariantLength, pad_M)),
+            make_tuple(Sequence<0>{}),
+            make_tuple(Sequence<0>{}));
+
+        return grid_desc_m_padded;
+    }
+
     using SrcGridDesc_M_K = decltype(MakeSrc2dDescriptor({1}, {1}, 1, 1));
     using Kernel1MeanVarGridDesc_M_KBlock =
-        decltype(MakeMeanVarDescriptor_M_K<Sequence<true, false>, 1, 1>(1, 1));
+        decltype(MakeWorkspaceMeanVarDescriptor_M_K<Sequence<true, false>, 1, 1>(1, 1));
 
     using Kernel2MeanVarGridDesc_M_KBlock =
-        decltype(MakeMeanVarDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+        decltype(MakeWorkspaceMeanVarDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
 
     using Kernel2CountGridDesc_M_KBlock =
-        decltype(MakeCountDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+        decltype(MakeWorkspaceCountDescriptor_M_K<Sequence<true, true>, 1, 1>(1, 1));
+
+    using SaveMeanInvStdGridDesc_M = decltype(MakeSaveMeanInvStdDescriptor_M({1}, {1}));
 
     using GridwiseWelford = GridwiseNormalizationSplitK1st<XDataType,
                                                            ComputeDataType,
-                                                           MeanVarDataType,
+                                                           WorkspaceMeanVarDataType,
                                                            SrcGridDesc_M_K,
                                                            Kernel1MeanVarGridDesc_M_KBlock,
                                                            BlockSize,
@@ -258,16 +307,18 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                                                            XSrcVectorSize>;
 
     using GridwiseWelfordNormalization =
-        GridwiseNormalizationSplitK2nd<MeanVarDataType,
+        GridwiseNormalizationSplitK2nd<WorkspaceMeanVarDataType,
                                        XDataType,
                                        GammaDataType,
                                        BetaDataType,
                                        YDataType,
+                                       SaveMeanInvStdDataType,
                                        ComputeDataType,
                                        YElementwiseOperation,
                                        Kernel2MeanVarGridDesc_M_KBlock,
                                        Kernel2CountGridDesc_M_KBlock,
                                        SrcGridDesc_M_K,
+                                       SaveMeanInvStdGridDesc_M,
                                        BlockSize,
                                        MThreadClusterSize,
                                        KThreadClusterSize,
@@ -280,7 +331,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                                        BetaSrcVectorDim,
                                        BetaSrcVectorSize,
                                        XYVectorDim,
-                                       YDstVectorSize>;
+                                       YDstVectorSize,
+                                       SaveMeanInvStdDstVectorSize>;
 
     struct Argument : public BaseArgument
     {
@@ -289,17 +341,23 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                  const std::vector<index_t> gammaStrides,
                  const std::vector<index_t> betaStrides,
                  const std::vector<index_t> yStrides,
+                 const std::vector<index_t> saveMeanStrides,
+                 const std::vector<index_t> saveInvStdStrides,
                  const std::vector<index_t> reduceDims,
                  YElementwiseOperation y_elementwise_op,
                  double epsilon,
                  const XDataType* p_x,
                  const GammaDataType* p_gamma,
                  const BetaDataType* p_beta,
-                 YDataType* p_y)
+                 YDataType* p_y,
+                 SaveMeanInvStdDataType* p_saveMean,
+                 SaveMeanInvStdDataType* p_saveInvStd)
             : p_x_(p_x),
               p_gamma_(p_gamma),
               p_beta_(p_beta),
               p_y_(p_y),
+              p_saveMean_(p_saveMean),
+              p_saveInvStd_(p_saveInvStd),
               p_workspace_mean_{nullptr},
               p_workspace_var_{nullptr},
               p_workspace_count_{nullptr},
@@ -312,6 +370,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
             yStrides_     = shuffle_tensor_dimensions<Rank, NumReduceDim>(yStrides, reduceDims);
             gammaStrides_ = shuffle_tensor_dimensions<Rank, NumReduceDim>(gammaStrides, reduceDims);
             betaStrides_  = shuffle_tensor_dimensions<Rank, NumReduceDim>(betaStrides, reduceDims);
+            saveMeanStrides_   = saveMeanStrides;
+            saveInvStdStrides_ = saveInvStdStrides;
 
             std::tie(MRaw_, KRaw_) = get_2d_lengths<Rank, NumReduceDim>(Lengths_);
 
@@ -346,20 +406,28 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
             y_grid_desc_m_k_ =
                 MakeSrc2dDescriptor(Lengths_, yStrides_, kGridSize_, numBlockTileIteration_);
 
+            save_mean_grid_desc_m_    = MakeSaveMeanInvStdDescriptor_M(Lengths_, saveMeanStrides);
+            save_inv_std_grid_desc_m_ = MakeSaveMeanInvStdDescriptor_M(Lengths_, saveInvStdStrides);
+
             // We don't need to pad in K dimension for Welford1. Set KPerTile 1.
             kernel1_mean_var_grid_desc_m_kblock_ =
-                MakeMeanVarDescriptor_M_K<Sequence<true, false>, M_BlockTileSize, 1>(MRaw_,
-                                                                                     kGridSize_);
+                MakeWorkspaceMeanVarDescriptor_M_K<Sequence<true, false>, M_BlockTileSize, 1>(
+                    MRaw_, kGridSize_);
 
             kernel2_mean_var_grid_desc_m_kblock_ =
-                MakeMeanVarDescriptor_M_K<Sequence<true, true>,
-                                          M_BlockTileSize,
-                                          K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+                MakeWorkspaceMeanVarDescriptor_M_K<Sequence<true, true>,
+                                                   M_BlockTileSize,
+                                                   K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
 
             kernel2_count_grid_desc_m_kblock_ =
-                MakeCountDescriptor_M_K<Sequence<true, true>,
-                                        M_BlockTileSize,
-                                        K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+                MakeWorkspaceCountDescriptor_M_K<Sequence<true, true>,
+                                                 M_BlockTileSize,
+                                                 K_MeanVarCountBlockTileSize>(MRaw_, kGridSize_);
+
+            if constexpr(NumInvariantDim == 0)
+                invariant_lowest_length_ = 1;
+            else
+                invariant_lowest_length_ = Lengths_[NumInvariantDim - 1];
         }
 
         ComputeDataType epsilon_;
@@ -368,6 +436,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
         const GammaDataType* p_gamma_;
         const BetaDataType* p_beta_;
         YDataType* p_y_;
+        SaveMeanInvStdDataType* p_saveMean_;
+        SaveMeanInvStdDataType* p_saveInvStd_;
         void* p_workspace_mean_;
         void* p_workspace_var_;
         void* p_workspace_count_;
@@ -377,6 +447,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
         std::vector<index_t> gammaStrides_;
         std::vector<index_t> betaStrides_;
         std::vector<index_t> yStrides_;
+        std::vector<index_t> saveMeanStrides_;
+        std::vector<index_t> saveInvStdStrides_;
 
         YElementwiseOperation y_elementwise_op_;
 
@@ -389,6 +461,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
         SrcGridDesc_M_K gamma_grid_desc_m_k_;
         SrcGridDesc_M_K beta_grid_desc_m_k_;
         SrcGridDesc_M_K y_grid_desc_m_k_;
+        SaveMeanInvStdGridDesc_M save_mean_grid_desc_m_;
+        SaveMeanInvStdGridDesc_M save_inv_std_grid_desc_m_;
 
         Kernel1MeanVarGridDesc_M_KBlock kernel1_mean_var_grid_desc_m_kblock_;
         Kernel2MeanVarGridDesc_M_KBlock kernel2_mean_var_grid_desc_m_kblock_;
@@ -396,6 +470,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
 
         index_t MRaw_; // invarient length
         index_t KRaw_; // reduce length
+
+        index_t invariant_lowest_length_;
     };
 
     struct Invoker : public BaseInvoker
@@ -408,60 +484,68 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
 
             auto kernel1 = kernel_normalizationSplitK1st<GridwiseWelford,
                                                          XDataType,
-                                                         MeanVarDataType,
+                                                         WorkspaceMeanVarDataType,
                                                          ComputeDataType,
                                                          SrcGridDesc_M_K,
                                                          Kernel1MeanVarGridDesc_M_KBlock>;
 
             auto kernel2 = kernel_normalizationSplitK2nd<GridwiseWelfordNormalization,
-                                                         MeanVarDataType,
+                                                         WorkspaceMeanVarDataType,
                                                          XDataType,
                                                          GammaDataType,
                                                          BetaDataType,
                                                          YDataType,
+                                                         SaveMeanInvStdDataType,
                                                          ComputeDataType,
                                                          YElementwiseOperation,
                                                          Kernel2MeanVarGridDesc_M_KBlock,
                                                          Kernel2CountGridDesc_M_KBlock,
-                                                         SrcGridDesc_M_K>;
+                                                         SrcGridDesc_M_K,
+                                                         SaveMeanInvStdGridDesc_M>;
 
             float avg_time = 0;
-            avg_time += launch_and_time_kernel(stream_config,
-                                               kernel1,
-                                               dim3(arg.gridSize_),
-                                               dim3(BlockSize),
-                                               0,
-                                               arg.x_grid_desc_m_k_,
-                                               arg.kernel1_mean_var_grid_desc_m_kblock_,
-                                               arg.numBlockTileIteration_,
-                                               arg.p_x_,
-                                               static_cast<MeanVarDataType*>(arg.p_workspace_mean_),
-                                               static_cast<MeanVarDataType*>(arg.p_workspace_var_),
-                                               static_cast<int32_t*>(arg.p_workspace_count_));
+            avg_time += launch_and_time_kernel(
+                stream_config,
+                kernel1,
+                dim3(arg.gridSize_),
+                dim3(BlockSize),
+                0,
+                arg.x_grid_desc_m_k_,
+                arg.kernel1_mean_var_grid_desc_m_kblock_,
+                arg.numBlockTileIteration_,
+                arg.p_x_,
+                static_cast<WorkspaceMeanVarDataType*>(arg.p_workspace_mean_),
+                static_cast<WorkspaceMeanVarDataType*>(arg.p_workspace_var_),
+                static_cast<int32_t*>(arg.p_workspace_count_));
 
-            avg_time += launch_and_time_kernel(stream_config,
-                                               kernel2,
-                                               dim3(arg.gridSize_),
-                                               dim3(BlockSize),
-                                               0,
-                                               arg.kernel2_mean_var_grid_desc_m_kblock_,
-                                               arg.kernel2_count_grid_desc_m_kblock_,
-                                               arg.x_grid_desc_m_k_,
-                                               arg.gamma_grid_desc_m_k_,
-                                               arg.beta_grid_desc_m_k_,
-                                               arg.y_grid_desc_m_k_,
-                                               arg.numMeanVarCountIteration_,
-                                               arg.numBlockTileIteration_,
-                                               arg.kGridSize_,
-                                               arg.epsilon_,
-                                               static_cast<MeanVarDataType*>(arg.p_workspace_mean_),
-                                               static_cast<MeanVarDataType*>(arg.p_workspace_var_),
-                                               static_cast<int32_t*>(arg.p_workspace_count_),
-                                               arg.p_x_,
-                                               arg.p_gamma_,
-                                               arg.p_beta_,
-                                               arg.p_y_,
-                                               arg.y_elementwise_op_);
+            avg_time += launch_and_time_kernel(
+                stream_config,
+                kernel2,
+                dim3(arg.gridSize_),
+                dim3(BlockSize),
+                0,
+                arg.kernel2_mean_var_grid_desc_m_kblock_,
+                arg.kernel2_count_grid_desc_m_kblock_,
+                arg.x_grid_desc_m_k_,
+                arg.gamma_grid_desc_m_k_,
+                arg.beta_grid_desc_m_k_,
+                arg.y_grid_desc_m_k_,
+                arg.save_mean_grid_desc_m_,
+                arg.save_inv_std_grid_desc_m_,
+                arg.numMeanVarCountIteration_,
+                arg.numBlockTileIteration_,
+                arg.kGridSize_,
+                arg.epsilon_,
+                static_cast<const WorkspaceMeanVarDataType*>(arg.p_workspace_mean_),
+                static_cast<const WorkspaceMeanVarDataType*>(arg.p_workspace_var_),
+                static_cast<const int32_t*>(arg.p_workspace_count_),
+                arg.p_x_,
+                arg.p_gamma_,
+                arg.p_beta_,
+                arg.p_y_,
+                arg.p_saveMean_,
+                arg.p_saveInvStd_,
+                arg.y_elementwise_op_);
 
             return avg_time;
         };
@@ -482,10 +566,10 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
         int welford_size = pArg_->MRaw_ * pArg_->kGridSize_;
 
         // workspace for welford intermediate mean
-        workspace_size += welford_size * sizeof(MeanVarDataType) + 64;
+        workspace_size += welford_size * sizeof(WorkspaceMeanVarDataType) + 64;
 
         // workspace for welford intermediate variance
-        workspace_size += welford_size * sizeof(MeanVarDataType) + 64;
+        workspace_size += welford_size * sizeof(WorkspaceMeanVarDataType) + 64;
 
         // workspace for welford intermediate count
         workspace_size += pArg_->kGridSize_ * sizeof(int32_t) + 64;
@@ -504,13 +588,13 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
         // setup buffer used for intermediate welford mean
         pArg_->p_workspace_mean_ = static_cast<char*>(pArg_->p_workspace_);
 
-        index_t mean_space_sz = welford_size * sizeof(MeanVarDataType);
+        index_t mean_space_sz = welford_size * sizeof(WorkspaceMeanVarDataType);
         mean_space_sz         = math::integer_least_multiple(mean_space_sz, 64);
 
         // setup buffer used for intermediate welford varirance
         pArg_->p_workspace_var_ = reinterpret_cast<char*>(pArg_->p_workspace_mean_) + mean_space_sz;
 
-        index_t variance_space_sz = welford_size * sizeof(MeanVarDataType);
+        index_t variance_space_sz = welford_size * sizeof(WorkspaceMeanVarDataType);
         variance_space_sz         = math::integer_least_multiple(variance_space_sz, 64);
 
         // setup buffer used for intermediate welford count
@@ -522,8 +606,6 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
     {
         const Argument* p_arg_ = dynamic_cast<const Argument*>(p_arg);
 
-        constexpr index_t NumInvariantDim = Rank - NumReduceDim;
-
         if constexpr(XYVectorDim == 0)
         {
             if constexpr(NumInvariantDim == 0)
@@ -535,10 +617,10 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                 if(p_arg_->xStrides_[NumInvariantDim - 1] != 1)
                     return false;
 
-                if(p_arg_->invariant_lowest_length % XSrcVectorSize != 0)
+                if(p_arg_->invariant_lowest_length_ % XSrcVectorSize != 0)
                     return false;
 
-                if(p_arg_->invariant_lowest_length % YDstVectorSize != 0)
+                if(p_arg_->invariant_lowest_length_ % YDstVectorSize != 0)
                     return false;
             };
         }
@@ -578,7 +660,7 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
             if(p_arg_->betaStrides_[NumInvariantDim - 1] != 1)
                 return false;
 
-            if(p_arg_->invariant_lowest_length % BetaSrcVectorSize != 0)
+            if(p_arg_->invariant_lowest_length_ % BetaSrcVectorSize != 0)
                 return false;
         }
         else // if fastest dim is reduced
@@ -593,6 +675,9 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
         if(p_arg_->kGridSize_ <= 1)
             return false;
 
+        if(p_arg_->invariant_lowest_length_ % SaveMeanInvStdDstVectorSize != 0)
+            return false;
+
         return true;
     };
 
@@ -602,6 +687,8 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                         const std::vector<index_t> gammaStrides,
                         const std::vector<index_t> betaStrides,
                         const std::vector<index_t> yStrides,
+                        const std::vector<index_t> saveMeanStrides,
+                        const std::vector<index_t> saveInvStdStrides,
                         const std::vector<index_t> reduceDims,
                         double epsilon,
                         const void* p_x,
@@ -609,27 +696,30 @@ struct DeviceNormalizationSplitKImpl : public DeviceNormalization<XDataType,
                         const void* p_beta,
                         void* p_y,
                         void* p_saveMean,
-                        void* p_saveInvVar,
+                        void* p_saveInvStd,
                         YElementwiseOperation y_elementwise_op) override
     {
-        // TODO
-        // Optional cache of the intermediate results (mean and InvVariance) during the
-        // forward pass could speedup in the backward
-        ignore = p_saveMean;
-        ignore = p_saveInvVar;
+        if(lengths.size() != Rank || xStrides.size() != Rank || gammaStrides.size() != Rank ||
+           betaStrides.size() != Rank || yStrides.size() != Rank ||
+           saveMeanStrides.size() != NumInvariantDim || saveInvStdStrides.size() != NumInvariantDim)
+            throw std::runtime_error("dimension is incorrect");
 
         return std::make_unique<Argument>(lengths,
                                           xStrides,
                                           gammaStrides,
                                           betaStrides,
                                           yStrides,
+                                          saveMeanStrides,
+                                          saveInvStdStrides,
                                           reduceDims,
                                           y_elementwise_op,
                                           epsilon,
                                           static_cast<const XDataType*>(p_x),
                                           static_cast<const GammaDataType*>(p_gamma),
                                           static_cast<const BetaDataType*>(p_beta),
-                                          static_cast<YDataType*>(p_y));
+                                          static_cast<YDataType*>(p_y),
+                                          static_cast<SaveMeanInvStdDataType*>(p_saveMean),
+                                          static_cast<SaveMeanInvStdDataType*>(p_saveInvStd));
     };
 
     std::unique_ptr<BaseInvoker> MakeInvokerPointer() override
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
index c3f122106d..1bee1b93f9 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_naive_variance.hpp
@@ -18,9 +18,11 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename ComputeDataType,
           typename YElementwiseOperation,
           typename GridDesc_M_K,
+          typename GridDesc_M,
           index_t BlockSize,
           index_t MThreadClusterSize,
           index_t KThreadClusterSize,
@@ -34,6 +36,7 @@ template <typename XDataType,
           index_t BetaSrcVectorSize,
           index_t YDstVectorDim,
           index_t YDstVectorSize,
+          index_t SaveMeanInvStdDstVectorSize,
           bool SweepOnce>
 struct GridwiseNormalizationNaiveVariance_mk_to_mk
 {
@@ -45,6 +48,10 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                       (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
                   "Invalid thread slice sizes and/or vector sizes configuration, please check!");
 
+    static_assert(MThreadSliceSize % SaveMeanInvStdDstVectorSize == 0,
+                  "Invalid thread slice sizes and/or save mean and inverse std vector sizes "
+                  "configuration, please check!");
+
     static_assert(XSrcVectorSize == YDstVectorSize);
     static_assert(XSrcVectorSize == GammaSrcVectorSize);
     static_assert(XSrcVectorSize == BetaSrcVectorSize);
@@ -66,6 +73,10 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
     static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
 
+    using ThreadBufferLengths_M = Sequence<MThreadSliceSize>;
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}));
+
     using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})));
     using ThreadReduceDstDesc_M =
@@ -84,6 +95,8 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                                                     reduce::Add,
                                                     true>;
 
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
@@ -98,12 +111,16 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                                const GridDesc_M_K& gamma_grid_desc_m_k,
                                const GridDesc_M_K& beta_grid_desc_m_k,
                                const GridDesc_M_K& y_grid_desc_m_k,
+                               const GridDesc_M& save_mean_grid_desc_m,
+                               const GridDesc_M& save_inv_std_grid_desc_m,
                                index_t num_k_block_tile_iteration,
                                ComputeDataType epsilon,
                                const XDataType* const __restrict__ p_x_global,
                                const GammaDataType* const __restrict__ p_gamma_global,
                                const BetaDataType* const __restrict__ p_beta_global,
                                YDataType* const __restrict__ p_y_global,
+                               SaveMeanInvStdDataType* const __restrict__ p_save_mean_global,
+                               SaveMeanInvStdDataType* const __restrict__ p_save_inv_std_global,
                                const YElementwiseOperation y_elementwise_op)
     {
         // LDS
@@ -115,6 +132,12 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
         auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
 
+        auto save_mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_save_mean_global, save_mean_grid_desc_m.GetElementSpaceSize());
+
+        auto save_inv_std_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_save_inv_std_global, save_inv_std_grid_desc_m.GetElementSpaceSize());
+
         auto x_thread_buf = generate_tuple(
             [&](auto) {
                 return StaticBuffer<AddressSpaceEnum::Vgpr,
@@ -152,6 +175,8 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
             mean_square_thread_buf;
         StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>&
             var_thread_buf = mean_square_thread_buf;
+        StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>&
+            inv_std_thread_buf = mean_square_thread_buf;
 
         const index_t thread_local_id = get_thread_local_1d_id();
         const index_t block_global_id = get_block_1d_id();
@@ -228,6 +253,42 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                                  thread_k_cluster_id * YDstVectorSize),
                 y_elementwise_op);
 
+        auto threadwise_mean_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               SaveMeanInvStdDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               GridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,                 // DimAccessOrder
+                                               0,                           // SrcVectorDim
+                                               SaveMeanInvStdDstVectorSize, // ScalarPerVector
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                save_mean_grid_desc_m,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
+        auto threadwise_inv_std_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               SaveMeanInvStdDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               GridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,                 // DimAccessOrder
+                                               0,                           // SrcVectorDim
+                                               SaveMeanInvStdDstVectorSize, // ScalarPerVector
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                save_inv_std_grid_desc_m,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
         constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
         constexpr auto thread_copy_bwd_step_m_k =
             make_multi_index(0, SweepOnce ? 0 : -K_BlockTileSize);
@@ -243,7 +304,8 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
 
         // E(x), E[x^2], var(x)
         // FIXME: Should not hack the transform from deviceOP
-        int reduce_length = x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0];
+        ComputeDataType reduce_length = type_convert<ComputeDataType>(
+            x_grid_desc_m_k.GetTransforms()[I2].GetUpperLengths()[I0]);
 
         static_for<0, MThreadSliceSize, 1>{}([&](auto I) {
             mean_thread_buf(I)        = reduce::Add::template GetIdentityValue<ComputeDataType>();
@@ -302,10 +364,34 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                 // var(x) = E[x^2] - E[x]^2
                 var_thread_buf(I) =
                     mean_square_thread_buf(I) - (mean_thread_buf(I) * mean_thread_buf(I));
+
+                inv_std_thread_buf(I) = type_convert<ComputeDataType>(1.0f) /
+                                        ck::math::sqrt(var_thread_buf(I) + epsilon);
             });
 
+            // save mean and inverse std for backward (optional)
+            if(thread_k_cluster_id == 0)
+            {
+                if(p_save_mean_global != nullptr)
+                {
+                    threadwise_mean_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              mean_thread_buf,
+                                              save_mean_grid_desc_m,
+                                              save_mean_global_val_buf);
+                }
+                if(p_save_inv_std_global != nullptr)
+                {
+                    threadwise_inv_std_store.Run(thread_buffer_desc_m,
+                                                 make_tuple(I0),
+                                                 inv_std_thread_buf,
+                                                 save_inv_std_grid_desc_m,
+                                                 save_inv_std_global_val_buf);
+                }
+            }
+
+            // normalization
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                 static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =
@@ -314,7 +400,7 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                         // normalize
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
                             (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
-                            divisor;
+                            inv_std_thread_buf(iM);
 
                         // gamma & beta
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
@@ -404,8 +490,30 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                 // var(x) = E[x^2] - E[x]^2
                 var_thread_buf(I) =
                     mean_square_thread_buf(I) - (mean_thread_buf(I) * mean_thread_buf(I));
+
+                inv_std_thread_buf(I) = 1 / ck::math::sqrt(var_thread_buf(I) + epsilon);
             });
 
+            if(thread_k_cluster_id == 0)
+            {
+                if(p_save_mean_global != nullptr)
+                {
+                    threadwise_mean_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              mean_thread_buf,
+                                              save_mean_grid_desc_m,
+                                              save_mean_global_val_buf);
+                }
+                if(p_save_inv_std_global != nullptr)
+                {
+                    threadwise_inv_std_store.Run(thread_buffer_desc_m,
+                                                 make_tuple(I0),
+                                                 inv_std_thread_buf,
+                                                 save_inv_std_grid_desc_m,
+                                                 save_inv_std_global_val_buf);
+                }
+            }
+
             auto thread_copy_tail_m_k =
                 (num_k_block_tile_iteration - 1) * ThreadBufferNumber * thread_copy_fwd_step_m_k;
 
@@ -437,7 +545,6 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                 });
 
                 static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                    auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                     static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                         static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                             constexpr auto offset_m_k =
@@ -446,7 +553,7 @@ struct GridwiseNormalizationNaiveVariance_mk_to_mk
                             // normalize
                             y_thread_buf(iK0)(Number<offset_m_k>{}) =
                                 (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
-                                divisor;
+                                inv_std_thread_buf(iM);
 
                             // gamma
                             y_thread_buf(iK0)(Number<offset_m_k>{}) =
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
index e50fb98133..8157b4fbc3 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_selector.hpp
@@ -12,31 +12,42 @@ template <typename GridwiseReduction,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename ComputeDataType,
           typename YElementwiseOperation,
-          typename GridDesc_M_K>
-__global__ void kernel_normalization(const GridDesc_M_K x_grid_desc_m_k,
-                                     const GridDesc_M_K gamma_grid_desc_m_k,
-                                     const GridDesc_M_K beta_grid_desc_m_k,
-                                     const GridDesc_M_K y_grid_desc_m_k,
-                                     index_t num_k_block_tile_iteration,
-                                     ComputeDataType epsilon,
-                                     const XDataType* const __restrict__ p_x_global,
-                                     const GammaDataType* const __restrict__ p_gamma_global,
-                                     const BetaDataType* const __restrict__ p_beta_global,
-                                     YDataType* const __restrict__ p_y_global,
-                                     const YElementwiseOperation y_elementwise_op)
+          typename GridDesc_M_K,
+          typename GridDesc_M>
+__global__ void
+kernel_normalization(const GridDesc_M_K x_grid_desc_m_k,
+                     const GridDesc_M_K gamma_grid_desc_m_k,
+                     const GridDesc_M_K beta_grid_desc_m_k,
+                     const GridDesc_M_K y_grid_desc_m_k,
+                     const GridDesc_M save_mean_grid_desc_m,
+                     const GridDesc_M save_inv_std_grid_desc_m,
+                     index_t num_k_block_tile_iteration,
+                     ComputeDataType epsilon,
+                     const XDataType* const __restrict__ p_x_global,
+                     const GammaDataType* const __restrict__ p_gamma_global,
+                     const BetaDataType* const __restrict__ p_beta_global,
+                     YDataType* const __restrict__ p_y_global,
+                     SaveMeanInvStdDataType* const __restrict__ p_save_mean_global,
+                     SaveMeanInvStdDataType* const __restrict__ p_save_inv_std_global,
+                     const YElementwiseOperation y_elementwise_op)
 {
     GridwiseReduction::Run(x_grid_desc_m_k,
                            gamma_grid_desc_m_k,
                            beta_grid_desc_m_k,
                            y_grid_desc_m_k,
+                           save_mean_grid_desc_m,
+                           save_inv_std_grid_desc_m,
                            num_k_block_tile_iteration,
                            epsilon,
                            p_x_global,
                            p_gamma_global,
                            p_beta_global,
                            p_y_global,
+                           p_save_mean_global,
+                           p_save_inv_std_global,
                            y_elementwise_op);
 };
 
@@ -44,9 +55,11 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename ComputeDataType,
           typename YElementwiseOperation,
           typename GridDesc_M_K,
+          typename GridDesc_M,
           index_t BlockSize,
           index_t MThreadClusterSize,
           index_t KThreadClusterSize,
@@ -60,6 +73,7 @@ template <typename XDataType,
           index_t BetaSrcVectorSize,
           index_t YDstVectorDim,
           index_t YDstVectorSize,
+          index_t SaveMeanInvStdDstVectorSize,
           bool UseWelford>
 auto NormalizationKernelSelector(bool isSweepOnce)
 {
@@ -68,9 +82,11 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                     GammaDataType,
                                                     BetaDataType,
                                                     YDataType,
+                                                    SaveMeanInvStdDataType,
                                                     ComputeDataType,
                                                     YElementwiseOperation,
                                                     GridDesc_M_K,
+                                                    GridDesc_M,
                                                     BlockSize,
                                                     MThreadClusterSize,
                                                     KThreadClusterSize,
@@ -84,15 +100,18 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                     BetaSrcVectorSize,
                                                     YDstVectorDim,
                                                     YDstVectorSize,
+                                                    SaveMeanInvStdDstVectorSize,
                                                     false>;
     using GridwiseNormalizationSweepOnceNaive =
         GridwiseNormalizationNaiveVariance_mk_to_mk<XDataType,
                                                     GammaDataType,
                                                     BetaDataType,
                                                     YDataType,
+                                                    SaveMeanInvStdDataType,
                                                     ComputeDataType,
                                                     YElementwiseOperation,
                                                     GridDesc_M_K,
+                                                    GridDesc_M,
                                                     BlockSize,
                                                     MThreadClusterSize,
                                                     KThreadClusterSize,
@@ -106,15 +125,18 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                     BetaSrcVectorSize,
                                                     YDstVectorDim,
                                                     YDstVectorSize,
+                                                    SaveMeanInvStdDstVectorSize,
                                                     true>;
     using GridwiseNormalizationGenericWelford =
         GridwiseNormalizationWelfordVariance_mk_to_mk<XDataType,
                                                       GammaDataType,
                                                       BetaDataType,
                                                       YDataType,
+                                                      SaveMeanInvStdDataType,
                                                       ComputeDataType,
                                                       YElementwiseOperation,
                                                       GridDesc_M_K,
+                                                      GridDesc_M,
                                                       BlockSize,
                                                       MThreadClusterSize,
                                                       KThreadClusterSize,
@@ -128,15 +150,18 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                       BetaSrcVectorSize,
                                                       YDstVectorDim,
                                                       YDstVectorSize,
+                                                      SaveMeanInvStdDstVectorSize,
                                                       false>;
     using GridwiseNormalizationSweepOnceWelford =
         GridwiseNormalizationWelfordVariance_mk_to_mk<XDataType,
                                                       GammaDataType,
                                                       BetaDataType,
                                                       YDataType,
+                                                      SaveMeanInvStdDataType,
                                                       ComputeDataType,
                                                       YElementwiseOperation,
                                                       GridDesc_M_K,
+                                                      GridDesc_M,
                                                       BlockSize,
                                                       MThreadClusterSize,
                                                       KThreadClusterSize,
@@ -150,6 +175,7 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                       BetaSrcVectorSize,
                                                       YDstVectorDim,
                                                       YDstVectorSize,
+                                                      SaveMeanInvStdDstVectorSize,
                                                       true>;
 
     if constexpr(UseWelford)
@@ -159,17 +185,21 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                   GammaDataType,
                                                   BetaDataType,
                                                   YDataType,
+                                                  SaveMeanInvStdDataType,
                                                   ComputeDataType,
                                                   YElementwiseOperation,
-                                                  GridDesc_M_K>
+                                                  GridDesc_M_K,
+                                                  GridDesc_M>
                            : kernel_normalization<GridwiseNormalizationGenericWelford,
                                                   XDataType,
                                                   GammaDataType,
                                                   BetaDataType,
                                                   YDataType,
+                                                  SaveMeanInvStdDataType,
                                                   ComputeDataType,
                                                   YElementwiseOperation,
-                                                  GridDesc_M_K>;
+                                                  GridDesc_M_K,
+                                                  GridDesc_M>;
     }
     else
     {
@@ -178,17 +208,21 @@ auto NormalizationKernelSelector(bool isSweepOnce)
                                                   GammaDataType,
                                                   BetaDataType,
                                                   YDataType,
+                                                  SaveMeanInvStdDataType,
                                                   ComputeDataType,
                                                   YElementwiseOperation,
-                                                  GridDesc_M_K>
+                                                  GridDesc_M_K,
+                                                  GridDesc_M>
                            : kernel_normalization<GridwiseNormalizationGenericNaive,
                                                   XDataType,
                                                   GammaDataType,
                                                   BetaDataType,
                                                   YDataType,
+                                                  SaveMeanInvStdDataType,
                                                   ComputeDataType,
                                                   YElementwiseOperation,
-                                                  GridDesc_M_K>;
+                                                  GridDesc_M_K,
+                                                  GridDesc_M>;
     }
 }
 
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
index 136ac94e7f..9e380f9638 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_splitk_2nd.hpp
@@ -17,11 +17,13 @@ template <typename MeanVarDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename ComputeDataType,
           typename YElementwiseOperation,
           typename MeanVarGridDesc_M_KBlock,
           typename CountGridDesc_M_KBlock,
           typename XYGammaBetaGridDesc_M_K,
+          typename SaveMeanInvStdGridDesc_M,
           index_t BlockSize,
           index_t MThreadClusterSize,
           index_t KThreadClusterSize,
@@ -34,7 +36,8 @@ template <typename MeanVarDataType,
           index_t BetaSrcVectorDim,
           index_t BetaSrcVectorSize,
           index_t YDstVectorDim,
-          index_t YDstVectorSize>
+          index_t YDstVectorSize,
+          index_t SaveMeanInvStdDstVectorSize>
 struct GridwiseNormalizationSplitK2nd
 {
     static_assert((XSrcVectorDim == 0 && MThreadSliceSize % XSrcVectorSize == 0) ||
@@ -45,6 +48,10 @@ struct GridwiseNormalizationSplitK2nd
                       (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
                   "Invalid thread slice sizes and/or vector sizes configuration, please check!");
 
+    static_assert(MThreadSliceSize % SaveMeanInvStdDstVectorSize == 0,
+                  "Invalid thread slice sizes and/or save mean and inverse std vector sizes "
+                  "configuration, please check!");
+
     static_assert(XSrcVectorSize == YDstVectorSize);
     static_assert(XSrcVectorSize == GammaSrcVectorSize);
     static_assert(XSrcVectorSize == BetaSrcVectorSize);
@@ -69,6 +76,10 @@ struct GridwiseNormalizationSplitK2nd
     static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
 
+    using ThreadBufferLengths_M = Sequence<MThreadSliceSize>;
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}));
+
     using ThreadBufferLengths_M_1 = Sequence<MThreadSliceSize, 1>;
     static constexpr auto thread_buffer_desc_m_1 =
         make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}, I1));
@@ -99,6 +110,8 @@ struct GridwiseNormalizationSplitK2nd
                                const XYGammaBetaGridDesc_M_K& gamma_grid_desc_m_k,
                                const XYGammaBetaGridDesc_M_K& beta_grid_desc_m_k,
                                const XYGammaBetaGridDesc_M_K& y_grid_desc_m_k,
+                               const SaveMeanInvStdGridDesc_M& save_mean_grid_desc_m,
+                               const SaveMeanInvStdGridDesc_M& save_inv_std_grid_desc_m,
                                index_t num_k_mean_var_count_iteration,
                                index_t num_k_block_tile_iteration,
                                index_t k_grid_size,
@@ -110,6 +123,8 @@ struct GridwiseNormalizationSplitK2nd
                                const GammaDataType* const __restrict__ p_gamma_global,
                                const BetaDataType* const __restrict__ p_beta_global,
                                YDataType* const __restrict__ p_y_global,
+                               SaveMeanInvStdDataType* const __restrict__ p_save_mean_global,
+                               SaveMeanInvStdDataType* const __restrict__ p_save_inv_std_global,
                                const YElementwiseOperation y_elementwise_op)
     {
         // Thread/Block id
@@ -145,6 +160,12 @@ struct GridwiseNormalizationSplitK2nd
         auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
 
+        auto save_mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_save_mean_global, save_mean_grid_desc_m.GetElementSpaceSize());
+
+        auto save_inv_std_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_save_inv_std_global, save_inv_std_grid_desc_m.GetElementSpaceSize());
+
         // VGPR
         StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
             in_mean_thread_buf;
@@ -158,6 +179,7 @@ struct GridwiseNormalizationSplitK2nd
             var_thread_buf;
         StaticBuffer<AddressSpaceEnum::Vgpr, int32_t, MThreadSliceSize, true>
             welford_count_thread_buf;
+        auto& inv_std_thread_buf = var_thread_buf;
 
         auto x_thread_buf = generate_tuple(
             [&](auto) {
@@ -283,6 +305,42 @@ struct GridwiseNormalizationSplitK2nd
                                      thread_k_cluster_id * YDstVectorSize),
                 y_elementwise_op);
 
+        auto threadwise_mean_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               SaveMeanInvStdDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               SaveMeanInvStdGridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,                 // DimAccessOrder
+                                               0,                           // SrcVectorDim
+                                               SaveMeanInvStdDstVectorSize, // ScalarPerVector
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                save_mean_grid_desc_m,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
+        auto threadwise_inv_std_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               SaveMeanInvStdDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               SaveMeanInvStdGridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,                 // DimAccessOrder
+                                               0,                           // SrcVectorDim
+                                               SaveMeanInvStdDstVectorSize, // ScalarPerVector
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                save_inv_std_grid_desc_m,
+                make_multi_index(block_m_cluster_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
         // step1: Merge mean and variance
         constexpr auto mean_var_count_thread_copy_step_I0_k =
             make_multi_index(I0, KThreadClusterSize);
@@ -332,9 +390,33 @@ struct GridwiseNormalizationSplitK2nd
 
             BlockwiseWelford::Run(
                 mean_thread_buf(I), var_thread_buf(I), welford_count_thread_buf(I));
+
+            inv_std_thread_buf(I) =
+                type_convert<ComputeDataType>(1.0f) / ck::math::sqrt(var_thread_buf(I) + epsilon);
         });
 
-        // step2: normalization
+        // step2: save mean and inverse std for backward (optional)
+        if(block_k_cluster_id == 0 && thread_k_cluster_id == 0)
+        {
+            if(p_save_mean_global != nullptr)
+            {
+                threadwise_mean_store.Run(thread_buffer_desc_m,
+                                          make_tuple(I0),
+                                          mean_thread_buf,
+                                          save_mean_grid_desc_m,
+                                          save_mean_global_val_buf);
+            }
+            if(p_save_inv_std_global != nullptr)
+            {
+                threadwise_inv_std_store.Run(thread_buffer_desc_m,
+                                             make_tuple(I0),
+                                             inv_std_thread_buf,
+                                             save_inv_std_grid_desc_m,
+                                             save_inv_std_global_val_buf);
+            }
+        }
+
+        // step3: normalization
         constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
 
         for(index_t k = 0; k < num_k_block_tile_iteration; ++k)
@@ -360,7 +442,6 @@ struct GridwiseNormalizationSplitK2nd
             });
 
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                 static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =
@@ -369,7 +450,7 @@ struct GridwiseNormalizationSplitK2nd
                         // normalize
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
                             (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
-                            divisor;
+                            inv_std_thread_buf(iM);
 
                         // gamma
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
diff --git a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
index ff9712276c..15b412fba4 100644
--- a/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
+++ b/include/ck/tensor_operation/gpu/grid/normalization/gridwise_normalization_welford_variance.hpp
@@ -16,9 +16,11 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           typename ComputeDataType,
           typename YElementwiseOperation,
           typename GridDesc_M_K,
+          typename GridDesc_M,
           index_t BlockSize,
           index_t MThreadClusterSize,
           index_t KThreadClusterSize,
@@ -32,6 +34,7 @@ template <typename XDataType,
           index_t BetaSrcVectorSize,
           index_t YDstVectorDim,
           index_t YDstVectorSize,
+          index_t SaveMeanInvStdDstVectorSize,
           bool SweepOnce>
 struct GridwiseNormalizationWelfordVariance_mk_to_mk
 {
@@ -43,6 +46,10 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                       (YDstVectorDim == 1 && KThreadSliceSize % YDstVectorSize == 0),
                   "Invalid thread slice sizes and/or vector sizes configuration, please check!");
 
+    static_assert(MThreadSliceSize % SaveMeanInvStdDstVectorSize == 0,
+                  "Invalid thread slice sizes and/or save mean and inverse std vector sizes "
+                  "configuration, please check!");
+
     static_assert(XSrcVectorSize == YDstVectorSize);
     static_assert(XSrcVectorSize == GammaSrcVectorSize);
     static_assert(XSrcVectorSize == BetaSrcVectorSize);
@@ -64,6 +71,10 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
     static constexpr auto thread_buffer_desc_m_k = make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{}));
 
+    using ThreadBufferLengths_M = Sequence<MThreadSliceSize>;
+    static constexpr auto thread_buffer_desc_m =
+        make_naive_tensor_descriptor_packed(make_tuple(Number<MThreadSliceSize>{}));
+
     using ThreadReduceSrcDesc_M_K = decltype(make_naive_tensor_descriptor_packed(
         make_tuple(Number<MThreadSliceSize>{}, Number<XSrcVectorSize>{})));
     using ThreadReduceDstDesc_M =
@@ -77,6 +88,8 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                                               ThreadClusterLengths_M_K,
                                               ThreadClusterArrangeOrder>;
 
+    using PassThroughOp = tensor_operation::element_wise::PassThrough;
+
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
     static constexpr auto I2 = Number<2>{};
@@ -114,17 +127,18 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                                const GridDesc_M_K& gamma_grid_desc_m_k,
                                const GridDesc_M_K& beta_grid_desc_m_k,
                                const GridDesc_M_K& y_grid_desc_m_k,
+                               const GridDesc_M& save_mean_grid_desc_m,
+                               const GridDesc_M& save_inv_std_grid_desc_m,
                                index_t num_k_block_tile_iteration,
                                ComputeDataType epsilon,
                                const XDataType* const __restrict__ p_x_global,
                                const GammaDataType* const __restrict__ p_gamma_global,
                                const BetaDataType* const __restrict__ p_beta_global,
                                YDataType* const __restrict__ p_y_global,
+                               SaveMeanInvStdDataType* const __restrict__ p_save_mean_global,
+                               SaveMeanInvStdDataType* const __restrict__ p_save_inv_std_global,
                                const YElementwiseOperation y_elementwise_op)
     {
-        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
-
         auto x_thread_buf = generate_tuple(
             [&](auto) {
                 return StaticBuffer<AddressSpaceEnum::Vgpr,
@@ -150,6 +164,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
             mean_thread_buf;
         StaticBuffer<AddressSpaceEnum::Vgpr, ComputeDataType, MThreadSliceSize, true>
             var_thread_buf;
+        auto& inv_std_thread_buf = var_thread_buf;
 
         const index_t thread_local_id = get_thread_local_1d_id();
         const index_t block_global_id = get_block_1d_id();
@@ -226,6 +241,42 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                                  thread_k_cluster_id * YDstVectorSize),
                 y_elementwise_op);
 
+        auto threadwise_mean_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               SaveMeanInvStdDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               GridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,                 // DimAccessOrder
+                                               0,                           // SrcVectorDim
+                                               SaveMeanInvStdDstVectorSize, // ScalarPerVector
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                save_mean_grid_desc_m,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
+        auto threadwise_inv_std_store =
+            ThreadwiseTensorSliceTransfer_v1r3<ComputeDataType,
+                                               SaveMeanInvStdDataType,
+                                               decltype(thread_buffer_desc_m),
+                                               GridDesc_M,
+                                               PassThroughOp,
+                                               ThreadBufferLengths_M,
+                                               Sequence<0>,                 // DimAccessOrder
+                                               0,                           // SrcVectorDim
+                                               SaveMeanInvStdDstVectorSize, // ScalarPerVector
+                                               InMemoryDataOperationEnum::Set,
+                                               1,
+                                               true>(
+                save_inv_std_grid_desc_m,
+                make_multi_index(block_global_id * M_BlockTileSize +
+                                 thread_m_cluster_id * MThreadSliceSize),
+                PassThroughOp{});
+
         constexpr auto thread_copy_fwd_step_m_k = make_multi_index(0, K_BlockTileStepSize);
         constexpr auto thread_copy_bwd_step_m_k =
             make_multi_index(0, SweepOnce ? 0 : -K_BlockTileSize);
@@ -239,6 +290,15 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
         const auto beta_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
             p_beta_global, beta_grid_desc_m_k.GetElementSpaceSize());
 
+        auto y_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_y_global, y_grid_desc_m_k.GetElementSpaceSize());
+
+        auto save_mean_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_save_mean_global, save_mean_grid_desc_m.GetElementSpaceSize());
+
+        auto save_inv_std_global_val_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_save_inv_std_global, save_inv_std_grid_desc_m.GetElementSpaceSize());
+
         auto threadwise_welford       = ThreadwiseWelford();
         threadwise_welford.max_count_ = GetKPerThread(x_grid_desc_m_k, thread_k_cluster_id);
 
@@ -279,10 +339,33 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
 
                 int count = threadwise_welford.cur_count_;
                 BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
+                inv_std_thread_buf(I) = type_convert<ComputeDataType>(1.0f) /
+                                        ck::math::sqrt(var_thread_buf(I) + epsilon);
             });
 
+            // save mean and inverse std for backward (optional)
+            if(thread_k_cluster_id == 0)
+            {
+                if(p_save_mean_global != nullptr)
+                {
+                    threadwise_mean_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              mean_thread_buf,
+                                              save_mean_grid_desc_m,
+                                              save_mean_global_val_buf);
+                }
+                if(p_save_inv_std_global != nullptr)
+                {
+                    threadwise_inv_std_store.Run(thread_buffer_desc_m,
+                                                 make_tuple(I0),
+                                                 inv_std_thread_buf,
+                                                 save_inv_std_grid_desc_m,
+                                                 save_inv_std_global_val_buf);
+                }
+            }
+
+            // normalization
             static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                 static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                     static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                         constexpr auto offset_m_k =
@@ -291,7 +374,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                         // normalize
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
                             (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
-                            divisor;
+                            inv_std_thread_buf(iM);
 
                         // gamma & beta
                         y_thread_buf(iK0)(Number<offset_m_k>{}) =
@@ -360,8 +443,29 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
 
                 int count = threadwise_welford.cur_count_;
                 BlockwiseWelford::Run(mean_thread_buf(I), var_thread_buf(I), count);
+                inv_std_thread_buf(I) = 1 / ck::math::sqrt(var_thread_buf(I) + epsilon);
             });
 
+            if(thread_k_cluster_id == 0)
+            {
+                if(p_save_mean_global != nullptr)
+                {
+                    threadwise_mean_store.Run(thread_buffer_desc_m,
+                                              make_tuple(I0),
+                                              mean_thread_buf,
+                                              save_mean_grid_desc_m,
+                                              save_mean_global_val_buf);
+                }
+                if(p_save_inv_std_global != nullptr)
+                {
+                    threadwise_inv_std_store.Run(thread_buffer_desc_m,
+                                                 make_tuple(I0),
+                                                 inv_std_thread_buf,
+                                                 save_inv_std_grid_desc_m,
+                                                 save_inv_std_global_val_buf);
+                }
+            }
+
             auto thread_copy_tail_m_k =
                 (num_k_block_tile_iteration - 1) * ThreadBufferNumber * thread_copy_fwd_step_m_k;
 
@@ -393,7 +497,6 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                 });
 
                 static_for<0, MThreadSliceSize, 1>{}([&](auto iM) {
-                    auto divisor = 1 / ck::math::sqrt(var_thread_buf(iM) + epsilon);
                     static_for<0, ThreadBufferNumber, 1>{}([&](auto iK0) {
                         static_for<0, XSrcVectorSize, 1>{}([&](auto iK1) {
                             constexpr auto offset_m_k =
@@ -402,7 +505,7 @@ struct GridwiseNormalizationWelfordVariance_mk_to_mk
                             // normalize
                             y_thread_buf(iK0)(Number<offset_m_k>{}) =
                                 (x_thread_buf(iK0)(Number<offset_m_k>{}) - mean_thread_buf(iM)) *
-                                divisor;
+                                inv_std_thread_buf(iM);
 
                             // gamma
                             y_thread_buf(iK0)(Number<offset_m_k>{}) =
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
index 6a48528c54..62920fc03e 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_groupnorm.hpp
@@ -20,8 +20,9 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
-          typename AccDataType,
-          typename AccElementwiseOperation>
+          typename SaveMeanInvStdDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation>
 struct ReferenceGroupnorm : public device::BaseOperator
 {
     // x = [N, H, W, G, C]
@@ -35,14 +36,18 @@ struct ReferenceGroupnorm : public device::BaseOperator
                  const Tensor<GammaDataType>& gamma,
                  const Tensor<BetaDataType>& beta,
                  Tensor<YDataType>& y,
-                 AccElementwiseOperation acc_elementwise_op,
+                 Tensor<SaveMeanInvStdDataType>& save_mean,
+                 Tensor<SaveMeanInvStdDataType>& save_inv_std,
+                 YElementwiseOperation y_elementwise_op,
                  const std::vector<index_t> lengths,
-                 AccDataType epsilon)
+                 ComputeDataType epsilon)
             : x_(x),
               gamma_(gamma),
               beta_(beta),
               y_(y),
-              acc_elementwise_op_(acc_elementwise_op),
+              save_mean_(save_mean),
+              save_inv_std_(save_inv_std),
+              y_elementwise_op_(y_elementwise_op),
               lengths_(lengths),
               epsilon_(epsilon)
         {
@@ -52,9 +57,11 @@ struct ReferenceGroupnorm : public device::BaseOperator
         const Tensor<XDataType> gamma_;
         const Tensor<XDataType> beta_;
         Tensor<YDataType>& y_;
-        AccElementwiseOperation acc_elementwise_op_;
+        Tensor<SaveMeanInvStdDataType>& save_mean_;
+        Tensor<SaveMeanInvStdDataType>& save_inv_std_;
+        YElementwiseOperation y_elementwise_op_;
         std::vector<index_t> lengths_;
-        AccDataType epsilon_;
+        ComputeDataType epsilon_;
     };
 
     // Invoker
@@ -68,8 +75,8 @@ struct ReferenceGroupnorm : public device::BaseOperator
             int G = arg.lengths_[3];
             int C = arg.lengths_[4];
 
-            Tensor<AccDataType> mean({N, G});
-            Tensor<AccDataType> var({N, G});
+            Tensor<ComputeDataType> mean({N, G});
+            Tensor<ComputeDataType> var({N, G});
 
             // Compute mean & var in [H, W, C] by Welford Algorithm
             // TODO - parallel for each HWC
@@ -78,9 +85,9 @@ struct ReferenceGroupnorm : public device::BaseOperator
             {
                 for(int g = 0; g < G; ++g)
                 {
-                    AccDataType mean_val = type_convert<AccDataType>(0.0f);
-                    AccDataType var_val  = type_convert<AccDataType>(0.0f);
-                    int32_t curr_count   = 0;
+                    ComputeDataType mean_val = type_convert<ComputeDataType>(0.0f);
+                    ComputeDataType var_val  = type_convert<ComputeDataType>(0.0f);
+                    int32_t curr_count       = 0;
 
                     for(int h = 0; h < H; ++h)
                     {
@@ -89,10 +96,11 @@ struct ReferenceGroupnorm : public device::BaseOperator
                             for(int c = 0; c < C; ++c)
                             {
                                 curr_count++;
-                                AccDataType x = type_convert<AccDataType>(arg.x_(n, h, w, g, c));
-                                AccDataType delta = x - mean_val;
+                                ComputeDataType x =
+                                    type_convert<ComputeDataType>(arg.x_(n, h, w, g, c));
+                                ComputeDataType delta = x - mean_val;
                                 mean_val += delta / curr_count;
-                                AccDataType delta2 = x - mean_val;
+                                ComputeDataType delta2 = x - mean_val;
                                 var_val += delta * delta2;
                             }
                         }
@@ -100,6 +108,12 @@ struct ReferenceGroupnorm : public device::BaseOperator
 
                     mean(n, g) = mean_val;
                     var(n, g)  = var_val / curr_count;
+
+                    arg.save_mean_(n, g) = ck::type_convert<SaveMeanInvStdDataType>(mean(n, g));
+
+                    ComputeDataType divisor =
+                        static_cast<ComputeDataType>(1) / ck::math::sqrt(var(n, g) + arg.epsilon_);
+                    arg.save_inv_std_(n, g) = ck::type_convert<SaveMeanInvStdDataType>(divisor);
                 }
             }
 
@@ -114,15 +128,19 @@ struct ReferenceGroupnorm : public device::BaseOperator
                         {
                             for(int c = 0; c < C; ++c)
                             {
-                                AccDataType x = type_convert<AccDataType>(arg.x_(n, h, w, g, c));
-                                AccDataType gamma    = type_convert<AccDataType>(arg.gamma_(g, c));
-                                AccDataType beta     = type_convert<AccDataType>(arg.beta_(g, c));
-                                AccDataType mean_val = type_convert<AccDataType>(mean(n, g));
-                                AccDataType var_val  = type_convert<AccDataType>(var(n, g));
-                                AccDataType y        = gamma * (x - mean_val) /
-                                                    ck::math::sqrt(arg.epsilon_ + var_val) +
-                                                beta;
-                                arg.acc_elementwise_op_(y, y);
+                                ComputeDataType x =
+                                    type_convert<ComputeDataType>(arg.x_(n, h, w, g, c));
+                                ComputeDataType gamma =
+                                    type_convert<ComputeDataType>(arg.gamma_(g, c));
+                                ComputeDataType beta =
+                                    type_convert<ComputeDataType>(arg.beta_(g, c));
+                                ComputeDataType mean_val =
+                                    type_convert<ComputeDataType>(mean(n, g));
+                                ComputeDataType var_val = type_convert<ComputeDataType>(var(n, g));
+                                ComputeDataType y       = gamma * (x - mean_val) /
+                                                        ck::math::sqrt(arg.epsilon_ + var_val) +
+                                                    beta;
+                                arg.y_elementwise_op_(y, y);
                                 arg.y_(n, h, w, g, c) = type_convert<YDataType>(y);
                             }
                         }
@@ -159,11 +177,14 @@ struct ReferenceGroupnorm : public device::BaseOperator
                              const Tensor<GammaDataType>& gamma,
                              const Tensor<BetaDataType>& beta,
                              Tensor<YDataType>& y,
-                             AccElementwiseOperation acc_elementwise_op,
+                             Tensor<SaveMeanInvStdDataType>& save_mean,
+                             Tensor<SaveMeanInvStdDataType>& save_inv_std,
+                             YElementwiseOperation y_elementwise_op,
                              const std::vector<index_t> lengths,
-                             AccDataType epsilon)
+                             ComputeDataType epsilon)
     {
-        return Argument{x, gamma, beta, y, acc_elementwise_op, lengths, epsilon};
+        return Argument{
+            x, gamma, beta, y, save_mean, save_inv_std, y_elementwise_op, lengths, epsilon};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
index 9994a2f9f7..444ae970c1 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_layernorm.hpp
@@ -20,8 +20,9 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
-          typename AccDataType,
-          typename AccElementwiseOperation,
+          typename SaveMeanInvStdDataType,
+          typename ComputeDataType,
+          typename YElementwiseOperation,
           index_t Rank,
           index_t NumReduceDim>
 struct ReferenceLayernorm : public device::BaseOperator
@@ -36,15 +37,19 @@ struct ReferenceLayernorm : public device::BaseOperator
                  const Tensor<GammaDataType>& gamma_n,
                  const Tensor<BetaDataType>& beta_n,
                  Tensor<YDataType>& y_m_n,
-                 AccElementwiseOperation acc_elementwise_op,
+                 Tensor<SaveMeanInvStdDataType>& save_mean_m,
+                 Tensor<SaveMeanInvStdDataType>& save_inv_std_m,
+                 YElementwiseOperation y_elementwise_op,
                  const std::vector<index_t> lengths,
                  const std::vector<index_t> reduceDims,
-                 AccDataType epsilon)
+                 ComputeDataType epsilon)
             : x_m_n_(x_m_n),
               gamma_n_(gamma_n),
               beta_n_(beta_n),
               y_m_n_(y_m_n),
-              acc_elementwise_op_(acc_elementwise_op),
+              save_mean_m_(save_mean_m),
+              save_inv_std_m_(save_inv_std_m),
+              y_elementwise_op_(y_elementwise_op),
               lengths_(lengths),
               reduceDims_(reduceDims),
               epsilon_(epsilon)
@@ -55,10 +60,12 @@ struct ReferenceLayernorm : public device::BaseOperator
         const Tensor<XDataType> gamma_n_;
         const Tensor<XDataType> beta_n_;
         Tensor<YDataType>& y_m_n_;
-        AccElementwiseOperation acc_elementwise_op_;
+        Tensor<SaveMeanInvStdDataType>& save_mean_m_;
+        Tensor<SaveMeanInvStdDataType>& save_inv_std_m_;
+        YElementwiseOperation y_elementwise_op_;
         std::vector<index_t> lengths_;
         std::vector<index_t> reduceDims_;
-        AccDataType epsilon_;
+        ComputeDataType epsilon_;
     };
 
     // Invoker
@@ -69,8 +76,8 @@ struct ReferenceLayernorm : public device::BaseOperator
             int M = arg.lengths_[0];
             int N = arg.lengths_[1];
 
-            Tensor<AccDataType> mean({M});
-            Tensor<AccDataType> var({M});
+            Tensor<ComputeDataType> mean({M});
+            Tensor<ComputeDataType> var({M});
 
             for(int m = 0; m < M; ++m)
             {
@@ -79,7 +86,7 @@ struct ReferenceLayernorm : public device::BaseOperator
 
                 for(int n = 0; n < N; ++n)
                 {
-                    auto x_val = ck::type_convert<AccDataType>(arg.x_m_n_(m, n));
+                    auto x_val = ck::type_convert<ComputeDataType>(arg.x_m_n_(m, n));
                     mean(m) += x_val;
                     var(m) += x_val * x_val;
                 }
@@ -90,17 +97,21 @@ struct ReferenceLayernorm : public device::BaseOperator
 
             for(int m = 0; m < M; ++m)
             {
-                AccDataType divisor =
-                    static_cast<AccDataType>(1) / ck::math::sqrt(var(m) + arg.epsilon_);
+                ComputeDataType divisor =
+                    static_cast<ComputeDataType>(1) / ck::math::sqrt(var(m) + arg.epsilon_);
 
                 for(int n = 0; n < N; ++n)
                 {
-                    auto x_val = ck::type_convert<AccDataType>(arg.x_m_n_(m, n));
-                    auto y_val = (x_val - mean(m)) * divisor;
-                    y_val      = (y_val * arg.gamma_n_(n)) + arg.beta_n_(n);
-                    arg.acc_elementwise_op_(y_val, y_val);
+                    auto x_val     = ck::type_convert<ComputeDataType>(arg.x_m_n_(m, n));
+                    auto gamma_val = ck::type_convert<ComputeDataType>(arg.gamma_n_(n));
+                    auto beta_val  = ck::type_convert<ComputeDataType>(arg.beta_n_(n));
+                    auto y_val     = (x_val - mean(m)) * divisor;
+                    y_val          = (y_val * gamma_val) + beta_val;
+                    arg.y_elementwise_op_(y_val, y_val);
                     arg.y_m_n_(m, n) = ck::type_convert<YDataType>(y_val);
                 }
+                arg.save_mean_m_(m)    = ck::type_convert<SaveMeanInvStdDataType>(mean(m));
+                arg.save_inv_std_m_(m) = ck::type_convert<SaveMeanInvStdDataType>(divisor);
             }
 
             return 0;
@@ -140,13 +151,23 @@ struct ReferenceLayernorm : public device::BaseOperator
                              const Tensor<GammaDataType>& gamma_n,
                              const Tensor<BetaDataType>& beta_n,
                              Tensor<YDataType>& y_m_n,
-                             AccElementwiseOperation acc_elementwise_op,
+                             Tensor<SaveMeanInvStdDataType>& save_mean_m,
+                             Tensor<SaveMeanInvStdDataType>& save_inv_std_m,
+                             YElementwiseOperation y_elementwise_op,
                              const std::vector<index_t> lengths,
                              const std::vector<index_t> reduceDims,
-                             AccDataType epsilon)
+                             ComputeDataType epsilon)
     {
-        return Argument{
-            x_m_n, gamma_n, beta_n, y_m_n, acc_elementwise_op, lengths, reduceDims, epsilon};
+        return Argument{x_m_n,
+                        gamma_n,
+                        beta_n,
+                        y_m_n,
+                        save_mean_m,
+                        save_inv_std_m,
+                        y_elementwise_op,
+                        lengths,
+                        reduceDims,
+                        epsilon};
     }
 
     static auto MakeInvoker() { return Invoker{}; }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
index 8e90a7ea98..229de41b5e 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization.hpp
@@ -19,13 +19,13 @@ namespace instance {
 #ifdef CK_ENABLE_FP16
 // FP16
 void add_device_normalization_rank_2_1_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 2, 1>>>&);
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, PassThrough, 2, 1>>>&);
 
 void add_device_normalization_rank_4_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 4, 3>>>&);
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, PassThrough, 4, 3>>>&);
 
 void add_device_normalization_rank_5_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, PassThrough, 5, 3>>>&);
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, PassThrough, 5, 3>>>&);
 #endif
 #ifdef CK_ENABLE_FP32
 // FP32
@@ -42,14 +42,15 @@ template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           index_t Rank,
           index_t NumReduceDim>
 struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormalization<
     XDataType,
     GammaDataType,
     BetaDataType,
-    F32,
     YDataType,
+    SaveMeanInvStdDataType,
     ck::tensor_operation::element_wise::PassThrough,
     Rank,
     NumReduceDim>>
@@ -57,8 +58,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
     using DeviceOp = DeviceNormalization<XDataType,
                                          GammaDataType,
                                          BetaDataType,
-                                         F32,
                                          YDataType,
+                                         SaveMeanInvStdDataType,
                                          ck::tensor_operation::element_wise::PassThrough,
                                          Rank,
                                          NumReduceDim>;
@@ -68,7 +69,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 #ifdef CK_ENABLE_FP16
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
-                     is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
+                     is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16> &&
+                     is_same_v<SaveMeanInvStdDataType, F32>)
         {
             if constexpr(Rank == 2 && NumReduceDim == 1)
             {
@@ -86,7 +88,8 @@ struct DeviceOperationInstanceFactory<ck::tensor_operation::device::DeviceNormal
 #endif
 #ifdef CK_ENABLE_FP32
         if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
-                     is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
+                     is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32> &&
+                     is_same_v<SaveMeanInvStdDataType, F32>)
         {
             if constexpr(Rank == 2 && NumReduceDim == 1)
             {
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp b/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
index 2391775299..ae9c09cacb 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp
@@ -19,7 +19,7 @@ namespace instance {
 
 // FP16
 void add_device_normalization_rank_5_3_swish_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Swish, 5, 3>>>&);
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Swish, 5, 3>>>&);
 
 // FP32
 void add_device_normalization_rank_5_3_swish_f32_instances(
@@ -27,20 +27,21 @@ void add_device_normalization_rank_5_3_swish_f32_instances(
 
 // [x, gamma, beta, y] = [f16, f32, f32, f16]
 void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&);
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F16, F32, Swish, 5, 3>>>&);
 
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
           index_t Rank,
           index_t NumReduceDim>
 struct DeviceOperationInstanceFactory<
     ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                       GammaDataType,
                                                       BetaDataType,
-                                                      F32,
                                                       YDataType,
+                                                      SaveMeanInvStdDataType,
                                                       ck::tensor_operation::element_wise::Swish,
                                                       Rank,
                                                       NumReduceDim>>
@@ -48,8 +49,8 @@ struct DeviceOperationInstanceFactory<
     using DeviceOp = DeviceNormalization<XDataType,
                                          GammaDataType,
                                          BetaDataType,
-                                         F32,
                                          YDataType,
+                                         SaveMeanInvStdDataType,
                                          ck::tensor_operation::element_wise::Swish,
                                          Rank,
                                          NumReduceDim>;
@@ -59,7 +60,8 @@ struct DeviceOperationInstanceFactory<
         std::vector<std::unique_ptr<DeviceOp>> op_ptrs;
 
         if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F16> &&
-                     is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16>)
+                     is_same_v<BetaDataType, F16> && is_same_v<YDataType, F16> &&
+                     is_same_v<SaveMeanInvStdDataType, F32>)
         {
             if constexpr(Rank == 5 && NumReduceDim == 3)
             {
@@ -67,7 +69,8 @@ struct DeviceOperationInstanceFactory<
             }
         }
         else if constexpr(is_same_v<XDataType, F32> && is_same_v<GammaDataType, F32> &&
-                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32>)
+                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F32> &&
+                          is_same_v<SaveMeanInvStdDataType, F32>)
         {
             if constexpr(Rank == 5 && NumReduceDim == 3)
             {
@@ -75,7 +78,8 @@ struct DeviceOperationInstanceFactory<
             }
         }
         else if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F32> &&
-                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F16>)
+                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F16> &&
+                          is_same_v<SaveMeanInvStdDataType, F32>)
         {
             if constexpr(Rank == 5 && NumReduceDim == 3)
             {
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
index 762da1c6ae..439e724199 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_f16_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
 void add_device_normalization_rank_5_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 5, 3>>>&
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Pass, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
index aa662b7dfe..5f42d073ff 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 using Swish = ck::tensor_operation::element_wise::Swish;
 
 void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F16, F32, Swish, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
index bc5cd801ae..63aea024da 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 using Swish = ck::tensor_operation::element_wise::Swish;
 
 void add_device_normalization_rank_5_3_swish_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Swish, 5, 3>>>&
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Swish, 5, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
index 0d235f1fa7..e15ff4b6d0 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm2d_f16_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
 void add_device_normalization_rank_2_1_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 2, 1>>>&
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Pass, 2, 1>>>&
         instances)
 {
     add_device_operation_instances(instances,
diff --git a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
index 6bc3950062..4152c6ebbf 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/device_layernorm4d_f16_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 using Pass = ck::tensor_operation::element_wise::PassThrough;
 
 void add_device_normalization_rank_4_3_f16_instances(
-    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F32, F16, Pass, 4, 3>>>&
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F16, F16, F16, F32, Pass, 4, 3>>>&
         instances)
 {
     add_device_operation_instances(instances,
diff --git a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
index 7aa3da8eed..488f34b4b3 100644
--- a/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
+++ b/library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp
@@ -22,25 +22,25 @@ template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_instances =
     // clang-format off
     std::tuple <
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,   // irregular size
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8>
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8, 2>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>
         // clang-format on
         >;
 
@@ -48,150 +48,150 @@ template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_splitk_f16_instances =
     // clang-format off
     std::tuple <
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8>,
-        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8>
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 8, 1, 8, 1, 8, 8, 2>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 8, 1, 8, 1, 8, 8, 1>,
+        DeviceNormalizationSplitKImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 16, 1, 8, 1, 8, 1, 8, 8, 1>
         // clang-format on
         >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_generic_instance = std::tuple<
     // clang-format off
-        DeviceNormalizationImpl<F16, F16, F16, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+        DeviceNormalizationImpl<F16, F16, F16, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
     // clang-format on
     >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f32_instances = std::tuple<
     // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
     // clang-format on
     >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_splitk_f32_instances = std::tuple<
     // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
     // clang-format on
     >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f32_generic_instance = std::tuple<
     // clang-format off
-        DeviceNormalizationImpl<F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+        DeviceNormalizationImpl<F32, F32, F32, F32, F32, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
     // clang-format on
     >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_f32_f32_f16_instances = std::tuple<
     // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
     // clang-format on
     >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_splitk_f16_f32_f32_f16_instances = std::tuple<
     // clang-format off
-        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
-        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, SaveMeanInvStdDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorDim, GammaSrcVectorSize, BetaSrcVectorDim, BetaSrcVectorSize, YDstVectorSize, SaveMeanInvStdScalarPerVector>
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2, 1>,   // irregular size
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4, 2>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4, 1>,
+        DeviceNormalizationSplitKImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4, 1>
     // clang-format on
     >;
 
 template <typename OutElementwise, index_t Rank, index_t Reduce>
 using device_normalization_f16_f32_f32_f16_generic_instance = std::tuple<
     // clang-format off
-        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1>
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, F32, OutElementwise, Rank, Reduce, 64, 1, 64, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1>
     // clang-format on
     >;
 
diff --git a/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp b/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
index 1fd9c81109..ae42919db6 100644
--- a/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_elementwise_layernorm_impl.hpp
@@ -80,6 +80,8 @@ bool profile_elementwise_layernorm_impl(int do_verification,
     Tensor<BetaDataType> beta(gammaBetaLength);
     Tensor<YDataType> y(length);
     Tensor<YDataType> host_y(length);
+    Tensor<AccDataType> host_save_mean({M});
+    Tensor<AccDataType> host_save_inv_std({M});
 
     switch(init_method)
     {
@@ -152,14 +154,23 @@ bool profile_elementwise_layernorm_impl(int do_verification,
                                                                                  BetaDataType,
                                                                                  YDataType,
                                                                                  AccDataType,
+                                                                                 AccDataType,
                                                                                  PassThrough,
                                                                                  Rank,
                                                                                  NumReduceDim>;
 
         ReferenceInstance ref;
-        auto ref_argument =
-            ref.MakeArgument(x, gamma, beta, host_y, PassThrough{}, {M, N}, {1}, 1e-4);
-        auto ref_invoker = ref.MakeInvoker();
+        auto ref_argument = ref.MakeArgument(x,
+                                             gamma,
+                                             beta,
+                                             host_y,
+                                             host_save_mean,
+                                             host_save_inv_std,
+                                             PassThrough{},
+                                             {M, N},
+                                             {1},
+                                             1e-4);
+        auto ref_invoker  = ref.MakeInvoker();
         ref_invoker.Run(ref_argument);
     }
 
diff --git a/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp b/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
index 4c3d0a0450..46591a3525 100644
--- a/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_add_relu_add_layernorm_impl.hpp
@@ -66,12 +66,15 @@ void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
                                                                               BetaDataType,
                                                                               HDataType,
                                                                               AccDataType,
+                                                                              AccDataType,
                                                                               HElementOp,
                                                                               2,
                                                                               1>;
 
     Tensor<EMeanVarDataType> e_m_n(HostTensorDescriptor{M, N});
     Tensor<AccDataType> c_m_n(HostTensorDescriptor{M, N});
+    Tensor<AccDataType> save_mean({M});
+    Tensor<AccDataType> save_inv_std({M});
 
     auto ref_gemm         = ReferenceGemm{};
     auto ref_gemm_invoker = ref_gemm.MakeInvoker();
@@ -97,7 +100,7 @@ void host_gemm_layernorm(Tensor<HDataType>& h_m_n,
     auto ref_layernorm_invoker = ref_layernorm.MakeInvoker();
 
     auto ref_layernorm_argument = ref_layernorm.MakeArgument(
-        e_m_n, gamma_n, beta_n, h_m_n, h_element_op, {M, N}, {1}, epsilon);
+        e_m_n, gamma_n, beta_n, h_m_n, save_mean, save_inv_std, h_element_op, {M, N}, {1}, epsilon);
     ref_layernorm_invoker.Run(ref_layernorm_argument);
 }
 
diff --git a/profiler/include/profiler/profile_groupnorm_impl.hpp b/profiler/include/profiler/profile_groupnorm_impl.hpp
index f88ba8453c..4715853d2a 100644
--- a/profiler/include/profiler/profile_groupnorm_impl.hpp
+++ b/profiler/include/profiler/profile_groupnorm_impl.hpp
@@ -21,8 +21,10 @@ namespace profiler {
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
-          typename AccDataType,
-          typename YDataType>
+          typename ComputeDataType,
+          typename YDataType,
+          typename SaveMeanInvStdDataType,
+          bool SaveMeanInvStd>
 bool profile_groupnorm_impl(int do_verification,
                             int init_method,
                             bool do_log,
@@ -34,6 +36,7 @@ bool profile_groupnorm_impl(int do_verification,
     if(length.size() != 5)
         return false;
 
+    index_t N = length[0];
     index_t G = length[3];
     index_t C = length[4];
 
@@ -45,7 +48,14 @@ bool profile_groupnorm_impl(int do_verification,
     Tensor<GammaDataType> gamma(gammaBetaLength);
     Tensor<BetaDataType> beta(gammaBetaLength);
     Tensor<YDataType> y(length);
+    Tensor<SaveMeanInvStdDataType> save_mean({N, G});
+    Tensor<SaveMeanInvStdDataType> save_inv_std({N, G});
+
     Tensor<YDataType> host_y(length);
+    Tensor<SaveMeanInvStdDataType> host_save_mean({N, G});
+    Tensor<SaveMeanInvStdDataType> host_save_inv_std({N, G});
+
+    std::vector<index_t> strideSaveMeanInvStd = {1};
 
     switch(init_method)
     {
@@ -69,6 +79,9 @@ bool profile_groupnorm_impl(int do_verification,
     DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
     DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
     DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem save_mean_dev(sizeof(SaveMeanInvStdDataType) * save_mean.mDesc.GetElementSpaceSize());
+    DeviceMem save_inv_std_dev(sizeof(SaveMeanInvStdDataType) *
+                               save_inv_std.mDesc.GetElementSpaceSize());
 
     x_dev.ToDevice(x.mData.data());
     gamma_dev.ToDevice(gamma.mData.data());
@@ -78,8 +91,8 @@ bool profile_groupnorm_impl(int do_verification,
     using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                                        GammaDataType,
                                                                        BetaDataType,
-                                                                       AccDataType,
                                                                        YDataType,
+                                                                       SaveMeanInvStdDataType,
                                                                        PassThrough,
                                                                        5,
                                                                        3>;
@@ -97,38 +110,70 @@ bool profile_groupnorm_impl(int do_verification,
 
     if(do_verification)
     {
-        using ReferenceInstance = ck::tensor_operation::host::ReferenceGroupnorm<XDataType,
-                                                                                 GammaDataType,
-                                                                                 BetaDataType,
-                                                                                 YDataType,
-                                                                                 AccDataType,
-                                                                                 PassThrough>;
+        using ReferenceInstance =
+            ck::tensor_operation::host::ReferenceGroupnorm<XDataType,
+                                                           GammaDataType,
+                                                           BetaDataType,
+                                                           YDataType,
+                                                           SaveMeanInvStdDataType,
+                                                           ComputeDataType,
+                                                           PassThrough>;
 
         ReferenceInstance ref;
-        auto ref_argument = ref.MakeArgument(x, gamma, beta, host_y, PassThrough{}, length, 1e-6);
-        auto ref_invoker  = ref.MakeInvoker();
+        auto ref_argument = ref.MakeArgument(
+            x, gamma, beta, host_y, host_save_mean, host_save_inv_std, PassThrough{}, length, 1e-6);
+        auto ref_invoker = ref.MakeInvoker();
         ref_invoker.Run(ref_argument);
     }
 
     int num_kernel = 0;
 
+    auto f_get_argument = [&](auto& inst_ptr) {
+        if constexpr(SaveMeanInvStd)
+            return inst_ptr->MakeArgumentPointer(
+                length,
+                std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()},
+                gammaBetaStride,
+                gammaBetaStride,
+                std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
+                std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
+                                         save_mean.mDesc.GetStrides().end()},
+                std::vector<ck::index_t>{save_inv_std.mDesc.GetStrides().begin(),
+                                         save_inv_std.mDesc.GetStrides().end()},
+                reduce_dim,
+                1e-6,
+                x_dev.GetDeviceBuffer(),
+                gamma_dev.GetDeviceBuffer(),
+                beta_dev.GetDeviceBuffer(),
+                y_dev.GetDeviceBuffer(),
+                save_mean_dev.GetDeviceBuffer(),
+                save_inv_std_dev.GetDeviceBuffer(),
+                PassThrough{});
+        else
+            return inst_ptr->MakeArgumentPointer(
+                length,
+                std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()},
+                gammaBetaStride,
+                gammaBetaStride,
+                std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
+                std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
+                                         save_mean.mDesc.GetStrides().end()},
+                std::vector<ck::index_t>{save_inv_std.mDesc.GetStrides().begin(),
+                                         save_inv_std.mDesc.GetStrides().end()},
+                reduce_dim,
+                1e-6,
+                x_dev.GetDeviceBuffer(),
+                gamma_dev.GetDeviceBuffer(),
+                beta_dev.GetDeviceBuffer(),
+                y_dev.GetDeviceBuffer(),
+                nullptr,
+                nullptr,
+                PassThrough{});
+    };
+
     for(auto& inst_ptr : instance_ptrs)
     {
-        auto argument_ptr = inst_ptr->MakeArgumentPointer(
-            length,
-            std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()},
-            gammaBetaStride,
-            gammaBetaStride,
-            std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
-            reduce_dim,
-            1e-6,
-            x_dev.GetDeviceBuffer(),
-            gamma_dev.GetDeviceBuffer(),
-            beta_dev.GetDeviceBuffer(),
-            y_dev.GetDeviceBuffer(),
-            nullptr,
-            nullptr,
-            PassThrough{});
+        auto argument_ptr = f_get_argument(inst_ptr);
 
         if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
         {
@@ -152,6 +197,10 @@ bool profile_groupnorm_impl(int do_verification,
                                 beta.mDesc.GetElementSize() * sizeof(BetaDataType) +
                                 y.mDesc.GetElementSize() * sizeof(YDataType);
 
+        if constexpr(SaveMeanInvStd)
+            num_bytes += save_mean.mDesc.GetElementSpaceSize() * sizeof(SaveMeanInvStdDataType) +
+                         save_inv_std.mDesc.GetElementSpaceSize() * sizeof(SaveMeanInvStdDataType);
+
         float gb_per_sec = num_bytes / 1.E6 / avg_time;
 
         if(time_kernel)
@@ -168,9 +217,22 @@ bool profile_groupnorm_impl(int do_verification,
         if(do_verification)
         {
             y_dev.FromDevice(y.mData.data());
-
             bool pass = ck::utils::check_err(y, host_y, "Error: Incorrect results", 1e-3, 1e-3);
 
+            if constexpr(SaveMeanInvStd)
+            {
+                save_mean_dev.FromDevice(save_mean.mData.data());
+                pass &= ck::utils::check_err(
+                    save_mean.mData, host_save_mean.mData, "Error: Incorrect results", 1e-3, 1e-3);
+
+                save_inv_std_dev.FromDevice(save_inv_std.mData.data());
+                pass &= ck::utils::check_err(save_inv_std.mData,
+                                             host_save_inv_std.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+            }
+
             if(do_log)
             {
                 LogRangeAsType<float>(std::cout << "x  : ", x.mData, ",") << std::endl;
diff --git a/profiler/include/profiler/profile_layernorm_impl.hpp b/profiler/include/profiler/profile_layernorm_impl.hpp
index f969646c2f..7c214af019 100644
--- a/profiler/include/profiler/profile_layernorm_impl.hpp
+++ b/profiler/include/profiler/profile_layernorm_impl.hpp
@@ -21,6 +21,8 @@ template <typename XDataType,
           typename BetaDataType,
           typename ComputeDataType,
           typename YDataType,
+          typename SaveMeanInvStdDataType,
+          bool SaveMeanInvStd,
           index_t Rank>
 bool profile_layernorm_impl(int do_verification,
                             int init_method,
@@ -43,13 +45,19 @@ bool profile_layernorm_impl(int do_verification,
     Tensor<GammaDataType> gamma(reduce_length);
     Tensor<BetaDataType> beta(reduce_length);
     Tensor<YDataType> y(length);
+    Tensor<SaveMeanInvStdDataType> save_mean({length[0]});
+    Tensor<SaveMeanInvStdDataType> save_inv_std({length[0]});
     Tensor<YDataType> host_y(length);
+    Tensor<SaveMeanInvStdDataType> host_save_mean({length[0]});
+    Tensor<SaveMeanInvStdDataType> host_save_inv_std({length[0]});
 
     std::vector<index_t> strideXY =
         std::vector<ck::index_t>{x.mDesc.GetStrides().begin(), x.mDesc.GetStrides().end()};
     std::vector<index_t> strideGammaBeta = strideXY;
     strideGammaBeta[0]                   = 0;
 
+    std::vector<index_t> strideSaveMeanInvStd = {1};
+
     switch(init_method)
     {
     case 0:
@@ -75,6 +83,9 @@ bool profile_layernorm_impl(int do_verification,
     DeviceMem gamma_dev(sizeof(GammaDataType) * gamma.mDesc.GetElementSpaceSize());
     DeviceMem beta_dev(sizeof(BetaDataType) * beta.mDesc.GetElementSpaceSize());
     DeviceMem y_dev(sizeof(YDataType) * y.mDesc.GetElementSpaceSize());
+    DeviceMem save_mean_dev(sizeof(SaveMeanInvStdDataType) * save_mean.mDesc.GetElementSpaceSize());
+    DeviceMem save_inv_std_dev(sizeof(SaveMeanInvStdDataType) *
+                               save_inv_std.mDesc.GetElementSpaceSize());
 
     x_dev.ToDevice(x.mData.data());
     gamma_dev.ToDevice(gamma.mData.data());
@@ -86,8 +97,8 @@ bool profile_layernorm_impl(int do_verification,
     using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
                                                                        GammaDataType,
                                                                        BetaDataType,
-                                                                       ComputeDataType,
                                                                        YDataType,
+                                                                       SaveMeanInvStdDataType,
                                                                        PassThrough,
                                                                        Rank,
                                                                        NumReduceDim>;
@@ -105,40 +116,74 @@ bool profile_layernorm_impl(int do_verification,
 
     if(do_verification)
     {
-        using ReferenceInstance = ck::tensor_operation::host::ReferenceLayernorm<XDataType,
-                                                                                 GammaDataType,
-                                                                                 BetaDataType,
-                                                                                 YDataType,
-                                                                                 ComputeDataType,
-                                                                                 PassThrough,
-                                                                                 Rank,
-                                                                                 NumReduceDim>;
+        using ReferenceInstance =
+            ck::tensor_operation::host::ReferenceLayernorm<XDataType,
+                                                           GammaDataType,
+                                                           BetaDataType,
+                                                           YDataType,
+                                                           SaveMeanInvStdDataType,
+                                                           ComputeDataType,
+                                                           PassThrough,
+                                                           Rank,
+                                                           NumReduceDim>;
 
         ReferenceInstance ref;
-        auto ref_argument =
-            ref.MakeArgument(x, gamma, beta, host_y, PassThrough{}, length, reduce_dim, 1e-4);
-        auto ref_invoker = ref.MakeInvoker();
+        auto ref_argument = ref.MakeArgument(x,
+                                             gamma,
+                                             beta,
+                                             host_y,
+                                             host_save_mean,
+                                             host_save_inv_std,
+                                             PassThrough{},
+                                             length,
+                                             reduce_dim,
+                                             1e-4);
+        auto ref_invoker  = ref.MakeInvoker();
         ref_invoker.Run(ref_argument);
     }
 
     int num_kernel = 0;
 
+    auto f_get_argument = [&](auto& inst_ptr) {
+        if constexpr(SaveMeanInvStd)
+            return inst_ptr->MakeArgumentPointer(length,
+                                                 strideXY,
+                                                 strideGammaBeta,
+                                                 strideGammaBeta,
+                                                 strideXY,
+                                                 strideSaveMeanInvStd,
+                                                 strideSaveMeanInvStd,
+                                                 reduce_dim,
+                                                 1e-4,
+                                                 x_dev.GetDeviceBuffer(),
+                                                 gamma_dev.GetDeviceBuffer(),
+                                                 beta_dev.GetDeviceBuffer(),
+                                                 y_dev.GetDeviceBuffer(),
+                                                 save_mean_dev.GetDeviceBuffer(),
+                                                 save_inv_std_dev.GetDeviceBuffer(),
+                                                 PassThrough{});
+        else
+            return inst_ptr->MakeArgumentPointer(length,
+                                                 strideXY,
+                                                 strideGammaBeta,
+                                                 strideGammaBeta,
+                                                 strideXY,
+                                                 strideSaveMeanInvStd,
+                                                 strideSaveMeanInvStd,
+                                                 reduce_dim,
+                                                 1e-4,
+                                                 x_dev.GetDeviceBuffer(),
+                                                 gamma_dev.GetDeviceBuffer(),
+                                                 beta_dev.GetDeviceBuffer(),
+                                                 y_dev.GetDeviceBuffer(),
+                                                 nullptr,
+                                                 nullptr,
+                                                 PassThrough{});
+    };
+
     for(auto& inst_ptr : instance_ptrs)
     {
-        auto argument_ptr = inst_ptr->MakeArgumentPointer(length,
-                                                          strideXY,
-                                                          strideGammaBeta,
-                                                          strideGammaBeta,
-                                                          strideXY,
-                                                          reduce_dim,
-                                                          1e-4,
-                                                          x_dev.GetDeviceBuffer(),
-                                                          gamma_dev.GetDeviceBuffer(),
-                                                          beta_dev.GetDeviceBuffer(),
-                                                          y_dev.GetDeviceBuffer(),
-                                                          nullptr,
-                                                          nullptr,
-                                                          PassThrough{});
+        auto argument_ptr = f_get_argument(inst_ptr);
 
         if(inst_ptr->IsSupportedArgument(argument_ptr.get()))
         {
@@ -168,6 +213,10 @@ bool profile_layernorm_impl(int do_verification,
                                 beta.mDesc.GetElementSize() * sizeof(BetaDataType) +
                                 y.mDesc.GetElementSize() * sizeof(YDataType);
 
+        if constexpr(SaveMeanInvStd)
+            num_bytes += save_mean.mDesc.GetElementSpaceSize() * sizeof(SaveMeanInvStdDataType) +
+                         save_inv_std.mDesc.GetElementSpaceSize() * sizeof(SaveMeanInvStdDataType);
+
         float gb_per_sec = num_bytes / 1.E6 / avg_time;
 
         if(time_kernel)
@@ -184,10 +233,23 @@ bool profile_layernorm_impl(int do_verification,
         if(do_verification)
         {
             y_dev.FromDevice(y.mData.data());
-
             bool pass =
                 ck::utils::check_err(y.mData, host_y.mData, "Error: Incorrect results", 1e-3, 1e-3);
 
+            if constexpr(SaveMeanInvStd)
+            {
+                save_mean_dev.FromDevice(save_mean.mData.data());
+                pass &= ck::utils::check_err(
+                    save_mean.mData, host_save_mean.mData, "Error: Incorrect results", 1e-3, 1e-3);
+
+                save_inv_std_dev.FromDevice(save_inv_std.mData.data());
+                pass &= ck::utils::check_err(save_inv_std.mData,
+                                             host_save_inv_std.mData,
+                                             "Error: Incorrect results",
+                                             1e-3,
+                                             1e-3);
+            }
+
             if(do_log)
             {
                 LogRangeAsType<float>(std::cout << "x  : ", x.mData, ",") << std::endl;
diff --git a/profiler/src/profile_groupnorm.cpp b/profiler/src/profile_groupnorm.cpp
index d55784ff0a..079f6f0db7 100644
--- a/profiler/src/profile_groupnorm.cpp
+++ b/profiler/src/profile_groupnorm.cpp
@@ -93,12 +93,12 @@ int profile_groupnorm(int argc, char* argv[])
 
     if(data_type == ck::DataTypeEnum::Float)
     {
-        ck::profiler::profile_groupnorm_impl<F32, F32, F32, F32, F32>(
+        ck::profiler::profile_groupnorm_impl<F32, F32, F32, F32, F32, F32, false>(
             do_verification, init_method, do_log, time_kernel, length);
     }
     else if(data_type == ck::DataTypeEnum::Half)
     {
-        ck::profiler::profile_groupnorm_impl<F16, F16, F16, F32, F16>(
+        ck::profiler::profile_groupnorm_impl<F16, F16, F16, F32, F16, F32, false>(
             do_verification, init_method, do_log, time_kernel, length);
     }
     else
diff --git a/profiler/src/profile_layernorm.cpp b/profiler/src/profile_layernorm.cpp
index 7bf210e678..fdeaa036b2 100644
--- a/profiler/src/profile_layernorm.cpp
+++ b/profiler/src/profile_layernorm.cpp
@@ -82,12 +82,12 @@ int profile_layernorm(int argc, char* argv[])
 
     if(data_type == ck::DataTypeEnum::Half)
     {
-        ck::profiler::profile_layernorm_impl<F16, F16, F16, F32, F16, rank>(
+        ck::profiler::profile_layernorm_impl<F16, F16, F16, F32, F16, F32, false, rank>(
             do_verification, init_method, do_log, time_kernel, length);
     }
     else if(data_type == ck::DataTypeEnum::Float)
     {
-        ck::profiler::profile_layernorm_impl<F32, F32, F32, F32, F32, rank>(
+        ck::profiler::profile_layernorm_impl<F32, F32, F32, F32, F32, F32, false, rank>(
             do_verification, init_method, do_log, time_kernel, length);
     }
     else
diff --git a/test/normalization/test_groupnorm_fp16.cpp b/test/normalization/test_groupnorm_fp16.cpp
index 325ea75fe5..67387ad40b 100644
--- a/test/normalization/test_groupnorm_fp16.cpp
+++ b/test/normalization/test_groupnorm_fp16.cpp
@@ -12,11 +12,12 @@ template <typename Tuple>
 class TestGroupnorm : public ::testing::Test
 {
     protected:
-    using XDataType       = std::tuple_element_t<0, Tuple>;
-    using GammaDataType   = std::tuple_element_t<1, Tuple>;
-    using BetaDataType    = std::tuple_element_t<2, Tuple>;
-    using ComputeDataType = std::tuple_element_t<3, Tuple>;
-    using YDataType       = std::tuple_element_t<4, Tuple>;
+    using XDataType              = std::tuple_element_t<0, Tuple>;
+    using GammaDataType          = std::tuple_element_t<1, Tuple>;
+    using BetaDataType           = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType        = std::tuple_element_t<3, Tuple>;
+    using YDataType              = std::tuple_element_t<4, Tuple>;
+    using SaveMeanInvStdDataType = std::tuple_element_t<5, Tuple>;
 
     void Run()
     {
@@ -37,7 +38,9 @@ class TestGroupnorm : public ::testing::Test
                                                      GammaDataType,
                                                      BetaDataType,
                                                      ComputeDataType,
-                                                     YDataType>(true, 2, false, false, length);
+                                                     YDataType,
+                                                     SaveMeanInvStdDataType,
+                                                     true>(true, 2, false, false, length);
             EXPECT_TRUE(success);
         }
     }
@@ -45,7 +48,7 @@ class TestGroupnorm : public ::testing::Test
 
 using KernelTypes = ::testing::Types<
     // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
-    std::tuple<F16, F16, F16, F32, F16>>;
+    std::tuple<F16, F16, F16, F32, F16, F32>>;
 
 TYPED_TEST_SUITE(TestGroupnorm, KernelTypes);
 TYPED_TEST(TestGroupnorm, Test_FP16) { this->Run(); }
diff --git a/test/normalization/test_groupnorm_fp32.cpp b/test/normalization/test_groupnorm_fp32.cpp
index ec88442fc0..136079f104 100644
--- a/test/normalization/test_groupnorm_fp32.cpp
+++ b/test/normalization/test_groupnorm_fp32.cpp
@@ -12,11 +12,12 @@ template <typename Tuple>
 class TestGroupnorm : public ::testing::Test
 {
     protected:
-    using XDataType       = std::tuple_element_t<0, Tuple>;
-    using GammaDataType   = std::tuple_element_t<1, Tuple>;
-    using BetaDataType    = std::tuple_element_t<2, Tuple>;
-    using ComputeDataType = std::tuple_element_t<3, Tuple>;
-    using YDataType       = std::tuple_element_t<4, Tuple>;
+    using XDataType              = std::tuple_element_t<0, Tuple>;
+    using GammaDataType          = std::tuple_element_t<1, Tuple>;
+    using BetaDataType           = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType        = std::tuple_element_t<3, Tuple>;
+    using YDataType              = std::tuple_element_t<4, Tuple>;
+    using SaveMeanInvStdDataType = std::tuple_element_t<5, Tuple>;
 
     void Run()
     {
@@ -35,7 +36,9 @@ class TestGroupnorm : public ::testing::Test
                                                      GammaDataType,
                                                      BetaDataType,
                                                      ComputeDataType,
-                                                     YDataType>(true, 2, false, false, length);
+                                                     YDataType,
+                                                     SaveMeanInvStdDataType,
+                                                     true>(true, 2, false, false, length);
             EXPECT_TRUE(success);
         }
     }
@@ -43,7 +46,7 @@ class TestGroupnorm : public ::testing::Test
 
 using KernelTypes = ::testing::Types<
     // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
-    std::tuple<F32, F32, F32, F32, F32>>;
+    std::tuple<F32, F32, F32, F32, F32, F32>>;
 
 TYPED_TEST_SUITE(TestGroupnorm, KernelTypes);
 TYPED_TEST(TestGroupnorm, Test_FP32) { this->Run(); }
diff --git a/test/normalization/test_layernorm2d_fp16.cpp b/test/normalization/test_layernorm2d_fp16.cpp
index 2222740fcc..54bab25257 100644
--- a/test/normalization/test_layernorm2d_fp16.cpp
+++ b/test/normalization/test_layernorm2d_fp16.cpp
@@ -12,11 +12,12 @@ template <typename Tuple>
 class TestLayernorm2d : public ::testing::Test
 {
     protected:
-    using XDataType       = std::tuple_element_t<0, Tuple>;
-    using GammaDataType   = std::tuple_element_t<1, Tuple>;
-    using BetaDataType    = std::tuple_element_t<2, Tuple>;
-    using ComputeDataType = std::tuple_element_t<3, Tuple>;
-    using YDataType       = std::tuple_element_t<4, Tuple>;
+    using XDataType              = std::tuple_element_t<0, Tuple>;
+    using GammaDataType          = std::tuple_element_t<1, Tuple>;
+    using BetaDataType           = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType        = std::tuple_element_t<3, Tuple>;
+    using YDataType              = std::tuple_element_t<4, Tuple>;
+    using SaveMeanInvStdDataType = std::tuple_element_t<5, Tuple>;
 
     void Run()
     {
@@ -31,6 +32,8 @@ class TestLayernorm2d : public ::testing::Test
                                                                 BetaDataType,
                                                                 ComputeDataType,
                                                                 YDataType,
+                                                                SaveMeanInvStdDataType,
+                                                                true,
                                                                 2>(true, 2, false, false, length);
             EXPECT_TRUE(success);
         }
@@ -39,7 +42,7 @@ class TestLayernorm2d : public ::testing::Test
 
 using KernelTypes = ::testing::Types<
     // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
-    std::tuple<F16, F16, F16, F32, F16>>;
+    std::tuple<F16, F16, F16, F32, F16, F32>>;
 
 TYPED_TEST_SUITE(TestLayernorm2d, KernelTypes);
 TYPED_TEST(TestLayernorm2d, Test_FP16) { this->Run(); }
diff --git a/test/normalization/test_layernorm2d_fp32.cpp b/test/normalization/test_layernorm2d_fp32.cpp
index 30fbe06c60..ee9646a4d5 100644
--- a/test/normalization/test_layernorm2d_fp32.cpp
+++ b/test/normalization/test_layernorm2d_fp32.cpp
@@ -12,11 +12,12 @@ template <typename Tuple>
 class TestLayernorm2d : public ::testing::Test
 {
     protected:
-    using XDataType       = std::tuple_element_t<0, Tuple>;
-    using GammaDataType   = std::tuple_element_t<1, Tuple>;
-    using BetaDataType    = std::tuple_element_t<2, Tuple>;
-    using ComputeDataType = std::tuple_element_t<3, Tuple>;
-    using YDataType       = std::tuple_element_t<4, Tuple>;
+    using XDataType              = std::tuple_element_t<0, Tuple>;
+    using GammaDataType          = std::tuple_element_t<1, Tuple>;
+    using BetaDataType           = std::tuple_element_t<2, Tuple>;
+    using ComputeDataType        = std::tuple_element_t<3, Tuple>;
+    using YDataType              = std::tuple_element_t<4, Tuple>;
+    using SaveMeanInvStdDataType = std::tuple_element_t<5, Tuple>;
 
     void Run()
     {
@@ -31,6 +32,8 @@ class TestLayernorm2d : public ::testing::Test
                                                                 BetaDataType,
                                                                 ComputeDataType,
                                                                 YDataType,
+                                                                SaveMeanInvStdDataType,
+                                                                true,
                                                                 2>(true, 2, false, false, length);
             EXPECT_TRUE(success);
         }
@@ -39,7 +42,7 @@ class TestLayernorm2d : public ::testing::Test
 
 using KernelTypes = ::testing::Types<
     // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType>
-    std::tuple<F32, F32, F32, F32, F32>>;
+    std::tuple<F32, F32, F32, F32, F32, F32>>;
 
 TYPED_TEST_SUITE(TestLayernorm2d, KernelTypes);
 TYPED_TEST(TestLayernorm2d, Test_FP32) { this->Run(); }

From 0abc0f87db861cf25395ce9540ecb8e7f4821e4f Mon Sep 17 00:00:00 2001
From: Bartlomiej Wroblewski <bwroblewski10@gmail.com>
Date: Thu, 19 Oct 2023 16:53:18 +0200
Subject: [PATCH 180/194] Change 1d,2d,... to 1D,2D,... (#997)

---
 CHANGELOG.md                                     | 2 +-
 example/12_reduce/README.md                      | 4 ++--
 example/30_grouped_conv_fwd_multiple_d/README.md | 2 +-
 profiler/README.md                               | 8 ++++----
 4 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/CHANGELOG.md b/CHANGELOG.md
index f88df081da..3e46a4ab4b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -17,7 +17,7 @@ None
 - Support for 3D grouped convolution on RDNA 3 GPUs (#935, #950, #985)
 - Grouped convolution support for small K and C (#822 #879 #897)
 - Support for NHWGC (2D and 3D) grouped convolution backward weight (#769 #804)
-- Support for bf16/f32/f16 and NHWGC (2D and 3d) grouped convolution backward data (#757 #799)
+- Support for bf16/f32/f16 and NHWGC (2D and 3D) grouped convolution backward data (#757 #799)
 - Support for Batched Gemm DL (#732)
 
 ### Changes
diff --git a/example/12_reduce/README.md b/example/12_reduce/README.md
index 76d28527bb..bcffa684c8 100644
--- a/example/12_reduce/README.md
+++ b/example/12_reduce/README.md
@@ -2,7 +2,7 @@
 
 ## Run ```example_reduce_blockwise```
 ```bash
-# -D <xxx> : input 3d/4d/5d tensor lengths
+# -D <xxx> : input 3D/4D/5D tensor lengths
 # -R <xxx> : reduce dimension ids
 # -v <x> :   verification (0=no, 1=yes)
 #arg1: data type (0: fp16, 1: fp32, 3: int8, 5: bp16, 6: fp64, 7: int4)
@@ -22,7 +22,7 @@ Perf: 0.238063 ms, 264.285 GB/s, DeviceReduceBlockWise<256,M_C4_S1,K_C64_S1,InSr
 
 ## Run ```example_reduce_multiblock_atomic_add```
 ```bash
-# -D <xxx> : input 3d/4d/5d tensor lengths
+# -D <xxx> : input 3D/4D/5D tensor lengths
 # -R <xxx> : reduce dimension ids
 # -v <x> :   verification (0=no, 1=yes)
 #arg1: data type (0: fp32, 1: fp64)
diff --git a/example/30_grouped_conv_fwd_multiple_d/README.md b/example/30_grouped_conv_fwd_multiple_d/README.md
index 739a0425a8..4718795c57 100644
--- a/example/30_grouped_conv_fwd_multiple_d/README.md
+++ b/example/30_grouped_conv_fwd_multiple_d/README.md
@@ -4,7 +4,7 @@ arg1: verification (0=no, 1=yes)
 arg2: initialization (0=no init, 1=integer value, 2=decimal value)
 arg3: time kernel (0=no, 1=yes)
 Following arguments (depending on number of spatial dims):
- Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+ Number of spatial dimensions (1=Conv1D, 2=Conv2D, 3=Conv3D)
  G, N, K, C,
  <filter spatial dimensions>, (ie Y, X for 2D)
  <input image spatial dimensions>, (ie Hi, Wi for 2D)
diff --git a/profiler/README.md b/profiler/README.md
index a2d370b210..98eae3a763 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -22,7 +22,7 @@ c_m_n: dim 2, lengths {3840, 4096}, strides {4096, 1}
 Best Perf: 1.1933 ms, 107.977 TFlops, 79.0848 GB/s
 ```
 
-## Profile 2d forward convolution kernels
+## Profile 2D forward convolution kernels
 ```bash
 #arg1: tensor operation (conv=Convolution)
 #arg2: data type (0=fp32, 1=fp16)
@@ -115,7 +115,7 @@ Best Perf: 58.0306 ms, 37.8942 TFlops, 27.7545 GB/s
 # arg6: print tensor value (0: no; 1: yes)
 # arg7: time kernel (0: no, 1: yes)
 # Following arguments (depending on number of spatial dims):
-#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  Number of spatial dimensions (1=Conv1D, 2=Conv2D, 3=Conv3D)
 #  G, N, K, C, 
 #  <filter spatial dimensions>, (ie Y, X for 2D)
 #  <input image spatial dimensions>, (ie Hi, Wi for 2D)
@@ -158,7 +158,7 @@ GB/s: 127.947
 # arg6: print tensor value (0: no; 1: yes)
 # arg7: time kernel (0: no, 1: yes)
 # Following arguments (depending on number of spatial dims):
-#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  Number of spatial dimensions (1=Conv1D, 2=Conv2D, 3=Conv3D)
 #  G, N, K, C, 
 #  <filter spatial dimensions>, (ie Y, X for 2D)
 #  <input image spatial dimensions>, (ie Hi, Wi for 2D)
@@ -201,7 +201,7 @@ Note: This kernel use atomic add, this will cause output buffer to be accumulate
 # arg7: time kernel (0: no, 1: yes)
 # arg8: operation type (0: ImageToColumn, 1: ColumnToImage)
 # Following arguments (depending on number of spatial dims):
-#  Number of spatial dimensions (1=Conv1d, 2=Conv2d, 3=Conv3d)
+#  Number of spatial dimensions (1=Conv1D, 2=Conv2D, 3=Conv3D)
 #  G, N, K, C, 
 #  <filter spatial dimensions>, (ie Y, X for 2D)
 #  <input image spatial dimensions>, (ie Hi, Wi for 2D)

From deef92d5d0c6d1afbef2e6567273b818ce8085bc Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Thu, 19 Oct 2023 23:19:07 +0800
Subject: [PATCH 181/194] Avoid force setting ENABLE_PIPELINE_V2_OPT to OFF
 (#961)

* Avoid force setting ENABLE_PIPELINE_V2_OPT to OFF

* Remove compilation option variable MAX_ILP_OPTS
---
 .../tensor_operation_instance/gpu/gemm/CMakeLists.txt  | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 89820775c3..5edd6c85f4 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -96,13 +96,9 @@ list(APPEND GEMM_INSTANCES device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instan
 
 add_instance_library(device_gemm_instance ${GEMM_INSTANCES})
 
-set(ENABLE_PIPELINE_V2_OPT OFF)
+set(ENABLE_PIPELINE_V2_OPT)
 
 if (ENABLE_PIPELINE_V2_OPT)
-    set(MAX_ILP_OPTS
-       -mllvm
-       -amdgpu-enable-max-ilp-scheduling-strategy
-    )
     set(WAVES_PER_EU_DEFS
        CK_USE_WAVES_PER_EU=1
        CK_MIN_WAVES_PER_EU=1
@@ -118,7 +114,7 @@ if (ENABLE_PIPELINE_V2_OPT)
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
     # layout=NN
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-       COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+       COMPILE_OPTIONS ";;"
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
     # layout=TT
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
@@ -126,7 +122,7 @@ if (ENABLE_PIPELINE_V2_OPT)
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
     # layout=TN
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-       COMPILE_OPTIONS "${MAX_ILP_OPTS}"
+       COMPILE_OPTIONS ";;"
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
 endif(ENABLE_PIPELINE_V2_OPT)
 

From 82f3a835d57d371c4fcf4476769e9dcd7faa6073 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Thu, 19 Oct 2023 17:23:19 +0200
Subject: [PATCH 182/194]  Extend available elementwise operations with conv
 examples (#995)

* Extend available elementwise operations with conv examples

* Fixes

* Remove not needed convert

* Update CMakeFile and dir name
---
 example/62_conv_fwd_activ/CMakeLists.txt      |  35 +++
 .../convnd_fwd_activ_common.hpp               | 238 ++++++++++++++++++
 .../convnd_fwd_xdl_abs_fp16.cpp               |  11 +
 .../convnd_fwd_xdl_clippedrelu_fp16.cpp       |  11 +
 .../convnd_fwd_xdl_elu_fp16.cpp               |  11 +
 .../convnd_fwd_xdl_leakyrelu_fp16.cpp         |  11 +
 .../convnd_fwd_xdl_pow_fp16.cpp               |  11 +
 .../convnd_fwd_xdl_relu_fp16.cpp              |  11 +
 .../convnd_fwd_xdl_sigmoid_fp16.cpp           |  11 +
 .../convnd_fwd_xdl_softrelu_fp16.cpp          |  11 +
 .../convnd_fwd_xdl_tanh_fp16.cpp              |  11 +
 .../run_convnd_fwd_activ_example.inc          |  91 +++++++
 .../element/unary_element_wise_operation.hpp  | 106 +++++++-
 include/ck/utility/math.hpp                   |  22 --
 include/ck/utility/math_v2.hpp                | 192 +++++++++++++-
 .../statically_indexed_array_multi_index.hpp  |   1 +
 .../cpu/reference_conv_fwd.hpp                |  24 +-
 17 files changed, 758 insertions(+), 50 deletions(-)
 create mode 100644 example/62_conv_fwd_activ/CMakeLists.txt
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_activ_common.hpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_abs_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_clippedrelu_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_elu_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_leakyrelu_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_pow_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_relu_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_sigmoid_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_softrelu_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/convnd_fwd_xdl_tanh_fp16.cpp
 create mode 100644 example/62_conv_fwd_activ/run_convnd_fwd_activ_example.inc

diff --git a/example/62_conv_fwd_activ/CMakeLists.txt b/example/62_conv_fwd_activ/CMakeLists.txt
new file mode 100644
index 0000000000..ea38216fa9
--- /dev/null
+++ b/example/62_conv_fwd_activ/CMakeLists.txt
@@ -0,0 +1,35 @@
+list(APPEND gpu_list gfx908 gfx90a gfx940 gfx941 gfx942)
+set(target 0)
+foreach(gpu IN LISTS GPU_TARGETS)
+ if(gpu IN_LIST gpu_list AND target EQUAL 0)
+      add_custom_target(example_convnd_fwd_activ_xdl)
+      # Sigmoid
+      add_example_executable(example_convnd_fwd_xdl_sigmoid_fp16 convnd_fwd_xdl_sigmoid_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_sigmoid_fp16)
+      # Tanh
+      add_example_executable(example_convnd_fwd_xdl_tanh_fp16 convnd_fwd_xdl_tanh_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_tanh_fp16)
+      # Relu
+      add_example_executable(example_convnd_fwd_xdl_relu_fp16 convnd_fwd_xdl_relu_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_relu_fp16)
+      # SoftRelu
+      add_example_executable(example_convnd_fwd_xdl_softrelu_fp16 convnd_fwd_xdl_softrelu_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_softrelu_fp16)
+      # Abs
+      add_example_executable(example_convnd_fwd_xdl_abs_fp16 convnd_fwd_xdl_abs_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_abs_fp16)
+      # Pow
+      add_example_executable(example_convnd_fwd_xdl_pow_fp16 convnd_fwd_xdl_pow_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_pow_fp16)
+      # Clipped Relu
+      add_example_executable(example_convnd_fwd_xdl_clippedrelu_fp16 convnd_fwd_xdl_clippedrelu_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_clippedrelu_fp16)
+      # Leaky Relu
+      add_example_executable(example_convnd_fwd_xdl_leakyrelu_fp16 convnd_fwd_xdl_leakyrelu_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_leakyrelu_fp16)
+      # Elu
+      add_example_executable(example_convnd_fwd_xdl_elu_fp16 convnd_fwd_xdl_elu_fp16.cpp)
+      add_example_dependencies(example_convnd_fwd_activ_xdl example_convnd_fwd_xdl_elu_fp16)
+   set(target 1)
+ endif()
+endforeach()
diff --git a/example/62_conv_fwd_activ/convnd_fwd_activ_common.hpp b/example/62_conv_fwd_activ/convnd_fwd_activ_common.hpp
new file mode 100644
index 0000000000..185026b1e3
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_activ_common.hpp
@@ -0,0 +1,238 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include <cstdlib>
+#include <iostream>
+#include <numeric>
+#include <type_traits>
+
+#include "ck/ck.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+
+#include "ck/library/utility/algorithm.hpp"
+#include "ck/library/utility/check_err.hpp"
+#include "ck/library/utility/device_memory.hpp"
+#include "ck/library/utility/host_tensor.hpp"
+#include "ck/library/utility/host_tensor_generator.hpp"
+#include "ck/library/utility/convolution_parameter.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+#include "ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp"
+#include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
+
+constexpr ck::index_t NDimSpatial = 3;
+using InDataType                  = ck::half_t;
+using WeiDataType                 = ck::half_t;
+using AccDataType                 = float;
+using CShuffleDataType            = ck::half_t;
+using OutDataType                 = ck::half_t;
+
+template <ck::index_t... Is>
+using S = ck::Sequence<Is...>;
+
+using InLayout  = ck::tensor_layout::convolution::GNDHWC;
+using WeiLayout = ck::tensor_layout::convolution::GKZYXC;
+using OutLayout = ck::tensor_layout::convolution::GNDHWK;
+
+using InElementOp  = ck::tensor_operation::element_wise::PassThrough;
+using WeiElementOp = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr auto ConvSpec =
+    ck::tensor_operation::device::ConvolutionForwardSpecialization::Default;
+
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <typename OutElementOp>
+using DeviceGroupedConvNDFwdInstance =
+    ck::tensor_operation::device::DeviceGroupedConvFwdMultipleD_Xdl_CShuffle<
+        NDimSpatial,
+        InLayout,
+        WeiLayout,
+        ck::Tuple<>,
+        OutLayout,
+        InDataType,
+        WeiDataType,
+        AccDataType,
+        CShuffleDataType,
+        ck::Tuple<>,
+        OutDataType,
+        InElementOp,
+        WeiElementOp,
+        OutElementOp,
+        ConvSpec,    // ConvForwardSpecialization
+        GemmSpec,    // GemmSpecialization
+        1,           //
+        256,         // BlockSize
+        128,         // MPerBlock
+        256,         // NPerBlock
+        32,          // KPerBlock
+        8,           // AK1
+        8,           // BK1
+        32,          // MPerXdl
+        32,          // NPerXdl
+        2,           // MXdlPerWave
+        4,           // NXdlPerWave
+        S<4, 64, 1>, // ABlockTransferThreadClusterLengths_AK0_M_AK1
+        S<1, 0, 2>,  // ABlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // ABlockTransferSrcAccessOrder
+        2,           // ABlockTransferSrcVectorDim
+        8,           // ABlockTransferSrcScalarPerVector
+        8,           // ABlockTransferDstScalarPerVector_AK1
+        1,           // ABlockLdsExtraM
+        S<4, 64, 1>, // BBlockTransferThreadClusterLengths_BK0_N_BK1
+        S<1, 0, 2>,  // BBlockTransferThreadClusterArrangeOrder
+        S<1, 0, 2>,  // BBlockTransferSrcAccessOrder
+        2,           // BBlockTransferSrcVectorDim
+        8,           // BBlockTransferSrcScalarPerVector
+        8,           // BBlockTransferDstScalarPerVector_BK1
+        1,           // BBlockLdsExtraN
+        1,
+        1,
+        S<1, 32, 1, 8>,
+        8>;
+
+template <ck::index_t NDimSpatial,
+          typename InDataType,
+          typename WeiDataType,
+          typename OutDataType,
+          typename InElementOp,
+          typename WeiElementOp,
+          typename OutElementOp,
+          typename DeviceConvNDFwdInstance>
+bool run_grouped_conv_fwd(bool do_verification,
+                          int init_method,
+                          bool time_kernel,
+                          const ck::utils::conv::ConvParam& conv_param,
+                          const HostTensorDescriptor& in_g_n_c_wis_desc,
+                          const HostTensorDescriptor& wei_g_k_c_xs_desc,
+                          const HostTensorDescriptor& out_g_n_k_wos_desc,
+                          const InElementOp& in_element_op,
+                          const WeiElementOp& wei_element_op,
+                          const OutElementOp& out_element_op)
+{
+    Tensor<InDataType> in(in_g_n_c_wis_desc);
+    Tensor<WeiDataType> wei(wei_g_k_c_xs_desc);
+    Tensor<OutDataType> out_host(out_g_n_k_wos_desc);
+    Tensor<OutDataType> out_device(out_g_n_k_wos_desc);
+
+    std::cout << "in: " << in.mDesc << std::endl;
+    std::cout << "wei: " << wei.mDesc << std::endl;
+    std::cout << "out: " << out_host.mDesc << std::endl;
+
+    switch(init_method)
+    {
+    case 0: break;
+    case 1:
+        in.GenerateTensorValue(GeneratorTensor_2<InDataType>{-2, 2});
+        wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-2, 2});
+        break;
+    default:
+        in.GenerateTensorValue(GeneratorTensor_3<InDataType>{-1.0, 1.0});
+        wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.05, 0.05});
+    }
+
+    DeviceMem in_device_buf(sizeof(InDataType) * in.mDesc.GetElementSpaceSize());
+    DeviceMem wei_device_buf(sizeof(WeiDataType) * wei.mDesc.GetElementSpaceSize());
+    DeviceMem out_device_buf(sizeof(OutDataType) * out_device.mDesc.GetElementSpaceSize());
+
+    in_device_buf.ToDevice(in.mData.data());
+    wei_device_buf.ToDevice(wei.mData.data());
+
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> a_g_n_c_wis_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> b_g_k_c_xs_strides{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_lengths{};
+    std::array<ck::index_t, NDimSpatial + 3> e_g_n_k_wos_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
+    std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
+    std::array<ck::index_t, NDimSpatial> input_left_pads{};
+    std::array<ck::index_t, NDimSpatial> input_right_pads{};
+
+    auto copy = [](const auto& x, auto& y) { ck::ranges::copy(x, y.begin()); };
+
+    copy(in_g_n_c_wis_desc.GetLengths(), a_g_n_c_wis_lengths);
+    copy(in_g_n_c_wis_desc.GetStrides(), a_g_n_c_wis_strides);
+    copy(wei_g_k_c_xs_desc.GetLengths(), b_g_k_c_xs_lengths);
+    copy(wei_g_k_c_xs_desc.GetStrides(), b_g_k_c_xs_strides);
+    copy(out_g_n_k_wos_desc.GetLengths(), e_g_n_k_wos_lengths);
+    copy(out_g_n_k_wos_desc.GetStrides(), e_g_n_k_wos_strides);
+    copy(conv_param.conv_filter_strides_, conv_filter_strides);
+    copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
+    copy(conv_param.input_left_pads_, input_left_pads);
+    copy(conv_param.input_right_pads_, input_right_pads);
+
+    // do Conv
+    auto conv     = DeviceConvNDFwdInstance{};
+    auto invoker  = conv.MakeInvoker();
+    auto argument = conv.MakeArgument(in_device_buf.GetDeviceBuffer(),
+                                      wei_device_buf.GetDeviceBuffer(),
+                                      std::array<const void*, 0>{},
+                                      out_device_buf.GetDeviceBuffer(),
+                                      a_g_n_c_wis_lengths,
+                                      a_g_n_c_wis_strides,
+                                      b_g_k_c_xs_lengths,
+                                      b_g_k_c_xs_strides,
+                                      std::array<std::array<ck::index_t, NDimSpatial + 3>, 0>{{}},
+                                      std::array<std::array<ck::index_t, NDimSpatial + 3>, 0>{{}},
+                                      e_g_n_k_wos_lengths,
+                                      e_g_n_k_wos_strides,
+                                      conv_filter_strides,
+                                      conv_filter_dilations,
+                                      input_left_pads,
+                                      input_right_pads,
+                                      in_element_op,
+                                      wei_element_op,
+                                      out_element_op);
+
+    if(!conv.IsSupportedArgument(argument))
+    {
+        throw std::runtime_error(
+            "wrong! device_conv with the specified compilation parameters does "
+            "not support this Conv problem");
+    }
+
+    float avg_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel});
+
+    std::size_t flop      = conv_param.GetFlops();
+    std::size_t num_btype = conv_param.GetByte<InDataType, WeiDataType, OutDataType>();
+
+    float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+    float gb_per_sec = num_btype / 1.E6 / avg_time;
+    std::cout << "Perf: " << avg_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
+              << conv.GetTypeString() << std::endl;
+
+    if(do_verification)
+    {
+        auto ref_conv = ck::tensor_operation::host::ReferenceConvFwd<NDimSpatial,
+                                                                     InDataType,
+                                                                     WeiDataType,
+                                                                     OutDataType,
+                                                                     InElementOp,
+                                                                     WeiElementOp,
+                                                                     OutElementOp>();
+
+        auto ref_invoker  = ref_conv.MakeInvoker();
+        auto ref_argument = ref_conv.MakeArgument(in,
+                                                  wei,
+                                                  out_host,
+                                                  conv_param.conv_filter_strides_,
+                                                  conv_param.conv_filter_dilations_,
+                                                  conv_param.input_left_pads_,
+                                                  conv_param.input_right_pads_,
+                                                  in_element_op,
+                                                  wei_element_op,
+                                                  out_element_op);
+
+        ref_invoker.Run(ref_argument);
+
+        out_device_buf.FromDevice(out_device.mData.data());
+
+        return ck::utils::check_err(out_device, out_host, "Error: incorrect results!");
+    }
+
+    return true;
+}
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_abs_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_abs_fp16.cpp
new file mode 100644
index 0000000000..4fe0c857fa
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_abs_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::UnaryAbs;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_clippedrelu_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_clippedrelu_fp16.cpp
new file mode 100644
index 0000000000..feabacc5c9
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_clippedrelu_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::ClippedRelu;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_elu_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_elu_fp16.cpp
new file mode 100644
index 0000000000..793102dbc6
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_elu_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::Elu;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_leakyrelu_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_leakyrelu_fp16.cpp
new file mode 100644
index 0000000000..a77408db7e
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_leakyrelu_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::LeakyRelu;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_pow_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_pow_fp16.cpp
new file mode 100644
index 0000000000..2b695cf8c3
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_pow_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::Power;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_relu_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_relu_fp16.cpp
new file mode 100644
index 0000000000..e1b6e3f0cc
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_relu_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::Relu;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_sigmoid_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_sigmoid_fp16.cpp
new file mode 100644
index 0000000000..350c15a787
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_sigmoid_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::Sigmoid;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_softrelu_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_softrelu_fp16.cpp
new file mode 100644
index 0000000000..ec52e1a3c4
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_softrelu_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::SoftRelu;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/convnd_fwd_xdl_tanh_fp16.cpp b/example/62_conv_fwd_activ/convnd_fwd_xdl_tanh_fp16.cpp
new file mode 100644
index 0000000000..dca405669a
--- /dev/null
+++ b/example/62_conv_fwd_activ/convnd_fwd_xdl_tanh_fp16.cpp
@@ -0,0 +1,11 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "convnd_fwd_activ_common.hpp"
+
+using OutElementOp = ck::tensor_operation::element_wise::TanH;
+
+using DeviceGroupedConvNDFwdActivInstance = DeviceGroupedConvNDFwdInstance<OutElementOp>;
+#include "run_convnd_fwd_activ_example.inc"
+
+int main(int argc, char* argv[]) { return !run_convnd_fwd_example(argc, argv); }
diff --git a/example/62_conv_fwd_activ/run_convnd_fwd_activ_example.inc b/example/62_conv_fwd_activ/run_convnd_fwd_activ_example.inc
new file mode 100644
index 0000000000..7c20c01066
--- /dev/null
+++ b/example/62_conv_fwd_activ/run_convnd_fwd_activ_example.inc
@@ -0,0 +1,91 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+void print_helper_msg()
+{
+    std::cout << "arg1: verification (0=no, 1=yes)\n"
+              << "arg2: initialization (0=no init, 1=integer value, 2=decimal value)\n"
+              << "arg3: time kernel (0=no, 1=yes)\n"
+              << ck::utils::conv::get_conv_param_parser_helper_msg() << std::endl;
+}
+
+bool run_convnd_fwd_example(int argc, char* argv[])
+{
+    print_helper_msg();
+
+    bool do_verification = true;
+    // Use floats for SoftRelu by default to avoid overflow after e^x.
+    int init_method =
+        std::is_same_v<OutElementOp, ck::tensor_operation::element_wise::SoftRelu> ? 2 : 1;
+    bool time_kernel = false;
+
+    // Following shapes are selected to avoid overflow. Expect inf in case of
+    // size increase for some elementwise ops.
+    ck::utils::conv::ConvParam conv_param{
+        3, 1, 16, 128, 8, {3, 3, 3}, {17, 17, 17}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}};
+
+    if(argc == 1)
+    {
+        // use default
+    }
+    else if(argc == 4)
+    {
+        do_verification = std::stoi(argv[1]);
+        init_method     = std::stoi(argv[2]);
+        time_kernel     = std::stoi(argv[3]);
+    }
+    else
+    {
+        do_verification                   = std::stoi(argv[1]);
+        init_method                       = std::stoi(argv[2]);
+        time_kernel                       = std::stoi(argv[3]);
+        const ck::index_t num_dim_spatial = std::stoi(argv[4]);
+
+        conv_param = ck::utils::conv::parse_conv_param(num_dim_spatial, 5, argv);
+    }
+
+    const auto in_element_op  = InElementOp{};
+    const auto wei_element_op = WeiElementOp{};
+    const auto out_element_op = OutElementOp{};
+
+    const auto run = [&]() {
+        const auto in_g_n_c_wis_desc =
+            ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InLayout>(
+                conv_param);
+
+        const auto wei_g_k_c_xs_desc =
+            ck::utils::conv::make_weight_host_tensor_descriptor_g_k_c_xs_packed<WeiLayout>(
+                conv_param);
+
+        const auto out_g_n_k_wos_desc =
+            ck::utils::conv::make_output_host_tensor_descriptor_g_n_k_wos_packed<OutLayout>(
+                conv_param);
+
+        return run_grouped_conv_fwd<NDimSpatial,
+                                    InDataType,
+                                    WeiDataType,
+                                    OutDataType,
+                                    InElementOp,
+                                    WeiElementOp,
+                                    OutElementOp,
+                                    DeviceGroupedConvNDFwdActivInstance>(do_verification,
+                                                                         init_method,
+                                                                         time_kernel,
+                                                                         conv_param,
+                                                                         in_g_n_c_wis_desc,
+                                                                         wei_g_k_c_xs_desc,
+                                                                         out_g_n_k_wos_desc,
+                                                                         in_element_op,
+                                                                         wei_element_op,
+                                                                         out_element_op);
+    };
+
+    if(conv_param.num_dim_spatial_ == 3)
+    {
+        return run();
+    }
+
+    return false;
+}
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index 23fdd23e10..dabdf649e4 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -442,10 +442,11 @@ struct Sigmoid
     __host__ __device__ void operator()(T& y, const T& x) const
     {
         static_assert(is_same<T, float>::value || is_same<T, double>::value ||
-                          is_same<T, ck::half_t>::value,
+                          is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
+                          is_same<T, int32_t>::value,
                       "Data type is not supported by this operation!");
-
-        y = 1 / (ck::type_convert<T>(1) + exp(-x));
+        constexpr T one = type_convert<T>(1);
+        y               = one / (one + ck::math::exp(-x));
     };
 };
 
@@ -455,7 +456,8 @@ struct TanH
     __host__ __device__ void operator()(T& y, const T& x) const
     {
         static_assert(is_same<T, float>::value || is_same<T, double>::value ||
-                          is_same<T, ck::half_t>::value,
+                          is_same<T, ck::half_t>::value || is_same<T, int8_t>::value ||
+                          is_same<T, int32_t>::value,
                       "Data type is not supported by this operation!");
 
         y = ck::math::tanh(x);
@@ -481,7 +483,101 @@ struct Swish
         y        = type_convert<Y>(x / (1.f + ck::math::exp(bx)));
     };
 
-    float beta_ = 1.0f;
+    const float beta_;
+};
+
+struct SoftRelu
+{
+    SoftRelu(float alpha = 1.f) : alpha_(alpha){};
+
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, half_t>::value || is_same<T, int32_t>::value ||
+                          is_same<T, int8_t>::value,
+                      "Data type is not supported by this operation!");
+        T casted_alpha  = type_convert<T>(alpha_);
+        constexpr T one = type_convert<T>(1);
+        y               = ck::math::log(one + ck::math::exp(x * casted_alpha)) / casted_alpha;
+    }
+    const float alpha_;
+};
+
+struct Power
+{
+    Power(float alpha = 0.f, float beta = 1.f, float gamma = 2.f)
+        : alpha_(alpha), beta_(beta), gamma_(gamma){};
+
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, half_t>::value || is_same<T, int32_t>::value ||
+                          is_same<T, int8_t>::value,
+                      "Data type is not supported by this operation!");
+        T casted_alpha     = type_convert<T>(alpha_);
+        T casted_beta      = type_convert<T>(beta_);
+        T casted_gamma     = type_convert<T>(gamma_);
+        T shifted_scaled_x = casted_alpha + casted_beta * x;
+        y                  = ck::math::pow(shifted_scaled_x, casted_gamma);
+    }
+    const float alpha_;
+    const float beta_;
+    const float gamma_;
+};
+
+struct ClippedRelu
+{
+    ClippedRelu(float alpha = 0.f, float beta = 1.f) : alpha_(alpha), beta_(beta){};
+
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, half_t>::value || is_same<T, int32_t>::value ||
+                          is_same<T, int8_t>::value,
+                      "Data type is not supported by this operation!");
+        T casted_alpha = type_convert<T>(alpha_);
+        T casted_beta  = type_convert<T>(beta_);
+        y              = ck::math::min(casted_beta, ck::math::max(casted_alpha, x));
+    }
+    const float alpha_;
+    const float beta_;
+};
+
+struct LeakyRelu
+{
+    LeakyRelu(float alpha = 0.01f) : alpha_(alpha){};
+
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, half_t>::value || is_same<T, int32_t>::value ||
+                          is_same<T, int8_t>::value,
+                      "Data type is not supported by this operation!");
+        T casted_alpha = type_convert<T>(alpha_);
+        y              = x >= 0 ? x : x * casted_alpha;
+    }
+    const float alpha_;
+};
+
+struct Elu
+{
+    Elu(float alpha = 1.f) : alpha_(alpha){};
+
+    template <typename T>
+    __host__ __device__ void operator()(T& y, const T& x) const
+    {
+        static_assert(is_same<T, float>::value || is_same<T, double>::value ||
+                          is_same<T, half_t>::value || is_same<T, int32_t>::value ||
+                          is_same<T, int8_t>::value,
+                      "Data type is not supported by this operation!");
+        T casted_alpha = type_convert<T>(alpha_);
+        y              = x > 0 ? x : casted_alpha * ck::math::expm1(x);
+    }
+    const float alpha_;
 };
 
 } // namespace element_wise
diff --git a/include/ck/utility/math.hpp b/include/ck/utility/math.hpp
index c5e967c8f4..7efbb3e63a 100644
--- a/include/ck/utility/math.hpp
+++ b/include/ck/utility/math.hpp
@@ -150,28 +150,6 @@ __host__ __device__ constexpr T clamp(const T& x, const T& lowerbound, const T&
     return min(max(x, lowerbound), upperbound);
 }
 
-// disallow implicit type casting
-template <typename T>
-__device__ T exp(T x);
-
-// TODO: add f16 support using v_exp_f16
-
-template <>
-__device__ float exp<float>(float x)
-{
-    return __expf(x);
-}
-
-template <>
-__device__ double exp<double>(double x)
-{
-    return exp(x);
-}
-
-static inline __host__ float exp(float x) { return std::expf(x); }
-
-static inline __host__ double exp(double x) { return std::exp(x); }
-
 // greatest common divisor, aka highest common factor
 __host__ __device__ constexpr index_t gcd(index_t x, index_t y)
 {
diff --git a/include/ck/utility/math_v2.hpp b/include/ck/utility/math_v2.hpp
index 1cac2cc0c7..a07fde3da3 100644
--- a/include/ck/utility/math_v2.hpp
+++ b/include/ck/utility/math_v2.hpp
@@ -9,6 +9,7 @@
 
 #include "ck/utility/data_type.hpp"
 #include "ck/utility/type.hpp"
+#include "ck/utility/type_convert.hpp"
 
 namespace ck {
 namespace math {
@@ -92,14 +93,96 @@ static inline __host__ float sqrt(float x) { return std::sqrt(x); };
 
 static inline __host__ double sqrt(double x) { return std::sqrt(x); };
 
-static inline __host__ half_t tanh(half_t x)
+template <typename T>
+inline __host__ T tanh(T x)
 {
-    return static_cast<half_t>(std::tanh(static_cast<float>(x)));
+    return ck::type_convert<T>(std::tanhf(ck::type_convert<float>(x)));
 };
 
-static inline __host__ float tanh(float x) { return std::tanh(x); };
+template <>
+inline __host__ float tanh<float>(float x)
+{
+    return std::tanhf(x);
+};
 
-static inline __host__ double tanh(double x) { return std::tanh(x); };
+template <>
+inline __host__ double tanh<double>(double x)
+{
+    return std::tanh(x);
+};
+
+template <typename T>
+inline __host__ T exp(T x)
+{
+    return ck::type_convert<T>(std::expf(ck::type_convert<float>(x)));
+}
+
+template <>
+inline __host__ float exp<float>(float x)
+{
+    return std::expf(x);
+}
+
+template <>
+inline __host__ double exp<double>(double x)
+{
+    return std::exp(x);
+}
+
+template <typename T>
+inline __host__ T log(T x)
+{
+    return ck::type_convert<T>(std::logf(ck::type_convert<float>(x)));
+}
+
+template <>
+inline __host__ float log<float>(float x)
+{
+    return std::logf(x);
+}
+
+template <>
+inline __host__ double log<double>(double x)
+{
+    return std::log(x);
+}
+
+template <typename T>
+inline __host__ T pow(T x, T gamma)
+{
+    return ck::type_convert<T>(
+        std::powf(ck::type_convert<float>(x), ck::type_convert<float>(gamma)));
+}
+
+template <>
+inline __host__ float pow<float>(float x, float gamma)
+{
+    return std::powf(x, gamma);
+}
+
+template <>
+inline __host__ double pow<double>(double x, double gamma)
+{
+    return std::pow(x, gamma);
+}
+
+template <typename T>
+inline __host__ T expm1(T x)
+{
+    return ck::type_convert<T>(std::expm1f(ck::type_convert<float>(x)));
+}
+
+template <>
+inline __host__ float expm1<float>(float x)
+{
+    return std::expm1f(x);
+}
+
+template <>
+inline __host__ double expm1<double>(double x)
+{
+    return std::expm1(x);
+}
 
 // math functions for the HIP kernel,  some are implemented by calling hip builtin functions
 
@@ -181,14 +264,107 @@ static inline __device__ float sqrt(float x) { return __builtin_amdgcn_sqrtf(x);
 
 static inline __device__ double sqrt(double x) { return __builtin_amdgcn_sqrt(x); };
 
-static inline __device__ half_t tanh(half_t x)
+template <typename T>
+inline __device__ T tanh(T x)
 {
-    return static_cast<half_t>(::tanhf(static_cast<float>(x)));
+    return ck::type_convert<T>(::tanhf(ck::type_convert<float>(x)));
 };
 
-static inline __device__ float tanh(float x) { return ::tanhf(x); };
+template <>
+inline __device__ float tanh<float>(float x)
+{
+    return ::tanhf(x);
+};
 
-static inline __device__ double tanh(double x) { return ::tanh(x); };
+template <>
+inline __device__ double tanh<double>(double x)
+{
+    return ::tanh(x);
+};
+
+template <typename T>
+inline __device__ T exp(T x)
+{
+    return ck::type_convert<T>(__expf(ck::type_convert<float>(x)));
+};
+
+template <>
+inline __device__ half_t exp<half_t>(half_t x)
+{
+    return hexp(x);
+};
+
+template <>
+inline __device__ float exp<float>(float x)
+{
+    return __expf(x);
+};
+
+template <>
+inline __device__ double exp<double>(double x)
+{
+    return exp(x);
+};
+
+template <typename T>
+inline __device__ T log(T x)
+{
+    return ck::type_convert<T>(__logf(ck::type_convert<float>(x)));
+};
+
+template <>
+inline __device__ half_t log<half_t>(half_t x)
+{
+    return hlog(x);
+};
+
+template <>
+inline __device__ float log<float>(float x)
+{
+    return __logf(x);
+};
+
+template <>
+inline __device__ double log<double>(double x)
+{
+    return log(x);
+};
+
+template <typename T>
+inline __device__ T pow(T x, T gamma)
+{
+    return ck::type_convert<T>(powf(ck::type_convert<float>(x), ck::type_convert<float>(gamma)));
+};
+
+template <>
+inline __device__ float pow<float>(float x, float gamma)
+{
+    return powf(x, gamma);
+};
+
+template <>
+inline __device__ double pow<double>(double x, double gamma)
+{
+    return pow(x, gamma);
+};
+
+template <typename T>
+inline __device__ T expm1(T x)
+{
+    return ck::type_convert<T>(expm1f(ck::type_convert<float>(x)));
+};
+
+template <>
+inline __device__ float expm1<float>(float x)
+{
+    return expm1f(x);
+};
+
+template <>
+inline __device__ double expm1<double>(double x)
+{
+    return expm1(x);
+};
 
 } // namespace math
 } // namespace ck
diff --git a/include/ck/utility/statically_indexed_array_multi_index.hpp b/include/ck/utility/statically_indexed_array_multi_index.hpp
index 4a8b96ae8a..80a865ed87 100644
--- a/include/ck/utility/statically_indexed_array_multi_index.hpp
+++ b/include/ck/utility/statically_indexed_array_multi_index.hpp
@@ -5,6 +5,7 @@
 #define CK_STATICALLY_INDEXED_ARRAY_MULTI_INDEX_HPP
 
 #include "common_header.hpp"
+#include "ck/utility/math_v2.hpp"
 
 namespace ck {
 
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
index 8f4182a231..0be9d83ad3 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_conv_fwd.hpp
@@ -128,11 +128,9 @@ struct ReferenceConvFwd : public device::BaseOperator
                         }
                     }
 
-                    float v_out;
-
-                    arg.out_element_op_(v_out, v_acc);
-
-                    arg.output_(g, n, k, wo) = ck::type_convert<OutDataType>(v_out);
+                    OutDataType v_out;
+                    arg.out_element_op_(v_out, ck::type_convert<OutDataType>(v_acc));
+                    arg.output_(g, n, k, wo) = v_out;
                 };
 
                 make_ParallelTensorFunctor(func,
@@ -184,11 +182,9 @@ struct ReferenceConvFwd : public device::BaseOperator
                         }
                     }
 
-                    float v_out;
-
-                    arg.out_element_op_(v_out, v_acc);
-
-                    arg.output_(g, n, k, ho, wo) = ck::type_convert<OutDataType>(v_out);
+                    OutDataType v_out;
+                    arg.out_element_op_(v_out, ck::type_convert<OutDataType>(v_acc));
+                    arg.output_(g, n, k, ho, wo) = v_out;
                 };
 
                 make_ParallelTensorFunctor(func,
@@ -253,11 +249,9 @@ struct ReferenceConvFwd : public device::BaseOperator
                         }
                     }
 
-                    float v_out;
-
-                    arg.out_element_op_(v_out, v_acc);
-
-                    arg.output_(g, n, k, d_o, ho, wo) = ck::type_convert<OutDataType>(v_out);
+                    OutDataType v_out;
+                    arg.out_element_op_(v_out, ck::type_convert<OutDataType>(v_acc));
+                    arg.output_(g, n, k, d_o, ho, wo) = v_out;
                 };
 
                 make_ParallelTensorFunctor(func,

From b4fc4d0b8d247f013e53c04aede0cc45000f13a3 Mon Sep 17 00:00:00 2001
From: Qianfeng <qianfeng.zhang@amd.com>
Date: Fri, 20 Oct 2023 00:26:04 +0800
Subject: [PATCH 183/194] Misc fixes  (#994)

* reinterpret_cast to const char* in dumpBufferToFile to be compatible with both const and non-const input pointers

* Add seed input to GeneratorTensor_4 for normal_distribution generator

* Add GetTypeString() for DeviceElementwiseImpl

* Add HIP_CHECK_ERROR macro
---
 include/ck/host_utility/hip_check_error.hpp   | 15 +++++++++++++
 .../device/impl/device_elementwise_impl.hpp   | 22 +++++++++++++++++++
 .../ck/library/utility/host_common_util.hpp   |  2 +-
 .../library/utility/host_tensor_generator.hpp |  5 +++--
 4 files changed, 41 insertions(+), 3 deletions(-)

diff --git a/include/ck/host_utility/hip_check_error.hpp b/include/ck/host_utility/hip_check_error.hpp
index af7bebd9d6..3e44faecb6 100644
--- a/include/ck/host_utility/hip_check_error.hpp
+++ b/include/ck/host_utility/hip_check_error.hpp
@@ -3,8 +3,10 @@
 
 #pragma once
 
+#include <sstream>
 #include <hip/hip_runtime.h>
 
+// To be removed, which really does not tell the location of failed HIP functional call
 inline void hip_check_error(hipError_t x)
 {
     if(x != hipSuccess)
@@ -15,3 +17,16 @@ inline void hip_check_error(hipError_t x)
         throw std::runtime_error(ss.str());
     }
 }
+
+#define HIP_CHECK_ERROR(retval_or_funcall)                                         \
+    do                                                                             \
+    {                                                                              \
+        hipError_t _tmpVal = retval_or_funcall;                                    \
+        if(_tmpVal != hipSuccess)                                                  \
+        {                                                                          \
+            std::ostringstream ostr;                                               \
+            ostr << "HIP Function Failed (" << __FILE__ << "," << __LINE__ << ") " \
+                 << hipGetErrorString(_tmpVal);                                    \
+            throw std::runtime_error(ostr.str());                                  \
+        }                                                                          \
+    } while(0)
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
index e203468c61..37867f1eaa 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_elementwise_impl.hpp
@@ -296,6 +296,28 @@ struct DeviceElementwiseImpl
     {
         return std::make_unique<Invoker>();
     };
+
+    std::string GetTypeString() const override
+    {
+        auto str = std::stringstream();
+
+        // clang-format off
+        str << "DeviceElementwiseImpl<" ;
+        str << "NumDim_" << NumDim << ","; 
+	str << "MPerThread_" << MPerThread << ","; 
+
+        str << "InScalarPerVector"; 
+        static_for<0, InScalarPerVectorSeq::Size(), 1>{}([&](auto i) { str << "_" << InScalarPerVectorSeq::At(i).value; });
+        str << ","; 
+        str << "OutScalarPerVector"; 
+        static_for<0, OutScalarPerVectorSeq::Size(), 1>{}([&](auto i) { str << "_" << OutScalarPerVectorSeq::At(i).value; });
+
+        str << ">";
+        // clang-format on
+
+        return str.str();
+    }
+
 }; // namespace device
 
 } // namespace device
diff --git a/library/include/ck/library/utility/host_common_util.hpp b/library/include/ck/library/utility/host_common_util.hpp
index 20a8f234db..16d6e2233a 100644
--- a/library/include/ck/library/utility/host_common_util.hpp
+++ b/library/include/ck/library/utility/host_common_util.hpp
@@ -22,7 +22,7 @@ static inline void dumpBufferToFile(const char* fileName, T* data, size_t dataNu
     std::ofstream outFile(fileName, std::ios::binary);
     if(outFile)
     {
-        outFile.write(reinterpret_cast<char*>(data), dataNumItems * sizeof(T));
+        outFile.write(reinterpret_cast<const char*>(data), dataNumItems * sizeof(T));
         outFile.close();
         std::cout << "Write output to file " << fileName << std::endl;
     }
diff --git a/library/include/ck/library/utility/host_tensor_generator.hpp b/library/include/ck/library/utility/host_tensor_generator.hpp
index 6b531c4c6f..6fd7ed8aa8 100644
--- a/library/include/ck/library/utility/host_tensor_generator.hpp
+++ b/library/include/ck/library/utility/host_tensor_generator.hpp
@@ -200,10 +200,11 @@ struct GeneratorTensor_3<ck::bf8_t>
 template <typename T>
 struct GeneratorTensor_4
 {
-    std::default_random_engine generator;
+    std::mt19937 generator;
     std::normal_distribution<float> distribution;
 
-    GeneratorTensor_4(float mean, float stddev) : generator(1), distribution(mean, stddev){};
+    GeneratorTensor_4(float mean, float stddev, unsigned int seed = 1)
+        : generator(seed), distribution(mean, stddev){};
 
     template <typename... Is>
     T operator()(Is...)

From f7331c603b87b940419715360cb81600d3338b0e Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Thu, 19 Oct 2023 09:34:39 -0700
Subject: [PATCH 184/194] Fix the DL kernel issues on Navi3x. (#998)

* apply the patch for dl kernels on gfx11

* build DL kernels on navi32 CI
---
 Jenkinsfile                          | 4 ++--
 include/ck/ck.hpp                    | 4 ++++
 include/ck/utility/inner_product.hpp | 2 ++
 3 files changed, 8 insertions(+), 2 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index 38bd3d7df4..021c19f150 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -790,8 +790,8 @@ pipeline {
                     }
                     agent{ label rocmnode("navi32") }
                     environment{
-                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1101" """
-                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
+                        setup_args = """ -DCMAKE_INSTALL_PREFIX=../install -DGPU_TARGETS="gfx1101" -DDL_KERNELS=ON """
+                        execute_args = """ cd ../client_example && rm -rf build && mkdir build && cd build && cmake -D CMAKE_PREFIX_PATH="${env.WORKSPACE}/install;/opt/rocm" -DGPU_TARGETS="gfx1101" -DDL_KERNELS=ON -D CMAKE_CXX_COMPILER="${build_compiler()}" .. && make -j """
                     }
                     steps{
                         Build_CK_and_Reboot(setup_args: setup_args, config_targets: "install", no_reboot:true, build_type: 'Release', execute_cmd: execute_args, prefixpath: '/usr/local')
diff --git a/include/ck/ck.hpp b/include/ck/ck.hpp
index 069ff7fc74..1e41404192 100644
--- a/include/ck/ck.hpp
+++ b/include/ck/ck.hpp
@@ -66,6 +66,10 @@
 #define CK_USE_AMD_V_FMAC_F32
 #define CK_USE_AMD_V_DOT2_F32_F16
 #define CK_USE_AMD_V_DOT4_I32_I8
+#elif defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__)
+#define CK_USE_AMD_V_FMAC_F32
+#define CK_USE_AMD_V_DOT2_F32_F16
+#define CK_USE_AMD_V_DOT4_I32_I8_GFX11
 #endif
 
 // MFMA instruction
diff --git a/include/ck/utility/inner_product.hpp b/include/ck/utility/inner_product.hpp
index 16c139c008..65efaf388a 100644
--- a/include/ck/utility/inner_product.hpp
+++ b/include/ck/utility/inner_product.hpp
@@ -192,6 +192,8 @@ inner_product<int8x4_t, int8x4_t, int32_t>(const int8x4_t& a, const int8x4_t& b,
 #else
     c = __builtin_amdgcn_sdot4(bit_cast<int32_t>(a), bit_cast<int32_t>(b), c, false);
 #endif
+#elif defined(CK_USE_AMD_V_DOT4_I32_I8_GFX11)
+    c = __builtin_amdgcn_sudot4(true, bit_cast<int32_t>(a), true, bit_cast<int32_t>(b), c, false);
 #else
     const vector_type<int8_t, 4> a_vector{a};
     const vector_type<int8_t, 4> b_vector{b};

From 1fd27d520f6aaa54c20cd6cc4439fb9c9f60c508 Mon Sep 17 00:00:00 2001
From: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
Date: Fri, 20 Oct 2023 08:00:45 -0500
Subject: [PATCH 185/194] Fix bf8 conversion issues (#1003)

* Fix the conversion

* Add bf8 functionality

* Enable example on MI200 as well
---
 .../20_grouped_conv_bwd_weight/CMakeLists.txt |   6 +-
 .../element/unary_element_wise_operation.hpp  |   3 +-
 include/ck/utility/data_type.hpp              |   7 +-
 include/ck/utility/f8_utils.hpp               | 151 ++++++++++++------
 include/ck/utility/type_convert.hpp           |  24 +--
 5 files changed, 116 insertions(+), 75 deletions(-)

diff --git a/example/20_grouped_conv_bwd_weight/CMakeLists.txt b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
index 2b0c4a28ce..c28fca6fa2 100644
--- a/example/20_grouped_conv_bwd_weight/CMakeLists.txt
+++ b/example/20_grouped_conv_bwd_weight/CMakeLists.txt
@@ -10,10 +10,8 @@ foreach(gpu IN LISTS GPU_TARGETS)
         add_example_executable(example_grouped_conv_bwd_weight_xdl_bf16 grouped_conv_bwd_weight_xdl_bf16.cpp)
         add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_bf16)
 
-        if(GPU_TARGETS MATCHES "gfx940" OR GPU_TARGETS MATCHES "gfx941" OR GPU_TARGETS MATCHES "gfx942")
-            add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
-            add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
-        endif()
+        add_example_executable(example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8 grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8.cpp)
+        add_example_dependencies(example_grouped_conv_bwd_weight example_grouped_conv_bwd_weight_xdl_fp16_comp_bf8_fp8)
         set(target 1)
     endif()
 
diff --git a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
index dabdf649e4..af48c6c1af 100644
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
@@ -207,7 +207,8 @@ struct ConvertF8SR
     __host__ __device__ void operator()(Y& y, const X& x) const
     {
         // check Y datatype
-        static_assert(is_same<Y, f8_t>::value, "Data type is not supported by this operation!");
+        static_assert(is_same<Y, f8_t>::value || is_same<Y, bf8_t>::value,
+                      "Data type is not supported by this operation!");
 
         // check X datatype
         static_assert(is_same<X, float>::value || is_same<X, half_t>::value,
diff --git a/include/ck/utility/data_type.hpp b/include/ck/utility/data_type.hpp
index ceaca27a45..d367ad8df5 100644
--- a/include/ck/utility/data_type.hpp
+++ b/include/ck/utility/data_type.hpp
@@ -1075,6 +1075,7 @@ struct NumericUtils<float>
 {
     static constexpr int exp            = 8;
     static constexpr int mant           = 23;
+    static constexpr int bias           = 127;
     static constexpr uint32_t nan_mask  = 0x7F800000;
     static constexpr uint32_t head_mask = 0xFF800000;
     static constexpr uint32_t mant_mask = 0x7FFFFF;
@@ -1091,6 +1092,7 @@ struct NumericUtils<half_t>
 {
     static constexpr int exp            = 5;
     static constexpr int mant           = 10;
+    static constexpr int bias           = 15;
     static constexpr uint16_t nan_mask  = 0x7C00;
     static constexpr uint16_t head_mask = 0xFC00;
     static constexpr uint16_t mant_mask = 0x3FF;
@@ -1107,6 +1109,8 @@ struct NumericUtils<f8_t>
 {
     static constexpr int exp  = 4;
     static constexpr int mant = 3;
+    static constexpr int bias = 8; // negative zero nan mode
+    // static constexpr int bias = 7; // ieee mode
 };
 
 template <>
@@ -1114,6 +1118,7 @@ struct NumericUtils<bf8_t>
 {
     static constexpr int exp  = 5;
     static constexpr int mant = 2;
+    static constexpr int bias = 16; // negative zero nan mode
+    // static constexpr int bias = 15; // ieee mode
 };
-//
 } // namespace ck
diff --git a/include/ck/utility/f8_utils.hpp b/include/ck/utility/f8_utils.hpp
index b63c82fe9a..1960667732 100644
--- a/include/ck/utility/f8_utils.hpp
+++ b/include/ck/utility/f8_utils.hpp
@@ -5,7 +5,6 @@
 
 #include "ck/utility/data_type.hpp"
 
-// these conversions are disabled if native conversions available
 namespace ck {
 
 // fp8 rounding modes
@@ -17,6 +16,9 @@ enum class f8_rounding_mode
     stochastic
 };
 
+__host__ inline int clz(uint32_t x) { return __builtin_clz(x); }
+__device__ inline int clz(uint32_t x) { return __clz(x); }
+
 } // namespace ck
 
 namespace ck::utils {
@@ -34,7 +36,7 @@ __host__ __device__ Y run_cast_to_f8(X x, uint32_t rng)
     constexpr int in_exp  = NumericUtils<X>::exp;
     constexpr int in_mant = NumericUtils<X>::mant;
 
-    int exponent;
+    int exponent, bias;
     uint32_t head, mantissa, sign;
     // nan code is same for float and half
     constexpr Y nan_code        = 0x80;
@@ -49,12 +51,11 @@ __host__ __device__ Y run_cast_to_f8(X x, uint32_t rng)
     mantissa = x_bitwise & NumericUtils<X>::mant_mask;
     exponent = (head >> in_mant) & NumericUtils<X>::exp_mask;
     sign     = head >> (in_exp + in_mant);
+    bias     = NumericUtils<X>::bias;
 
     uint32_t signed_inf   = (sign << (in_exp + in_mant)) + (((1 << in_exp) - 1) << in_mant);
     uint32_t drop_mask    = (1 << (in_mant - out_mant)) - 1;
     constexpr int max_exp = (1 << out_exp) - (negative_zero_nan ? 1 : 2);
-    constexpr int exp_low_cutoff =
-        (1 << (in_exp - 1)) - (1 << (out_exp - 1)) + 1 - (negative_zero_nan ? 1 : 0);
 
     if constexpr(negative_zero_nan)
     {
@@ -67,56 +68,107 @@ __host__ __device__ Y run_cast_to_f8(X x, uint32_t rng)
             return signed_inf + (mantissa != 0 ? 1 : 0);
     }
 
-    // if input is half and output is bf8
-    if((NumericUtils<X>::mant == 10) && (NumericUtils<Y>::mant == 2) && negative_zero_nan &&
-       exponent == 0)
-    {
-        exponent += 1;
-        while(mantissa < (1 << in_mant))
-        {
-            mantissa <<= 1;
-            exponent -= 1;
-        }
-        mantissa &= ~(1 << in_mant);
-    }
-
     // check if x is 0.0
     if(x_bitwise == 0)
         return 0;
 
-    exponent -= exp_low_cutoff - 1;
-    if(exponent <= 0)
-        drop_mask = (1 << (in_mant - out_mant + 1 - exponent)) - 1;
-    mantissa += 1 << in_mant;
-    // apply random number if needed
-    mantissa += (stoch ? rng : mantissa) & drop_mask;
-    if(mantissa >= (2 << in_mant))
-    {
-        mantissa >>= 1;
-        exponent++;
-    }
-    mantissa >>= (in_mant - out_mant);
+    // First need to check if it is normal or denorm as there is a difference of implict 1
+    // Then need to adjust the exponent to align with the F8 exponent, in the meanwhile, shift
+    // The mantissa. Then for stochastic rounding, add rng to mantissa and truncate. And for
+    // RNE, no need to add rng. Then probably need to check whether there is carry and adjust
+    // exponent and mantissa again3
 
-    // check negative exponent
-    if(exponent <= 0)
-    {
-        if(x_bitwise == 0)
-            return 0;
-        else
+    // For IEEE bias mode, the bias is 2^(k-1)-1 where k is the width of exponent bits
+    const int out_bias                  = (1 << (out_exp - 1)) - 1 + (negative_zero_nan ? 1 : 0);
+    const int out_denormal_act_exponent = 1 - out_bias; // actual exponent of f8 denormal
+    // act_exponent is the actual exponent of fp32/fp16 (after subtracting bias)
+    // out_exponent is the converted f8 exponent with bias encoding
+    // exponent_diff is the diff between fp32/fp16 exponent and f8 exponent,
+    // the difference needs to be adjusted and mantissa shifted
+    int act_exponent, out_exponent, exponent_diff;
+
+    if(exponent == 0)
+    { // fp32/fp16 is in denormal.
+        /* fp32 denormal is below 2^-127 so it is usually not a concern here, we mostly concern fp16
+here. In this case, f8 is usually in denormal. But there could be exceptions. fp16 denormal has
+exponent bias 15 while bf8 with NANOO has exponent bias 16. It means that there are some numbers in
+fp16 denormal but they are bf8 (NANOO) normals - smallest bf8 (NANOO) normal is 2^-15. fp16 numbers
+where exponent==0 (actual exponent -14) and highest bit of mantissa is 1 are bf8 (NANOO) normal.
+In this case, the fp16 mantissa should be shift left by 1 */
+        act_exponent  = exponent - bias + 1;
+        exponent_diff = out_denormal_act_exponent -
+                        act_exponent; // actual exponent is exponent-bias+1 as it is denormal
+    }
+    else
+    { // fp32/fp16 is normal with implicit 1
+        act_exponent = exponent - bias;
+        if(act_exponent <= out_denormal_act_exponent)
         {
-            // subnormal range; represented by a subnormal float8 (exponent 0)
-            // and involves loss of accuracy
-            mantissa >>= 1 - exponent;
-            exponent = 0;
+            /* This is the case where fp32/fp16 is normal but it is in f8 denormal range.
+   For example fp8 nanoo mode, denormal exponent is -7, but if the fp32/fp16
+   actual exponent is -7, it is actually larger due to the implict 1,
+   Therefore it needs to be adjust to -6 and mantissa shift right by 1.
+   So for fp32/fp16, exponent -8 is the cut point to convert to fp8 nanoo */
+            exponent_diff = out_denormal_act_exponent - act_exponent;
+        }
+        else
+        { // both fp32/fp16 and f8 are in normal range
+            exponent_diff =
+                0; // exponent_diff=0 does not mean there is no difference for this case,
+            // act_exponent could be larger. Just that it does not need shift mantissa
+        }
+        mantissa += (1 << in_mant); // Add the implicit 1 into mantissa
+    }
+
+    bool midpoint = (mantissa & ((1 << (in_mant - out_mant + exponent_diff)) - 1)) ==
+                    (1 << (in_mant - out_mant + exponent_diff - 1));
+    /* This part is a bit tricky. The judgment of whether it is a tie needs to be done before we
+ shift right as shift right could rip off some residual part and make something not midpoint look
+ like midpoint. For example, the fp16 number 0x1002 (0 00100 0000000010), it is larger than
+ midpoint, but after shift right by 4 bits, it would look like midpoint. */
+
+    if(exponent_diff > 0)
+        mantissa >>= exponent_diff;
+    else if(exponent_diff == -1)
+        mantissa <<= -exponent_diff;
+    bool implicit_one = mantissa & (1 << in_mant);
+    // if there is no implict 1, it  means the f8 is denormal and need to adjust to denorm exponent
+    out_exponent =
+        (act_exponent + exponent_diff) /*actual f8 exponent*/ + out_bias - (implicit_one ? 0 : 1);
+
+    // Now we have the exponent and mantissa adjusted
+    bool odd =
+        mantissa &
+        (1 << (in_mant - out_mant)); // if the least significant bit that is not truncated is 1
+    mantissa += (stoch ? rng : (midpoint ? (odd ? mantissa : mantissa - 1) : mantissa)) & drop_mask;
+
+    // Now we deal with overflow
+    if(out_exponent == 0)
+    {
+        if((1 << in_mant) & mantissa)
+        {
+            out_exponent = 1; // denormal overflow to become normal, promote exponent
+            // No need to make 1 implicit now as it will be addressed later
         }
     }
-    // above range: quantize to maximum possible float of the same sign
-    else if(exponent > max_exp)
+    else
+    {
+        if((1 << (in_mant + 1)) & mantissa)
+        {
+            mantissa >>= 1;
+            out_exponent++;
+            // No need to make 1 implicit now as it will be addressed later
+        }
+    }
+
+    mantissa >>= (in_mant - out_mant);
+
+    if(out_exponent > max_exp)
     {
         if(clip)
         {
-            mantissa = (1 << out_mant) - 1;
-            exponent = max_exp;
+            mantissa     = (1 << out_mant) - 1;
+            out_exponent = max_exp;
         }
         else
         {
@@ -125,10 +177,10 @@ __host__ __device__ Y run_cast_to_f8(X x, uint32_t rng)
     }
 
     // check if x is 0.0 or -0.0
-    if(exponent == 0 && mantissa == 0)
+    if(out_exponent == 0 && mantissa == 0)
         return negative_zero_nan ? 0 : (sign << (out_exp + out_mant));
     mantissa &= (1 << out_mant) - 1;
-    return (sign << (out_exp + out_mant)) | (exponent << out_mant) | mantissa;
+    return (sign << (out_exp + out_mant)) | (out_exponent << out_mant) | mantissa;
 }
 
 template <typename X, typename Y, bool negative_zero_nan>
@@ -194,12 +246,9 @@ __host__ __device__ Y run_cast_from_f8(X x)
     if(exponent == 0)
     {
         // guaranteed mantissa!=0 since cases 0x0 and 0x80 are handled above
-        exponent++;
-        while(mantissa < (1 << in_mant))
-        {
-            mantissa <<= 1;
-            exponent--;
-        }
+        int sh = 1 + clz(mantissa) - (32 - in_mant);
+        mantissa <<= sh;
+        exponent += 1 - sh;
         mantissa &= ((1 << in_mant) - 1);
     }
     exponent += exp_low_cutoff - 1;
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index aba8baa593..12d628a4bb 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -145,7 +145,7 @@ inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return type_convert<f8_t>(type_convert<float>(x));
-#elif 0
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -153,8 +153,6 @@ inline __host__ __device__ f8_t type_convert<f8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
-#else
-    return type_convert<f8_t>(type_convert<float>(x));
 #endif
 }
 
@@ -165,11 +163,9 @@ inline __host__ __device__ half_t type_convert<half_t, f8_t>(f8_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // use native conversion to float and convert to fp16
     return type_convert<half_t>(type_convert<float>(x));
-#elif 0
+#else
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<f8_t, half_t, negative_zero_nan>(x);
-#else
-    return type_convert<half_t>(type_convert<float>(x));
 #endif
 }
 
@@ -223,7 +219,7 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return type_convert<bf8_t>(type_convert<float>(x));
-#elif 0
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::standard;
@@ -231,8 +227,6 @@ inline __host__ __device__ bf8_t type_convert<bf8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
-#else
-    return type_convert<bf8_t>(type_convert<float>(x));
 #endif
 }
 
@@ -243,11 +237,9 @@ inline __host__ __device__ half_t type_convert<half_t, bf8_t>(bf8_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // use native conversion to float and convert to fp16
     return type_convert<half_t>(type_convert<float>(x));
-#elif 0
+#else
     constexpr bool negative_zero_nan = true;
     return utils::cast_from_f8<bf8_t, half_t, negative_zero_nan>(x);
-#else
-    return type_convert<half_t>(type_convert<float>(x));
 #endif
 }
 
@@ -347,7 +339,7 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return f8_convert_sr<f8_t>(type_convert<float>(x));
-#elif 0
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
@@ -356,8 +348,6 @@ inline __host__ __device__ f8_t f8_convert_sr<f8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, f8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
-#else
-    return f8_convert_sr<f8_t>(type_convert<float>(x));
 #endif
 }
 
@@ -396,7 +386,7 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
     // convert to float and use native converion
     return f8_convert_sr<f8_t>(type_convert<float>(x));
-#elif 0
+#else
     constexpr bool negative_zero_nan = true;
     constexpr bool clip              = true;
     constexpr f8_rounding_mode rm    = f8_rounding_mode::stochastic;
@@ -406,8 +396,6 @@ inline __host__ __device__ bf8_t f8_convert_sr<bf8_t, half_t>(half_t x)
     return utils::
         cast_to_f8<half_t, bf8_t, negative_zero_nan, clip, (rm == f8_rounding_mode::stochastic)>(
             x, rng);
-#else
-    return f8_convert_sr<bf8_t>(type_convert<float>(x));
 #endif
 }
 

From ac0e006766185e5fb6342197f4a00599551beb26 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Sat, 21 Oct 2023 22:19:43 +0200
Subject: [PATCH 186/194] Fix cmake dtype check (#989)

* Fix instances dtype check

* Fix source dtypes seletor for examples and tests

* Sync with new cmakefile changes

* Remove not needed ifdefs

* Remove not needed ifdefs
---
 example/CMakeLists.txt                        | 92 +++++++++----------
 .../gpu/CMakeLists.txt                        |  4 +
 .../grouped_conv3d_bwd_weight/CMakeLists.txt  |  8 +-
 test/CMakeLists.txt                           | 92 +++++++++----------
 test/data_type/CMakeLists.txt                 |  6 +-
 test/data_type/{bf8.cpp => test_bf8.cpp}      |  0
 test/data_type/{fp8.cpp => test_fp8.cpp}      |  0
 test/data_type/{int4.cpp => test_int4.cpp}    |  0
 8 files changed, 97 insertions(+), 105 deletions(-)
 rename test/data_type/{bf8.cpp => test_bf8.cpp} (100%)
 rename test/data_type/{fp8.cpp => test_fp8.cpp} (100%)
 rename test/data_type/{int4.cpp => test_int4.cpp} (100%)

diff --git a/example/CMakeLists.txt b/example/CMakeLists.txt
index 5a53982bb2..c19ba93b69 100644
--- a/example/CMakeLists.txt
+++ b/example/CMakeLists.txt
@@ -11,31 +11,27 @@ function(add_example_executable EXAMPLE_NAME FILE_NAME)
     if(DEFINED DTYPES)
     foreach(source IN LISTS FILE_NAME)
         set(test 0)
-        foreach(type IN LISTS DTYPES)
-            if(type MATCHES "fp16")
-                set(type1 "_f16")
-            elseif(type MATCHES "fp32")
-                set(type1 "_f32")
-            elseif(type MATCHES "fp8")
-                set(type1 "_f8")
-            elseif(type MATCHES "bf16")
-                set(type1 "_b16")
-            elseif(type MATCHES "fp64")
-                set(type1 "_f64")
-            elseif(type MATCHES "int8")
-                set(type1 "_i8")
-            endif()
-            if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
-                #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                set(test 0)
-                break()
-            elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                NOT(source MATCHES type OR source MATCHES type1))
-                    #if filename contains a type which doesn't match any selected type, mark it for removal
-                    set(test 1)
-            endif()
-        endforeach()
+        if((source MATCHES "_fp16" OR source MATCHES "_f16") AND NOT "fp16" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_fp32" OR source MATCHES "_f32") AND NOT "fp32" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_fp64" OR source MATCHES "_f64") AND NOT "fp64" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_fp8" OR source MATCHES "_f8") AND NOT "fp8" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_bf8" OR source MATCHES "_bf8") AND NOT "bf8" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_bf16" OR source MATCHES "_b16") AND NOT "bf16" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_int8" OR source MATCHES "_i8") AND NOT "int8" IN_LIST DTYPES)
+            set(test 1)
+        endif()
         if(test EQUAL 1)
             message("removing example source file ${source} ")
             list(REMOVE_ITEM FILE_NAME "${source}")
@@ -74,31 +70,27 @@ function(add_example_executable_no_testing EXAMPLE_NAME FILE_NAME)
     if(DEFINED DTYPES)
     foreach(source IN LISTS FILE_NAME)
         set(test 0)
-        foreach(type IN LISTS DTYPES)
-                if(type MATCHES "fp16")
-                    set(type1 "_f16")
-                elseif(type MATCHES "fp32")
-                    set(type1 "_f32")
-                elseif(type MATCHES "fp8")
-                    set(type1 "_f8")
-                elseif(type MATCHES "bf16")
-                    set(type1 "_b16")
-                elseif(type MATCHES "fp64")
-                    set(type1 "_f64")
-                elseif(type MATCHES "int8")
-                    set(type1 "_i8")
-                endif()
-                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
-                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                    set(test 0)
-                    break()
-                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                  source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                  NOT(source MATCHES type OR source MATCHES type1))
-                    #if filename contains a type which doesn't match any selected type, mark it for removal
-                    set(test 1)
-                endif()
-        endforeach()
+        if((source MATCHES "_fp16" OR source MATCHES "_f16") AND NOT "fp16" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_fp32" OR source MATCHES "_f32") AND NOT "fp32" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_fp64" OR source MATCHES "_f64") AND NOT "fp64" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_fp8" OR source MATCHES "_f8") AND NOT "fp8" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_bf8" OR source MATCHES "_bf8") AND NOT "bf8" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_bf16" OR source MATCHES "_b16") AND NOT "bf16" IN_LIST DTYPES)
+            set(test 1)
+        endif()
+        if((source MATCHES "_int8" OR source MATCHES "_i8") AND NOT "int8" IN_LIST DTYPES)
+            set(test 1)
+        endif()
         if(test EQUAL 1)
             message("removing example ${source} ")
             list(REMOVE_ITEM FILE_NAME "${source}")
diff --git a/library/src/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
index e1952a886c..9cb5d0e9aa 100644
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
@@ -69,6 +69,10 @@ FOREACH(subdir_path ${dir_list})
             message("fp8 instance found!")
             set(add_inst 1)
         endif()
+        if(("${cmake_instance}" MATCHES "_bf8" OR "${cmake_instance}" MATCHES "_b8") AND DTYPES MATCHES "bf8")
+            message("bf8 instance found!")
+            set(add_inst 1)
+        endif()
         if(("${cmake_instance}" MATCHES "_fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
             message("fp16 instance found!")
             set(add_inst 1)
diff --git a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
index a2c4b1f80b..968e8dea2f 100644
--- a/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/grouped_conv3d_bwd_weight/CMakeLists.txt
@@ -4,8 +4,7 @@ set(GROUPED_CONV3D_BWD_WEIGHT
     xdl/device_grouped_conv3d_bwd_weight_xdl_gndhwc_gkzyxc_gndhwk_bf16_instance.cpp
     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_instance.cpp
     xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f32_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp
-    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp)
+    xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_bf16_instance.cpp)
 
 if(DL_KERNELS)
     list(APPEND GROUPED_CONV3D_BWD_WEIGHT
@@ -27,4 +26,9 @@ list(APPEND GROUPED_CONV3D_BWD_WEIGHT
     wmma/device_grouped_conv3d_bwd_weight_wmma_gndhwc_gkzyxc_gndhwk_i8_instance.cpp
     wmma/device_grouped_conv3d_bwd_weight_wmma_ndhwgc_gkzyxc_ndhwgk_i8_instance.cpp)
 
+if((DTYPES MATCHES "fp8" AND DTYPES MATCHES "bf8" AND DTYPES MATCHES "fp16") OR NOT DEFINED DTYPES)
+    list(APPEND GROUPED_CONV3D_BWD_WEIGHT
+      xdl/device_grouped_conv3d_bwd_weight_xdl_ndhwgc_gkzyxc_ndhwgk_f16_comp_bf8_fp8_instance.cpp)
+endif()
+
 add_instance_library(device_grouped_conv3d_bwd_weight_instance ${GROUPED_CONV3D_BWD_WEIGHT})
diff --git a/test/CMakeLists.txt b/test/CMakeLists.txt
index 265f428b55..1567d8bc69 100644
--- a/test/CMakeLists.txt
+++ b/test/CMakeLists.txt
@@ -13,31 +13,27 @@ function(add_test_executable TEST_NAME)
     if(DEFINED DTYPES)
         foreach(source IN LISTS ARGN)
             set(test 0)
-            foreach(type IN LISTS DTYPES)
-                if(type MATCHES "fp16")
-                    set(type1 "_f16")
-                elseif(type MATCHES "fp32")
-                    set(type1 "_f32")
-                elseif(type MATCHES "fp8")
-                    set(type1 "_f8")
-                elseif(type MATCHES "bf16")
-                    set(type1 "_b16")
-                elseif(type MATCHES "fp64")
-                    set(type1 "_f64")
-                elseif(type MATCHES "int8")
-                    set(type1 "_i8")
-                endif()
-                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
-                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                    set(test 0)
-                    break()
-                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                    source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                    NOT(source MATCHES type OR source MATCHES type1))
-                    #if filename contains a type which doesn't match any selected type, mark it for removal
-                    set(test 1)
-                endif()
-            endforeach()
+            if((source MATCHES "_fp16" OR source MATCHES "_f16") AND NOT "fp16" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_fp32" OR source MATCHES "_f32") AND NOT "fp32" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_fp64" OR source MATCHES "_f64") AND NOT "fp64" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_fp8" OR source MATCHES "_f8") AND NOT "fp8" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_bf8" OR source MATCHES "_bf8") AND NOT "bf8" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_bf16" OR source MATCHES "_b16") AND NOT "bf16" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_int8" OR source MATCHES "_i8") AND NOT "int8" IN_LIST DTYPES)
+                set(test 1)
+            endif()
             if(test EQUAL 1)
                 message("removing test ${source} ")
                 list(REMOVE_ITEM ARGN "${source}")
@@ -72,31 +68,27 @@ function(add_gtest_executable TEST_NAME)
     if(DEFINED DTYPES)
         foreach(source IN LISTS ARGN)
             set(test 0)
-            foreach(type IN LISTS DTYPES)
-                if(type MATCHES "fp16")
-                    set(type1 "_f16")
-                elseif(type MATCHES "fp32")
-                    set(type1 "_f32")
-                elseif(type MATCHES "fp8")
-                    set(type1 "_f8")
-                elseif(type MATCHES "bf16")
-                    set(type1 "_b16")
-                elseif(type MATCHES "fp64")
-                    set(type1 "_f64")
-                elseif(type MATCHES "int8")
-                    set(type1 "_i8")
-                endif()
-                if("${source}" MATCHES "${type}" OR "${source}" MATCHES "${type1}")
-                    #if filename matches any selected type, exit type loop and do no exclude the file from the list
-                    set(test 0)
-                    break()
-                elseif((source MATCHES "fp8" OR source MATCHES "fp32" OR source MATCHES "fp64" OR source MATCHES "bf16" OR source MATCHES "int8" OR source MATCHES "fp16" OR
-                    source MATCHES "_f8" OR source MATCHES "_f32" OR source MATCHES "_f64" OR source MATCHES "_i8" OR source MATCHES "_f16" OR source MATCHES "_b16") AND
-                    NOT(source MATCHES type OR source MATCHES type1))
-                    #if filename contains a type which doesn't match any selected type, mark it for removal
-                    set(test 1)
-                endif()
-            endforeach()
+            if((source MATCHES "_fp16" OR source MATCHES "_f16") AND NOT "fp16" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_fp32" OR source MATCHES "_f32") AND NOT "fp32" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_fp64" OR source MATCHES "_f64") AND NOT "fp64" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_fp8" OR source MATCHES "_f8") AND NOT "fp8" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_bf8" OR source MATCHES "_bf8") AND NOT "bf8" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_bf16" OR source MATCHES "_b16") AND NOT "bf16" IN_LIST DTYPES)
+                set(test 1)
+            endif()
+            if((source MATCHES "_int8" OR source MATCHES "_i8") AND NOT "int8" IN_LIST DTYPES)
+                set(test 1)
+            endif()
             if(test EQUAL 1)
                 message("removing gtest ${source} ")
                 list(REMOVE_ITEM ARGN "${source}")
diff --git a/test/data_type/CMakeLists.txt b/test/data_type/CMakeLists.txt
index 2409ca05c2..0ebfc931ac 100644
--- a/test/data_type/CMakeLists.txt
+++ b/test/data_type/CMakeLists.txt
@@ -1,15 +1,15 @@
 if (USE_BITINT_EXTENSION_INT4)
-  add_gtest_executable(test_int4 int4.cpp)
+  add_gtest_executable(test_int4 test_int4.cpp)
   if(result EQUAL 0)
     target_link_libraries(test_int4 PRIVATE utility)
   endif()
 endif()
 
-add_gtest_executable(test_fp8 fp8.cpp)
+add_gtest_executable(test_fp8 test_fp8.cpp)
 if(result EQUAL 0)
   target_link_libraries(test_fp8 PRIVATE utility)
 endif()
-add_gtest_executable(test_bf8 bf8.cpp)
+add_gtest_executable(test_bf8 test_bf8.cpp)
 if(result EQUAL 0)
   target_link_libraries(test_bf8 PRIVATE utility)
 endif()
diff --git a/test/data_type/bf8.cpp b/test/data_type/test_bf8.cpp
similarity index 100%
rename from test/data_type/bf8.cpp
rename to test/data_type/test_bf8.cpp
diff --git a/test/data_type/fp8.cpp b/test/data_type/test_fp8.cpp
similarity index 100%
rename from test/data_type/fp8.cpp
rename to test/data_type/test_fp8.cpp
diff --git a/test/data_type/int4.cpp b/test/data_type/test_int4.cpp
similarity index 100%
rename from test/data_type/int4.cpp
rename to test/data_type/test_int4.cpp

From bec84efbb73f19464dca9b2b2d7aadb2c070a977 Mon Sep 17 00:00:00 2001
From: zjing14 <zhangjing14@gmail.com>
Date: Mon, 23 Oct 2023 16:52:53 -0500
Subject: [PATCH 187/194] Enabled padding for regular gemm (#1004)

* add mnk padding for fp8

* add padding for row_col layout

* added padding for fp32

---------

Co-authored-by: Jing Zhang <jizha@amd.com>
---
 ..._shuffle_f16_f16_f16_mk_kn_mn_instance.cpp | 106 +++++++++---------
 ..._shuffle_f16_f16_f16_mk_nk_mn_instance.cpp |  88 ++++++++-------
 ..._shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp |  42 ++++---
 ..._shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp |  36 +++---
 4 files changed, 148 insertions(+), 124 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
index 31a9abe53a..ef797b13a7 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instance.cpp
@@ -29,7 +29,10 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+template <ck::tensor_operation::device::GemmSpecialization GemmSpec>
 using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple<
     // clang-format off
         //#####################| ALayout| BLayout| CLayout| AData| BData| CData| AccData| CShuffle|           A|           B|           C|           GEMM| NumGemmK| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|          LoopScheduler|                    Pipeline|
@@ -37,61 +40,61 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances = std::tuple<
         //#####################|        |        |        |      |      |      |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
         //#####################|        |        |        |      |      |      |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
         // pipeline v1, 1 wave
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>
 #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
         // pipeline v1, 2 waves
         ,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>
 #endif
 #if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
         // pipeline v2, 1 wave
         ,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   2,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   2,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   2,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   2,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   2,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              2,         0,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,      Row,    Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>
 #endif
     // clang-format on
     >;
@@ -101,8 +104,11 @@ void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances(
         DeviceGemm<Row, Row, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances{});
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances<GemmDefault>{});
+
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f16_f16_f16_mk_kn_mn_instances<MNKPadding>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
index 201fd93110..5f796829df 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instance.cpp
@@ -29,6 +29,9 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
+template <ck::tensor_operation::device::GemmSpecialization GemmSpec>
 // Compilation parameters for a[m, k] * b[n, k] = c[m, n]
 using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple<
     // clang-format off
@@ -37,52 +40,52 @@ using device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances = std::tuple<
         //#####################|        |        |        |      |      |      |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|                       |                            |
         //#####################|        |        |        |      |      |      |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |                       |                            |
         // pipeline v1, 1 wave
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v1>
 #if CK_EXPERIMENTAL_INTER_WAVE_INSTANCES        
         // pipeline v1, 2 waves
         ,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Interwave,        PipelineVersion::v1>
 #endif
 #if CK_EXPERIMENTAL_PIPELINE_V2_INSTANCES        
         // pipeline v2, 1 wave
         ,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    32,   8,   8,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    32,   8,   8,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    32,   8,   8,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    64,    32,   8,   8,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    32,   8,   8,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    32,   128,    32,   8,   8,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 8>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    32,    32,   8,   8,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,   F16,   F16,   F16,     F32,      F16, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    32,    64,    32,   8,   8,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              8,              8,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,              8,  LoopScheduler::Default,        PipelineVersion::v2>
 #endif
     // clang-format on
     >;
@@ -92,8 +95,11 @@ void add_device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances(
         DeviceGemm<Row, Col, Row, F16, F16, F16, PassThrough, PassThrough, PassThrough>>>&
         instances)
 {
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances{});
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances<GemmDefault>{});
+
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f16_f16_f16_mk_nk_mn_instances<MNKPadding>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
index e1aac45134..0b1bbe3af9 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_kn_mn_instance.cpp
@@ -27,7 +27,10 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // Compilation parameters for a[m, k] * b[k, n] = c[m, n]
+template <ck::tensor_operation::device::GemmSpecialization GemmSpec>
 using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances =
     std::tuple<
         // clang-format off
@@ -35,22 +38,22 @@ using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances =
         //#####################|        |        |        |   Type|   Type|   Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //#####################|        |        |        |       |       |       |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //#####################|        |        |        |       |       |       |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 2>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 2>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    64,  16,   4,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    64,  16,   4,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    64,  16,   4,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,             16,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    64,  16,   4,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<16,16, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    64,  16,   4,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<8, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              4,         0,           1,           1,               S<1, 64, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Row,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,             16,         1,           1,           1,               S<1, 64, 1, 4>,             16>
         // clang-format on
         >;
 
@@ -58,8 +61,11 @@ void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Row, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
 {
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances{});
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances<GemmDefault>{});
+
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f8_f8_f8_mk_kn_mn_instances<MNKPadding>{});
 }
 
 } // namespace instance
diff --git a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp
index 5c557312aa..2bef8ebbaf 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/gemm/device_gemm_xdl_c_shuffle_fp8_fp8_fp8_mk_nk_mn_instance.cpp
@@ -27,7 +27,10 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 
 static constexpr auto GemmDefault = ck::tensor_operation::device::GemmSpecialization::Default;
 
+static constexpr auto MNKPadding = ck::tensor_operation::device::GemmSpecialization::MNKPadding;
+
 // Compilation parameters for a[m, k] * b[n, k] = c[m, n]
+template <ck::tensor_operation::device::GemmSpecialization GemmSpec>
 using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances =
     std::tuple<
         // clang-format off
@@ -35,19 +38,19 @@ using device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances =
         //#####################|        |        |        |   Type|   Type|   Type|    Type| DataType| Elementwise| Elementwise| Elementwise| Specialization| Prefetch|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
         //#####################|        |        |        |       |       |       |        |         |   Operation|   Operation|   Operation|               |    Stage|      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
         //#####################|        |        |        |       |       |       |        |         |            |            |            |               |         |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |                |
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>,
-        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,    GemmDefault,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   256,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   256,    64,  16,  16,   32,   32,    2,    4,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,   128,    64,  16,  16,   32,   32,    4,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    64,   128,    64,  16,  16,   32,   32,    2,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    64,    64,  16,  16,   32,   32,    2,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,   128,    64,    64,  16,  16,   32,   32,    2,    1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   256,    64,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 64, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,   128,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 32, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,   128,    32,   128,    64,  16,  16,   32,   32,    1,    2,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 32, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 4>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    64,    32,    64,  16,  16,   32,   32,    2,    1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>,
+        DeviceGemm_Xdl_CShuffle<     Row,     Col,     Row,     F8,     F8,     F8,     F32,       F8, PassThrough, PassThrough, PassThrough,       GemmSpec,        1,    64,    32,    64,    64,  16,  16,   32,   32,    1,    2,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,             16,             16,         1,     S<4, 16, 1>,     S<1, 0, 2>,     S<1, 0, 2>,             2,              8,              8,         1,           1,           1,               S<1, 16, 1, 2>,             16>
         // clang-format on
         >;
 
@@ -55,8 +58,11 @@ void add_device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances(
     std::vector<std::unique_ptr<
         DeviceGemm<Row, Col, Row, F8, F8, F8, PassThrough, PassThrough, PassThrough>>>& instances)
 {
-    add_device_operation_instances(instances,
-                                   device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances{});
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances<GemmDefault>{});
+
+    add_device_operation_instances(
+        instances, device_gemm_xdl_c_shuffle_f8_f8_f8_mk_nk_mn_instances<MNKPadding>{});
 }
 
 } // namespace instance

From 6fe0bc7e72f297086ab4014a317f8dff5e147cd2 Mon Sep 17 00:00:00 2001
From: "dependabot[bot]" <49699333+dependabot[bot]@users.noreply.github.com>
Date: Thu, 26 Oct 2023 14:30:32 -0600
Subject: [PATCH 188/194] Bump sphinxcontrib-bibtex from 2.5.0 to 2.6.1 in
 /docs/sphinx (#871)

Bumps [sphinxcontrib-bibtex](https://github.com/mcmtroffaes/sphinxcontrib-bibtex) from 2.5.0 to 2.6.1.
- [Changelog](https://github.com/mcmtroffaes/sphinxcontrib-bibtex/blob/develop/CHANGELOG.rst)
- [Commits](https://github.com/mcmtroffaes/sphinxcontrib-bibtex/compare/2.5.0...2.6.1)

---
updated-dependencies:
- dependency-name: sphinxcontrib-bibtex
  dependency-type: direct:production
  update-type: version-update:semver-minor
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Sam Wu <sam.wu2@amd.com>
---
 docs/sphinx/requirements.in  | 2 +-
 docs/sphinx/requirements.txt | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/docs/sphinx/requirements.in b/docs/sphinx/requirements.in
index a12a00a2b2..c4ce8be79a 100644
--- a/docs/sphinx/requirements.in
+++ b/docs/sphinx/requirements.in
@@ -1,2 +1,2 @@
 rocm-docs-core>=0.20.0
-sphinxcontrib-bibtex==2.5.0
+sphinxcontrib-bibtex==2.6.1
diff --git a/docs/sphinx/requirements.txt b/docs/sphinx/requirements.txt
index 1a648a7493..a83a4e2738 100644
--- a/docs/sphinx/requirements.txt
+++ b/docs/sphinx/requirements.txt
@@ -139,7 +139,7 @@ sphinx-notfound-page==0.8.3
     # via rocm-docs-core
 sphinxcontrib-applehelp==1.0.4
     # via sphinx
-sphinxcontrib-bibtex==2.5.0
+sphinxcontrib-bibtex==2.6.1
     # via -r requirements.in
 sphinxcontrib-devhelp==1.0.2
     # via sphinx

From f46a6ffad83ab0245b8087df602f5fcfceb37ad2 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Fri, 27 Oct 2023 21:10:47 -0700
Subject: [PATCH 189/194] Fix the fp8 gemm for large tensors on MI300. (#1011)

* Fix the fp8 conversion

* Try clipping value before conversion

* Fix return

* Simplify with a const

* reduce the gemm input tensor values to reduce round-off error

* replace if-else with lambda

* fix syntax

---------

Co-authored-by: Rostyslav Geyyer <rosty.geyyer@amd.com>
---
 example/27_layernorm/run_layernorm_example.inc  | 8 ++++----
 example/42_groupnorm/run_groupnorm_example.inc  | 4 ++--
 include/ck/utility/type_convert.hpp             | 2 ++
 profiler/include/profiler/profile_gemm_impl.hpp | 4 ++--
 4 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/example/27_layernorm/run_layernorm_example.inc b/example/27_layernorm/run_layernorm_example.inc
index c8f599a39c..399165c36e 100644
--- a/example/27_layernorm/run_layernorm_example.inc
+++ b/example/27_layernorm/run_layernorm_example.inc
@@ -8,8 +8,8 @@ int run_groupnorm_example()
 {
     bool time_kernel = false;
 
-    ck::index_t M      = 1024;
-    ck::index_t N      = 1024;
+    ck::index_t M = 1024;
+    ck::index_t N = 1024;
 
     Tensor<XDataType> x({M, N});
     Tensor<GammaDataType> gamma({N});
@@ -44,9 +44,9 @@ int run_groupnorm_example()
         {0, 1},
         std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
         std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
-                                    save_mean.mDesc.GetStrides().end()},
+                                 save_mean.mDesc.GetStrides().end()},
         std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
-                                    save_mean.mDesc.GetStrides().end()},
+                                 save_mean.mDesc.GetStrides().end()},
         {1},
         1e-4,
         x_dev.GetDeviceBuffer(),
diff --git a/example/42_groupnorm/run_groupnorm_example.inc b/example/42_groupnorm/run_groupnorm_example.inc
index da41e90639..89117a9b94 100644
--- a/example/42_groupnorm/run_groupnorm_example.inc
+++ b/example/42_groupnorm/run_groupnorm_example.inc
@@ -65,9 +65,9 @@ int run_groupnorm_example(int argc, char* argv[])
         {0, 0, 0, C, 1},
         std::vector<ck::index_t>{y.mDesc.GetStrides().begin(), y.mDesc.GetStrides().end()},
         std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
-                                    save_mean.mDesc.GetStrides().end()},
+                                 save_mean.mDesc.GetStrides().end()},
         std::vector<ck::index_t>{save_mean.mDesc.GetStrides().begin(),
-                                    save_mean.mDesc.GetStrides().end()},
+                                 save_mean.mDesc.GetStrides().end()},
         {1, 2, 4}, // reduction dimension: [H, W, C]
         1e-6,
         x_dev.GetDeviceBuffer(),
diff --git a/include/ck/utility/type_convert.hpp b/include/ck/utility/type_convert.hpp
index 12d628a4bb..89b7dc5789 100644
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
@@ -100,6 +100,8 @@ template <>
 inline __host__ __device__ f8_t type_convert<f8_t, float>(float x)
 {
 #if defined(__gfx940__) || defined(__gfx941__) || defined(__gfx942__)
+    float max_fp8 = 240.0f;
+    x             = x > max_fp8 ? max_fp8 : (x < -max_fp8 ? -max_fp8 : x);
     union
     {
         float fval;
diff --git a/profiler/include/profiler/profile_gemm_impl.hpp b/profiler/include/profiler/profile_gemm_impl.hpp
index ccebb020c4..0ed225c934 100644
--- a/profiler/include/profiler/profile_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_impl.hpp
@@ -75,8 +75,8 @@ int profile_gemm_impl(int do_verification,
         b_k_n.GenerateTensorValue(GeneratorTensor_2<BDataType>{-5, 5});
         break;
     default:
-        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
-        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
+        a_m_k.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 0.1});
+        b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.05, 0.05});
     }
 
     using AElementOp = ck::tensor_operation::element_wise::PassThrough;

From 4e44a9e8da15ce9e2f32b490f008da8ad592c87e Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Mon, 30 Oct 2023 13:16:29 -0700
Subject: [PATCH 190/194] Enable sccache in the default docker and CI. (#1009)

* replace ccache with sccache, pin package versions

* put ccache back temporarily to avoid breaking other CI jobs

* add sccashe_wrapper.sh script

* fix the package version syntax

* fix the pymysql package issue

* run sccache_wrapper before build if ccache server found

* set the paths before calling the sccache_wrapper

* use /tmp instead of /usr/local for cache

* try using sccache --start-server instead of wrapper

* try using redis server with sccache

* define SCCACHE_REDIS

* add redis and ping packages, and redis port

* use the new sccache redis server

* do not use sccache with staging compiler

* fix the condition syntax

* add stunnel to redis

* add tunnel verification

* separate caches for different architectures

* fix syntax for the cache tag

* quse double brackets for conditions

* add bash line to the script

* add a switch for sccache and only use it in build stage

* run check_host function when enabling sccache

* fix the invocation tags for sccache

* fix groovy syntax

* set the invocation tag in groovy

* disable sccache in clang-format stage

* try another syntax for invocation tags

* use local sccache server if can't connect to redis

* fix script syntax

* update README

* refresh readme

* readme updates

* remove the timing and verification caveat from readme

---------

Co-authored-by: Lisa Delaney <lisa.delaney@amd.com>
---
 CMakeLists.txt            |  13 +--
 Dockerfile                |  22 +++-
 Jenkinsfile               |  95 +++++++++-------
 README.md                 | 229 ++++++++++++++++++++++----------------
 script/redis-cli.conf     |  10 ++
 script/sccache_wrapper.sh |  56 ++++++++++
 6 files changed, 274 insertions(+), 151 deletions(-)
 create mode 100644 script/redis-cli.conf
 create mode 100755 script/sccache_wrapper.sh

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 1d4ac3f14f..04674124cc 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -373,9 +373,10 @@ include_directories(BEFORE
 
 SET(BUILD_DEV ON CACHE BOOL "BUILD_DEV")
 if(BUILD_DEV)
-    add_compile_options(-Werror)
-    add_compile_options(-Weverything)
+    add_compile_options(-Werror -Weverything)
 endif()
+#add flags to reduce the size of binaries
+add_compile_options(-Oz -flto=thin)
 message("CMAKE_CXX_FLAGS: ${CMAKE_CXX_FLAGS}")
 
 add_custom_target(check COMMAND ${CMAKE_CTEST_COMMAND} --output-on-failure -C ${CMAKE_CFG_INTDIR})
@@ -390,35 +391,27 @@ IF(IS_DIRECTORY "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu
     file(READ "${PROJECT_SOURCE_DIR}/library/src/tensor_operation_instance/gpu/${subdir_path}/CMakeLists.txt" cmake_instance)
     set(add_inst 0)
     if(("${cmake_instance}" MATCHES "fp8" OR "${cmake_instance}" MATCHES "_f8") AND DTYPES MATCHES "fp8")
-        #message("fp8 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "bf8" OR "${cmake_instance}" MATCHES "_b8") AND DTYPES MATCHES "bf8")
-        #message("bf8 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "fp16" OR "${cmake_instance}" MATCHES "_f16") AND DTYPES MATCHES "fp16")
-        #message("fp16 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "fp32" OR "${cmake_instance}" MATCHES "_f32") AND DTYPES MATCHES "fp32")
-        #message("fp32 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "fp64" OR "${cmake_instance}" MATCHES "_f64") AND DTYPES MATCHES "fp64")
-        #message("fp64 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "bf16" OR "${cmake_instance}" MATCHES "_b16") AND DTYPES MATCHES "bf16")
-        #message("bf16 instance found!")
         set(add_inst 1)
     endif()
     if(("${cmake_instance}" MATCHES "int8" OR "${cmake_instance}" MATCHES "_i8") AND DTYPES MATCHES "int8")
-        #message("int8 instance found!")
         set(add_inst 1)
     endif()
     if(NOT "${cmake_instance}" MATCHES "DTYPES")
-        #message("instance should be built for all types!")
         set(add_inst 1)
     endif()
     if(add_inst EQUAL 1 OR NOT DEFINED DTYPES)
diff --git a/Dockerfile b/Dockerfile
index eb3305a42e..7134e206c1 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -26,25 +26,37 @@ RUN wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - && \
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
 RUN amdgpu-install -y --usecase=rocm --no-dkms
 
+## Sccache binary built from source for ROCm
+ARG SCCACHE_REPO_URL=http://compute-artifactory.amd.com/artifactory/rocm-generic-experimental/rocm-sccache
+ENV SCCACHE_INSTALL_LOCATION=/usr/local/.cargo/bin
+RUN mkdir -p ${SCCACHE_INSTALL_LOCATION} && \
+curl ${SCCACHE_REPO_URL}/portable/0.2.16/sccache-0.2.16-alpha.1-rocm --output ${SCCACHE_INSTALL_LOCATION}/sccache && \
+chmod +x ${SCCACHE_INSTALL_LOCATION}/sccache
+ENV PATH=$PATH:${SCCACHE_INSTALL_LOCATION}
+
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     build-essential \
-    ccache \
     cmake \
+    ccache \
     git \
     hip-rocclr \
+    iputils-ping \
     jq \
     libelf-dev \
     libncurses5-dev \
     libnuma-dev \
     libpthread-stubs0-dev \
     llvm-amdgpu \
+    net-tools \
     pkg-config \
     python \
     python3 \
     python3-dev \
     python3-pip \
+    redis \
     sshpass \
+    stunnel \
     software-properties-common \
     vim \
     nano \
@@ -62,7 +74,7 @@ RUN gunzip /usr/local/bin/ninja.gz
 RUN chmod a+x /usr/local/bin/ninja
 RUN git clone https://github.com/nico/ninjatracing.git
 # Update the cmake to the latest version
-RUN pip install --upgrade cmake
+RUN pip install --upgrade cmake==3.27.5
 
 # Setup ubsan environment to printstacktrace
 RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
@@ -77,9 +89,9 @@ ARG PREFIX=/opt/rocm
 RUN pip3 install --upgrade pip
 RUN pip3 install sqlalchemy==1.4.46
 RUN pip3 install pymysql
-RUN pip3 install pandas
+RUN pip3 install pandas==2.0.3
 RUN pip3 install setuptools-rust
-RUN pip3 install sshtunnel
+RUN pip3 install sshtunnel==0.4.0
 # Setup ubsan environment to printstacktrace
 ENV UBSAN_OPTIONS=print_stacktrace=1
 
@@ -115,6 +127,8 @@ RUN if [ "$compiler_version" = "amd-stg-open" ] && [ "$compiler_commit" != "" ];
     else echo "using the release compiler"; \
     fi
 
+#clean-up the deb package
+RUN sh -c "rm -rf amdgpu-install*"
 
 #ENV HIP_CLANG_PATH='/llvm-project/build/bin'
 #RUN sh -c "echo HIP_CLANG_PATH = '$HIP_CLANG_PATH'"
diff --git a/Jenkinsfile b/Jenkinsfile
index 021c19f150..ddcbc53bff 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -65,10 +65,10 @@ def getDockerImageName(){
 }
 
 def check_host() {
-    if ("${env.CK_CCACHE}" != "null"){
-        def CCACHE_SERVER="${env.CK_CCACHE.split(':')[0]}"
-        echo "ccache server: ${CCACHE_SERVER}"
-        sh '''ping -c 1 -p 6379 "${CCACHE_SERVER}" | echo $? > tmp.txt'''
+    if ("${env.CK_SCCACHE}" != "null"){
+        def SCCACHE_SERVER="${env.CK_SCCACHE.split(':')[0]}"
+        echo "sccache server: ${SCCACHE_SERVER}"
+        sh '''ping -c 1 -p 6379 "${SCCACHE_SERVER}" | echo $? > tmp.txt'''
         def output = readFile(file: "tmp.txt")
         echo "tmp.txt contents: \$output"
         return (output != "0")
@@ -96,24 +96,9 @@ def build_compiler(){
 
 def getDockerImage(Map conf=[:]){
     env.DOCKER_BUILDKIT=1
-    def prefixpath = conf.get("prefixpath", "/opt/rocm") // prefix:/opt/rocm
+    def prefixpath = conf.get("prefixpath", "/opt/rocm")
     def no_cache = conf.get("no_cache", false)
     def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-    echo "ccache server: ${env.CK_CCACHE}"
-    if(env.CK_CCACHE)
-    {
-        if(check_host())
-        {
-            echo "FOUND CCACHE SERVER: ${env.CK_CCACHE}"
-        }
-        else 
-        {
-            echo "CCACHE SERVER: ${env.CK_CCACHE} NOT FOUND, got ${check_host} response"
-        }
-        dockerArgs = dockerArgs + " --build-arg CCACHE_SECONDARY_STORAGE='redis://${env.CK_CCACHE}' --build-arg COMPILER_LAUNCHER='ccache' "
-        env.CCACHE_DIR = """/tmp/ccache_store"""
-        env.CCACHE_SECONDARY_STORAGE="""redis://${env.CK_CCACHE}"""
-    }
     if(no_cache)
     {
         dockerArgs = dockerArgs + " --no-cache "
@@ -142,21 +127,6 @@ def buildDocker(install_prefix){
     def image_name = getDockerImageName()
     echo "Building Docker for ${image_name}"
     def dockerArgs = "--build-arg BUILDKIT_INLINE_CACHE=1 --build-arg PREFIX=${install_prefix} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-    echo "ccache server: ${env.CK_CCACHE}"
-    if(env.CK_CCACHE)
-    {
-        if(check_host())
-        {
-            echo "FOUND CCACHE SERVER: ${env.CK_CCACHE}"
-        }
-        else 
-        {
-            echo "CCACHE SERVER: ${env.CK_CCACHE} NOT FOUND, got ${check_host} response"
-        }
-        dockerArgs = dockerArgs + " --build-arg CCACHE_SECONDARY_STORAGE='redis://${env.CK_CCACHE}' --build-arg COMPILER_LAUNCHER='ccache' "
-        env.CCACHE_DIR = """/tmp/ccache_store"""
-        env.CCACHE_SECONDARY_STORAGE="""redis://${env.CK_CCACHE}"""
-    }
 
     echo "Build Args: ${dockerArgs}"
     try{
@@ -219,13 +189,9 @@ def cmake_build(Map conf=[:]){
     }else{
         setup_args = " -DCMAKE_BUILD_TYPE=release" + setup_args
     }
-    if(env.CK_CCACHE)
-    {
-        setup_args = " -DCMAKE_CXX_COMPILER_LAUNCHER='ccache' -DCMAKE_C_COMPILER_LAUNCHER='ccache' " + setup_args
-    }
-    echo "ccache server: ${env.CK_CCACHE}"
 
     def pre_setup_cmd = """
+            #!/bin/bash
             echo \$HSA_ENABLE_SDMA
             ulimit -c unlimited
             rm -rf build
@@ -234,6 +200,46 @@ def cmake_build(Map conf=[:]){
             mkdir install
             cd build
         """
+    def invocation_tag=""
+    if (setup_args.contains("gfx11")){
+        invocation_tag="gfx11"
+    }
+    if (setup_args.contains("gfx10")){
+        invocation_tag="gfx10"
+    }
+    if (setup_args.contains("gfx90")){
+        invocation_tag="gfx90"
+    }
+    if (setup_args.contains("gfx94")){
+        invocation_tag="gfx94"
+    }
+    if(check_host() && params.USE_SCCACHE && "${env.CK_SCCACHE}" != "null" && "${invocation_tag}" != "") {
+        pre_setup_cmd = pre_setup_cmd + """
+            #!/bin/bash
+            export ROCM_PATH=/opt/rocm
+            export SCCACHE_ENABLED=true
+            export SCCACHE_LOG_LEVEL=debug
+            export SCCACHE_IDLE_TIMEOUT=14400
+            export COMPILERS_HASH_DIR=/tmp/.sccache
+            export SCCACHE_BIN=/usr/local/.cargo/bin/sccache
+            export SCCACHE_EXTRAFILES=/tmp/.sccache/rocm_compilers_hash_file
+            export SCCACHE_REDIS="redis://${env.CK_SCCACHE}"
+            echo "connect = ${env.CK_SCCACHE}" >> ../script/redis-cli.conf
+            export SCCACHE_C_CUSTOM_CACHE_BUSTER="${invocation_tag}"
+            echo \$SCCACHE_C_CUSTOM_CACHE_BUSTER
+            stunnel ../script/redis-cli.conf
+            (
+                set -e
+                ../script/sccache_wrapper.sh --enforce_redis
+            )
+            error_code=\$?
+            if [ \$error_code -ne 0 ]; then
+                echo "could not connect to the redis server. using sccache locally."
+                ../script/sccache_wrapper.sh
+            fi
+        """
+        setup_args = " -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache " + setup_args
+    }
     def setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
     // reduce parallelism when compiling, clang uses too much memory
     def nt = nthreads()
@@ -251,7 +257,7 @@ def cmake_build(Map conf=[:]){
     sh cmd
 
     // Only archive from master or develop
-    if (package_build == true && (env.BRANCH_NAME == "develop" || env.BRANCH_NAME == "master")) {
+    if (package_build == true && (env.BRANCH_NAME == "develop" || env.BRANCH_NAME == "amd-master")) {
         archiveArtifacts artifacts: "build/*.deb", allowEmptyArchive: true, fingerprint: true
     }
 }
@@ -635,7 +641,7 @@ def process_results(Map conf=[:]){
 //launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
 CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCMVERSION=5.7;COMPILER_VERSION=
                                               0 21 * * * % ROCMVERSION=5.7;COMPILER_VERSION=;COMPILER_COMMIT=
-                                              0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=''' : ""
+                                              0 19 * * * % BUILD_DOCKER=true;DL_KERNELS=true;COMPILER_VERSION=amd-stg-open;COMPILER_COMMIT=;USE_SCCACHE=false''' : ""
 
 pipeline {
     agent none
@@ -682,7 +688,10 @@ pipeline {
             name: 'hipTensor_branch',
             defaultValue: 'mainline',
             description: 'Specify which branch of hipTensor to use (default: mainline)')
-
+        booleanParam(
+            name: "USE_SCCACHE",
+            defaultValue: true,
+            description: "Use the sccache for building CK (default: ON)")
     }
     environment{
         dbuser = "${dbuser}"
diff --git a/README.md b/README.md
index c2b493db11..e5a20f143f 100644
--- a/README.md
+++ b/README.md
@@ -1,139 +1,189 @@
 # Composable Kernel
 
-## Methodology
+The Composable Kernel (CK) library provides a programming model for writing performance-critical
+kernels for machine learning workloads across multiple architectures (GPUs, CPUs, etc.). The CK library
+uses general purpose kernel languages, such as HIP C++.
 
-Composable Kernel (CK) library aims to provide a programming model for writing performance critical kernels for machine learning workloads across multiple architectures including GPUs, CPUs, etc, through general purpose kernel languages, like HIP C++.
+CK uses two concepts to achieve performance portability and code maintainability:
 
-CK utilizes two concepts to achieve performance portability and code maintainability:
 * A tile-based programming model
-* Algorithm complexity reduction for complex ML operators, using innovative technique we call "Tensor Coordinate Transformation".
+* Algorithm complexity reduction for complex machine learning (ML) operators. This uses an innovative
+   technique called *Tensor Coordinate Transformation*.
 
 ![ALT](/docs/data/ck_component.png "CK Components")
 
-## Code Structure
+The current CK library is structured into four layers:
 
-Current CK library are structured into 4 layers:
-* "Templated Tile Operators" layer
-* "Templated Kernel and Invoker" layer
-* "Instantiated Kernel and Invoker" layer
-* "Client API" layer
+* Templated Tile Operators
+* Templated Kernel and Invoker
+* Instantiated Kernel and Invoker
+* Client API
 
 ![ALT](/docs/data/ck_layer.png "CK Layers")
 
-## Documentation
+## General information
 
-Run the steps below to build documentation locally.
+To build our documentation locally, use the following code:
 
-```
+``` bash
 cd docs
 pip3 install -r sphinx/requirements.txt
 python3 -m sphinx -T -E -b html -d _build/doctrees -D language=en . _build/html
 ```
 
-## Contributors
+You can find a list of our developers and contributors on our [Contributors](/CONTRIBUTORS.md) page.
+page.
 
-The list of developers and contributors is here: [Contributors](/CONTRIBUTORS.md)
+```note
+If you use CK, cite us as follows:
 
-## Citation
-
-If you use CK, please use following citations:
-* CK paper will be freely available on arXiv soon: [Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming](???)
+* [Realizing Tensor Operators Using Coordinate Transformations and Tile Based Programming](???):
+  This paper will be available on arXiv soon.
 * [CITATION.cff](/CITATION.cff)
-
-## License
-
-CK is released under the MIT license. [License File](/LICENSE)
-
-
-# Build CK
-
-## Build docker image
-
-```bash
-DOCKER_BUILDKIT=1 docker build -t ck:latest -f Dockerfile .
-```
-Pre-built dockers are available from this public repo: 
-https://hub.docker.com/r/rocm/composable_kernel/tags
-
-## Launch docker
-
-```bash
-docker run                                     \
--it                                            \
---privileged                                   \
---group-add sudo                               \
--w /root/workspace                             \
--v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace  \
-ck:latest                                      \
-/bin/bash
 ```
 
-## Build CK
+CK is released under the **[MIT license](/LICENSE)**.
 
-```bash
-mkdir build && cd build
+## Building CK
 
-# Need to specify target ID, example below is for gfx908 and gfx90a
+We recommend building CK inside Docker containers, which include all necessary packages. Pre-built
+Docker images are available on [DockerHub](https://hub.docker.com/r/rocm/composable_kernel/tags).
 
-cmake                                                                                             \
--D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
--D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
--D CMAKE_BUILD_TYPE=Release                                                                       \
--D GPU_TARGETS="gfx908;gfx90a"                                                                    \
-..
-```
+1. To build a new Docker image, use the Dockerfile provided with the source code:
 
-If GPU_TARGETS is not set on the cmake command line, CK will be built for all targets supported by the 
-current compiler.
+    ```bash
+    DOCKER_BUILDKIT=1 docker build -t ck:latest -f Dockerfile .
+    ```
 
+2. Launch the Docker container:
+
+    ```bash
+    docker run                                     \
+    -it                                            \
+    --privileged                                   \
+    --group-add sudo                               \
+    -w /root/workspace                             \
+    -v ${PATH_TO_LOCAL_WORKSPACE}:/root/workspace  \
+    ck:latest                                      \
+    /bin/bash
+    ```
+
+3. Clone CK source code from the GitHub repository and start the build:
+
+    ```bash
+    git clone https://github.com/ROCmSoftwarePlatform/composable_kernel.git && \
+    cd composable_kernel && \
+    mkdir build && \
+    cd build
+    ```
+
+    You must set the `GPU_TARGETS` macro to specify the GPU target architecture(s) you want
+    to run CK on. You can specify single or multiple architectures. If you specify multiple architectures,
+    use a semicolon between each; for example, `gfx908;gfx90a;gfx940`.
+
+    ```bash
+    cmake                                                                                             \
+    -D CMAKE_PREFIX_PATH=/opt/rocm                                                                    \
+    -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                         \
+    -D CMAKE_BUILD_TYPE=Release                                                                       \
+    -D GPU_TARGETS="gfx908;gfx90a"                                                                    \
+    ..
+    ```
+
+    If you don't set `GPU_TARGETS` on the cmake command line, CK is built for all GPU targets
+    supported by the current compiler (this may take a long time).
+
+4. Build the entire CK library:
+
+    ```bash
+    make -j
+    ```
+
+5. Install CK:
+
+    ```bash
+    make -j install
+    ```
+
+## Optional post-install steps
+
+* Build examples and tests:
+
+    ```bash
+    make -j examples tests
+    ```
+
+* Build and run all examples and tests:
+
+    ```bash
+    make -j check
+    ```
+
+    You can find instructions for running each individual example in [example](/example).
+
+* Build ckProfiler:
+
+    ```bash
+    make -j ckProfiler
+    ```
+
+    You can find instructions for running ckProfiler in [profiler](/profiler).
+
+Note the `-j` option for building with multiple threads in parallel. This speeds up the build significantly.
+Depending on the number of CPU cores and the amount of RAM on your system, you may want to
+limit the number of threads. For example, if you have a 128-core CPU and 64 Gb of RAM.
+
+By default, `-j` launches one thread per CPU core, which can cause the build to run out of memory and
+crash. In such cases, you can reduce the number of threads to 32 by using `-j32`.
 
 Additional cmake flags can be used to significantly speed-up the build:
 
-INSTANCES_ONLY (by default is OFF) must be set to ON in order to build only the instances and library
-while skipping all tests, examples, and profiler. This is useful for libraries that use CK as a dependency.
+* `INSTANCES_ONLY` (default is OFF) must be set to ON in order to build only the instances and library
+  while skipping all tests, examples, and profiler. This is useful in cases when you plan to use CK as a
+  dependency and don't plan to run any examples or tests.
 
-DTYPES (by default not set) can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build instances 
-of select data types only. Currently, building of int8 instances is taking a lot of time (the compiler fix is in the works).
+* `DTYPES` (default is not set) can be set to any subset of "fp64;fp32;fp16;fp8;bf16;int8" to build
+  instances of select data types only. The main default data types are fp32 and fp16; you can safely skip
+  other data types.
 
-DL_KERNELS (by default is OFF) must be set to ON in order to build the gemm_dl and batched_gemm_multi_d_dl 
-instances. Those instances are only needed for the NAVI2x platforms.
+* `DL_KERNELS` (default is OFF) must be set to ON in order to build instances, such as `gemm_dl` or
+  `batched_gemm_multi_d_dl`. These instances are useful on architectures like the NAVI2x, as most
+  other platforms have faster instances, such as `xdl` or `wmma`, available.
 
-### Build examples and tests
+## Using sccache for building
+
+The default CK Docker images come with a pre-installed version of sccache, which supports clang
+being used as hip-compiler (" -x hip"). Using sccache can help reduce the time to re-build code from
+hours to 1-2 minutes. In order to invoke sccache, you need to run:
 
 ```bash
- make -j examples tests
- make test
+ sccache --start-server
 ```
 
-Instructions for running each individual examples are under [example](/example)
-
-
-## Build ckProfiler
+then add the following flags to the cmake command line:
 
 ```bash
- make -j ckProfiler
+ -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache
 ```
-Instructions for running ckProfiler are under [profiler](/profiler)
 
-## Install CK
-
-```bash
-make install
-```
+You may need to clean up the build folder and repeat the cmake and make steps in order to take
+advantage of the sccache during subsequent builds.
 
 ## Using CK as pre-built kernel library
 
-Instructions for using CK as a pre-built kernel library are under [client_example](/client_example)
+You can find instructions for using CK as a pre-built kernel library in [client_example](/client_example).
 
-## Contributing
+## Contributing to CK
 
-When you contribute to Composable Kernel, make sure to run `clang-format` on all the changed files. We highly recommend using git hooks that are managed by the `pre-commit` framework. To install hooks, run:
+When you contribute to CK, make sure you run `clang-format` on all changed files. We highly
+recommend using git hooks that are managed by the `pre-commit` framework. To install hooks, run:
 
 ```bash
 sudo script/install_precommit.sh
 ```
 
-This way, `pre-commit` will add the appropriate hooks to your local repository and automatically run `clang-format` (and possibly additional checks) before any commit is created.
+With this approach, `pre-commit` adds the appropriate hooks to your local repository and
+automatically runs `clang-format` (and possibly additional checks) before any commit is created.
 
 If you need to uninstall hooks from the repository, you can do so by running the following command:
 
@@ -141,14 +191,5 @@ If you need to uninstall hooks from the repository, you can do so by running the
 script/uninstall_precommit.sh
 ```
 
-If for any reason, you need to temporarily disable precommit hooks, you can add the `--no-verify` option to the `git commit` command.
-
-## Caveat
-### Kernel Timing and Verification
-
-CK's own kernel timer will warn up kernel once, and then run it multiple times
-to get average kernel time. For some kernels that use atomic add, this will cause
-output buffer to be accumulated multiple times, causing verification failure.
-To work around it, do not use CK's own timer and do verification at the same time.
-CK's own timer and verification in each example and ckProfiler can be enabled or
-disabled from command line.
+If you need to temporarily disable pre-commit hooks, you can add the `--no-verify` option to the
+`git commit` command.
diff --git a/script/redis-cli.conf b/script/redis-cli.conf
new file mode 100644
index 0000000000..c17bc66202
--- /dev/null
+++ b/script/redis-cli.conf
@@ -0,0 +1,10 @@
+fips = no
+setuid = root
+setgid = root
+pid = /var/run/stunnel.pid
+debug = 7
+options = NO_SSLv2
+options = NO_SSLv3
+[redis-cli]
+client = yes
+accept = 127.0.0.1:6379
diff --git a/script/sccache_wrapper.sh b/script/sccache_wrapper.sh
new file mode 100755
index 0000000000..b0ec08de45
--- /dev/null
+++ b/script/sccache_wrapper.sh
@@ -0,0 +1,56 @@
+#!/bin/bash
+set -e
+COMPILERS_HASH_DIR=${COMPILERS_HASH_DIR:-"/tmp/.sccache"}
+SCCACHE_EXTRAFILES=${SCCACHE_EXTRAFILES:-"${COMPILERS_HASH_DIR}/rocm_compilers_hash_file"}
+SCCACHE_BIN=${SCCACHE_BIN:-"${SCCACHE_INSTALL_LOCATION}/sccache"}
+ENFORCE_REDIS="false"
+while [ "$1" != "" ];
+do
+    case $1 in
+        --enforce_redis )
+            shift; ENFORCE_REDIS="true" ;;
+        --no-hipcc )
+            shift ;;
+        *)
+            break ;;
+    esac
+done
+setup_rocm_compilers_hash_file() {
+    mkdir -p "$COMPILERS_HASH_DIR"
+    HIPCC_MD5="$(md5sum "${ROCM_PATH}/bin/hipcc")"
+    pushd "${ROCM_PATH}/amdgcn/bitcode"
+        DEVICELIBS_BITCODES_MD5="$(find . -type f -exec md5sum {} \; | sort | md5sum)"
+    popd
+    HIPCC_HASH_VALUE="${HIPCC_MD5%% *}"
+    DEVICELIBS_BITCODES_HASH_VALUE="${DEVICELIBS_BITCODES_MD5%% *}"
+    # MD5 checksums of clang and clang-offload-bundler cannot be used since they will keep changing
+    # if the ROCM_PATH changes, ie; for every mainline build.
+    # This is because ROCM_PATH gets encoded into the clang/clang-offload-bundler binaries as part
+    # of RPATH.
+    # The versions themselves contain the commit hash of the compiler repo at the time of building.
+    # Hence, this should be a viable alternative to using the binary checksum itself.
+    CLANG_VERSION="$("${ROCM_PATH}/llvm/bin/clang" --version | head -n 1)"
+    CLANG_OFFLOAD_BUNDLER_VERSION="$("${ROCM_PATH}/llvm/bin/clang-offload-bundler" --version | head -n 1)"
+    printf '%s: %s\n' 'clang version' "${CLANG_VERSION}" | tee -a "$SCCACHE_EXTRAFILES"
+    printf '%s: %s\n' 'clang-offload-bundler version' "${CLANG_OFFLOAD_BUNDLER_VERSION}" | tee -a "$SCCACHE_EXTRAFILES"
+    printf '%s: %s\n' 'hipcc md5sum' "${HIPCC_HASH_VALUE}" | tee -a "$SCCACHE_EXTRAFILES"
+    printf '%s: %s\n' 'devicelibs bitcode md5sum' "${DEVICELIBS_BITCODES_HASH_VALUE}" | tee -a "$SCCACHE_EXTRAFILES"
+    echo "sccache-wrapper: compilers hash file set up at ${SCCACHE_EXTRAFILES}"
+    cat "$SCCACHE_EXTRAFILES"
+}
+if [ "${ENFORCE_REDIS}" == "true" ]; then
+    if [ -z "${SCCACHE_REDIS}" ]; then
+        echo "SCCACHE_REDIS not set. Not wrapping compilers with sccache."
+        exit 10
+    else
+        response=$(redis-cli -u ${SCCACHE_REDIS} ping) || true
+        if [ "${response}" != "PONG" ]; then
+            echo "Redis server unreachable. Not wrapping compilers with sccache."
+            exit 20
+        fi
+    fi
+fi
+setup_rocm_compilers_hash_file
+$SCCACHE_BIN --version
+$SCCACHE_BIN --start-server
+

From 2e824c6d46acfd8690a155186be5c2dd4c6648ab Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Bart=C5=82omiej=20Kocot?= <barkocot@amd.com>
Date: Tue, 31 Oct 2023 10:46:32 +0100
Subject: [PATCH 191/194] Add support for groups in Img2Col/Col2Img (#1007)

* Add support for groups in Img2Col/Col2Img

* Fix interface test

* Fix interface test G to N

* Improve performance

* Change gemm layout to 3d

* Fixes
---
 .../22_im2col_col2im/column_to_image.cpp      |  18 +-
 .../22_im2col_col2im/image_to_column.cpp      |  18 +-
 .../52_im2col_col2im/column_to_image_f32.cpp  |  13 +-
 .../52_im2col_col2im/image_to_column_f32.cpp  |  13 +-
 .../device/device_conv_tensor_rearrange.hpp   |  17 +-
 .../impl/device_column_to_image_impl.hpp      | 102 ++++--
 .../impl/device_image_to_column_impl.hpp      |  91 +++--
 .../gpu/grid/gridwise_tensor_rearrange.hpp    |  43 ++-
 .../cpu/reference_column_to_image.hpp         |  41 +--
 .../cpu/reference_image_to_column.hpp         |  37 +-
 .../gpu/conv_tensor_rearrange.hpp             | 333 +++++++++++++++---
 .../gpu/column_to_image/CMakeLists.txt        |   9 +-
 ...ce_column_to_image_gndhwc_3d_instance.cpp} |   8 +-
 ...ice_column_to_image_gnhwc_2d_instance.cpp} |   8 +-
 ...vice_column_to_image_gnwc_1d_instance.cpp} |   8 +-
 ...ice_column_to_image_ndhwgc_3d_instance.cpp |  62 ++++
 ...vice_column_to_image_nhwgc_2d_instance.cpp |  62 ++++
 ...evice_column_to_image_nwgc_1d_instance.cpp |  61 ++++
 .../gpu/image_to_column/CMakeLists.txt        |   9 +-
 ...ce_image_to_column_gndhwc_3d_instance.cpp} |  10 +-
 ...ice_image_to_column_gnhwc_2d_instance.cpp} |  10 +-
 ...vice_image_to_column_gnwc_1d_instance.cpp} |  10 +-
 ...ice_image_to_column_ndhwgc_3d_instance.cpp |  62 ++++
 ...vice_image_to_column_nhwgc_2d_instance.cpp |  62 ++++
 ...evice_image_to_column_nwgc_1d_instance.cpp |  61 ++++
 profiler/README.md                            |   3 +-
 .../profile_conv_tensor_rearrange_impl.hpp    |  31 +-
 .../src/profile_conv_tensor_rearrange.cpp     | 133 ++++++-
 .../test_conv_tensor_rearrange.cpp            |  46 +--
 .../test_conv_tensor_rearrange_interface.cpp  |  14 +-
 30 files changed, 1114 insertions(+), 281 deletions(-)
 rename library/src/tensor_operation_instance/gpu/column_to_image/{device_column_to_image_nhwc_3d_instance.cpp => device_column_to_image_gndhwc_3d_instance.cpp} (87%)
 rename library/src/tensor_operation_instance/gpu/column_to_image/{device_column_to_image_nhwc_2d_instance.cpp => device_column_to_image_gnhwc_2d_instance.cpp} (88%)
 rename library/src/tensor_operation_instance/gpu/column_to_image/{device_column_to_image_nhwc_1d_instance.cpp => device_column_to_image_gnwc_1d_instance.cpp} (88%)
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_ndhwgc_3d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwgc_2d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nwgc_1d_instance.cpp
 rename library/src/tensor_operation_instance/gpu/image_to_column/{device_image_to_column_nhwc_3d_instance.cpp => device_image_to_column_gndhwc_3d_instance.cpp} (83%)
 rename library/src/tensor_operation_instance/gpu/image_to_column/{device_image_to_column_nhwc_2d_instance.cpp => device_image_to_column_gnhwc_2d_instance.cpp} (84%)
 rename library/src/tensor_operation_instance/gpu/image_to_column/{device_image_to_column_nhwc_1d_instance.cpp => device_image_to_column_gnwc_1d_instance.cpp} (84%)
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_ndhwgc_3d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwgc_2d_instance.cpp
 create mode 100644 library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nwgc_1d_instance.cpp

diff --git a/client_example/22_im2col_col2im/column_to_image.cpp b/client_example/22_im2col_col2im/column_to_image.cpp
index 43338ce408..9ebe63198f 100644
--- a/client_example/22_im2col_col2im/column_to_image.cpp
+++ b/client_example/22_im2col_col2im/column_to_image.cpp
@@ -16,10 +16,10 @@
 using InDataType  = ck::half_t;
 using OutDataType = ck::half_t;
 
-using ImageLayout = ck::tensor_layout::convolution::GNHWC;
+using ImageLayout = ck::tensor_layout::convolution::NHWGC;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t G             = 2;
 static constexpr ck::index_t N             = 32; // batch size
 static constexpr ck::index_t C             = 32; // input channel (per group)
 static constexpr ck::index_t Y             = 3;  // filter H
@@ -52,18 +52,18 @@ int main()
     std::array<ck::index_t, 2> wei_spatial_lengths{Y, X};
     std::array<ck::index_t, 2> out_spatial_lengths{Ho, Wo};
 
-    // We have NHWGC in memory space (G is dummy)
-    // However, CK's API only accept length and stride with order of GNCHW
-    // Hence, we need to adjust the order of stride
+    // We have NHWGC in memory space
+    // However, CK's API only accepts lengths and strides with order of GNCHW.
+    // Hence, we need to adjust the order of strides.
     std::array<ck::index_t, 5> image_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
-    std::array<ck::index_t, 2> gemm_strides{Y * X * C, 1};
+    std::array<ck::index_t, 3> gemm_strides{Y * X * C, G * Y * X * C, 1};
 
     std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
     std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
     std::array<ck::index_t, NumDimSpatial> input_left_pads{1, 1};
     std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
 
-    SimpleDeviceMem in(sizeof(InDataType) * N * Ho * Wo * Y * X * C);
+    SimpleDeviceMem in(sizeof(InDataType) * G * N * Ho * Wo * Y * X * C);
     SimpleDeviceMem out(sizeof(OutDataType) * N * Hi * Wi * G * C);
 
     using namespace ck::conv_tensor_rearrange_op;
@@ -93,6 +93,7 @@ int main()
         auto& op_ptr        = op_ptrs[i];
         auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
+                                                        G,
                                                         N,
                                                         C,
                                                         in_spatial_lengths,
@@ -112,7 +113,7 @@ int main()
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
             std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
-                                    sizeof(OutDataType) * N * Ho * Wo * Y * X * C;
+                                    sizeof(OutDataType) * G * N * Ho * Wo * Y * X * C;
 
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
 
@@ -149,6 +150,7 @@ int main()
                   << std::endl;
         auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
+                                                        G,
                                                         N,
                                                         C,
                                                         in_spatial_lengths,
diff --git a/client_example/22_im2col_col2im/image_to_column.cpp b/client_example/22_im2col_col2im/image_to_column.cpp
index a1447abf64..8eafbdc5bb 100644
--- a/client_example/22_im2col_col2im/image_to_column.cpp
+++ b/client_example/22_im2col_col2im/image_to_column.cpp
@@ -16,10 +16,10 @@
 using InDataType  = ck::half_t;
 using OutDataType = ck::half_t;
 
-using ImageLayout = ck::tensor_layout::convolution::GNHWC;
+using ImageLayout = ck::tensor_layout::convolution::NHWGC;
 
 static constexpr ck::index_t NumDimSpatial = 2;
-static constexpr ck::index_t G             = 1;
+static constexpr ck::index_t G             = 2;
 static constexpr ck::index_t N             = 32; // batch size
 static constexpr ck::index_t C             = 32; // input channel (per group)
 static constexpr ck::index_t Y             = 3;  // filter H
@@ -52,11 +52,11 @@ int main()
     std::array<ck::index_t, 2> wei_spatial_lengths{Y, X};
     std::array<ck::index_t, 2> out_spatial_lengths{Ho, Wo};
 
-    // We have NHWGC in memory space (G is dummy)
-    // However, CK's API only accept length and stride with order of GNCHW
-    // Hence, we need to adjust the order of stride
+    // We have NHWGC in memory space
+    // However, CK's API only accepts lengths and strides with order of GNCHW.
+    // Hence, we need to adjust the order of strides.
     std::array<ck::index_t, 5> image_strides{C, Hi * Wi * G * C, 1, Wi * G * C, G * C};
-    std::array<ck::index_t, 2> gemm_strides{Y * X * C, 1};
+    std::array<ck::index_t, 3> gemm_strides{Y * X * C, G * Y * X * C, 1};
 
     std::array<ck::index_t, NumDimSpatial> filter_strides{1, 1};
     std::array<ck::index_t, NumDimSpatial> filter_dilations{1, 1};
@@ -64,7 +64,7 @@ int main()
     std::array<ck::index_t, NumDimSpatial> input_right_pads{1, 1};
 
     SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * G * C);
-    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * G * N * Ho * Wo * Y * X * C);
 
     using namespace ck::conv_tensor_rearrange_op;
 
@@ -93,6 +93,7 @@ int main()
         auto& op_ptr        = op_ptrs[i];
         auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
+                                                        G,
                                                         N,
                                                         C,
                                                         in_spatial_lengths,
@@ -112,7 +113,7 @@ int main()
             float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
 
             std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * G * C +
-                                    sizeof(OutDataType) * N * Ho * Wo * Y * X * C;
+                                    sizeof(OutDataType) * G * N * Ho * Wo * Y * X * C;
 
             float gb_per_sec = num_bytes / 1.E6 / avg_time;
 
@@ -149,6 +150,7 @@ int main()
                   << std::endl;
         auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
                                                         out.GetDeviceBuffer(),
+                                                        G,
                                                         N,
                                                         C,
                                                         in_spatial_lengths,
diff --git a/example/52_im2col_col2im/column_to_image_f32.cpp b/example/52_im2col_col2im/column_to_image_f32.cpp
index 52144e6885..047f2a2118 100644
--- a/example/52_im2col_col2im/column_to_image_f32.cpp
+++ b/example/52_im2col_col2im/column_to_image_f32.cpp
@@ -20,7 +20,7 @@ using DeviceColToImgInstance = ck::tensor_operation::device::DeviceColumnToImage
 
 bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
 {
-
+    const auto G = conv_params.G_;
     const auto N = conv_params.N_;
     const auto C = conv_params.C_;
 
@@ -31,7 +31,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
         C * ck::accumulate_n<ck::index_t>(
                 conv_params.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
 
-    const auto in_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+    const auto in_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
     const auto out_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params);
 
@@ -39,7 +39,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
-    std::array<ck::index_t, 2> gemm_m_k_strides{};
+    std::array<ck::index_t, 3> gemm_g_m_k_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -50,7 +50,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
     copy(conv_params.input_spatial_lengths_, input_spatial_lengths);
     copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
     copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
-    copy(in_desc.GetStrides(), gemm_m_k_strides);
+    copy(in_desc.GetStrides(), gemm_g_m_k_strides);
     copy(out_desc.GetStrides(), image_g_n_c_wis_strides);
     copy(conv_params.conv_filter_strides_, conv_filter_strides);
     copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
@@ -86,13 +86,14 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
     auto invoker  = col2img.MakeInvoker();
     auto argument = col2img.MakeArgument(in_device_buf.GetDeviceBuffer(),
                                          out_device_buf.GetDeviceBuffer(),
+                                         G,
                                          N,
                                          C,
                                          input_spatial_lengths,
                                          filter_spatial_lengths,
                                          output_spatial_lengths,
                                          image_g_n_c_wis_strides,
-                                         gemm_m_k_strides,
+                                         gemm_g_m_k_strides,
                                          conv_filter_strides,
                                          conv_filter_dilations,
                                          input_left_pads,
@@ -108,7 +109,7 @@ bool RunColumnToImage(const ExecutionConfig& config, const ck::utils::conv::Conv
     }
 
     float ave_time        = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
-    std::size_t num_btype = NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
+    std::size_t num_btype = G * NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
     float gb_per_sec      = num_btype / 1.E6 / ave_time;
     std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
 
diff --git a/example/52_im2col_col2im/image_to_column_f32.cpp b/example/52_im2col_col2im/image_to_column_f32.cpp
index 6d883460cc..140bdf8062 100644
--- a/example/52_im2col_col2im/image_to_column_f32.cpp
+++ b/example/52_im2col_col2im/image_to_column_f32.cpp
@@ -20,7 +20,7 @@ using DeviceImgToColInstance = ck::tensor_operation::device::DeviceImageToColumn
 
 bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::ConvParam& conv_params)
 {
-
+    const auto G = conv_params.G_;
     const auto N = conv_params.N_;
     const auto C = conv_params.C_;
 
@@ -33,13 +33,13 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
 
     const auto in_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(conv_params);
-    const auto out_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+    const auto out_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
 
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
-    std::array<ck::index_t, 2> gemm_m_k_strides{};
+    std::array<ck::index_t, 3> gemm_g_m_k_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -51,7 +51,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
     copy(conv_params.filter_spatial_lengths_, filter_spatial_lengths);
     copy(conv_params.output_spatial_lengths_, output_spatial_lengths);
     copy(in_desc.GetStrides(), image_g_n_c_wis_strides);
-    copy(out_desc.GetStrides(), gemm_m_k_strides);
+    copy(out_desc.GetStrides(), gemm_g_m_k_strides);
     copy(conv_params.conv_filter_strides_, conv_filter_strides);
     copy(conv_params.conv_filter_dilations_, conv_filter_dilations);
     copy(conv_params.input_left_pads_, input_left_pads);
@@ -86,13 +86,14 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
     auto invoker  = img2col.MakeInvoker();
     auto argument = img2col.MakeArgument(in_device_buf.GetDeviceBuffer(),
                                          out_device_buf.GetDeviceBuffer(),
+                                         G,
                                          N,
                                          C,
                                          input_spatial_lengths,
                                          filter_spatial_lengths,
                                          output_spatial_lengths,
                                          image_g_n_c_wis_strides,
-                                         gemm_m_k_strides,
+                                         gemm_g_m_k_strides,
                                          conv_filter_strides,
                                          conv_filter_dilations,
                                          input_left_pads,
@@ -108,7 +109,7 @@ bool RunImageToColumn(const ExecutionConfig& config, const ck::utils::conv::Conv
     }
 
     float ave_time        = invoker.Run(argument, StreamConfig{nullptr, config.time_kernel});
-    std::size_t num_btype = NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
+    std::size_t num_btype = G * NDoHoWo * CZYX * (sizeof(OutDataType) + sizeof(InDataType));
     float gb_per_sec      = num_btype / 1.E6 / ave_time;
     std::cout << "Perf: " << ave_time << " ms, " << gb_per_sec << " GB/s" << std::endl;
 
diff --git a/include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp b/include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp
index 898cfe0f2c..b28204fd84 100644
--- a/include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp
+++ b/include/ck/tensor_operation/gpu/device/device_conv_tensor_rearrange.hpp
@@ -14,11 +14,12 @@ namespace device {
 /**
  * \brief Convolution Tensor Rearrange.
  *
- * This Device operator supports conversion image ([G, N, Di, Hi, Wi, C]) to
- * the gemm problem([N * Do * Ho * Wo, Z *  Y * X * C]) (Image to Column) and
- * conversion gemm form to the image (Column to Image).
- *
- * Note that G must be equal to 1.
+ * This Device operator supports converting an image to
+ * the GEMM representation (Image to Column) and
+ * converting a GEMM form to the image (Column to Image).
+ * Supported layouts:
+ * [G, N, Di, Hi, Wi, C] <-> [G, N * Do * Ho * Wo, Z *  Y * X * C]
+ * [N, Di, Hi, Wi, G, C] <-> [N * Do * Ho * Wo, G, Z *  Y * X * C]
  *
  * \tparam NDimSpatial Number of spatial dimensions.
  * \tparam ImageLayout Input Layout.
@@ -39,13 +40,14 @@ struct DeviceConvTensorRearrange : public BaseOperator
      *
      * \param p_in A pointer to the device memory of the input image.
      * \param p_out A pointer to the device memory of the output.
+     * \param G Convolution number of groups.
      * \param N Convolution batch size.
      * \param C Convolution number of channels.
      * \param input_spatial_lengths Input spatial lengths.
      * \param filter_spatial_lengths Filter spatial lengths.
      * \param output_spatial_lengths Output spatial lengths.
      * \param image_g_n_c_wis_strides Image strides in order [G, N, C, D, H, W].
-     * \param gemm_m_k_strides Gemm form strides.
+     * \param gemm_g_m_k_strides Gemm form strides.
      * \param conv_filter_strides Convolution filter strides.
      * \param conv_filter_dilations Convolution filter dilations.
      * \param input_left_pads Convolution left pads.
@@ -55,13 +57,14 @@ struct DeviceConvTensorRearrange : public BaseOperator
     virtual std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in,
                         void* p_out,
+                        const ck::index_t G,
                         const ck::index_t N,
                         const ck::index_t C,
                         const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                         const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                        const std::array<index_t, 2>& gemm_m_k_strides,
+                        const std::array<index_t, 3>& gemm_g_m_k_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<index_t, NDimSpatial>& input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
index f8b4a01681..567be5f364 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_column_to_image_impl.hpp
@@ -17,15 +17,18 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/io.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 
-// Image to column for input layout NDHWC:
-//   input : image converted to the gemm problem [N * Do * Ho * Wo, Z * Y * X * C]
-//   output : image [N, Di, Hi, Wi, C]
+// Column to Image:
+//   input : gemm form [G, N * Do * Ho * Wo, Z * Y * X * C]
+//   output : input image [G, N, Di, Hi, Wi, C]
+//   input : gemm form [N * Do * Ho * Wo, G, Z * Y * X * C]
+//   output : input image [N, Di, Hi, Wi, G, C]
 template <index_t NDimSpatial,
           typename ImageLayout,
           typename InputDataType,
@@ -43,6 +46,14 @@ struct DeviceColumnToImageImpl
                                        OutputDataType,
                                        conv_tensor_rearrange_op::ColumnToImage>
 {
+    static constexpr bool is_NSpatialGC =
+        std::is_same_v<ImageLayout, tensor_layout::convolution::NWGC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::NHWGC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::NDHWGC>;
+    static constexpr bool is_GNSpatialC =
+        std::is_same_v<ImageLayout, tensor_layout::convolution::GNWC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::GNHWC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::GNDHWC>;
 
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
@@ -90,7 +101,7 @@ struct DeviceColumnToImageImpl
                             const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                             const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                             const std::array<index_t, NDimSpatial>& conv_filter_strides,
-                            const std::array<index_t, 2>& gemm_m_k_strides,
+                            const std::array<index_t, 3>& gemm_g_m_k_strides,
                             const std::array<index_t, NDimSpatial>& independent_filters,
                             const std::array<index_t, NDimSpatial>& effs)
     {
@@ -100,23 +111,23 @@ struct DeviceColumnToImageImpl
             C * ck::accumulate_n<index_t>(
                     filter_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
 
-        const index_t NStride = DoHoWo * gemm_m_k_strides[I0] * gemm_m_k_strides[I1];
+        const index_t NStride = DoHoWo * gemm_g_m_k_strides[I1] * gemm_g_m_k_strides[I2];
         // Calculate the appropriate stride for each set of independent filters
         // in each dimension
-        const index_t WStride =
-            math::integer_divide_ceil(effs[XIdx], conv_filter_strides[XIdx]) * gemm_m_k_strides[I0];
+        const index_t WStride = math::integer_divide_ceil(effs[XIdx], conv_filter_strides[XIdx]) *
+                                gemm_g_m_k_strides[I1];
         const index_t HStride = math::integer_divide_ceil(effs[YIdx], conv_filter_strides[YIdx]) *
-                                output_spatial_lengths[XIdx] * gemm_m_k_strides[I0];
+                                output_spatial_lengths[XIdx] * gemm_g_m_k_strides[I1];
         const index_t DStride = math::integer_divide_ceil(effs[ZIdx], conv_filter_strides[ZIdx]) *
                                 output_spatial_lengths[YIdx] * output_spatial_lengths[XIdx] *
-                                gemm_m_k_strides[I0];
+                                gemm_g_m_k_strides[I1];
         // Create descriptor for independent filters in each dimension and
         // then merge them into column form
         if constexpr(NDimSpatial == 1)
         {
             const auto desc_gemm_form =
                 make_naive_tensor_descriptor(make_tuple(N, independent_filters[XIdx], CZYX),
-                                             make_tuple(NStride, WStride, gemm_m_k_strides[I1]));
+                                             make_tuple(NStride, WStride, gemm_g_m_k_strides[I2]));
             const auto desc_gemm_form_merged_filters = transform_tensor_descriptor(
                 desc_gemm_form,
                 make_tuple(make_merge_transform(make_tuple(N, independent_filters[XIdx])),
@@ -130,7 +141,7 @@ struct DeviceColumnToImageImpl
         {
             const auto desc_gemm_form = make_naive_tensor_descriptor(
                 make_tuple(N, independent_filters[YIdx], independent_filters[XIdx], CZYX),
-                make_tuple(NStride, HStride, WStride, gemm_m_k_strides[I1]));
+                make_tuple(NStride, HStride, WStride, gemm_g_m_k_strides[I2]));
             const auto desc_gemm_form_merged_filters = transform_tensor_descriptor(
                 desc_gemm_form,
                 make_tuple(make_merge_transform(
@@ -149,7 +160,7 @@ struct DeviceColumnToImageImpl
                            independent_filters[YIdx],
                            independent_filters[XIdx],
                            CZYX),
-                make_tuple(NStride, DStride, HStride, WStride, gemm_m_k_strides[I1]));
+                make_tuple(NStride, DStride, HStride, WStride, gemm_g_m_k_strides[I2]));
             const auto desc_gemm_form_merged_filters = transform_tensor_descriptor(
                 desc_gemm_form,
                 make_tuple(make_merge_transform(make_tuple(N,
@@ -252,34 +263,38 @@ struct DeviceColumnToImageImpl
         decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, InputGridDesc>(
             InputGridDesc{}))>;
 
-    using GridwiseTensorRearrangeKernel = GridwiseTensorRearrange<InputGridDesc,
-                                                                  InputDataType,
-                                                                  OutputGridDesc,
-                                                                  OutputDataType,
-                                                                  BlockSize,
-                                                                  MPerBlock,
-                                                                  KPerBlock,
-                                                                  ThreadClusterLengths,
-                                                                  ScalarPerVector,
-                                                                  InMemoryDataOperationEnum::Add,
-                                                                  Block2ETileMap>;
+    using GridwiseTensorRearrangeKernel =
+        GridwiseTensorRearrange<InputGridDesc,
+                                InputDataType,
+                                OutputGridDesc,
+                                OutputDataType,
+                                BlockSize,
+                                MPerBlock,
+                                KPerBlock,
+                                ThreadClusterLengths,
+                                ScalarPerVector,
+                                InMemoryDataOperationEnum::Add,
+                                Block2ETileMap,
+                                ComputePtrOffsetOfStridedBatch<I0>>;
 
     struct Argument : public BaseArgument
     {
         Argument(const void* p_in, // input image
                  void* p_out,      // output image
+                 const ck::index_t G,
                  const ck::index_t N,
                  const ck::index_t C,
                  const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                  const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                  const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                  const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                 const std::array<index_t, 2>& gemm_m_k_strides,
+                 const std::array<index_t, 3>& gemm_g_m_k_strides,
                  const std::array<index_t, NDimSpatial>& conv_filter_strides,
                  const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                  const std::array<index_t, NDimSpatial>& input_left_pads,
                  const std::array<index_t, NDimSpatial>& input_right_pads)
-            : C_(C),
+            : G_(G),
+              C_(C),
               X_(filter_spatial_lengths[NDimSpatial - I1]),
               p_in_{static_cast<const InputDataType*>(p_in)},
               p_out_{static_cast<OutputDataType*>(p_out)},
@@ -289,6 +304,9 @@ struct DeviceColumnToImageImpl
               input_left_pads_{input_left_pads},
               input_right_pads_{input_right_pads}
         {
+            compute_ptr_offset_of_batch_.BatchStrideA_ = gemm_g_m_k_strides[I0];
+            compute_ptr_offset_of_batch_.BatchStrideC_ = image_g_n_c_wis_strides[I0];
+
             const index_t x_eff =
                 (filter_spatial_lengths[XIdx] - 1) * conv_filter_dilations[XIdx] + 1;
             const index_t y_eff =
@@ -354,7 +372,7 @@ struct DeviceColumnToImageImpl
                                                     filter_spatial_lengths,
                                                     output_spatial_lengths,
                                                     conv_filter_strides,
-                                                    gemm_m_k_strides,
+                                                    gemm_g_m_k_strides,
                                                     independent_filters,
                                                     effs);
                         const auto out_grid_desc_m_k =
@@ -387,10 +405,9 @@ struct DeviceColumnToImageImpl
                         // Memory offsets to next set of independent filters,
                         // move to independent filters in each dimension
                         const index_t in_offset =
-                            x_idx * gemm_m_k_strides[0] +
-                            y_idx * gemm_m_k_strides[0] * output_spatial_lengths[XIdx] +
-                            z_idx * gemm_m_k_strides[0] * output_spatial_lengths[YIdx] *
-                                output_spatial_lengths[XIdx];
+                            (x_idx + y_idx * output_spatial_lengths[XIdx] +
+                             z_idx * output_spatial_lengths[YIdx] * output_spatial_lengths[XIdx]) *
+                            gemm_g_m_k_strides[I1];
                         // Move to independent filters in appropriate dimensions
                         const index_t out_offset =
                             x_offset_with_pad * image_g_n_c_wis_strides[spatial_offset + XIdx] +
@@ -417,6 +434,7 @@ struct DeviceColumnToImageImpl
             }
         }
 
+        const ck::index_t G_;
         const ck::index_t C_;
         const ck::index_t X_;
 
@@ -434,6 +452,8 @@ struct DeviceColumnToImageImpl
 
         std::vector<const InputDataType*> p_in_container_;
         std::vector<OutputDataType*> p_out_container_;
+
+        ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_;
     };
 
     struct Invoker : public BaseInvoker
@@ -451,6 +471,7 @@ struct DeviceColumnToImageImpl
                                                         OutputGridDesc,
                                                         OutputDataType,
                                                         Block2ETileMap,
+                                                        ComputePtrOffsetOfStridedBatch<I0>,
                                                         GridwiseTensorRearrangeKernel>;
 
             // Execute each set of independent filters
@@ -460,7 +481,7 @@ struct DeviceColumnToImageImpl
                     BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, InputGridDesc>(
                         arg.out_grid_desc_m_k_container_[i]);
                 const index_t grid_size =
-                    block_2_tile_map.CalculateGridSize(arg.in_grid_desc_m_k_container_[i]);
+                    block_2_tile_map.CalculateGridSize(arg.in_grid_desc_m_k_container_[i]) * arg.G_;
                 elapsed_time += launch_and_time_kernel(stream_config,
                                                        kernel,
                                                        dim3(grid_size),
@@ -470,7 +491,9 @@ struct DeviceColumnToImageImpl
                                                        arg.p_in_container_[i],
                                                        arg.out_grid_desc_m_k_container_[i],
                                                        arg.p_out_container_[i],
-                                                       block_2_tile_map);
+                                                       arg.G_,
+                                                       block_2_tile_map,
+                                                       arg.compute_ptr_offset_of_batch_);
             }
             return elapsed_time;
         }
@@ -485,8 +508,7 @@ struct DeviceColumnToImageImpl
     bool IsSupportedArgument(const Argument& arg)
     {
         using namespace tensor_layout::convolution;
-        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
-                       std::is_same_v<ImageLayout, GNDHWC>))
+        if constexpr(!(is_NSpatialGC || is_GNSpatialC))
         {
             return false;
         }
@@ -534,13 +556,14 @@ struct DeviceColumnToImageImpl
 
     static auto MakeArgument(const void* p_in, // input image
                              void* p_out,      // output image
+                             const ck::index_t G,
                              const ck::index_t N,
                              const ck::index_t C,
                              const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                              const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                              const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                              const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                             const std::array<index_t, 2>& gemm_m_k_strides,
+                             const std::array<index_t, 3>& gemm_g_m_k_strides,
                              const std::array<index_t, NDimSpatial>& conv_filter_strides,
                              const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                              const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -548,13 +571,14 @@ struct DeviceColumnToImageImpl
     {
         return Argument{static_cast<const InputDataType*>(p_in),
                         static_cast<OutputDataType*>(p_out),
+                        G,
                         N,
                         C,
                         input_spatial_lengths,
                         filter_spatial_lengths,
                         output_spatial_lengths,
                         image_g_n_c_wis_strides,
-                        gemm_m_k_strides,
+                        gemm_g_m_k_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -566,13 +590,14 @@ struct DeviceColumnToImageImpl
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in, // input image
                         void* p_out,      // output image
+                        const ck::index_t G,
                         const ck::index_t N,
                         const ck::index_t C,
                         const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                         const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                        const std::array<index_t, 2>& gemm_m_k_strides,
+                        const std::array<index_t, 3>& gemm_g_m_k_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -580,13 +605,14 @@ struct DeviceColumnToImageImpl
     {
         return std::make_unique<Argument>(static_cast<const InputDataType*>(p_in),
                                           static_cast<OutputDataType*>(p_out),
+                                          G,
                                           N,
                                           C,
                                           input_spatial_lengths,
                                           filter_spatial_lengths,
                                           output_spatial_lengths,
                                           image_g_n_c_wis_strides,
-                                          gemm_m_k_strides,
+                                          gemm_g_m_k_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
index 8b4ae7875c..c83ffdcd26 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_image_to_column_impl.hpp
@@ -15,15 +15,18 @@
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/conv_tensor_rearrange_op.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_utils.hpp"
 #include "ck/host_utility/io.hpp"
 
 namespace ck {
 namespace tensor_operation {
 namespace device {
 
-// Image to column for input layout NDHWC:
-//   input : input image [N, Di, Hi, Wi, C]
-//   output : gemm form [N * Do * Ho * Wo, Z * Y * X * C]
+// Image to column:
+//   input : input image [G, N, Di, Hi, Wi, C]
+//   output : gemm form [G * N * Do * Ho * Wo, Z * Y * X * C]
+//   input : input image [N, Di, Hi, Wi, G, C]
+//   output : gemm form [N * Do * Ho * Wo * G, Z * Y * X * C]
 template <index_t NDimSpatial,
           typename ImageLayout,
           typename InputDataType,
@@ -41,6 +44,14 @@ struct DeviceImageToColumnImpl
                                        OutputDataType,
                                        conv_tensor_rearrange_op::ImageToColumn>
 {
+    static constexpr bool is_NSpatialGC =
+        std::is_same_v<ImageLayout, tensor_layout::convolution::NWGC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::NHWGC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::NDHWGC>;
+    static constexpr bool is_GNSpatialC =
+        std::is_same_v<ImageLayout, tensor_layout::convolution::GNWC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::GNHWC> ||
+        std::is_same_v<ImageLayout, tensor_layout::convolution::GNDHWC>;
 
     static constexpr auto I0 = Number<0>{};
     static constexpr auto I1 = Number<1>{};
@@ -109,7 +120,7 @@ struct DeviceImageToColumnImpl
                           const ck::index_t C,
                           const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                           const std::array<index_t, NDimSpatial>& output_spatial_lengths,
-                          const std::array<index_t, 2>& gemm_m_k_strides)
+                          const std::array<index_t, 3>& gemm_g_m_k_strides)
     {
         const index_t NDoHoWo =
             N * ck::accumulate_n<index_t>(
@@ -117,11 +128,10 @@ struct DeviceImageToColumnImpl
         const index_t CZYX =
             C * ck::accumulate_n<index_t>(
                     filter_spatial_lengths.begin(), NDimSpatial, 1, std::multiplies<>());
-        const auto desc_mraw_kraw = make_naive_tensor_descriptor(
-            make_tuple(NDoHoWo, CZYX), make_tuple(gemm_m_k_strides[I0], gemm_m_k_strides[I1]));
 
-        const auto desc_m_k = matrix_padder.PadADescriptor_M_K(desc_mraw_kraw);
-        return desc_m_k;
+        const auto desc_mraw_kraw = make_naive_tensor_descriptor(
+            make_tuple(NDoHoWo, CZYX), make_tuple(gemm_g_m_k_strides[I1], gemm_g_m_k_strides[I2]));
+        return matrix_padder.PadADescriptor_M_K(desc_mraw_kraw);
     }
 
     using InputGridDesc =
@@ -132,34 +142,38 @@ struct DeviceImageToColumnImpl
         decltype(BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
             OutputGridDesc{}))>;
 
-    using GridwiseTensorRearrangeKernel = GridwiseTensorRearrange<InputGridDesc,
-                                                                  InputDataType,
-                                                                  OutputGridDesc,
-                                                                  OutputDataType,
-                                                                  BlockSize,
-                                                                  MPerBlock,
-                                                                  KPerBlock,
-                                                                  ThreadClusterLengths,
-                                                                  ScalarPerVector,
-                                                                  InMemoryDataOperationEnum::Set,
-                                                                  Block2ETileMap>;
+    using GridwiseTensorRearrangeKernel =
+        GridwiseTensorRearrange<InputGridDesc,
+                                InputDataType,
+                                OutputGridDesc,
+                                OutputDataType,
+                                BlockSize,
+                                MPerBlock,
+                                KPerBlock,
+                                ThreadClusterLengths,
+                                ScalarPerVector,
+                                InMemoryDataOperationEnum::Set,
+                                Block2ETileMap,
+                                ComputePtrOffsetOfStridedBatch<I0>>;
 
     struct Argument : public BaseArgument
     {
         Argument(const void* p_in, // input image
                  void* p_out,      // gemm form
+                 const ck::index_t G,
                  const ck::index_t N,
                  const ck::index_t C,
                  const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                  const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                  const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                  const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                 const std::array<index_t, 2>& gemm_m_k_strides,
+                 const std::array<index_t, 3>& gemm_g_m_k_strides,
                  const std::array<index_t, NDimSpatial>& conv_filter_strides,
                  const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                  const std::array<index_t, NDimSpatial>& input_left_pads,
                  const std::array<index_t, NDimSpatial>& input_right_pads)
-            : C_(C),
+            : G_(G),
+              C_(C),
               X_(filter_spatial_lengths[NDimSpatial - I1]),
               p_in_{static_cast<const InputDataType*>(p_in)},
               p_out_{static_cast<OutputDataType*>(p_out)},
@@ -176,14 +190,16 @@ struct DeviceImageToColumnImpl
                                                         filter_spatial_lengths,
                                                         output_spatial_lengths,
                                                         image_g_n_c_wis_strides,
-
                                                         conv_filter_strides,
                                                         conv_filter_dilations,
                                                         input_left_pads,
                                                         input_right_pads);
 
             out_grid_desc_m_k_ = MakeOutDescriptor_M_K(
-                N, C, filter_spatial_lengths, output_spatial_lengths, gemm_m_k_strides);
+                N, C, filter_spatial_lengths, output_spatial_lengths, gemm_g_m_k_strides);
+
+            compute_ptr_offset_of_batch_.BatchStrideA_ = image_g_n_c_wis_strides[I0];
+            compute_ptr_offset_of_batch_.BatchStrideC_ = gemm_g_m_k_strides[I0];
         }
 
         void Print() const
@@ -192,6 +208,7 @@ struct DeviceImageToColumnImpl
             std::cout << out_grid_desc_m_k_ << std::endl;
         }
 
+        const ck::index_t G_;
         const ck::index_t C_;
         const ck::index_t X_;
 
@@ -206,6 +223,8 @@ struct DeviceImageToColumnImpl
 
         InputGridDesc in_grid_desc_m_k_;
         OutputGridDesc out_grid_desc_m_k_;
+
+        ComputePtrOffsetOfStridedBatch<I0> compute_ptr_offset_of_batch_;
     };
 
     struct Invoker : public BaseInvoker
@@ -220,12 +239,14 @@ struct DeviceImageToColumnImpl
             const auto block_2_tile_map =
                 BlockToCTileMap_M00_N0_M01Adapt<MPerBlock, KPerBlock, OutputGridDesc>(
                     arg.out_grid_desc_m_k_);
-            const index_t grid_size = block_2_tile_map.CalculateGridSize(arg.out_grid_desc_m_k_);
-            const auto kernel       = kernel_tensor_rearrange<InputGridDesc,
+            const index_t grid_size =
+                block_2_tile_map.CalculateGridSize(arg.out_grid_desc_m_k_) * arg.G_;
+            const auto kernel = kernel_tensor_rearrange<InputGridDesc,
                                                         InputDataType,
                                                         OutputGridDesc,
                                                         OutputDataType,
                                                         Block2ETileMap,
+                                                        ComputePtrOffsetOfStridedBatch<I0>,
                                                         GridwiseTensorRearrangeKernel>;
 
             float elapsed_time = launch_and_time_kernel(stream_config,
@@ -237,7 +258,9 @@ struct DeviceImageToColumnImpl
                                                         arg.p_in_,
                                                         arg.out_grid_desc_m_k_,
                                                         arg.p_out_,
-                                                        block_2_tile_map);
+                                                        arg.G_,
+                                                        block_2_tile_map,
+                                                        arg.compute_ptr_offset_of_batch_);
             return elapsed_time;
         }
 
@@ -250,9 +273,7 @@ struct DeviceImageToColumnImpl
 
     bool IsSupportedArgument(const Argument& arg)
     {
-        using namespace tensor_layout::convolution;
-        if constexpr(!(std::is_same_v<ImageLayout, GNWC> || std::is_same_v<ImageLayout, GNHWC> ||
-                       std::is_same_v<ImageLayout, GNDHWC>))
+        if constexpr(!(is_NSpatialGC || is_GNSpatialC))
         {
             return false;
         }
@@ -295,13 +316,14 @@ struct DeviceImageToColumnImpl
 
     static auto MakeArgument(const void* p_in, // input image
                              void* p_out,      // gemm form
+                             const ck::index_t G,
                              const ck::index_t N,
                              const ck::index_t C,
                              const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                              const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                              const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                              const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                             const std::array<index_t, 2>& gemm_m_k_strides,
+                             const std::array<index_t, 3>& gemm_g_m_k_strides,
                              const std::array<index_t, NDimSpatial>& conv_filter_strides,
                              const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                              const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -309,13 +331,14 @@ struct DeviceImageToColumnImpl
     {
         return Argument{static_cast<const InputDataType*>(p_in),
                         static_cast<OutputDataType*>(p_out),
+                        G,
                         N,
                         C,
                         input_spatial_lengths,
                         filter_spatial_lengths,
                         output_spatial_lengths,
                         image_g_n_c_wis_strides,
-                        gemm_m_k_strides,
+                        gemm_g_m_k_strides,
                         conv_filter_strides,
                         conv_filter_dilations,
                         input_left_pads,
@@ -327,13 +350,14 @@ struct DeviceImageToColumnImpl
     std::unique_ptr<BaseArgument>
     MakeArgumentPointer(const void* p_in, // input image
                         void* p_out,      // gemm form
+                        const ck::index_t G,
                         const ck::index_t N,
                         const ck::index_t C,
                         const std::array<index_t, NDimSpatial>& input_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& filter_spatial_lengths,
                         const std::array<index_t, NDimSpatial>& output_spatial_lengths,
                         const std::array<index_t, NDimSpatial + 3>& image_g_n_c_wis_strides,
-                        const std::array<index_t, 2>& gemm_m_k_strides,
+                        const std::array<index_t, 3>& gemm_g_m_k_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_strides,
                         const std::array<index_t, NDimSpatial>& conv_filter_dilations,
                         const std::array<index_t, NDimSpatial>& input_left_pads,
@@ -341,13 +365,14 @@ struct DeviceImageToColumnImpl
     {
         return std::make_unique<Argument>(static_cast<const InputDataType*>(p_in),
                                           static_cast<OutputDataType*>(p_out),
+                                          G,
                                           N,
                                           C,
                                           input_spatial_lengths,
                                           filter_spatial_lengths,
                                           output_spatial_lengths,
                                           image_g_n_c_wis_strides,
-                                          gemm_m_k_strides,
+                                          gemm_g_m_k_strides,
                                           conv_filter_strides,
                                           conv_filter_dilations,
                                           input_left_pads,
diff --git a/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp b/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
index e4e47b4fae..f77ffff350 100644
--- a/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_tensor_rearrange.hpp
@@ -21,6 +21,7 @@ template <typename InputGridDesc,
           typename OutputGridDesc,
           typename OutputDataType,
           typename Block2ETileMap,
+          typename ComputePtrOffsetOfStridedBatch,
           typename GridwiseTensorRearrangeKernel>
 __global__ void
 #if CK_USE_LAUNCH_BOUNDS
@@ -30,13 +31,20 @@ __global__ void
                                 const InputDataType* __restrict__ p_in_global,
                                 const OutputGridDesc out_grid_desc,
                                 OutputDataType* __restrict__ p_out_global,
-                                const Block2ETileMap block_2_tile_map)
+                                const index_t batch_count,
+                                const Block2ETileMap block_2_tile_map,
+                                const ComputePtrOffsetOfStridedBatch compute_ptr_offset_of_batch)
 {
 #if(!defined(__HIP_DEVICE_COMPILE__) || defined(__gfx906__) || defined(__gfx908__) ||             \
     defined(__gfx90a__) || defined(__gfx940__) || defined(__gfx1030__) || defined(__gfx1100__) || \
     defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx941__) || defined(__gfx942__))
-    GridwiseTensorRearrangeKernel::Run(
-        in_grid_desc, p_in_global, out_grid_desc, p_out_global, block_2_tile_map);
+    GridwiseTensorRearrangeKernel::Run(in_grid_desc,
+                                       p_in_global,
+                                       out_grid_desc,
+                                       p_out_global,
+                                       batch_count,
+                                       block_2_tile_map,
+                                       compute_ptr_offset_of_batch);
 #else
     ignore = in_grid_desc;
     ignore = p_in_global;
@@ -56,7 +64,8 @@ template <typename InputGridDesc,
           typename ThreadClusterLengths,
           index_t ScalarPerVector,
           InMemoryDataOperationEnum DstInMemOp,
-          typename Block2ETileMap>
+          typename Block2ETileMap,
+          typename ComputePtrOffsetOfStridedBatch>
 struct GridwiseTensorRearrange
 {
 
@@ -69,7 +78,9 @@ struct GridwiseTensorRearrange
                                const InputDataType* __restrict__ p_in_global,
                                const OutputGridDesc& out_grid_desc,
                                OutputDataType* __restrict__ p_out_global,
-                               const Block2ETileMap& block_2_tile_map)
+                               const index_t batch_count,
+                               const Block2ETileMap& block_2_tile_map,
+                               const ComputePtrOffsetOfStridedBatch& compute_ptr_offset_of_batch)
     {
         const auto block_work_idx =
             block_2_tile_map.CalculateBottomIndex(make_multi_index(get_block_1d_id()));
@@ -80,12 +91,6 @@ struct GridwiseTensorRearrange
         const index_t k_block_data_idx_on_grid =
             __builtin_amdgcn_readfirstlane(block_work_idx[I1] * KPerBlock);
 
-        // Global Memory
-        const auto in_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_in_global, in_grid_desc.GetElementSpaceSize());
-        auto out_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
-            p_out_global, out_grid_desc.GetElementSpaceSize());
-
         auto copy_global_to_global =
             ThreadGroupTensorSliceTransfer_v7<ThisThreadBlock,
                                               Tuple<InputDataType>,
@@ -108,6 +113,22 @@ struct GridwiseTensorRearrange
                 make_tuple(make_multi_index(m_block_data_idx_on_grid, k_block_data_idx_on_grid)),
                 tensor_operation::element_wise::PassThrough{}};
 
+        const index_t num_blocks_per_batch =
+            __builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
+        const index_t g_idx =
+            __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);
+
+        // Global Memory
+        const index_t a_batch_offset =
+            __builtin_amdgcn_readfirstlane(compute_ptr_offset_of_batch.GetAPtrOffset(g_idx));
+        const index_t c_batch_offset =
+            __builtin_amdgcn_readfirstlane(compute_ptr_offset_of_batch.GetCPtrOffset(g_idx));
+
+        const auto in_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_in_global + a_batch_offset, in_grid_desc.GetElementSpaceSize());
+        auto out_global_buf = make_dynamic_buffer<AddressSpaceEnum::Global>(
+            p_out_global + c_batch_offset, out_grid_desc.GetElementSpaceSize());
+
         copy_global_to_global.Run(
             tie(in_grid_desc), tie(in_global_buf), tie(out_grid_desc), tie(out_global_buf));
     }
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp
index 7ccfc6eb77..45e35ec56d 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_column_to_image.hpp
@@ -19,9 +19,7 @@ namespace host {
  * \brief Reference implementation for column to image.
  *
  * Input tensor descriptor has [N * Do * Ho * Wo, Z * Y * X * C] data layout.
- * Memory layout is the same.
  * Output tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
- * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
  *
  * \tparam NDimSpatial Number of spatial dimensions.
  * \tparam ImageLayout Image Layout.
@@ -95,18 +93,19 @@ struct ReferenceColumnToImage : public device::BaseOperator
         float Run(const Argument& arg)
         {
             if(!(arg.output_.GetNumOfDimension() == NDimSpatial + 3 &&
-                 arg.input_.GetNumOfDimension() == 2))
+                 arg.input_.GetNumOfDimension() == 3))
             {
                 throw std::runtime_error("wrong! inconsistent dimension");
             }
 
+            const index_t G = arg.output_.GetLengths()[0];
             const index_t N = arg.output_.GetLengths()[1];
             const index_t C = arg.output_.GetLengths()[2];
 
             if constexpr(NDimSpatial == 1)
             {
                 const index_t Wo = arg.output_spatial_lengths_[0];
-                auto func        = [&](auto n) {
+                auto func        = [&](auto g, auto n) {
                     for(index_t wo = 0; wo < Wo; ++wo)
                     {
                         index_t row    = n * Wo + wo;
@@ -123,9 +122,10 @@ struct ReferenceColumnToImage : public device::BaseOperator
                                 if(wi >= 0 &&
                                    ck::type_convert<std::size_t>(wi) < arg.output_.GetLengths()[3])
                                 {
-                                    float v_in  = ck::type_convert<float>(arg.input_(row, column));
-                                    float v_out = ck::type_convert<float>(arg.output_(0, n, c, wi));
-                                    arg.output_(0, n, c, wi) =
+                                    float v_in =
+                                        ck::type_convert<float>(arg.input_(g, row, column));
+                                    float v_out = ck::type_convert<float>(arg.output_(g, n, c, wi));
+                                    arg.output_(g, n, c, wi) =
                                         ck::type_convert<OutDataType>(v_in + v_out);
                                 }
                                 column++;
@@ -134,7 +134,7 @@ struct ReferenceColumnToImage : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -143,7 +143,7 @@ struct ReferenceColumnToImage : public device::BaseOperator
                 const index_t Ho = arg.output_spatial_lengths_[0];
                 const index_t Wo = arg.output_spatial_lengths_[1];
 
-                auto func = [&](auto n) {
+                auto func = [&](auto g, auto n) {
                     for(index_t ho = 0; ho < Ho; ++ho)
                     {
                         for(index_t wo = 0; wo < Wo; ++wo)
@@ -176,10 +176,10 @@ struct ReferenceColumnToImage : public device::BaseOperator
                                                arg.output_.GetLengths()[4])
                                         {
                                             float v_in =
-                                                ck::type_convert<float>(arg.input_(row, column));
+                                                ck::type_convert<float>(arg.input_(g, row, column));
                                             float v_out = ck::type_convert<float>(
-                                                arg.output_(0, n, c, hi, wi));
-                                            arg.output_(0, n, c, hi, wi) =
+                                                arg.output_(g, n, c, hi, wi));
+                                            arg.output_(g, n, c, hi, wi) =
                                                 ck::type_convert<OutDataType>(v_in + v_out);
                                         }
                                         column++;
@@ -190,7 +190,7 @@ struct ReferenceColumnToImage : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -200,7 +200,7 @@ struct ReferenceColumnToImage : public device::BaseOperator
                 const index_t Ho = arg.output_spatial_lengths_[1];
                 const index_t Wo = arg.output_spatial_lengths_[2];
 
-                auto func = [&](auto n) {
+                auto func = [&](auto g, auto n) {
                     for(index_t d_o = 0; d_o < Do; ++d_o)
                     {
                         for(index_t ho = 0; ho < Ho; ++ho)
@@ -245,10 +245,10 @@ struct ReferenceColumnToImage : public device::BaseOperator
                                                        arg.output_.GetLengths()[5])
                                                 {
                                                     float v_in = ck::type_convert<float>(
-                                                        arg.input_(row, column));
+                                                        arg.input_(g, row, column));
                                                     float v_out = ck::type_convert<float>(
-                                                        arg.output_(0, n, c, di, hi, wi));
-                                                    arg.output_(0, n, c, di, hi, wi) =
+                                                        arg.output_(g, n, c, di, hi, wi));
+                                                    arg.output_(g, n, c, di, hi, wi) =
                                                         ck::type_convert<OutDataType>(v_in + v_out);
                                                 }
                                                 column++;
@@ -261,7 +261,7 @@ struct ReferenceColumnToImage : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -303,8 +303,9 @@ struct ReferenceColumnToImage : public device::BaseOperator
             C * ck::accumulate_n<index_t>(
                     arg.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
 
-        if(!(arg.input_.GetLengths()[0] == static_cast<std::size_t>(NDoHoWo) &&
-             arg.input_.GetLengths()[1] == static_cast<std::size_t>(CZYX)))
+        if(!(arg.input_.GetLengths()[0] == static_cast<std::size_t>(G) &&
+             arg.input_.GetLengths()[1] == static_cast<std::size_t>(NDoHoWo) &&
+             arg.input_.GetLengths()[2] == static_cast<std::size_t>(CZYX)))
         {
             return false;
         }
diff --git a/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp b/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
index 9e12d07844..56b0ce7914 100644
--- a/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
+++ b/library/include/ck/library/reference_tensor_operation/cpu/reference_image_to_column.hpp
@@ -19,9 +19,7 @@ namespace host {
  * \brief Reference implementation for image to column.
  *
  * Input tensor descriptor has [G, N, C, Di, Hi, Wi] data layout.
- * G must be equal to 1. Memory layout is [G, N, Di, Hi, Wi, C].
- * Output tensor descriptor has [N * Do * Ho * Wo, Z * Y * X * C] data layout.
- * Memory layout is the same.
+ * Output tensor descriptor has [G * N * Do * Ho * Wo, Z * Y * X * C] data layout.
  *
  * \tparam NDimSpatial Number of spatial dimensions.
  * \tparam ImageLayout Image Layout.
@@ -95,18 +93,19 @@ struct ReferenceImageToColumn : public device::BaseOperator
         float Run(const Argument& arg)
         {
             if(!(arg.input_.GetNumOfDimension() == NDimSpatial + 3 &&
-                 arg.output_.GetNumOfDimension() == 2))
+                 arg.output_.GetNumOfDimension() == 3))
             {
                 throw std::runtime_error("wrong! inconsistent dimension");
             }
 
+            const index_t G = arg.input_.GetLengths()[0];
             const index_t N = arg.input_.GetLengths()[1];
             const index_t C = arg.input_.GetLengths()[2];
 
             if constexpr(NDimSpatial == 1)
             {
                 const index_t Wo = arg.output_spatial_lengths_[0];
-                auto func        = [&](auto n, auto wo) {
+                auto func        = [&](auto g, auto n, auto wo) {
                     index_t row    = n * Wo + wo;
                     index_t column = 0;
 
@@ -121,15 +120,15 @@ struct ReferenceImageToColumn : public device::BaseOperator
                             if(wi >= 0 &&
                                ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[3])
                             {
-                                InDataType v_in          = arg.input_(0, n, c, wi);
-                                arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                                InDataType v_in             = arg.input_(g, n, c, wi);
+                                arg.output_(g, row, column) = ck::type_convert<OutDataType>(v_in);
                             }
                             column++;
                         }
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N, Wo)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N, Wo)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -138,7 +137,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                 const index_t Ho = arg.output_spatial_lengths_[0];
                 const index_t Wo = arg.output_spatial_lengths_[1];
 
-                auto func = [&](auto n, auto ho, auto wo) {
+                auto func = [&](auto g, auto n, auto ho, auto wo) {
                     index_t row    = n * Ho * Wo + ho * Wo + wo;
                     index_t column = 0;
 
@@ -162,8 +161,9 @@ struct ReferenceImageToColumn : public device::BaseOperator
                                    wi >= 0 &&
                                    ck::type_convert<std::size_t>(wi) < arg.input_.GetLengths()[4])
                                 {
-                                    InDataType v_in          = arg.input_(0, n, c, hi, wi);
-                                    arg.output_(row, column) = ck::type_convert<OutDataType>(v_in);
+                                    InDataType v_in = arg.input_(g, n, c, hi, wi);
+                                    arg.output_(g, row, column) =
+                                        ck::type_convert<OutDataType>(v_in);
                                 }
                                 column++;
                             }
@@ -171,7 +171,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N, Ho, Wo)(std::thread::hardware_concurrency());
+                make_ParallelTensorFunctor(func, G, N, Ho, Wo)(std::thread::hardware_concurrency());
 
                 return 0;
             }
@@ -181,7 +181,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                 const index_t Ho = arg.output_spatial_lengths_[1];
                 const index_t Wo = arg.output_spatial_lengths_[2];
 
-                auto func = [&](auto n, auto d_o, auto ho, auto wo) {
+                auto func = [&](auto g, auto n, auto d_o, auto ho, auto wo) {
                     index_t row    = n * Do * Ho * Wo + d_o * Ho * Wo + ho * Wo + wo;
                     index_t column = 0;
 
@@ -213,8 +213,8 @@ struct ReferenceImageToColumn : public device::BaseOperator
                                        ck::type_convert<std::size_t>(wi) <
                                            arg.input_.GetLengths()[5])
                                     {
-                                        InDataType v_in = arg.input_(0, n, c, di, hi, wi);
-                                        arg.output_(row, column) =
+                                        InDataType v_in = arg.input_(g, n, c, di, hi, wi);
+                                        arg.output_(g, row, column) =
                                             ck::type_convert<OutDataType>(v_in);
                                     }
                                     column++;
@@ -224,7 +224,7 @@ struct ReferenceImageToColumn : public device::BaseOperator
                     }
                 };
 
-                make_ParallelTensorFunctor(func, N, Do, Ho, Wo)(
+                make_ParallelTensorFunctor(func, G, N, Do, Ho, Wo)(
                     std::thread::hardware_concurrency());
 
                 return 0;
@@ -267,8 +267,9 @@ struct ReferenceImageToColumn : public device::BaseOperator
             C * ck::accumulate_n<index_t>(
                     arg.filter_spatial_lengths_.begin(), NDimSpatial, 1, std::multiplies<>());
 
-        if(!(arg.output_.GetLengths()[0] == static_cast<std::size_t>(NDoHoWo) &&
-             arg.output_.GetLengths()[1] == static_cast<std::size_t>(CZYX)))
+        if(!(arg.output_.GetLengths()[0] == static_cast<std::size_t>(G) &&
+             arg.output_.GetLengths()[1] == static_cast<std::size_t>(NDoHoWo) &&
+             arg.output_.GetLengths()[2] == static_cast<std::size_t>(CZYX)))
         {
             return false;
         }
diff --git a/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp
index 57c0f88567..0be50a334a 100644
--- a/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange.hpp
@@ -19,109 +19,214 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
+// GNWC/GNHWC/GNDHWC
 // Image to Column
-// nhwc, 1d
-void add_device_image_to_column_nwc_1d_bf16_instances(
+// GNWC, 1d
+void add_device_image_to_column_gnwc_1d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nwc_1d_f16_instances(
+void add_device_image_to_column_gnwc_1d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nwc_1d_f32_instances(
+void add_device_image_to_column_gnwc_1d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nwc_1d_i8_instances(
+void add_device_image_to_column_gnwc_1d_i8_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ImageToColumn>>>&
         instances);
-// nhwc, 2d
-void add_device_image_to_column_nhwc_2d_bf16_instances(
+// GNHWC, 2d
+void add_device_image_to_column_gnhwc_2d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nhwc_2d_f16_instances(
+void add_device_image_to_column_gnhwc_2d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nhwc_2d_f32_instances(
+void add_device_image_to_column_gnhwc_2d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_nhwc_2d_i8_instances(
+void add_device_image_to_column_gnhwc_2d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ImageToColumn>>>&
         instances);
-// nhwc, 3d
-void add_device_image_to_column_ndhwc_3d_bf16_instances(
+// GNDHWC, 3d
+void add_device_image_to_column_gndhwc_3d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_ndhwc_3d_f16_instances(
+void add_device_image_to_column_gndhwc_3d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_ndhwc_3d_f32_instances(
+void add_device_image_to_column_gndhwc_3d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ImageToColumn>>>&
         instances);
 
-void add_device_image_to_column_ndhwc_3d_i8_instances(
+void add_device_image_to_column_gndhwc_3d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ImageToColumn>>>&
         instances);
 
 // Column to Image
-// nhwc, 1d
-void add_device_column_to_image_nwc_1d_bf16_instances(
+// GNWC, 1d
+void add_device_column_to_image_gnwc_1d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nwc_1d_f16_instances(
+void add_device_column_to_image_gnwc_1d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nwc_1d_f32_instances(
+void add_device_column_to_image_gnwc_1d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nwc_1d_i8_instances(
+void add_device_column_to_image_gnwc_1d_i8_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
         instances);
-// nhwc, 2d
-void add_device_column_to_image_nhwc_2d_bf16_instances(
+// GNHWC, 2d
+void add_device_column_to_image_gnhwc_2d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nhwc_2d_f16_instances(
+void add_device_column_to_image_gnhwc_2d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nhwc_2d_f32_instances(
+void add_device_column_to_image_gnhwc_2d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_nhwc_2d_i8_instances(
+void add_device_column_to_image_gnhwc_2d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
         instances);
-// nhwc, 3d
-void add_device_column_to_image_ndhwc_3d_bf16_instances(
+// GNDHWC, 3d
+void add_device_column_to_image_gndhwc_3d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_ndhwc_3d_f16_instances(
+void add_device_column_to_image_gndhwc_3d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_ndhwc_3d_f32_instances(
+void add_device_column_to_image_gndhwc_3d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
         instances);
 
-void add_device_column_to_image_ndhwc_3d_i8_instances(
+void add_device_column_to_image_gndhwc_3d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
         instances);
+// NWGC/NHWGC/NDHWGC
+// Image to Column
+// NWGC, 1d
+void add_device_image_to_column_nwgc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwgc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwgc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nwgc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// NHWGC, 2d
+void add_device_image_to_column_nhwgc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwgc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwgc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_nhwgc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+// NDHWGC, 3d
+void add_device_image_to_column_ndhwgc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, BF16, BF16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwgc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F16, F16, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwgc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F32, F32, ImageToColumn>>>&
+        instances);
+
+void add_device_image_to_column_ndhwgc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances);
+
+// Column to Image
+// NWGC, 1d
+void add_device_column_to_image_nwgc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwgc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwgc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nwgc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// NHWGC, 2d
+void add_device_column_to_image_nhwgc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwgc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwgc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_nhwgc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
+// NDHWGC, 3d
+void add_device_column_to_image_ndhwgc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, BF16, BF16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwgc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F16, F16, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwgc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F32, F32, ColumnToImage>>>&
+        instances);
+
+void add_device_column_to_image_ndhwgc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances);
 
 template <ck::index_t NumDimSpatial,
           typename ImageLayout,
@@ -151,60 +256,120 @@ struct DeviceOperationInstanceFactory<
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_image_to_column_nwc_1d_f32_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_image_to_column_nwc_1d_f16_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_image_to_column_nwc_1d_bf16_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_image_to_column_nwc_1d_i8_instances(op_ptrs);
+                    add_device_image_to_column_gnwc_1d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_image_to_column_nhwc_2d_f32_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_image_to_column_nhwc_2d_f16_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_image_to_column_nhwc_2d_bf16_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_image_to_column_nhwc_2d_i8_instances(op_ptrs);
+                    add_device_image_to_column_gnhwc_2d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_image_to_column_ndhwc_3d_f32_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_image_to_column_ndhwc_3d_f16_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_image_to_column_ndhwc_3d_bf16_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_image_to_column_ndhwc_3d_i8_instances(op_ptrs);
+                    add_device_image_to_column_gndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, NWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nwgc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nwgc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nwgc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nwgc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, NHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_nhwgc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_nhwgc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_nhwgc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_nhwgc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, NDHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_image_to_column_ndhwgc_3d_i8_instances(op_ptrs);
                 }
             }
         }
@@ -214,60 +379,120 @@ struct DeviceOperationInstanceFactory<
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_column_to_image_nwc_1d_f32_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_column_to_image_nwc_1d_f16_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_column_to_image_nwc_1d_bf16_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_column_to_image_nwc_1d_i8_instances(op_ptrs);
+                    add_device_column_to_image_gnwc_1d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, GNHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_column_to_image_nhwc_2d_f32_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_column_to_image_nhwc_2d_f16_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_column_to_image_nhwc_2d_bf16_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_column_to_image_nhwc_2d_i8_instances(op_ptrs);
+                    add_device_column_to_image_gnhwc_2d_i8_instances(op_ptrs);
                 }
             }
             else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, GNDHWC>)
             {
                 if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
                 {
-                    add_device_column_to_image_ndhwc_3d_f32_instances(op_ptrs);
+                    add_device_column_to_image_gndhwc_3d_f32_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
                 {
-                    add_device_column_to_image_ndhwc_3d_f16_instances(op_ptrs);
+                    add_device_column_to_image_gndhwc_3d_f16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
                                   is_same_v<OutDataType, ck::bhalf_t>)
                 {
-                    add_device_column_to_image_ndhwc_3d_bf16_instances(op_ptrs);
+                    add_device_column_to_image_gndhwc_3d_bf16_instances(op_ptrs);
                 }
                 else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
                 {
-                    add_device_column_to_image_ndhwc_3d_i8_instances(op_ptrs);
+                    add_device_column_to_image_gndhwc_3d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 1 && is_same_v<ImageLayout, NWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nwgc_1d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nwgc_1d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nwgc_1d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nwgc_1d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 2 && is_same_v<ImageLayout, NHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_nhwgc_2d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_nhwgc_2d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_nhwgc_2d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_nhwgc_2d_i8_instances(op_ptrs);
+                }
+            }
+            else if constexpr(NumDimSpatial == 3 && is_same_v<ImageLayout, NDHWGC>)
+            {
+                if constexpr(is_same_v<InDataType, float> && is_same_v<OutDataType, float>)
+                {
+                    add_device_column_to_image_ndhwgc_3d_f32_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, half_t> && is_same_v<OutDataType, half_t>)
+                {
+                    add_device_column_to_image_ndhwgc_3d_f16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, ck::bhalf_t> &&
+                                  is_same_v<OutDataType, ck::bhalf_t>)
+                {
+                    add_device_column_to_image_ndhwgc_3d_bf16_instances(op_ptrs);
+                }
+                else if constexpr(is_same_v<InDataType, int8_t> && is_same_v<OutDataType, int8_t>)
+                {
+                    add_device_column_to_image_ndhwgc_3d_i8_instances(op_ptrs);
                 }
             }
         }
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt
index 5d1a554524..50855babb5 100644
--- a/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/CMakeLists.txt
@@ -1,5 +1,8 @@
 add_instance_library(device_column_to_image_instance
-   device_column_to_image_nhwc_1d_instance.cpp
-   device_column_to_image_nhwc_2d_instance.cpp
-   device_column_to_image_nhwc_3d_instance.cpp
+   device_column_to_image_gnwc_1d_instance.cpp
+   device_column_to_image_gnhwc_2d_instance.cpp
+   device_column_to_image_gndhwc_3d_instance.cpp
+   device_column_to_image_nwgc_1d_instance.cpp
+   device_column_to_image_nhwgc_2d_instance.cpp
+   device_column_to_image_ndhwgc_3d_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gndhwc_3d_instance.cpp
similarity index 87%
rename from library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp
rename to library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gndhwc_3d_instance.cpp
index 9762b46c43..8de2311251 100644
--- a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_3d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gndhwc_3d_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
-void add_device_column_to_image_ndhwc_3d_bf16_instances(
+void add_device_column_to_image_gndhwc_3d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ColumnToImage>>>&
         instances)
 {
@@ -22,7 +22,7 @@ void add_device_column_to_image_ndhwc_3d_bf16_instances(
 #endif
 }
 
-void add_device_column_to_image_ndhwc_3d_f16_instances(
+void add_device_column_to_image_gndhwc_3d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ColumnToImage>>>&
         instances)
 {
@@ -33,7 +33,7 @@ void add_device_column_to_image_ndhwc_3d_f16_instances(
 #endif
 }
 
-void add_device_column_to_image_ndhwc_3d_f32_instances(
+void add_device_column_to_image_gndhwc_3d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ColumnToImage>>>&
         instances)
 {
@@ -44,7 +44,7 @@ void add_device_column_to_image_ndhwc_3d_f32_instances(
 #endif
 }
 
-void add_device_column_to_image_ndhwc_3d_i8_instances(
+void add_device_column_to_image_gndhwc_3d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ColumnToImage>>>&
         instances)
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gnhwc_2d_instance.cpp
similarity index 88%
rename from library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp
rename to library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gnhwc_2d_instance.cpp
index 4de665a63f..611a43e452 100644
--- a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_2d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gnhwc_2d_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
-void add_device_column_to_image_nhwc_2d_bf16_instances(
+void add_device_column_to_image_gnhwc_2d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ColumnToImage>>>&
         instances)
 {
@@ -22,7 +22,7 @@ void add_device_column_to_image_nhwc_2d_bf16_instances(
 #endif
 }
 
-void add_device_column_to_image_nhwc_2d_f16_instances(
+void add_device_column_to_image_gnhwc_2d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ColumnToImage>>>&
         instances)
 {
@@ -33,7 +33,7 @@ void add_device_column_to_image_nhwc_2d_f16_instances(
 #endif
 }
 
-void add_device_column_to_image_nhwc_2d_f32_instances(
+void add_device_column_to_image_gnhwc_2d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ColumnToImage>>>&
         instances)
 {
@@ -44,7 +44,7 @@ void add_device_column_to_image_nhwc_2d_f32_instances(
 #endif
 }
 
-void add_device_column_to_image_nhwc_2d_i8_instances(
+void add_device_column_to_image_gnhwc_2d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ColumnToImage>>>&
         instances)
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gnwc_1d_instance.cpp
similarity index 88%
rename from library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp
rename to library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gnwc_1d_instance.cpp
index 8ba4d29775..c64a25df27 100644
--- a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwc_1d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_gnwc_1d_instance.cpp
@@ -11,7 +11,7 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
-void add_device_column_to_image_nwc_1d_bf16_instances(
+void add_device_column_to_image_gnwc_1d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ColumnToImage>>>&
         instances)
 {
@@ -22,7 +22,7 @@ void add_device_column_to_image_nwc_1d_bf16_instances(
 #endif
 }
 
-void add_device_column_to_image_nwc_1d_f16_instances(
+void add_device_column_to_image_gnwc_1d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ColumnToImage>>>&
         instances)
 {
@@ -33,7 +33,7 @@ void add_device_column_to_image_nwc_1d_f16_instances(
 #endif
 }
 
-void add_device_column_to_image_nwc_1d_f32_instances(
+void add_device_column_to_image_gnwc_1d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ColumnToImage>>>&
         instances)
 {
@@ -44,7 +44,7 @@ void add_device_column_to_image_nwc_1d_f32_instances(
 #endif
 }
 
-void add_device_column_to_image_nwc_1d_i8_instances(
+void add_device_column_to_image_gnwc_1d_i8_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ColumnToImage>>>&
         instances)
 {
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_ndhwgc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_ndhwgc_3d_instance.cpp
new file mode 100644
index 0000000000..eef5260e64
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_ndhwgc_3d_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_ndhwgc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwgc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwgc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_ndhwgc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwgc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwgc_2d_instance.cpp
new file mode 100644
index 0000000000..910782ca7d
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nhwgc_2d_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_nhwgc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwgc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwgc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nhwgc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nwgc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nwgc_1d_instance.cpp
new file mode 100644
index 0000000000..d806d91166
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/column_to_image/device_column_to_image_nwgc_1d_instance.cpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_column_to_image_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_column_to_image_nwgc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, BF16, BF16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_column_to_image_bf16_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwgc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F16, F16, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_column_to_image_f16_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwgc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F32, F32, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_column_to_image_f32_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_column_to_image_nwgc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, int8_t, int8_t, ColumnToImage>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_column_to_image_i8_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt
index de10369374..9e52a8157f 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/CMakeLists.txt
@@ -1,5 +1,8 @@
 add_instance_library(device_image_to_column_instance
-   device_image_to_column_nhwc_1d_instance.cpp
-   device_image_to_column_nhwc_2d_instance.cpp
-   device_image_to_column_nhwc_3d_instance.cpp
+   device_image_to_column_gnwc_1d_instance.cpp
+   device_image_to_column_gnhwc_2d_instance.cpp
+   device_image_to_column_gndhwc_3d_instance.cpp
+   device_image_to_column_nwgc_1d_instance.cpp
+   device_image_to_column_nhwgc_2d_instance.cpp
+   device_image_to_column_ndhwgc_3d_instance.cpp
 )
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gndhwc_3d_instance.cpp
similarity index 83%
rename from library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
rename to library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gndhwc_3d_instance.cpp
index 8290dae928..785019133e 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_3d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gndhwc_3d_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
@@ -11,7 +11,7 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
-void add_device_image_to_column_ndhwc_3d_bf16_instances(
+void add_device_image_to_column_gndhwc_3d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, BF16, BF16, ImageToColumn>>>&
         instances)
 {
@@ -22,7 +22,7 @@ void add_device_image_to_column_ndhwc_3d_bf16_instances(
 #endif
 }
 
-void add_device_image_to_column_ndhwc_3d_f16_instances(
+void add_device_image_to_column_gndhwc_3d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F16, F16, ImageToColumn>>>&
         instances)
 {
@@ -33,7 +33,7 @@ void add_device_image_to_column_ndhwc_3d_f16_instances(
 #endif
 }
 
-void add_device_image_to_column_ndhwc_3d_f32_instances(
+void add_device_image_to_column_gndhwc_3d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, F32, F32, ImageToColumn>>>&
         instances)
 {
@@ -44,7 +44,7 @@ void add_device_image_to_column_ndhwc_3d_f32_instances(
 #endif
 }
 
-void add_device_image_to_column_ndhwc_3d_i8_instances(
+void add_device_image_to_column_gndhwc_3d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<3, GNDHWC, int8_t, int8_t, ImageToColumn>>>&
         instances)
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gnhwc_2d_instance.cpp
similarity index 84%
rename from library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
rename to library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gnhwc_2d_instance.cpp
index 7115e75667..5125649c89 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_2d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gnhwc_2d_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
@@ -11,7 +11,7 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
-void add_device_image_to_column_nhwc_2d_bf16_instances(
+void add_device_image_to_column_gnhwc_2d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, BF16, BF16, ImageToColumn>>>&
         instances)
 {
@@ -22,7 +22,7 @@ void add_device_image_to_column_nhwc_2d_bf16_instances(
 #endif
 }
 
-void add_device_image_to_column_nhwc_2d_f16_instances(
+void add_device_image_to_column_gnhwc_2d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F16, F16, ImageToColumn>>>&
         instances)
 {
@@ -33,7 +33,7 @@ void add_device_image_to_column_nhwc_2d_f16_instances(
 #endif
 }
 
-void add_device_image_to_column_nhwc_2d_f32_instances(
+void add_device_image_to_column_gnhwc_2d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, F32, F32, ImageToColumn>>>&
         instances)
 {
@@ -44,7 +44,7 @@ void add_device_image_to_column_nhwc_2d_f32_instances(
 #endif
 }
 
-void add_device_image_to_column_nhwc_2d_i8_instances(
+void add_device_image_to_column_gnhwc_2d_i8_instances(
     std::vector<
         std::unique_ptr<DeviceConvTensorRearrange<2, GNHWC, int8_t, int8_t, ImageToColumn>>>&
         instances)
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gnwc_1d_instance.cpp
similarity index 84%
rename from library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
rename to library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gnwc_1d_instance.cpp
index 3a629f2346..92624d9b2c 100644
--- a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwc_1d_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_gnwc_1d_instance.cpp
@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
 #include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
@@ -11,7 +11,7 @@ namespace instance {
 
 using namespace ck::conv_tensor_rearrange_op;
 
-void add_device_image_to_column_nwc_1d_bf16_instances(
+void add_device_image_to_column_gnwc_1d_bf16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, BF16, BF16, ImageToColumn>>>&
         instances)
 {
@@ -22,7 +22,7 @@ void add_device_image_to_column_nwc_1d_bf16_instances(
 #endif
 }
 
-void add_device_image_to_column_nwc_1d_f16_instances(
+void add_device_image_to_column_gnwc_1d_f16_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F16, F16, ImageToColumn>>>&
         instances)
 {
@@ -33,7 +33,7 @@ void add_device_image_to_column_nwc_1d_f16_instances(
 #endif
 }
 
-void add_device_image_to_column_nwc_1d_f32_instances(
+void add_device_image_to_column_gnwc_1d_f32_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, F32, F32, ImageToColumn>>>&
         instances)
 {
@@ -44,7 +44,7 @@ void add_device_image_to_column_nwc_1d_f32_instances(
 #endif
 }
 
-void add_device_image_to_column_nwc_1d_i8_instances(
+void add_device_image_to_column_gnwc_1d_i8_instances(
     std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, GNWC, int8_t, int8_t, ImageToColumn>>>&
         instances)
 {
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_ndhwgc_3d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_ndhwgc_3d_instance.cpp
new file mode 100644
index 0000000000..89f33f356e
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_ndhwgc_3d_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_image_to_column_ndhwgc_3d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, BF16, BF16, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_image_to_column_bf16_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_ndhwgc_3d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F16, F16, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_image_to_column_f16_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_ndhwgc_3d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, F32, F32, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_image_to_column_f32_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_ndhwgc_3d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<3, NDHWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_image_to_column_i8_instances<3, NDHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwgc_2d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwgc_2d_instance.cpp
new file mode 100644
index 0000000000..eb0b5f14b0
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nhwgc_2d_instance.cpp
@@ -0,0 +1,62 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_image_to_column_nhwgc_2d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, BF16, BF16, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_image_to_column_bf16_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_nhwgc_2d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F16, F16, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_image_to_column_f16_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_nhwgc_2d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, F32, F32, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_image_to_column_f32_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_nhwgc_2d_i8_instances(
+    std::vector<
+        std::unique_ptr<DeviceConvTensorRearrange<2, NHWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_image_to_column_i8_instances<2, NHWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nwgc_1d_instance.cpp b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nwgc_1d_instance.cpp
new file mode 100644
index 0000000000..06723b4d04
--- /dev/null
+++ b/library/src/tensor_operation_instance/gpu/image_to_column/device_image_to_column_nwgc_1d_instance.cpp
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck/library/tensor_operation_instance/gpu/conv_tensor_rearrange/device_image_to_column_instance.hpp"
+#include "ck/library/tensor_operation_instance/add_device_operation_instance.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using namespace ck::conv_tensor_rearrange_op;
+
+void add_device_image_to_column_nwgc_1d_bf16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, BF16, BF16, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_BF16
+    add_device_operation_instances(instances, device_image_to_column_bf16_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_nwgc_1d_f16_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F16, F16, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP16
+    add_device_operation_instances(instances, device_image_to_column_f16_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_nwgc_1d_f32_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, F32, F32, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_FP32
+    add_device_operation_instances(instances, device_image_to_column_f32_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+void add_device_image_to_column_nwgc_1d_i8_instances(
+    std::vector<std::unique_ptr<DeviceConvTensorRearrange<1, NWGC, int8_t, int8_t, ImageToColumn>>>&
+        instances)
+{
+#ifdef CK_ENABLE_INT8
+    add_device_operation_instances(instances, device_image_to_column_i8_instances<1, NWGC>{});
+#else
+    ignore = instances;
+#endif
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck
diff --git a/profiler/README.md b/profiler/README.md
index 98eae3a763..0c745bf8f2 100644
--- a/profiler/README.md
+++ b/profiler/README.md
@@ -194,7 +194,8 @@ Note: This kernel use atomic add, this will cause output buffer to be accumulate
 #                  1: Input fp16, Weight fp16, Output fp16
 #                  2: Input bf16, Weight bf16, Output bf16
 #                  3: Input int8, Weight int8, Output int8)
-# arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])
+# arg3: tensor layout (0: Input[G, N, Hi, Wi, C], Output[G * N * Ho * Wo, Y * X * C],
+#                      1: Input[N, Hi, Wi, G, C], Output[N * Ho * Wo * G, Y * X * C])
 # arg4: verification (0: no, 1: yes)
 # arg5: initialization (0: no init, 1: integer value, 2: decimal value)
 # arg6: print tensor value (0: no; 1: yes)
diff --git a/profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp b/profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp
index dc3d748bf6..fa0a771962 100644
--- a/profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp
+++ b/profiler/include/profiler/profile_conv_tensor_rearrange_impl.hpp
@@ -93,6 +93,26 @@ static auto make_ref_op()
     }
 }
 
+template <typename InputLayout>
+static auto create_gemm_desc(const ck::index_t G, const ck::index_t NDoHoWo, const ck::index_t CZYX)
+{
+    using namespace ck::tensor_layout::convolution;
+    if constexpr(std::is_same_v<InputLayout, GNWC> || std::is_same_v<InputLayout, GNHWC> ||
+                 std::is_same_v<InputLayout, GNDHWC>)
+    {
+        return HostTensorDescriptor({G, NDoHoWo, CZYX});
+    }
+    else if constexpr(std::is_same_v<InputLayout, NWGC> || std::is_same_v<InputLayout, NHWGC> ||
+                      std::is_same_v<InputLayout, NDHWGC>)
+    {
+        return HostTensorDescriptor({G, NDoHoWo, CZYX}, {CZYX, CZYX * G, 1});
+    }
+    else
+    {
+        throw std::runtime_error("Unsupported layout!");
+    }
+}
+
 template <index_t NDimSpatial,
           typename InputLayout,
           typename InputDataType,
@@ -116,13 +136,13 @@ bool profile_conv_tensor_rearrange_impl(int do_verification,
     const auto image_desc =
         ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<InputLayout>(
             conv_param);
-    const auto gemm_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+    const auto gemm_desc = create_gemm_desc<InputLayout>(conv_param.G_, NDoHoWo, CZYX);
 
     std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
     std::array<ck::index_t, NDimSpatial + 3> image_g_n_c_wis_strides{};
-    std::array<ck::index_t, 2> gemm_m_k_strides{};
+    std::array<ck::index_t, 3> gemm_g_m_k_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
     std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
     std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -134,7 +154,7 @@ bool profile_conv_tensor_rearrange_impl(int do_verification,
     copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
     copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
     copy(image_desc.GetStrides(), image_g_n_c_wis_strides);
-    copy(gemm_desc.GetStrides(), gemm_m_k_strides);
+    copy(gemm_desc.GetStrides(), gemm_g_m_k_strides);
     copy(conv_param.conv_filter_strides_, conv_filter_strides);
     copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
     copy(conv_param.input_left_pads_, input_left_pads);
@@ -212,13 +232,14 @@ bool profile_conv_tensor_rearrange_impl(int do_verification,
         auto argument_ptr = op_ptr->MakeArgumentPointer(
             static_cast<InputDataType*>(in_device_buf.GetDeviceBuffer()),
             static_cast<OutputDataType*>(out_device_buf.GetDeviceBuffer()),
+            conv_param.G_,
             conv_param.N_,
             conv_param.C_,
             input_spatial_lengths,
             filter_spatial_lengths,
             output_spatial_lengths,
             image_g_n_c_wis_strides,
-            gemm_m_k_strides,
+            gemm_g_m_k_strides,
             conv_filter_strides,
             conv_filter_dilations,
             input_left_pads,
@@ -234,7 +255,7 @@ bool profile_conv_tensor_rearrange_impl(int do_verification,
             float avg_time =
                 invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
             std::size_t num_btype =
-                NDoHoWo * CZYX * (sizeof(OutputDataType) + sizeof(InputDataType));
+                conv_param.G_ * NDoHoWo * CZYX * (sizeof(OutputDataType) + sizeof(InputDataType));
             float gb_per_sec = num_btype / 1.E6 / avg_time;
             std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << gb_per_sec << " GB/s, "
                       << op_name << std::endl;
diff --git a/profiler/src/profile_conv_tensor_rearrange.cpp b/profiler/src/profile_conv_tensor_rearrange.cpp
index bad5ce40a5..6420698a28 100644
--- a/profiler/src/profile_conv_tensor_rearrange.cpp
+++ b/profiler/src/profile_conv_tensor_rearrange.cpp
@@ -19,7 +19,8 @@ enum struct RearrangeOp
 
 enum struct ConvLayout
 {
-    NHWC, // 0
+    GNHWC, // 0
+    NHWGC, // 1
 };
 
 enum struct DataType
@@ -42,7 +43,8 @@ static void print_helper_msg()
         << "                 1: Input fp16, Weight fp16, Output fp16\n"
         << "                 2: Input bf16, Weight bf16, Output bf16\n"
         << "                 3: Input int8, Weight int8, Output int8)\n"
-        << "arg3: tensor layout (0: Input[N, Hi, Wi, C], Output[N * Ho * Wo, Y * X * C])\n"
+        << "arg3: tensor layout (0: Input[G, N, Hi, Wi, C], Output[G * N * Ho * Wo, Y * X * C],\n"
+        << "                     1: Input[N, Hi, Wi, G, C], Output[N * Ho * Wo * G, Y * X * C])\n"
         << "arg4: verification (0: no, 1: yes)\n"
         << "arg5: initialization (0: no init, 1: integer value, 2: decimal value)\n"
         << "arg6: print tensor value (0: no; 1: yes)\n"
@@ -114,11 +116,9 @@ int profile_conv_tensor_rearrange(int argc, char* argv[])
         return pass ? 0 : 1;
     };
 
-    // Image To Column
     if(rearrange_op == RearrangeOp::ImageToColumn)
     {
-        // NHWC
-        if(layout == ConvLayout::NHWC)
+        if(layout == ConvLayout::GNHWC)
         {
             if(num_dim_spatial == 1)
             {
@@ -178,11 +178,70 @@ int profile_conv_tensor_rearrange(int argc, char* argv[])
                 }
             }
         }
+        else if(layout == ConvLayout::NHWGC)
+        {
+            if(num_dim_spatial == 1)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I1, NWGC{}, F32{}, F32{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I1, NWGC{}, F16{}, F16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I1, NWGC{}, BF16{}, BF16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I1, NWGC{}, INT8{}, INT8{}, ImageToColumn{});
+                }
+            }
+            else if(num_dim_spatial == 2)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I2, NHWGC{}, F32{}, F32{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I2, NHWGC{}, F16{}, F16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I2, NHWGC{}, BF16{}, BF16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I2, NHWGC{}, INT8{}, INT8{}, ImageToColumn{});
+                }
+            }
+            else if(num_dim_spatial == 3)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I3, NDHWGC{}, F32{}, F32{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I3, NDHWGC{}, F16{}, F16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I3, NDHWGC{}, BF16{}, BF16{}, ImageToColumn{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I3, NDHWGC{}, INT8{}, INT8{}, ImageToColumn{});
+                }
+            }
+        }
     }
     else if(rearrange_op == RearrangeOp::ColumnToImage)
     {
-        // NHWC
-        if(layout == ConvLayout::NHWC)
+        if(layout == ConvLayout::GNHWC)
         {
             if(num_dim_spatial == 1)
             {
@@ -242,6 +301,66 @@ int profile_conv_tensor_rearrange(int argc, char* argv[])
                 }
             }
         }
+        else if(layout == ConvLayout::NHWGC)
+        {
+            if(num_dim_spatial == 1)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I1, NWGC{}, F32{}, F32{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I1, NWGC{}, F16{}, F16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I1, NWGC{}, BF16{}, BF16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I1, NWGC{}, INT8{}, INT8{}, ColumnToImage{});
+                }
+            }
+            else if(num_dim_spatial == 2)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I2, NHWGC{}, F32{}, F32{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I2, NHWGC{}, F16{}, F16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I2, NHWGC{}, BF16{}, BF16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I2, NHWGC{}, INT8{}, INT8{}, ColumnToImage{});
+                }
+            }
+            else if(num_dim_spatial == 3)
+            {
+                if(data_type == DataType::F32_F32)
+                {
+                    return profile(I3, NDHWGC{}, F32{}, F32{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::F16_F16)
+                {
+                    return profile(I3, NDHWGC{}, F16{}, F16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::BF16_BF16)
+                {
+                    return profile(I3, NDHWGC{}, BF16{}, BF16{}, ColumnToImage{});
+                }
+                else if(data_type == DataType::INT8_INT8)
+                {
+                    return profile(I3, NDHWGC{}, INT8{}, INT8{}, ColumnToImage{});
+                }
+            }
+        }
     }
 
     std::cout << "this data_type & layout is not implemented" << std::endl;
diff --git a/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp b/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
index 7065b03e0d..5cb8731b26 100644
--- a/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
+++ b/test/conv_tensor_rearrange/test_conv_tensor_rearrange.cpp
@@ -45,14 +45,20 @@ class TestConvTensorRearrange : public ::testing::Test
 using namespace ck::tensor_layout::convolution;
 using namespace ck::conv_tensor_rearrange_op;
 
-using KernelTypes1d =
-    ::testing::Types<std::tuple<GNWC, ImageToColumn>, std::tuple<GNWC, ColumnToImage>>;
+using KernelTypes1d = ::testing::Types<std::tuple<GNWC, ImageToColumn>,
+                                       std::tuple<GNWC, ColumnToImage>,
+                                       std::tuple<NWGC, ImageToColumn>,
+                                       std::tuple<NWGC, ColumnToImage>>;
 
-using KernelTypes2d =
-    ::testing::Types<std::tuple<GNHWC, ImageToColumn>, std::tuple<GNHWC, ColumnToImage>>;
+using KernelTypes2d = ::testing::Types<std::tuple<GNHWC, ImageToColumn>,
+                                       std::tuple<GNHWC, ColumnToImage>,
+                                       std::tuple<NHWGC, ImageToColumn>,
+                                       std::tuple<NHWGC, ColumnToImage>>;
 
-using KernelTypes3d =
-    ::testing::Types<std::tuple<GNDHWC, ImageToColumn>, std::tuple<GNDHWC, ColumnToImage>>;
+using KernelTypes3d = ::testing::Types<std::tuple<GNDHWC, ImageToColumn>,
+                                       std::tuple<GNDHWC, ColumnToImage>,
+                                       std::tuple<NDHWGC, ImageToColumn>,
+                                       std::tuple<NDHWGC, ColumnToImage>>;
 
 template <typename Tuple>
 class TestConvTensorRearrange1d : public TestConvTensorRearrange<Tuple>
@@ -77,16 +83,16 @@ TYPED_TEST(TestConvTensorRearrange1d, Test1D)
 {
     this->conv_params.clear();
 
-    this->conv_params.push_back({1, 1, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
-    this->conv_params.push_back({1, 1, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
-    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {7}, {3}, {1}, {0}, {0}});
-    this->conv_params.push_back({1, 1, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 2, 4, 1, 192, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 64, 1, 64, {3}, {14}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 64, 1, 64, {1}, {7}, {3}, {1}, {0}, {0}});
+    this->conv_params.push_back({1, 2, 64, 1, 64, {1}, {3}, {1}, {1}, {0}, {0}});
     // ScalarPerVector should be 1
-    this->conv_params.push_back({1, 1, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 4, 1, 1, {3}, {28}, {1}, {1}, {1}, {1}});
     // stride != 1
-    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 1, 1, 4, {3}, {28}, {2}, {1}, {1}, {1}});
     // dilation != 1
-    this->conv_params.push_back({1, 1, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
+    this->conv_params.push_back({1, 2, 1, 1, 4, {3}, {28}, {1}, {2}, {1}, {1}});
 #ifdef CK_ENABLE_FP32
     this->template Run<1, float, float>();
 #endif
@@ -106,13 +112,13 @@ TYPED_TEST(TestConvTensorRearrange2d, Test2D)
     this->conv_params.clear();
 
     this->conv_params.push_back(
-        {2, 1, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+        {2, 2, 4, 1, 192, {3, 3}, {28, 28}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     this->conv_params.push_back(
-        {2, 1, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
+        {2, 2, 64, 1, 64, {3, 3}, {14, 14}, {1, 1}, {1, 1}, {1, 1}, {1, 1}});
     this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {7, 7}, {3, 3}, {1, 1}, {0, 0}, {0, 0}});
     this->conv_params.push_back({2, 1, 64, 1, 64, {1, 1}, {3, 3}, {1, 1}, {1, 1}, {0, 0}, {0, 0}});
     this->conv_params.push_back(
-        {2, 1, 64, 1, 64, {3, 3}, {28, 28}, {2, 2}, {2, 2}, {1, 1}, {1, 1}});
+        {2, 2, 64, 1, 64, {3, 3}, {28, 28}, {2, 2}, {2, 2}, {1, 1}, {1, 1}});
 #ifdef CK_ENABLE_FP32
     this->template Run<2, float, float>();
 #endif
@@ -131,13 +137,13 @@ TYPED_TEST(TestConvTensorRearrange3d, Test3D)
 {
     this->conv_params.clear();
     this->conv_params.push_back(
-        {3, 1, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {3, 3, 3}, {0, 0, 0}, {0, 0, 0}});
+        {3, 2, 16, 1, 64, {1, 1, 1}, {7, 7, 7}, {2, 2, 2}, {3, 3, 3}, {0, 0, 0}, {0, 0, 0}});
     this->conv_params.push_back(
-        {3, 1, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
+        {3, 2, 2, 1, 64, {3, 3, 3}, {14, 14, 3}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}, {1, 1, 1}});
     this->conv_params.push_back(
-        {3, 1, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
+        {3, 2, 32, 1, 64, {1, 1, 1}, {3, 3, 3}, {1, 1, 1}, {1, 1, 1}, {0, 0, 0}, {0, 0, 0}});
     this->conv_params.push_back(
-        {3, 1, 64, 1, 64, {3, 3, 3}, {14, 14, 14}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}});
+        {3, 2, 64, 1, 64, {3, 3, 3}, {14, 14, 14}, {2, 2, 2}, {2, 2, 2}, {1, 1, 1}, {1, 1, 1}});
 #ifdef CK_ENABLE_FP32
     this->template Run<3, float, float>();
 #endif
diff --git a/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp b/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
index 57fcdc4269..67c0f2698c 100644
--- a/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
+++ b/test/conv_tensor_rearrange/test_conv_tensor_rearrange_interface.cpp
@@ -53,7 +53,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
     template <typename ConvTensorRearrangeOp>
     bool Run()
     {
-
+        const auto G = conv_param.G_;
         const auto N = conv_param.N_;
         const auto C = conv_param.C_;
         const auto FakeC =
@@ -71,13 +71,13 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
         const auto image_desc =
             ck::utils::conv::make_input_host_tensor_descriptor_g_n_c_wis_packed<ImLayout>(
                 conv_param);
-        const auto gemm_desc = HostTensorDescriptor({NDoHoWo, CZYX});
+        const auto gemm_desc = HostTensorDescriptor({G, NDoHoWo, CZYX});
 
         std::array<ck::index_t, NDimSpatial> input_spatial_lengths{};
         std::array<ck::index_t, NDimSpatial> filter_spatial_lengths{};
         std::array<ck::index_t, NDimSpatial> output_spatial_lengths{};
         std::array<ck::index_t, NDimSpatial + 3> input_g_n_c_wis_strides{};
-        std::array<ck::index_t, 2> output_m_k_strides{};
+        std::array<ck::index_t, 3> output_g_m_k_strides{};
         std::array<ck::index_t, NDimSpatial> conv_filter_strides{};
         std::array<ck::index_t, NDimSpatial> conv_filter_dilations{};
         std::array<ck::index_t, NDimSpatial> input_left_pads{};
@@ -89,7 +89,7 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
         copy(conv_param.filter_spatial_lengths_, filter_spatial_lengths);
         copy(conv_param.output_spatial_lengths_, output_spatial_lengths);
         copy(image_desc.GetStrides(), input_g_n_c_wis_strides);
-        copy(gemm_desc.GetStrides(), output_m_k_strides);
+        copy(gemm_desc.GetStrides(), output_g_m_k_strides);
         copy(conv_param.conv_filter_strides_, conv_filter_strides);
         copy(conv_param.conv_filter_dilations_, conv_filter_dilations);
         copy(conv_param.input_left_pads_, input_left_pads);
@@ -100,13 +100,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
             auto img2col  = DeviceImgToColInstance{};
             auto argument = img2col.MakeArgument(nullptr,
                                                  nullptr,
+                                                 G,
                                                  N,
                                                  IsCPacked ? C : FakeC,
                                                  input_spatial_lengths,
                                                  filter_spatial_lengths,
                                                  output_spatial_lengths,
                                                  input_g_n_c_wis_strides,
-                                                 output_m_k_strides,
+                                                 output_g_m_k_strides,
                                                  conv_filter_strides,
                                                  conv_filter_dilations,
                                                  input_left_pads,
@@ -119,13 +120,14 @@ class TestConvTensorRearrangeInterface : public ::testing::Test
             auto col2img  = DeviceColToimgInstance{};
             auto argument = col2img.MakeArgument(nullptr,
                                                  nullptr,
+                                                 G,
                                                  N,
                                                  IsCPacked ? C : FakeC,
                                                  input_spatial_lengths,
                                                  filter_spatial_lengths,
                                                  output_spatial_lengths,
                                                  input_g_n_c_wis_strides,
-                                                 output_m_k_strides,
+                                                 output_g_m_k_strides,
                                                  conv_filter_strides,
                                                  conv_filter_dilations,
                                                  input_left_pads,

From 675b69782bbcb24637499a38dda2393f081d4100 Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Wed, 1 Nov 2023 03:17:31 +0800
Subject: [PATCH 192/194] Enable gfx941 & gfx942 support for DeviceGemmXdl<>
 device op (#1017)

* Enable gfx942 support for DeviceGemmXdl<> device op

* Enable gfx941 support for DeviceGemmXdl<> device op
---
 .../ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp    | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
index ebbb1d38c0..b008a6409e 100644
--- a/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp
@@ -184,7 +184,8 @@ struct DeviceGemmXdl : public DeviceGemm<ALayout,
                 return false;
             }
         }
-        else if(ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx940")
+        else if(ck::get_device_name() == "gfx90a" || ck::get_device_name() == "gfx940" ||
+                ck::get_device_name() == "gfx941" || ck::get_device_name() == "gfx942")
         {
             if constexpr(!(is_same_v<AccDataType, float> || is_same_v<AccDataType, float> ||
                            is_same_v<AccDataType, int32_t> || is_same_v<AccDataType, double>))

From db4461c142b563c8c5f723b32730c7989a4f870b Mon Sep 17 00:00:00 2001
From: Po Yen Chen <PoYen.Chen@amd.com>
Date: Wed, 1 Nov 2023 03:28:36 +0800
Subject: [PATCH 193/194] Disable the SLP vectorizer to prevent unnecessary
 wait (#1008)

* Disable the SLP vectorizer to prevent unnecessary wait

* Add comment to the reason of adding flag

* Fix wording
---
 .../src/tensor_operation_instance/gpu/gemm/CMakeLists.txt   | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
index 5edd6c85f4..86dfe4d190 100644
--- a/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/gemm/CMakeLists.txt
@@ -108,13 +108,15 @@ if (ENABLE_PIPELINE_V2_OPT)
        CK_EXPERIMENTAL_PIPELINE_V2_IGLP_OPT=1
     )
 
+    # TODO: The "-vectorize-slp=false" LLVM option is a workaround to prevent inefficient instruction scheduling
+    #       caused by the SLP Vectorizer. Remove this option after fix the SLP Vectorizer issue.
     # layout=NT
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-       COMPILE_OPTIONS ";;"
+       COMPILE_OPTIONS ";-mllvm;-vectorize-slp=false"
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
     # layout=NN
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/km_nk_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES
-       COMPILE_OPTIONS ";;"
+       COMPILE_OPTIONS ";-mllvm;-vectorize-slp=false"
        COMPILE_DEFINITIONS "${WAVES_PER_EU_DEFS};${IGLP_OPT_DEFS}")
     # layout=TT
     set_source_files_properties(device_gemm_xdl_f16_f16_f16/mk_kn_mn_default_pipeline_v2_opt_instance.cpp PROPERTIES

From 306fd506b1eeba5c6c16fc0c7ce47c827330f2d6 Mon Sep 17 00:00:00 2001
From: Illia Silin <98187287+illsilin@users.noreply.github.com>
Date: Wed, 1 Nov 2023 09:43:10 -0700
Subject: [PATCH 194/194] handle the exception when cannot connect to redis
 server (#1019)

---
 Jenkinsfile | 44 ++++++++++++++++++++++++++++++--------------
 1 file changed, 30 insertions(+), 14 deletions(-)

diff --git a/Jenkinsfile b/Jenkinsfile
index ddcbc53bff..66b05177b3 100644
--- a/Jenkinsfile
+++ b/Jenkinsfile
@@ -139,7 +139,7 @@ def buildDocker(install_prefix){
         else{
             echo "Checking for image: ${image_name}"
             sh "docker manifest inspect --insecure ${image_name}"
-            echo "Image: ${image_name} found!! Skipping building image"
+            echo "Image: ${image_name} found! Skipping building image"
         }
     }
     catch(Exception ex){
@@ -213,8 +213,10 @@ def cmake_build(Map conf=[:]){
     if (setup_args.contains("gfx94")){
         invocation_tag="gfx94"
     }
+    echo "invocation tag: ${invocation_tag}"
+    def redis_pre_setup_cmd = pre_setup_cmd
     if(check_host() && params.USE_SCCACHE && "${env.CK_SCCACHE}" != "null" && "${invocation_tag}" != "") {
-        pre_setup_cmd = pre_setup_cmd + """
+        redis_pre_setup_cmd = pre_setup_cmd + """
             #!/bin/bash
             export ROCM_PATH=/opt/rocm
             export SCCACHE_ENABLED=true
@@ -228,18 +230,30 @@ def cmake_build(Map conf=[:]){
             export SCCACHE_C_CUSTOM_CACHE_BUSTER="${invocation_tag}"
             echo \$SCCACHE_C_CUSTOM_CACHE_BUSTER
             stunnel ../script/redis-cli.conf
-            (
-                set -e
-                ../script/sccache_wrapper.sh --enforce_redis
-            )
-            error_code=\$?
-            if [ \$error_code -ne 0 ]; then
-                echo "could not connect to the redis server. using sccache locally."
-                ../script/sccache_wrapper.sh
-            fi
+            ../script/sccache_wrapper.sh --enforce_redis
         """
-        setup_args = " -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache " + setup_args
+        try {
+            def cmd1 = conf.get("cmd1", """
+                    ${redis_pre_setup_cmd}
+                """)
+            sh cmd1
+            setup_args = " -DCMAKE_CXX_COMPILER_LAUNCHER=sccache -DCMAKE_C_COMPILER_LAUNCHER=sccache " + setup_args
+        }
+        catch(Exception err){
+            echo "could not connect to redis server: ${err.getMessage()}. will not use sccache."
+            def cmd2 = conf.get("cmd2", """
+                    ${pre_setup_cmd}
+                """)
+            sh cmd2
+        }
     }
+    else{
+        def cmd3 = conf.get("cmd3",  """
+                ${pre_setup_cmd}
+            """)
+        sh cmd3
+    }
+
     def setup_cmd = conf.get("setup_cmd", "${cmake_envs} cmake ${setup_args}   .. ")
     // reduce parallelism when compiling, clang uses too much memory
     def nt = nthreads()
@@ -247,14 +261,16 @@ def cmake_build(Map conf=[:]){
     def execute_cmd = conf.get("execute_cmd", "")
 
     def cmd = conf.get("cmd", """
-            ${pre_setup_cmd}
             ${setup_cmd}
             ${build_cmd}
             ${execute_cmd}
         """)
 
     echo cmd
-    sh cmd
+
+    dir("build"){
+        sh cmd
+    }
 
     // Only archive from master or develop
     if (package_build == true && (env.BRANCH_NAME == "develop" || env.BRANCH_NAME == "amd-master")) {